Details | Last modification | View Log | RSS feed
| Rev | Author | Line No. | Line |
|---|---|---|---|
| 1 | pmbaty | 1 | /*++ |
| 2 | |||
| 3 | Copyright (c) Microsoft Corporation. All rights reserved. |
||
| 4 | |||
| 5 | Module Name: |
||
| 6 | |||
| 7 | xnamathconvert.inl |
||
| 8 | |||
| 9 | Abstract: |
||
| 10 | |||
| 11 | XNA math library for Windows and Xbox 360: Conversion, loading, and storing functions. |
||
| 12 | --*/ |
||
| 13 | |||
| 14 | #if defined(_MSC_VER) && (_MSC_VER > 1000) |
||
| 15 | #pragma once |
||
| 16 | #endif |
||
| 17 | |||
| 18 | #ifndef __XNAMATHCONVERT_INL__ |
||
| 19 | #define __XNAMATHCONVERT_INL__ |
||
| 20 | |||
| 21 | #define XM_PACK_FACTOR (FLOAT)(1 << 22) |
||
| 22 | #define XM_UNPACK_FACTOR_UNSIGNED (FLOAT)(1 << 23) |
||
| 23 | #define XM_UNPACK_FACTOR_SIGNED XM_PACK_FACTOR |
||
| 24 | |||
| 25 | #define XM_UNPACK_UNSIGNEDN_OFFSET(BitsX, BitsY, BitsZ, BitsW) \ |
||
| 26 | {-XM_UNPACK_FACTOR_UNSIGNED / (FLOAT)((1 << (BitsX)) - 1), \ |
||
| 27 | -XM_UNPACK_FACTOR_UNSIGNED / (FLOAT)((1 << (BitsY)) - 1), \ |
||
| 28 | -XM_UNPACK_FACTOR_UNSIGNED / (FLOAT)((1 << (BitsZ)) - 1), \ |
||
| 29 | -XM_UNPACK_FACTOR_UNSIGNED / (FLOAT)((1 << (BitsW)) - 1)} |
||
| 30 | |||
| 31 | #define XM_UNPACK_UNSIGNEDN_SCALE(BitsX, BitsY, BitsZ, BitsW) \ |
||
| 32 | {XM_UNPACK_FACTOR_UNSIGNED / (FLOAT)((1 << (BitsX)) - 1), \ |
||
| 33 | XM_UNPACK_FACTOR_UNSIGNED / (FLOAT)((1 << (BitsY)) - 1), \ |
||
| 34 | XM_UNPACK_FACTOR_UNSIGNED / (FLOAT)((1 << (BitsZ)) - 1), \ |
||
| 35 | XM_UNPACK_FACTOR_UNSIGNED / (FLOAT)((1 << (BitsW)) - 1)} |
||
| 36 | |||
| 37 | #define XM_UNPACK_SIGNEDN_SCALE(BitsX, BitsY, BitsZ, BitsW) \ |
||
| 38 | {-XM_UNPACK_FACTOR_SIGNED / (FLOAT)((1 << ((BitsX) - 1)) - 1), \ |
||
| 39 | -XM_UNPACK_FACTOR_SIGNED / (FLOAT)((1 << ((BitsY) - 1)) - 1), \ |
||
| 40 | -XM_UNPACK_FACTOR_SIGNED / (FLOAT)((1 << ((BitsZ) - 1)) - 1), \ |
||
| 41 | -XM_UNPACK_FACTOR_SIGNED / (FLOAT)((1 << ((BitsW) - 1)) - 1)} |
||
| 42 | |||
| 43 | //#define XM_UNPACK_SIGNEDN_OFFSET(BitsX, BitsY, BitsZ, BitsW) \ |
||
| 44 | // {-XM_UNPACK_FACTOR_SIGNED / (FLOAT)((1 << ((BitsX) - 1)) - 1) * 3.0f, \ |
||
| 45 | // -XM_UNPACK_FACTOR_SIGNED / (FLOAT)((1 << ((BitsY) - 1)) - 1) * 3.0f, \ |
||
| 46 | // -XM_UNPACK_FACTOR_SIGNED / (FLOAT)((1 << ((BitsZ) - 1)) - 1) * 3.0f, \ |
||
| 47 | // -XM_UNPACK_FACTOR_SIGNED / (FLOAT)((1 << ((BitsW) - 1)) - 1) * 3.0f} |
||
| 48 | |||
| 49 | #define XM_PACK_UNSIGNEDN_SCALE(BitsX, BitsY, BitsZ, BitsW) \ |
||
| 50 | {-(FLOAT)((1 << (BitsX)) - 1) / XM_PACK_FACTOR, \ |
||
| 51 | -(FLOAT)((1 << (BitsY)) - 1) / XM_PACK_FACTOR, \ |
||
| 52 | -(FLOAT)((1 << (BitsZ)) - 1) / XM_PACK_FACTOR, \ |
||
| 53 | -(FLOAT)((1 << (BitsW)) - 1) / XM_PACK_FACTOR} |
||
| 54 | |||
| 55 | #define XM_PACK_SIGNEDN_SCALE(BitsX, BitsY, BitsZ, BitsW) \ |
||
| 56 | {-(FLOAT)((1 << ((BitsX) - 1)) - 1) / XM_PACK_FACTOR, \ |
||
| 57 | -(FLOAT)((1 << ((BitsY) - 1)) - 1) / XM_PACK_FACTOR, \ |
||
| 58 | -(FLOAT)((1 << ((BitsZ) - 1)) - 1) / XM_PACK_FACTOR, \ |
||
| 59 | -(FLOAT)((1 << ((BitsW) - 1)) - 1) / XM_PACK_FACTOR} |
||
| 60 | |||
| 61 | #define XM_PACK_OFFSET XMVectorSplatConstant(3, 0) |
||
| 62 | //#define XM_UNPACK_OFFSET XM_PACK_OFFSET |
||
| 63 | |||
| 64 | /**************************************************************************** |
||
| 65 | * |
||
| 66 | * Data conversion |
||
| 67 | * |
||
| 68 | ****************************************************************************/ |
||
| 69 | |||
| 70 | //------------------------------------------------------------------------------ |
||
| 71 | |||
| 72 | XMFINLINE FLOAT XMConvertHalfToFloat |
||
| 73 | ( |
||
| 74 | HALF Value |
||
| 75 | ) |
||
| 76 | { |
||
| 77 | #if defined(_XM_NO_INTRINSICS_) || defined(_XM_SSE_INTRINSICS_) |
||
| 78 | |||
| 79 | UINT Mantissa; |
||
| 80 | UINT Exponent; |
||
| 81 | UINT Result; |
||
| 82 | |||
| 83 | Mantissa = (UINT)(Value & 0x03FF); |
||
| 84 | |||
| 85 | if ((Value & 0x7C00) != 0) // The value is normalized |
||
| 86 | { |
||
| 87 | Exponent = (UINT)((Value >> 10) & 0x1F); |
||
| 88 | } |
||
| 89 | else if (Mantissa != 0) // The value is denormalized |
||
| 90 | { |
||
| 91 | // Normalize the value in the resulting float |
||
| 92 | Exponent = 1; |
||
| 93 | |||
| 94 | do |
||
| 95 | { |
||
| 96 | Exponent--; |
||
| 97 | Mantissa <<= 1; |
||
| 98 | } while ((Mantissa & 0x0400) == 0); |
||
| 99 | |||
| 100 | Mantissa &= 0x03FF; |
||
| 101 | } |
||
| 102 | else // The value is zero |
||
| 103 | { |
||
| 104 | Exponent = (UINT)-112; |
||
| 105 | } |
||
| 106 | |||
| 107 | Result = ((Value & 0x8000) << 16) | // Sign |
||
| 108 | ((Exponent + 112) << 23) | // Exponent |
||
| 109 | (Mantissa << 13); // Mantissa |
||
| 110 | |||
| 111 | return *(FLOAT*)&Result; |
||
| 112 | |||
| 113 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 114 | #endif |
||
| 115 | } |
||
| 116 | |||
| 117 | //------------------------------------------------------------------------------ |
||
| 118 | |||
| 119 | XMINLINE FLOAT* XMConvertHalfToFloatStream |
||
| 120 | ( |
||
| 121 | FLOAT* pOutputStream, |
||
| 122 | UINT OutputStride, |
||
| 123 | CONST HALF* pInputStream, |
||
| 124 | UINT InputStride, |
||
| 125 | UINT HalfCount |
||
| 126 | ) |
||
| 127 | { |
||
| 128 | #if defined(_XM_NO_INTRINSICS_) || defined(_XM_SSE_INTRINSICS_) |
||
| 129 | |||
| 130 | UINT i; |
||
| 131 | BYTE* pHalf = (BYTE*)pInputStream; |
||
| 132 | BYTE* pFloat = (BYTE*)pOutputStream; |
||
| 133 | |||
| 134 | XMASSERT(pOutputStream); |
||
| 135 | XMASSERT(pInputStream); |
||
| 136 | |||
| 137 | for (i = 0; i < HalfCount; i++) |
||
| 138 | { |
||
| 139 | *(FLOAT*)pFloat = XMConvertHalfToFloat(*(HALF*)pHalf); |
||
| 140 | pHalf += InputStride; |
||
| 141 | pFloat += OutputStride; |
||
| 142 | } |
||
| 143 | |||
| 144 | return pOutputStream; |
||
| 145 | |||
| 146 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 147 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 148 | } |
||
| 149 | |||
| 150 | //------------------------------------------------------------------------------ |
||
| 151 | |||
| 152 | XMFINLINE HALF XMConvertFloatToHalf |
||
| 153 | ( |
||
| 154 | FLOAT Value |
||
| 155 | ) |
||
| 156 | { |
||
| 157 | #if defined(_XM_NO_INTRINSICS_) || defined(_XM_SSE_INTRINSICS_) |
||
| 158 | UINT Result; |
||
| 159 | |||
| 160 | UINT IValue = ((UINT *)(&Value))[0]; |
||
| 161 | UINT Sign = (IValue & 0x80000000U) >> 16U; |
||
| 162 | IValue = IValue & 0x7FFFFFFFU; // Hack off the sign |
||
| 163 | |||
| 164 | if (IValue > 0x47FFEFFFU) |
||
| 165 | { |
||
| 166 | // The number is too large to be represented as a half. Saturate to infinity. |
||
| 167 | Result = 0x7FFFU; |
||
| 168 | } |
||
| 169 | else |
||
| 170 | { |
||
| 171 | if (IValue < 0x38800000U) |
||
| 172 | { |
||
| 173 | // The number is too small to be represented as a normalized half. |
||
| 174 | // Convert it to a denormalized value. |
||
| 175 | UINT Shift = 113U - (IValue >> 23U); |
||
| 176 | IValue = (0x800000U | (IValue & 0x7FFFFFU)) >> Shift; |
||
| 177 | } |
||
| 178 | else |
||
| 179 | { |
||
| 180 | // Rebias the exponent to represent the value as a normalized half. |
||
| 181 | IValue += 0xC8000000U; |
||
| 182 | } |
||
| 183 | |||
| 184 | Result = ((IValue + 0x0FFFU + ((IValue >> 13U) & 1U)) >> 13U)&0x7FFFU; |
||
| 185 | } |
||
| 186 | return (HALF)(Result|Sign); |
||
| 187 | |||
| 188 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 189 | #endif |
||
| 190 | } |
||
| 191 | |||
| 192 | //------------------------------------------------------------------------------ |
||
| 193 | |||
| 194 | XMINLINE HALF* XMConvertFloatToHalfStream |
||
| 195 | ( |
||
| 196 | HALF* pOutputStream, |
||
| 197 | UINT OutputStride, |
||
| 198 | CONST FLOAT* pInputStream, |
||
| 199 | UINT InputStride, |
||
| 200 | UINT FloatCount |
||
| 201 | ) |
||
| 202 | { |
||
| 203 | #if defined(_XM_NO_INTRINSICS_) || defined(_XM_SSE_INTRINSICS_) |
||
| 204 | |||
| 205 | UINT i; |
||
| 206 | BYTE* pFloat = (BYTE*)pInputStream; |
||
| 207 | BYTE* pHalf = (BYTE*)pOutputStream; |
||
| 208 | |||
| 209 | XMASSERT(pOutputStream); |
||
| 210 | XMASSERT(pInputStream); |
||
| 211 | |||
| 212 | for (i = 0; i < FloatCount; i++) |
||
| 213 | { |
||
| 214 | *(HALF*)pHalf = XMConvertFloatToHalf(*(FLOAT*)pFloat); |
||
| 215 | pFloat += InputStride; |
||
| 216 | pHalf += OutputStride; |
||
| 217 | } |
||
| 218 | return pOutputStream; |
||
| 219 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 220 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 221 | } |
||
| 222 | |||
| 223 | //------------------------------------------------------------------------------ |
||
| 224 | |||
| 225 | #if defined(_XM_NO_INTRINSICS_) || defined(_XM_SSE_INTRINSICS_) |
||
| 226 | // For VMX128, these routines are all defines in the main header |
||
| 227 | |||
| 228 | #pragma warning(push) |
||
| 229 | #pragma warning(disable:4701) // Prevent warnings about 'Result' potentially being used without having been initialized |
||
| 230 | |||
| 231 | XMINLINE XMVECTOR XMConvertVectorIntToFloat |
||
| 232 | ( |
||
| 233 | FXMVECTOR VInt, |
||
| 234 | UINT DivExponent |
||
| 235 | ) |
||
| 236 | { |
||
| 237 | #if defined(_XM_NO_INTRINSICS_) |
||
| 238 | UINT ElementIndex; |
||
| 239 | FLOAT fScale; |
||
| 240 | XMVECTOR Result; |
||
| 241 | XMASSERT(DivExponent<32); |
||
| 242 | fScale = 1.0f / (FLOAT)(1U << DivExponent); |
||
| 243 | ElementIndex = 0; |
||
| 244 | do { |
||
| 245 | INT iTemp = (INT)VInt.vector4_u32[ElementIndex]; |
||
| 246 | Result.vector4_f32[ElementIndex] = ((FLOAT)iTemp) * fScale; |
||
| 247 | } while (++ElementIndex<4); |
||
| 248 | return Result; |
||
| 249 | #else // _XM_SSE_INTRINSICS_ |
||
| 250 | XMASSERT(DivExponent<32); |
||
| 251 | // Convert to floats |
||
| 252 | XMVECTOR vResult = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&VInt)[0]); |
||
| 253 | // Convert DivExponent into 1.0f/(1<<DivExponent) |
||
| 254 | UINT uScale = 0x3F800000U - (DivExponent << 23); |
||
| 255 | // Splat the scalar value |
||
| 256 | __m128i vScale = _mm_set1_epi32(uScale); |
||
| 257 | vResult = _mm_mul_ps(vResult,reinterpret_cast<const __m128 *>(&vScale)[0]); |
||
| 258 | return vResult; |
||
| 259 | #endif |
||
| 260 | } |
||
| 261 | |||
| 262 | //------------------------------------------------------------------------------ |
||
| 263 | |||
| 264 | XMINLINE XMVECTOR XMConvertVectorFloatToInt |
||
| 265 | ( |
||
| 266 | FXMVECTOR VFloat, |
||
| 267 | UINT MulExponent |
||
| 268 | ) |
||
| 269 | { |
||
| 270 | #if defined(_XM_NO_INTRINSICS_) |
||
| 271 | UINT ElementIndex; |
||
| 272 | XMVECTOR Result; |
||
| 273 | FLOAT fScale; |
||
| 274 | XMASSERT(MulExponent<32); |
||
| 275 | // Get the scalar factor. |
||
| 276 | fScale = (FLOAT)(1U << MulExponent); |
||
| 277 | ElementIndex = 0; |
||
| 278 | do { |
||
| 279 | INT iResult; |
||
| 280 | FLOAT fTemp = VFloat.vector4_f32[ElementIndex]*fScale; |
||
| 281 | if (fTemp <= -(65536.0f*32768.0f)) { |
||
| 282 | iResult = (-0x7FFFFFFF)-1; |
||
| 283 | } else if (fTemp > (65536.0f*32768.0f)-128.0f) { |
||
| 284 | iResult = 0x7FFFFFFF; |
||
| 285 | } else { |
||
| 286 | iResult = (INT)fTemp; |
||
| 287 | } |
||
| 288 | Result.vector4_u32[ElementIndex] = (UINT)iResult; |
||
| 289 | } while (++ElementIndex<4); |
||
| 290 | return Result; |
||
| 291 | #else // _XM_SSE_INTRINSICS_ |
||
| 292 | XMASSERT(MulExponent<32); |
||
| 293 | static const XMVECTORF32 MaxInt = {65536.0f*32768.0f-128.0f,65536.0f*32768.0f-128.0f,65536.0f*32768.0f-128.0f,65536.0f*32768.0f-128.0f}; |
||
| 294 | XMVECTOR vResult = _mm_set_ps1((FLOAT)(1U << MulExponent)); |
||
| 295 | vResult = _mm_mul_ps(vResult,VFloat); |
||
| 296 | // In case of positive overflow, detect it |
||
| 297 | XMVECTOR vOverflow = _mm_cmpgt_ps(vResult,MaxInt); |
||
| 298 | // Float to int conversion |
||
| 299 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 300 | // If there was positive overflow, set to 0x7FFFFFFF |
||
| 301 | vResult = _mm_and_ps(vOverflow,g_XMAbsMask); |
||
| 302 | vOverflow = _mm_andnot_ps(vOverflow,reinterpret_cast<const __m128 *>(&vResulti)[0]); |
||
| 303 | vOverflow = _mm_or_ps(vOverflow,vResult); |
||
| 304 | return vOverflow; |
||
| 305 | #endif |
||
| 306 | } |
||
| 307 | |||
| 308 | //------------------------------------------------------------------------------ |
||
| 309 | |||
| 310 | XMINLINE XMVECTOR XMConvertVectorUIntToFloat |
||
| 311 | ( |
||
| 312 | FXMVECTOR VUInt, |
||
| 313 | UINT DivExponent |
||
| 314 | ) |
||
| 315 | { |
||
| 316 | #if defined(_XM_NO_INTRINSICS_) |
||
| 317 | UINT ElementIndex; |
||
| 318 | FLOAT fScale; |
||
| 319 | XMVECTOR Result; |
||
| 320 | XMASSERT(DivExponent<32); |
||
| 321 | fScale = 1.0f / (FLOAT)(1U << DivExponent); |
||
| 322 | ElementIndex = 0; |
||
| 323 | do { |
||
| 324 | Result.vector4_f32[ElementIndex] = (FLOAT)VUInt.vector4_u32[ElementIndex] * fScale; |
||
| 325 | } while (++ElementIndex<4); |
||
| 326 | return Result; |
||
| 327 | #else // _XM_SSE_INTRINSICS_ |
||
| 328 | XMASSERT(DivExponent<32); |
||
| 329 | static const XMVECTORF32 FixUnsigned = {32768.0f*65536.0f,32768.0f*65536.0f,32768.0f*65536.0f,32768.0f*65536.0f}; |
||
| 330 | // For the values that are higher than 0x7FFFFFFF, a fixup is needed |
||
| 331 | // Determine which ones need the fix. |
||
| 332 | XMVECTOR vMask = _mm_and_ps(VUInt,g_XMNegativeZero); |
||
| 333 | // Force all values positive |
||
| 334 | XMVECTOR vResult = _mm_xor_ps(VUInt,vMask); |
||
| 335 | // Convert to floats |
||
| 336 | vResult = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vResult)[0]); |
||
| 337 | // Convert 0x80000000 -> 0xFFFFFFFF |
||
| 338 | __m128i iMask = _mm_srai_epi32(reinterpret_cast<const __m128i *>(&vMask)[0],31); |
||
| 339 | // For only the ones that are too big, add the fixup |
||
| 340 | vMask = _mm_and_ps(reinterpret_cast<const __m128 *>(&iMask)[0],FixUnsigned); |
||
| 341 | vResult = _mm_add_ps(vResult,vMask); |
||
| 342 | // Convert DivExponent into 1.0f/(1<<DivExponent) |
||
| 343 | UINT uScale = 0x3F800000U - (DivExponent << 23); |
||
| 344 | // Splat |
||
| 345 | iMask = _mm_set1_epi32(uScale); |
||
| 346 | vResult = _mm_mul_ps(vResult,reinterpret_cast<const __m128 *>(&iMask)[0]); |
||
| 347 | return vResult; |
||
| 348 | #endif |
||
| 349 | } |
||
| 350 | |||
| 351 | //------------------------------------------------------------------------------ |
||
| 352 | |||
| 353 | XMINLINE XMVECTOR XMConvertVectorFloatToUInt |
||
| 354 | ( |
||
| 355 | FXMVECTOR VFloat, |
||
| 356 | UINT MulExponent |
||
| 357 | ) |
||
| 358 | { |
||
| 359 | #if defined(_XM_NO_INTRINSICS_) |
||
| 360 | UINT ElementIndex; |
||
| 361 | XMVECTOR Result; |
||
| 362 | FLOAT fScale; |
||
| 363 | XMASSERT(MulExponent<32); |
||
| 364 | // Get the scalar factor. |
||
| 365 | fScale = (FLOAT)(1U << MulExponent); |
||
| 366 | ElementIndex = 0; |
||
| 367 | do { |
||
| 368 | UINT uResult; |
||
| 369 | FLOAT fTemp = VFloat.vector4_f32[ElementIndex]*fScale; |
||
| 370 | if (fTemp <= 0.0f) { |
||
| 371 | uResult = 0; |
||
| 372 | } else if (fTemp >= (65536.0f*65536.0f)) { |
||
| 373 | uResult = 0xFFFFFFFFU; |
||
| 374 | } else { |
||
| 375 | uResult = (UINT)fTemp; |
||
| 376 | } |
||
| 377 | Result.vector4_u32[ElementIndex] = uResult; |
||
| 378 | } while (++ElementIndex<4); |
||
| 379 | return Result; |
||
| 380 | #else // _XM_SSE_INTRINSICS_ |
||
| 381 | XMASSERT(MulExponent<32); |
||
| 382 | static const XMVECTORF32 MaxUInt = {65536.0f*65536.0f-256.0f,65536.0f*65536.0f-256.0f,65536.0f*65536.0f-256.0f,65536.0f*65536.0f-256.0f}; |
||
| 383 | static const XMVECTORF32 UnsignedFix = {32768.0f*65536.0f,32768.0f*65536.0f,32768.0f*65536.0f,32768.0f*65536.0f}; |
||
| 384 | XMVECTOR vResult = _mm_set_ps1(static_cast<float>(1U << MulExponent)); |
||
| 385 | vResult = _mm_mul_ps(vResult,VFloat); |
||
| 386 | // Clamp to >=0 |
||
| 387 | vResult = _mm_max_ps(vResult,g_XMZero); |
||
| 388 | // Any numbers that are too big, set to 0xFFFFFFFFU |
||
| 389 | XMVECTOR vOverflow = _mm_cmpgt_ps(vResult,MaxUInt); |
||
| 390 | XMVECTOR vValue = UnsignedFix; |
||
| 391 | // Too large for a signed integer? |
||
| 392 | XMVECTOR vMask = _mm_cmpge_ps(vResult,vValue); |
||
| 393 | // Zero for number's lower than 0x80000000, 32768.0f*65536.0f otherwise |
||
| 394 | vValue = _mm_and_ps(vValue,vMask); |
||
| 395 | // Perform fixup only on numbers too large (Keeps low bit precision) |
||
| 396 | vResult = _mm_sub_ps(vResult,vValue); |
||
| 397 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 398 | // Convert from signed to unsigned pnly if greater than 0x80000000 |
||
| 399 | vMask = _mm_and_ps(vMask,g_XMNegativeZero); |
||
| 400 | vResult = _mm_xor_ps(reinterpret_cast<const __m128 *>(&vResulti)[0],vMask); |
||
| 401 | // On those that are too large, set to 0xFFFFFFFF |
||
| 402 | vResult = _mm_or_ps(vResult,vOverflow); |
||
| 403 | return vResult; |
||
| 404 | #endif |
||
| 405 | } |
||
| 406 | |||
| 407 | #pragma warning(pop) |
||
| 408 | |||
| 409 | #endif // _XM_NO_INTRINSICS_ || _XM_SSE_INTRINSICS_ |
||
| 410 | |||
| 411 | /**************************************************************************** |
||
| 412 | * |
||
| 413 | * Vector and matrix load operations |
||
| 414 | * |
||
| 415 | ****************************************************************************/ |
||
| 416 | |||
| 417 | //------------------------------------------------------------------------------ |
||
| 418 | |||
| 419 | XMFINLINE XMVECTOR XMLoadInt(CONST UINT* pSource) |
||
| 420 | { |
||
| 421 | #if defined(_XM_NO_INTRINSICS_) |
||
| 422 | |||
| 423 | XMVECTOR V; |
||
| 424 | XMASSERT(pSource); |
||
| 425 | XMASSERT(((UINT_PTR)pSource & 3) == 0); |
||
| 426 | |||
| 427 | V.vector4_u32[0] = *pSource; |
||
| 428 | |||
| 429 | return V; |
||
| 430 | |||
| 431 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 432 | XMASSERT(pSource); |
||
| 433 | XMASSERT(((UINT_PTR)pSource & 3) == 0); |
||
| 434 | __m128i V = _mm_set_epi32( 0, 0, 0, *pSource ); |
||
| 435 | return reinterpret_cast<__m128 *>(&V)[0]; |
||
| 436 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 437 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 438 | } |
||
| 439 | |||
| 440 | //------------------------------------------------------------------------------ |
||
| 441 | |||
| 442 | XMFINLINE XMVECTOR XMLoadFloat(CONST FLOAT* pSource) |
||
| 443 | { |
||
| 444 | #if defined(_XM_NO_INTRINSICS_) |
||
| 445 | |||
| 446 | XMVECTOR V; |
||
| 447 | XMASSERT(pSource); |
||
| 448 | XMASSERT(((UINT_PTR)pSource & 3) == 0); |
||
| 449 | |||
| 450 | V.vector4_f32[0] = *pSource; |
||
| 451 | |||
| 452 | return V; |
||
| 453 | |||
| 454 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 455 | XMASSERT(pSource); |
||
| 456 | XMASSERT(((UINT_PTR)pSource & 3) == 0); |
||
| 457 | |||
| 458 | return _mm_load_ss( pSource ); |
||
| 459 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 460 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 461 | } |
||
| 462 | |||
| 463 | //------------------------------------------------------------------------------ |
||
| 464 | |||
| 465 | XMFINLINE XMVECTOR XMLoadInt2 |
||
| 466 | ( |
||
| 467 | CONST UINT* pSource |
||
| 468 | ) |
||
| 469 | { |
||
| 470 | #if defined(_XM_NO_INTRINSICS_) |
||
| 471 | |||
| 472 | XMVECTOR V; |
||
| 473 | |||
| 474 | XMASSERT(pSource); |
||
| 475 | |||
| 476 | V.vector4_u32[0] = pSource[0]; |
||
| 477 | V.vector4_u32[1] = pSource[1]; |
||
| 478 | |||
| 479 | return V; |
||
| 480 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 481 | |||
| 482 | XMASSERT(pSource); |
||
| 483 | __m128i V = _mm_set_epi32( 0, 0, *(pSource+1), *pSource ); |
||
| 484 | return reinterpret_cast<__m128 *>(&V)[0]; |
||
| 485 | |||
| 486 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 487 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 488 | } |
||
| 489 | |||
| 490 | //------------------------------------------------------------------------------ |
||
| 491 | |||
| 492 | XMFINLINE XMVECTOR XMLoadInt2A |
||
| 493 | ( |
||
| 494 | CONST UINT* pSource |
||
| 495 | ) |
||
| 496 | { |
||
| 497 | #if defined(_XM_NO_INTRINSICS_) |
||
| 498 | |||
| 499 | XMVECTOR V; |
||
| 500 | |||
| 501 | XMASSERT(pSource); |
||
| 502 | XMASSERT(((UINT_PTR)pSource & 0xF) == 0); |
||
| 503 | |||
| 504 | V.vector4_u32[0] = pSource[0]; |
||
| 505 | V.vector4_u32[1] = pSource[1]; |
||
| 506 | |||
| 507 | return V; |
||
| 508 | |||
| 509 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 510 | |||
| 511 | XMASSERT(pSource); |
||
| 512 | __m128i V = _mm_loadl_epi64( (const __m128i*)pSource ); |
||
| 513 | return reinterpret_cast<__m128 *>(&V)[0]; |
||
| 514 | |||
| 515 | #else // _XM_VMX128_INTRINSICS_ |
||
| 516 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 517 | } |
||
| 518 | |||
| 519 | //------------------------------------------------------------------------------ |
||
| 520 | |||
| 521 | XMFINLINE XMVECTOR XMLoadFloat2 |
||
| 522 | ( |
||
| 523 | CONST XMFLOAT2* pSource |
||
| 524 | ) |
||
| 525 | { |
||
| 526 | #if defined(_XM_NO_INTRINSICS_) |
||
| 527 | XMVECTOR V; |
||
| 528 | XMASSERT(pSource); |
||
| 529 | ((UINT *)(&V.vector4_f32[0]))[0] = ((const UINT *)(&pSource->x))[0]; |
||
| 530 | ((UINT *)(&V.vector4_f32[1]))[0] = ((const UINT *)(&pSource->y))[0]; |
||
| 531 | V.vector4_f32[2] = V.vector4_f32[3] = 0.0f; |
||
| 532 | return V; |
||
| 533 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 534 | XMASSERT(pSource); |
||
| 535 | #ifdef _XM_X86_ |
||
| 536 | __m128 x = _mm_load_ss( &pSource->x ); |
||
| 537 | __m128 y = _mm_load_ss( &pSource->y ); |
||
| 538 | return _mm_unpacklo_ps( x, y ); |
||
| 539 | #else // _XM_X64_ |
||
| 540 | // This reads 2 floats past the memory that should be ignored. |
||
| 541 | return _mm_loadu_ps( &pSource->x ); |
||
| 542 | #endif |
||
| 543 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 544 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 545 | } |
||
| 546 | |||
| 547 | //------------------------------------------------------------------------------ |
||
| 548 | |||
| 549 | XMFINLINE XMVECTOR XMLoadFloat2A |
||
| 550 | ( |
||
| 551 | CONST XMFLOAT2A* pSource |
||
| 552 | ) |
||
| 553 | { |
||
| 554 | #if defined(_XM_NO_INTRINSICS_) |
||
| 555 | |||
| 556 | XMVECTOR V; |
||
| 557 | |||
| 558 | XMASSERT(pSource); |
||
| 559 | XMASSERT(((UINT_PTR)pSource & 0xF) == 0); |
||
| 560 | |||
| 561 | V.vector4_f32[0] = pSource->x; |
||
| 562 | V.vector4_f32[1] = pSource->y; |
||
| 563 | |||
| 564 | return V; |
||
| 565 | |||
| 566 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 567 | XMASSERT(pSource); |
||
| 568 | #ifdef _XM_X86_ |
||
| 569 | __m128 x = _mm_load_ss( &pSource->x ); |
||
| 570 | __m128 y = _mm_load_ss( &pSource->y ); |
||
| 571 | return _mm_unpacklo_ps( x, y ); |
||
| 572 | #else // _XM_X64_ |
||
| 573 | // This reads 2 floats past the memory that should be ignored. |
||
| 574 | return _mm_load_ps( &pSource->x ); |
||
| 575 | #endif |
||
| 576 | #else // _XM_VMX128_INTRINSICS_ |
||
| 577 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 578 | } |
||
| 579 | |||
| 580 | //------------------------------------------------------------------------------ |
||
| 581 | |||
| 582 | XMFINLINE XMVECTOR XMLoadHalf2 |
||
| 583 | ( |
||
| 584 | CONST XMHALF2* pSource |
||
| 585 | ) |
||
| 586 | { |
||
| 587 | #if defined(_XM_NO_INTRINSICS_) |
||
| 588 | XMASSERT(pSource); |
||
| 589 | { |
||
| 590 | XMVECTOR vResult = { |
||
| 591 | XMConvertHalfToFloat(pSource->x), |
||
| 592 | XMConvertHalfToFloat(pSource->y), |
||
| 593 | 0.0f, |
||
| 594 | 0.0f |
||
| 595 | }; |
||
| 596 | return vResult; |
||
| 597 | } |
||
| 598 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 599 | XMASSERT(pSource); |
||
| 600 | XMVECTOR vResult = { |
||
| 601 | XMConvertHalfToFloat(pSource->x), |
||
| 602 | XMConvertHalfToFloat(pSource->y), |
||
| 603 | 0.0f, |
||
| 604 | 0.0f |
||
| 605 | }; |
||
| 606 | return vResult; |
||
| 607 | |||
| 608 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 609 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 610 | } |
||
| 611 | |||
| 612 | //------------------------------------------------------------------------------ |
||
| 613 | |||
| 614 | XMFINLINE XMVECTOR XMLoadShortN2 |
||
| 615 | ( |
||
| 616 | CONST XMSHORTN2* pSource |
||
| 617 | ) |
||
| 618 | { |
||
| 619 | #if defined(_XM_NO_INTRINSICS_) |
||
| 620 | XMASSERT(pSource); |
||
| 621 | XMASSERT(pSource->x != -32768); |
||
| 622 | XMASSERT(pSource->y != -32768); |
||
| 623 | { |
||
| 624 | XMVECTOR vResult = { |
||
| 625 | (FLOAT)pSource->x * (1.0f/32767.0f), |
||
| 626 | (FLOAT)pSource->y * (1.0f/32767.0f), |
||
| 627 | 0.0f, |
||
| 628 | 0.0f |
||
| 629 | }; |
||
| 630 | return vResult; |
||
| 631 | } |
||
| 632 | |||
| 633 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 634 | XMASSERT(pSource); |
||
| 635 | XMASSERT(pSource->x != -32768); |
||
| 636 | XMASSERT(pSource->y != -32768); |
||
| 637 | // Splat the two shorts in all four entries (WORD alignment okay, |
||
| 638 | // DWORD alignment preferred) |
||
| 639 | __m128 vTemp = _mm_load_ps1(reinterpret_cast<const float *>(&pSource->x)); |
||
| 640 | // Mask x&0xFFFF, y&0xFFFF0000,z&0,w&0 |
||
| 641 | vTemp = _mm_and_ps(vTemp,g_XMMaskX16Y16); |
||
| 642 | // x needs to be sign extended |
||
| 643 | vTemp = _mm_xor_ps(vTemp,g_XMFlipX16Y16); |
||
| 644 | // Convert to floating point numbers |
||
| 645 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 646 | // x - 0x8000 to undo the signed order. |
||
| 647 | vTemp = _mm_add_ps(vTemp,g_XMFixX16Y16); |
||
| 648 | // Convert 0-32767 to 0.0f-1.0f |
||
| 649 | return _mm_mul_ps(vTemp,g_XMNormalizeX16Y16); |
||
| 650 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 651 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 652 | } |
||
| 653 | |||
| 654 | //------------------------------------------------------------------------------ |
||
| 655 | |||
| 656 | XMFINLINE XMVECTOR XMLoadShort2 |
||
| 657 | ( |
||
| 658 | CONST XMSHORT2* pSource |
||
| 659 | ) |
||
| 660 | { |
||
| 661 | #if defined(_XM_NO_INTRINSICS_) |
||
| 662 | |||
| 663 | XMVECTOR V; |
||
| 664 | |||
| 665 | XMASSERT(pSource); |
||
| 666 | XMASSERT(pSource->x != -32768); |
||
| 667 | XMASSERT(pSource->y != -32768); |
||
| 668 | |||
| 669 | V.vector4_f32[0] = (FLOAT)pSource->x; |
||
| 670 | V.vector4_f32[1] = (FLOAT)pSource->y; |
||
| 671 | |||
| 672 | return V; |
||
| 673 | |||
| 674 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 675 | XMASSERT(pSource); |
||
| 676 | XMASSERT(pSource->x != -32768); |
||
| 677 | XMASSERT(pSource->y != -32768); |
||
| 678 | // Splat the two shorts in all four entries (WORD alignment okay, |
||
| 679 | // DWORD alignment preferred) |
||
| 680 | __m128 vTemp = _mm_load_ps1(reinterpret_cast<const float *>(&pSource->x)); |
||
| 681 | // Mask x&0xFFFF, y&0xFFFF0000,z&0,w&0 |
||
| 682 | vTemp = _mm_and_ps(vTemp,g_XMMaskX16Y16); |
||
| 683 | // x needs to be sign extended |
||
| 684 | vTemp = _mm_xor_ps(vTemp,g_XMFlipX16Y16); |
||
| 685 | // Convert to floating point numbers |
||
| 686 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 687 | // x - 0x8000 to undo the signed order. |
||
| 688 | vTemp = _mm_add_ps(vTemp,g_XMFixX16Y16); |
||
| 689 | // Y is 65536 too large |
||
| 690 | return _mm_mul_ps(vTemp,g_XMFixupY16); |
||
| 691 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 692 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 693 | } |
||
| 694 | |||
| 695 | //------------------------------------------------------------------------------ |
||
| 696 | |||
| 697 | XMFINLINE XMVECTOR XMLoadUShortN2 |
||
| 698 | ( |
||
| 699 | CONST XMUSHORTN2* pSource |
||
| 700 | ) |
||
| 701 | { |
||
| 702 | #if defined(_XM_NO_INTRINSICS_) |
||
| 703 | |||
| 704 | XMVECTOR V; |
||
| 705 | |||
| 706 | XMASSERT(pSource); |
||
| 707 | |||
| 708 | V.vector4_f32[0] = (FLOAT)pSource->x / 65535.0f; |
||
| 709 | V.vector4_f32[1] = (FLOAT)pSource->y / 65535.0f; |
||
| 710 | |||
| 711 | return V; |
||
| 712 | |||
| 713 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 714 | static const XMVECTORF32 FixupY16 = {1.0f/65535.0f,1.0f/(65535.0f*65536.0f),0.0f,0.0f}; |
||
| 715 | static const XMVECTORF32 FixaddY16 = {0,32768.0f*65536.0f,0,0}; |
||
| 716 | XMASSERT(pSource); |
||
| 717 | // Splat the two shorts in all four entries (WORD alignment okay, |
||
| 718 | // DWORD alignment preferred) |
||
| 719 | __m128 vTemp = _mm_load_ps1(reinterpret_cast<const float *>(&pSource->x)); |
||
| 720 | // Mask x&0xFFFF, y&0xFFFF0000,z&0,w&0 |
||
| 721 | vTemp = _mm_and_ps(vTemp,g_XMMaskX16Y16); |
||
| 722 | // y needs to be sign flipped |
||
| 723 | vTemp = _mm_xor_ps(vTemp,g_XMFlipY); |
||
| 724 | // Convert to floating point numbers |
||
| 725 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 726 | // y + 0x8000 to undo the signed order. |
||
| 727 | vTemp = _mm_add_ps(vTemp,FixaddY16); |
||
| 728 | // Y is 65536 times too large |
||
| 729 | vTemp = _mm_mul_ps(vTemp,FixupY16); |
||
| 730 | return vTemp; |
||
| 731 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 732 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 733 | } |
||
| 734 | |||
| 735 | //------------------------------------------------------------------------------ |
||
| 736 | |||
| 737 | XMFINLINE XMVECTOR XMLoadUShort2 |
||
| 738 | ( |
||
| 739 | CONST XMUSHORT2* pSource |
||
| 740 | ) |
||
| 741 | { |
||
| 742 | #if defined(_XM_NO_INTRINSICS_) |
||
| 743 | |||
| 744 | XMVECTOR V; |
||
| 745 | |||
| 746 | XMASSERT(pSource); |
||
| 747 | |||
| 748 | V.vector4_f32[0] = (FLOAT)pSource->x; |
||
| 749 | V.vector4_f32[1] = (FLOAT)pSource->y; |
||
| 750 | |||
| 751 | return V; |
||
| 752 | |||
| 753 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 754 | static const XMVECTORF32 FixaddY16 = {0,32768.0f,0,0}; |
||
| 755 | XMASSERT(pSource); |
||
| 756 | // Splat the two shorts in all four entries (WORD alignment okay, |
||
| 757 | // DWORD alignment preferred) |
||
| 758 | __m128 vTemp = _mm_load_ps1(reinterpret_cast<const float *>(&pSource->x)); |
||
| 759 | // Mask x&0xFFFF, y&0xFFFF0000,z&0,w&0 |
||
| 760 | vTemp = _mm_and_ps(vTemp,g_XMMaskX16Y16); |
||
| 761 | // y needs to be sign flipped |
||
| 762 | vTemp = _mm_xor_ps(vTemp,g_XMFlipY); |
||
| 763 | // Convert to floating point numbers |
||
| 764 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 765 | // Y is 65536 times too large |
||
| 766 | vTemp = _mm_mul_ps(vTemp,g_XMFixupY16); |
||
| 767 | // y + 0x8000 to undo the signed order. |
||
| 768 | vTemp = _mm_add_ps(vTemp,FixaddY16); |
||
| 769 | return vTemp; |
||
| 770 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 771 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 772 | } |
||
| 773 | |||
| 774 | //------------------------------------------------------------------------------ |
||
| 775 | |||
| 776 | XMFINLINE XMVECTOR XMLoadInt3 |
||
| 777 | ( |
||
| 778 | CONST UINT* pSource |
||
| 779 | ) |
||
| 780 | { |
||
| 781 | #if defined(_XM_NO_INTRINSICS_) |
||
| 782 | |||
| 783 | XMVECTOR V; |
||
| 784 | |||
| 785 | XMASSERT(pSource); |
||
| 786 | |||
| 787 | V.vector4_u32[0] = pSource[0]; |
||
| 788 | V.vector4_u32[1] = pSource[1]; |
||
| 789 | V.vector4_u32[2] = pSource[2]; |
||
| 790 | |||
| 791 | return V; |
||
| 792 | |||
| 793 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 794 | XMASSERT(pSource); |
||
| 795 | __m128i V = _mm_set_epi32( 0, *(pSource+2), *(pSource+1), *pSource ); |
||
| 796 | return reinterpret_cast<__m128 *>(&V)[0]; |
||
| 797 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 798 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 799 | } |
||
| 800 | |||
| 801 | //------------------------------------------------------------------------------ |
||
| 802 | |||
| 803 | XMFINLINE XMVECTOR XMLoadInt3A |
||
| 804 | ( |
||
| 805 | CONST UINT* pSource |
||
| 806 | ) |
||
| 807 | { |
||
| 808 | #if defined(_XM_NO_INTRINSICS_) |
||
| 809 | |||
| 810 | XMVECTOR V; |
||
| 811 | |||
| 812 | XMASSERT(pSource); |
||
| 813 | XMASSERT(((UINT_PTR)pSource & 0xF) == 0); |
||
| 814 | |||
| 815 | V.vector4_u32[0] = pSource[0]; |
||
| 816 | V.vector4_u32[1] = pSource[1]; |
||
| 817 | V.vector4_u32[2] = pSource[2]; |
||
| 818 | |||
| 819 | return V; |
||
| 820 | |||
| 821 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 822 | XMASSERT(pSource); |
||
| 823 | |||
| 824 | // Reads an extra integer that is 'undefined' |
||
| 825 | |||
| 826 | __m128i V = _mm_load_si128( (const __m128i*)pSource ); |
||
| 827 | return reinterpret_cast<__m128 *>(&V)[0]; |
||
| 828 | #else // _XM_VMX128_INTRINSICS_ |
||
| 829 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 830 | } |
||
| 831 | |||
| 832 | //------------------------------------------------------------------------------ |
||
| 833 | |||
| 834 | XMFINLINE XMVECTOR XMLoadFloat3 |
||
| 835 | ( |
||
| 836 | CONST XMFLOAT3* pSource |
||
| 837 | ) |
||
| 838 | { |
||
| 839 | #if defined(_XM_NO_INTRINSICS_) |
||
| 840 | XMVECTOR V; |
||
| 841 | XMASSERT(pSource); |
||
| 842 | ((UINT *)(&V.vector4_f32[0]))[0] = ((const UINT *)(&pSource->x))[0]; |
||
| 843 | ((UINT *)(&V.vector4_f32[1]))[0] = ((const UINT *)(&pSource->y))[0]; |
||
| 844 | ((UINT *)(&V.vector4_f32[2]))[0] = ((const UINT *)(&pSource->z))[0]; |
||
| 845 | V.vector4_f32[3] = 0.0f; |
||
| 846 | return V; |
||
| 847 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 848 | XMASSERT(pSource); |
||
| 849 | // This reads 1 floats past the memory that should be ignored. |
||
| 850 | return _mm_loadu_ps( &pSource->x ); |
||
| 851 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 852 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 853 | } |
||
| 854 | |||
| 855 | //------------------------------------------------------------------------------ |
||
| 856 | |||
| 857 | XMFINLINE XMVECTOR XMLoadFloat3A |
||
| 858 | ( |
||
| 859 | CONST XMFLOAT3A* pSource |
||
| 860 | ) |
||
| 861 | { |
||
| 862 | #if defined(_XM_NO_INTRINSICS_) |
||
| 863 | |||
| 864 | XMVECTOR V; |
||
| 865 | |||
| 866 | XMASSERT(pSource); |
||
| 867 | XMASSERT(((UINT_PTR)pSource & 0xF) == 0); |
||
| 868 | |||
| 869 | V.vector4_f32[0] = pSource->x; |
||
| 870 | V.vector4_f32[1] = pSource->y; |
||
| 871 | V.vector4_f32[2] = pSource->z; |
||
| 872 | |||
| 873 | return V; |
||
| 874 | |||
| 875 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 876 | XMASSERT(pSource); |
||
| 877 | |||
| 878 | // This reads 1 floats past the memory that should be ignored. |
||
| 879 | |||
| 880 | return _mm_load_ps( &pSource->x ); |
||
| 881 | #else // _XM_VMX128_INTRINSICS_ |
||
| 882 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 883 | } |
||
| 884 | |||
| 885 | //------------------------------------------------------------------------------ |
||
| 886 | |||
| 887 | XMFINLINE XMVECTOR XMLoadUHenDN3 |
||
| 888 | ( |
||
| 889 | CONST XMUHENDN3* pSource |
||
| 890 | ) |
||
| 891 | { |
||
| 892 | #if defined(_XM_NO_INTRINSICS_) |
||
| 893 | |||
| 894 | XMVECTOR V; |
||
| 895 | UINT Element; |
||
| 896 | |||
| 897 | XMASSERT(pSource); |
||
| 898 | |||
| 899 | Element = pSource->v & 0x7FF; |
||
| 900 | V.vector4_f32[0] = (FLOAT)Element / 2047.0f; |
||
| 901 | Element = (pSource->v >> 11) & 0x7FF; |
||
| 902 | V.vector4_f32[1] = (FLOAT)Element / 2047.0f; |
||
| 903 | Element = (pSource->v >> 22) & 0x3FF; |
||
| 904 | V.vector4_f32[2] = (FLOAT)Element / 1023.0f; |
||
| 905 | |||
| 906 | return V; |
||
| 907 | |||
| 908 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 909 | static const XMVECTORF32 UHenDN3Mul = {1.0f/2047.0f,1.0f/(2047.0f*2048.0f),1.0f/(1023.0f*2048.0f*2048.0f),0}; |
||
| 910 | XMASSERT(pSource); |
||
| 911 | // Get the 32 bit value and splat it |
||
| 912 | XMVECTOR vResult = _mm_load_ps1(reinterpret_cast<const float *>(&pSource->v)); |
||
| 913 | // Mask off x, y and z |
||
| 914 | vResult = _mm_and_ps(vResult,g_XMMaskHenD3); |
||
| 915 | // Convert x and y to unsigned |
||
| 916 | vResult = _mm_xor_ps(vResult,g_XMFlipZ); |
||
| 917 | // Convert to float |
||
| 918 | vResult = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vResult)[0]); |
||
| 919 | // Convert x and y back to signed |
||
| 920 | vResult = _mm_add_ps(vResult,g_XMAddUHenD3); |
||
| 921 | // Normalize x,y and z to -1.0f-1.0f |
||
| 922 | vResult = _mm_mul_ps(vResult,UHenDN3Mul); |
||
| 923 | return vResult; |
||
| 924 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 925 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 926 | } |
||
| 927 | |||
| 928 | //------------------------------------------------------------------------------ |
||
| 929 | |||
| 930 | XMFINLINE XMVECTOR XMLoadUHenD3 |
||
| 931 | ( |
||
| 932 | CONST XMUHEND3* pSource |
||
| 933 | ) |
||
| 934 | { |
||
| 935 | #if defined(_XM_NO_INTRINSICS_) |
||
| 936 | |||
| 937 | XMVECTOR V; |
||
| 938 | UINT Element; |
||
| 939 | |||
| 940 | XMASSERT(pSource); |
||
| 941 | |||
| 942 | Element = pSource->v & 0x7FF; |
||
| 943 | V.vector4_f32[0] = (FLOAT)Element; |
||
| 944 | Element = (pSource->v >> 11) & 0x7FF; |
||
| 945 | V.vector4_f32[1] = (FLOAT)Element; |
||
| 946 | Element = (pSource->v >> 22) & 0x3FF; |
||
| 947 | V.vector4_f32[2] = (FLOAT)Element; |
||
| 948 | |||
| 949 | return V; |
||
| 950 | |||
| 951 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 952 | XMASSERT(pSource); |
||
| 953 | // Get the 32 bit value and splat it |
||
| 954 | XMVECTOR vResult = _mm_load_ps1(reinterpret_cast<const float *>(&pSource->v)); |
||
| 955 | // Mask off x, y and z |
||
| 956 | vResult = _mm_and_ps(vResult,g_XMMaskHenD3); |
||
| 957 | // Convert x and y to unsigned |
||
| 958 | vResult = _mm_xor_ps(vResult,g_XMFlipZ); |
||
| 959 | // Convert to float |
||
| 960 | vResult = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vResult)[0]); |
||
| 961 | // Convert x and y back to signed |
||
| 962 | vResult = _mm_add_ps(vResult,g_XMAddUHenD3); |
||
| 963 | // Normalize x and y to -1024-1023.0f and z to -512-511.0f |
||
| 964 | vResult = _mm_mul_ps(vResult,g_XMMulHenD3); |
||
| 965 | return vResult; |
||
| 966 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 967 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 968 | } |
||
| 969 | |||
| 970 | //------------------------------------------------------------------------------ |
||
| 971 | |||
| 972 | XMFINLINE XMVECTOR XMLoadHenDN3 |
||
| 973 | ( |
||
| 974 | CONST XMHENDN3* pSource |
||
| 975 | ) |
||
| 976 | { |
||
| 977 | #if defined(_XM_NO_INTRINSICS_) |
||
| 978 | |||
| 979 | XMVECTOR V; |
||
| 980 | UINT Element; |
||
| 981 | static CONST UINT SignExtendXY[] = {0x00000000, 0xFFFFF800}; |
||
| 982 | static CONST UINT SignExtendZ[] = {0x00000000, 0xFFFFFC00}; |
||
| 983 | |||
| 984 | XMASSERT(pSource); |
||
| 985 | XMASSERT((pSource->v & 0x7FF) != 0x400); |
||
| 986 | XMASSERT(((pSource->v >> 11) & 0x7FF) != 0x400); |
||
| 987 | XMASSERT(((pSource->v >> 22) & 0x3FF) != 0x200); |
||
| 988 | |||
| 989 | Element = pSource->v & 0x7FF; |
||
| 990 | V.vector4_f32[0] = (FLOAT)(SHORT)(Element | SignExtendXY[Element >> 10]) / 1023.0f; |
||
| 991 | Element = (pSource->v >> 11) & 0x7FF; |
||
| 992 | V.vector4_f32[1] = (FLOAT)(SHORT)(Element | SignExtendXY[Element >> 10]) / 1023.0f; |
||
| 993 | Element = (pSource->v >> 22) & 0x3FF; |
||
| 994 | V.vector4_f32[2] = (FLOAT)(SHORT)(Element | SignExtendZ[Element >> 9]) / 511.0f; |
||
| 995 | |||
| 996 | return V; |
||
| 997 | |||
| 998 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 999 | static const XMVECTORF32 HenDN3Mul = {1.0f/1023.0f,1.0f/(1023.0f*2048.0f),1.0f/(511.0f*2048.0f*2048.0f),0}; |
||
| 1000 | XMASSERT(pSource); |
||
| 1001 | XMASSERT((pSource->v & 0x7FF) != 0x400); |
||
| 1002 | XMASSERT(((pSource->v >> 11) & 0x7FF) != 0x400); |
||
| 1003 | XMASSERT(((pSource->v >> 22) & 0x3FF) != 0x200); |
||
| 1004 | // Get the 32 bit value and splat it |
||
| 1005 | XMVECTOR vResult = _mm_load_ps1(reinterpret_cast<const float *>(&pSource->v)); |
||
| 1006 | // Mask off x, y and z |
||
| 1007 | vResult = _mm_and_ps(vResult,g_XMMaskHenD3); |
||
| 1008 | // Convert x and y to unsigned |
||
| 1009 | vResult = _mm_xor_ps(vResult,g_XMXorHenD3); |
||
| 1010 | // Convert to float |
||
| 1011 | vResult = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vResult)[0]); |
||
| 1012 | // Convert x and y back to signed |
||
| 1013 | vResult = _mm_add_ps(vResult,g_XMAddHenD3); |
||
| 1014 | // Normalize x,y and z to -1.0f-1.0f |
||
| 1015 | vResult = _mm_mul_ps(vResult,HenDN3Mul); |
||
| 1016 | return vResult; |
||
| 1017 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 1018 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 1019 | } |
||
| 1020 | |||
| 1021 | //------------------------------------------------------------------------------ |
||
| 1022 | |||
| 1023 | XMFINLINE XMVECTOR XMLoadHenD3 |
||
| 1024 | ( |
||
| 1025 | CONST XMHEND3* pSource |
||
| 1026 | ) |
||
| 1027 | { |
||
| 1028 | #if defined(_XM_NO_INTRINSICS_) |
||
| 1029 | |||
| 1030 | XMVECTOR V; |
||
| 1031 | UINT Element; |
||
| 1032 | static CONST UINT SignExtendXY[] = {0x00000000, 0xFFFFF800}; |
||
| 1033 | static CONST UINT SignExtendZ[] = {0x00000000, 0xFFFFFC00}; |
||
| 1034 | |||
| 1035 | XMASSERT(pSource); |
||
| 1036 | XMASSERT((pSource->v & 0x7FF) != 0x400); |
||
| 1037 | XMASSERT(((pSource->v >> 11) & 0x7FF) != 0x400); |
||
| 1038 | XMASSERT(((pSource->v >> 22) & 0x3FF) != 0x200); |
||
| 1039 | |||
| 1040 | Element = pSource->v & 0x7FF; |
||
| 1041 | V.vector4_f32[0] = (FLOAT)(SHORT)(Element | SignExtendXY[Element >> 10]); |
||
| 1042 | Element = (pSource->v >> 11) & 0x7FF; |
||
| 1043 | V.vector4_f32[1] = (FLOAT)(SHORT)(Element | SignExtendXY[Element >> 10]); |
||
| 1044 | Element = (pSource->v >> 22) & 0x3FF; |
||
| 1045 | V.vector4_f32[2] = (FLOAT)(SHORT)(Element | SignExtendZ[Element >> 9]); |
||
| 1046 | |||
| 1047 | return V; |
||
| 1048 | |||
| 1049 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 1050 | XMASSERT(pSource); |
||
| 1051 | XMASSERT((pSource->v & 0x7FF) != 0x400); |
||
| 1052 | XMASSERT(((pSource->v >> 11) & 0x7FF) != 0x400); |
||
| 1053 | XMASSERT(((pSource->v >> 22) & 0x3FF) != 0x200); |
||
| 1054 | // Get the 32 bit value and splat it |
||
| 1055 | XMVECTOR vResult = _mm_load_ps1(reinterpret_cast<const float *>(&pSource->v)); |
||
| 1056 | // Mask off x, y and z |
||
| 1057 | vResult = _mm_and_ps(vResult,g_XMMaskHenD3); |
||
| 1058 | // Convert x and y to unsigned |
||
| 1059 | vResult = _mm_xor_ps(vResult,g_XMXorHenD3); |
||
| 1060 | // Convert to float |
||
| 1061 | vResult = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vResult)[0]); |
||
| 1062 | // Convert x and y back to signed |
||
| 1063 | vResult = _mm_add_ps(vResult,g_XMAddHenD3); |
||
| 1064 | // Normalize x and y to -1024-1023.0f and z to -512-511.0f |
||
| 1065 | vResult = _mm_mul_ps(vResult,g_XMMulHenD3); |
||
| 1066 | return vResult; |
||
| 1067 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 1068 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 1069 | } |
||
| 1070 | |||
| 1071 | //------------------------------------------------------------------------------ |
||
| 1072 | |||
| 1073 | XMFINLINE XMVECTOR XMLoadUDHenN3 |
||
| 1074 | ( |
||
| 1075 | CONST XMUDHENN3* pSource |
||
| 1076 | ) |
||
| 1077 | { |
||
| 1078 | #if defined(_XM_NO_INTRINSICS_) |
||
| 1079 | |||
| 1080 | XMVECTOR V; |
||
| 1081 | UINT Element; |
||
| 1082 | |||
| 1083 | XMASSERT(pSource); |
||
| 1084 | |||
| 1085 | Element = pSource->v & 0x3FF; |
||
| 1086 | V.vector4_f32[0] = (FLOAT)Element / 1023.0f; |
||
| 1087 | Element = (pSource->v >> 10) & 0x7FF; |
||
| 1088 | V.vector4_f32[1] = (FLOAT)Element / 2047.0f; |
||
| 1089 | Element = (pSource->v >> 21) & 0x7FF; |
||
| 1090 | V.vector4_f32[2] = (FLOAT)Element / 2047.0f; |
||
| 1091 | |||
| 1092 | return V; |
||
| 1093 | |||
| 1094 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 1095 | static const XMVECTORF32 UDHenN3Mul = {1.0f/1023.0f,1.0f/(2047.0f*1024.0f),1.0f/(2047.0f*1024.0f*2048.0f),0}; |
||
| 1096 | XMASSERT(pSource); |
||
| 1097 | // Get the 32 bit value and splat it |
||
| 1098 | XMVECTOR vResult = _mm_load_ps1(reinterpret_cast<const float *>(&pSource->v)); |
||
| 1099 | // Mask off x, y and z |
||
| 1100 | vResult = _mm_and_ps(vResult,g_XMMaskDHen3); |
||
| 1101 | // Convert x and y to unsigned |
||
| 1102 | vResult = _mm_xor_ps(vResult,g_XMFlipZ); |
||
| 1103 | // Convert to float |
||
| 1104 | vResult = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vResult)[0]); |
||
| 1105 | // Convert x and y back to signed |
||
| 1106 | vResult = _mm_add_ps(vResult,g_XMAddUHenD3); |
||
| 1107 | // Normalize x,y and z to -1.0f-1.0f |
||
| 1108 | vResult = _mm_mul_ps(vResult,UDHenN3Mul); |
||
| 1109 | return vResult; |
||
| 1110 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 1111 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 1112 | } |
||
| 1113 | |||
| 1114 | //------------------------------------------------------------------------------ |
||
| 1115 | |||
| 1116 | XMFINLINE XMVECTOR XMLoadUDHen3 |
||
| 1117 | ( |
||
| 1118 | CONST XMUDHEN3* pSource |
||
| 1119 | ) |
||
| 1120 | { |
||
| 1121 | #if defined(_XM_NO_INTRINSICS_) |
||
| 1122 | |||
| 1123 | XMVECTOR V; |
||
| 1124 | UINT Element; |
||
| 1125 | |||
| 1126 | XMASSERT(pSource); |
||
| 1127 | |||
| 1128 | Element = pSource->v & 0x3FF; |
||
| 1129 | V.vector4_f32[0] = (FLOAT)Element; |
||
| 1130 | Element = (pSource->v >> 10) & 0x7FF; |
||
| 1131 | V.vector4_f32[1] = (FLOAT)Element; |
||
| 1132 | Element = (pSource->v >> 21) & 0x7FF; |
||
| 1133 | V.vector4_f32[2] = (FLOAT)Element; |
||
| 1134 | |||
| 1135 | return V; |
||
| 1136 | |||
| 1137 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 1138 | XMASSERT(pSource); |
||
| 1139 | // Get the 32 bit value and splat it |
||
| 1140 | XMVECTOR vResult = _mm_load_ps1(reinterpret_cast<const float *>(&pSource->v)); |
||
| 1141 | // Mask off x, y and z |
||
| 1142 | vResult = _mm_and_ps(vResult,g_XMMaskDHen3); |
||
| 1143 | // Convert x and y to unsigned |
||
| 1144 | vResult = _mm_xor_ps(vResult,g_XMFlipZ); |
||
| 1145 | // Convert to float |
||
| 1146 | vResult = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vResult)[0]); |
||
| 1147 | // Convert x and y back to signed |
||
| 1148 | vResult = _mm_add_ps(vResult,g_XMAddUHenD3); |
||
| 1149 | // Normalize x to 0-1023.0f and y and z to 0-2047.0f |
||
| 1150 | vResult = _mm_mul_ps(vResult,g_XMMulDHen3); |
||
| 1151 | return vResult; |
||
| 1152 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 1153 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 1154 | } |
||
| 1155 | |||
| 1156 | //------------------------------------------------------------------------------ |
||
| 1157 | |||
| 1158 | XMFINLINE XMVECTOR XMLoadDHenN3 |
||
| 1159 | ( |
||
| 1160 | CONST XMDHENN3* pSource |
||
| 1161 | ) |
||
| 1162 | { |
||
| 1163 | #if defined(_XM_NO_INTRINSICS_) |
||
| 1164 | |||
| 1165 | XMVECTOR V; |
||
| 1166 | UINT Element; |
||
| 1167 | static CONST UINT SignExtendX[] = {0x00000000, 0xFFFFFC00}; |
||
| 1168 | static CONST UINT SignExtendYZ[] = {0x00000000, 0xFFFFF800}; |
||
| 1169 | |||
| 1170 | XMASSERT(pSource); |
||
| 1171 | XMASSERT((pSource->v & 0x3FF) != 0x200); |
||
| 1172 | XMASSERT(((pSource->v >> 10) & 0x7FF) != 0x400); |
||
| 1173 | XMASSERT(((pSource->v >> 21) & 0x7FF) != 0x400); |
||
| 1174 | |||
| 1175 | Element = pSource->v & 0x3FF; |
||
| 1176 | V.vector4_f32[0] = (FLOAT)(SHORT)(Element | SignExtendX[Element >> 9]) / 511.0f; |
||
| 1177 | Element = (pSource->v >> 10) & 0x7FF; |
||
| 1178 | V.vector4_f32[1] = (FLOAT)(SHORT)(Element | SignExtendYZ[Element >> 10]) / 1023.0f; |
||
| 1179 | Element = (pSource->v >> 21) & 0x7FF; |
||
| 1180 | V.vector4_f32[2] = (FLOAT)(SHORT)(Element | SignExtendYZ[Element >> 10]) / 1023.0f; |
||
| 1181 | |||
| 1182 | return V; |
||
| 1183 | |||
| 1184 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 1185 | static const XMVECTORF32 DHenN3Mul = {1.0f/511.0f,1.0f/(1023.0f*1024.0f),1.0f/(1023.0f*1024.0f*2048.0f),0}; |
||
| 1186 | XMASSERT(pSource); |
||
| 1187 | XMASSERT((pSource->v & 0x3FF) != 0x200); |
||
| 1188 | XMASSERT(((pSource->v >> 10) & 0x7FF) != 0x400); |
||
| 1189 | XMASSERT(((pSource->v >> 21) & 0x7FF) != 0x400); |
||
| 1190 | // Get the 32 bit value and splat it |
||
| 1191 | XMVECTOR vResult = _mm_load_ps1(reinterpret_cast<const float *>(&pSource->v)); |
||
| 1192 | // Mask off x, y and z |
||
| 1193 | vResult = _mm_and_ps(vResult,g_XMMaskDHen3); |
||
| 1194 | // Convert x and y to unsigned |
||
| 1195 | vResult = _mm_xor_ps(vResult,g_XMXorDHen3); |
||
| 1196 | // Convert to float |
||
| 1197 | vResult = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vResult)[0]); |
||
| 1198 | // Convert x and y back to signed |
||
| 1199 | vResult = _mm_add_ps(vResult,g_XMAddDHen3); |
||
| 1200 | // Normalize x,y and z to -1.0f-1.0f |
||
| 1201 | vResult = _mm_mul_ps(vResult,DHenN3Mul); |
||
| 1202 | return vResult; |
||
| 1203 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 1204 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 1205 | } |
||
| 1206 | |||
| 1207 | //------------------------------------------------------------------------------ |
||
| 1208 | |||
| 1209 | XMFINLINE XMVECTOR XMLoadDHen3 |
||
| 1210 | ( |
||
| 1211 | CONST XMDHEN3* pSource |
||
| 1212 | ) |
||
| 1213 | { |
||
| 1214 | #if defined(_XM_NO_INTRINSICS_) |
||
| 1215 | |||
| 1216 | XMVECTOR V; |
||
| 1217 | UINT Element; |
||
| 1218 | static CONST UINT SignExtendX[] = {0x00000000, 0xFFFFFC00}; |
||
| 1219 | static CONST UINT SignExtendYZ[] = {0x00000000, 0xFFFFF800}; |
||
| 1220 | |||
| 1221 | XMASSERT(pSource); |
||
| 1222 | XMASSERT((pSource->v & 0x3FF) != 0x200); |
||
| 1223 | XMASSERT(((pSource->v >> 10) & 0x7FF) != 0x400); |
||
| 1224 | XMASSERT(((pSource->v >> 21) & 0x7FF) != 0x400); |
||
| 1225 | |||
| 1226 | Element = pSource->v & 0x3FF; |
||
| 1227 | V.vector4_f32[0] = (FLOAT)(SHORT)(Element | SignExtendX[Element >> 9]); |
||
| 1228 | Element = (pSource->v >> 10) & 0x7FF; |
||
| 1229 | V.vector4_f32[1] = (FLOAT)(SHORT)(Element | SignExtendYZ[Element >> 10]); |
||
| 1230 | Element = (pSource->v >> 21) & 0x7FF; |
||
| 1231 | V.vector4_f32[2] = (FLOAT)(SHORT)(Element | SignExtendYZ[Element >> 10]); |
||
| 1232 | |||
| 1233 | return V; |
||
| 1234 | |||
| 1235 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 1236 | XMASSERT(pSource); |
||
| 1237 | XMASSERT((pSource->v & 0x3FF) != 0x200); |
||
| 1238 | XMASSERT(((pSource->v >> 10) & 0x7FF) != 0x400); |
||
| 1239 | XMASSERT(((pSource->v >> 21) & 0x7FF) != 0x400); |
||
| 1240 | // Get the 32 bit value and splat it |
||
| 1241 | XMVECTOR vResult = _mm_load_ps1(reinterpret_cast<const float *>(&pSource->v)); |
||
| 1242 | // Mask off x, y and z |
||
| 1243 | vResult = _mm_and_ps(vResult,g_XMMaskDHen3); |
||
| 1244 | // Convert x and y to unsigned |
||
| 1245 | vResult = _mm_xor_ps(vResult,g_XMXorDHen3); |
||
| 1246 | // Convert to float |
||
| 1247 | vResult = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vResult)[0]); |
||
| 1248 | // Convert x and y back to signed |
||
| 1249 | vResult = _mm_add_ps(vResult,g_XMAddDHen3); |
||
| 1250 | // Normalize x to -210-511.0f and y and z to -1024-1023.0f |
||
| 1251 | vResult = _mm_mul_ps(vResult,g_XMMulDHen3); |
||
| 1252 | return vResult; |
||
| 1253 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 1254 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 1255 | } |
||
| 1256 | |||
| 1257 | //------------------------------------------------------------------------------ |
||
| 1258 | |||
| 1259 | XMFINLINE XMVECTOR XMLoadU565 |
||
| 1260 | ( |
||
| 1261 | CONST XMU565* pSource |
||
| 1262 | ) |
||
| 1263 | { |
||
| 1264 | #if defined(_XM_SSE_INTRINSICS_) && !defined(_XM_NO_INTRINSICS_) |
||
| 1265 | static const XMVECTORI32 U565And = {0x1F,0x3F<<5,0x1F<<11,0}; |
||
| 1266 | static const XMVECTORF32 U565Mul = {1.0f,1.0f/32.0f,1.0f/2048.f,0}; |
||
| 1267 | XMASSERT(pSource); |
||
| 1268 | // Get the 32 bit value and splat it |
||
| 1269 | XMVECTOR vResult = _mm_load_ps1(reinterpret_cast<const float *>(&pSource->v)); |
||
| 1270 | // Mask off x, y and z |
||
| 1271 | vResult = _mm_and_ps(vResult,U565And); |
||
| 1272 | // Convert to float |
||
| 1273 | vResult = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vResult)[0]); |
||
| 1274 | // Normalize x, y, and z |
||
| 1275 | vResult = _mm_mul_ps(vResult,U565Mul); |
||
| 1276 | return vResult; |
||
| 1277 | #else |
||
| 1278 | XMVECTOR V; |
||
| 1279 | UINT Element; |
||
| 1280 | |||
| 1281 | XMASSERT(pSource); |
||
| 1282 | |||
| 1283 | Element = pSource->v & 0x1F; |
||
| 1284 | V.vector4_f32[0] = (FLOAT)Element; |
||
| 1285 | Element = (pSource->v >> 5) & 0x3F; |
||
| 1286 | V.vector4_f32[1] = (FLOAT)Element; |
||
| 1287 | Element = (pSource->v >> 11) & 0x1F; |
||
| 1288 | V.vector4_f32[2] = (FLOAT)Element; |
||
| 1289 | |||
| 1290 | return V; |
||
| 1291 | #endif // !_XM_SSE_INTRINSICS_ |
||
| 1292 | } |
||
| 1293 | |||
| 1294 | //------------------------------------------------------------------------------ |
||
| 1295 | |||
| 1296 | XMFINLINE XMVECTOR XMLoadFloat3PK |
||
| 1297 | ( |
||
| 1298 | CONST XMFLOAT3PK* pSource |
||
| 1299 | ) |
||
| 1300 | { |
||
| 1301 | UINT Mantissa; |
||
| 1302 | UINT Exponent; |
||
| 1303 | UINT Result[3]; |
||
| 1304 | |||
| 1305 | XMASSERT(pSource); |
||
| 1306 | |||
| 1307 | // X Channel (6-bit mantissa) |
||
| 1308 | Mantissa = pSource->xm; |
||
| 1309 | |||
| 1310 | if ( pSource->xe == 0x1f ) // INF or NAN |
||
| 1311 | { |
||
| 1312 | Result[0] = 0x7f800000 | (pSource->xm << 17); |
||
| 1313 | } |
||
| 1314 | else |
||
| 1315 | { |
||
| 1316 | if ( pSource->xe != 0 ) // The value is normalized |
||
| 1317 | { |
||
| 1318 | Exponent = pSource->xe; |
||
| 1319 | } |
||
| 1320 | else if (Mantissa != 0) // The value is denormalized |
||
| 1321 | { |
||
| 1322 | // Normalize the value in the resulting float |
||
| 1323 | Exponent = 1; |
||
| 1324 | |||
| 1325 | do |
||
| 1326 | { |
||
| 1327 | Exponent--; |
||
| 1328 | Mantissa <<= 1; |
||
| 1329 | } while ((Mantissa & 0x40) == 0); |
||
| 1330 | |||
| 1331 | Mantissa &= 0x3F; |
||
| 1332 | } |
||
| 1333 | else // The value is zero |
||
| 1334 | { |
||
| 1335 | Exponent = (UINT)-112; |
||
| 1336 | } |
||
| 1337 | |||
| 1338 | Result[0] = ((Exponent + 112) << 23) | (Mantissa << 17); |
||
| 1339 | } |
||
| 1340 | |||
| 1341 | // Y Channel (6-bit mantissa) |
||
| 1342 | Mantissa = pSource->ym; |
||
| 1343 | |||
| 1344 | if ( pSource->ye == 0x1f ) // INF or NAN |
||
| 1345 | { |
||
| 1346 | Result[1] = 0x7f800000 | (pSource->ym << 17); |
||
| 1347 | } |
||
| 1348 | else |
||
| 1349 | { |
||
| 1350 | if ( pSource->ye != 0 ) // The value is normalized |
||
| 1351 | { |
||
| 1352 | Exponent = pSource->ye; |
||
| 1353 | } |
||
| 1354 | else if (Mantissa != 0) // The value is denormalized |
||
| 1355 | { |
||
| 1356 | // Normalize the value in the resulting float |
||
| 1357 | Exponent = 1; |
||
| 1358 | |||
| 1359 | do |
||
| 1360 | { |
||
| 1361 | Exponent--; |
||
| 1362 | Mantissa <<= 1; |
||
| 1363 | } while ((Mantissa & 0x40) == 0); |
||
| 1364 | |||
| 1365 | Mantissa &= 0x3F; |
||
| 1366 | } |
||
| 1367 | else // The value is zero |
||
| 1368 | { |
||
| 1369 | Exponent = (UINT)-112; |
||
| 1370 | } |
||
| 1371 | |||
| 1372 | Result[1] = ((Exponent + 112) << 23) | (Mantissa << 17); |
||
| 1373 | } |
||
| 1374 | |||
| 1375 | // Z Channel (5-bit mantissa) |
||
| 1376 | Mantissa = pSource->zm; |
||
| 1377 | |||
| 1378 | if ( pSource->ze == 0x1f ) // INF or NAN |
||
| 1379 | { |
||
| 1380 | Result[2] = 0x7f800000 | (pSource->zm << 17); |
||
| 1381 | } |
||
| 1382 | else |
||
| 1383 | { |
||
| 1384 | if ( pSource->ze != 0 ) // The value is normalized |
||
| 1385 | { |
||
| 1386 | Exponent = pSource->ze; |
||
| 1387 | } |
||
| 1388 | else if (Mantissa != 0) // The value is denormalized |
||
| 1389 | { |
||
| 1390 | // Normalize the value in the resulting float |
||
| 1391 | Exponent = 1; |
||
| 1392 | |||
| 1393 | do |
||
| 1394 | { |
||
| 1395 | Exponent--; |
||
| 1396 | Mantissa <<= 1; |
||
| 1397 | } while ((Mantissa & 0x20) == 0); |
||
| 1398 | |||
| 1399 | Mantissa &= 0x1F; |
||
| 1400 | } |
||
| 1401 | else // The value is zero |
||
| 1402 | { |
||
| 1403 | Exponent = (UINT)-112; |
||
| 1404 | } |
||
| 1405 | |||
| 1406 | Result[2] = ((Exponent + 112) << 23) | (Mantissa << 18); |
||
| 1407 | } |
||
| 1408 | |||
| 1409 | return XMLoadFloat3( (XMFLOAT3*)&Result ); |
||
| 1410 | } |
||
| 1411 | |||
| 1412 | //------------------------------------------------------------------------------ |
||
| 1413 | |||
| 1414 | XMFINLINE XMVECTOR XMLoadFloat3SE |
||
| 1415 | ( |
||
| 1416 | CONST XMFLOAT3SE* pSource |
||
| 1417 | ) |
||
| 1418 | { |
||
| 1419 | UINT Mantissa; |
||
| 1420 | UINT Exponent, ExpBits; |
||
| 1421 | UINT Result[3]; |
||
| 1422 | |||
| 1423 | XMASSERT(pSource); |
||
| 1424 | |||
| 1425 | if ( pSource->e == 0x1f ) // INF or NAN |
||
| 1426 | { |
||
| 1427 | Result[0] = 0x7f800000 | (pSource->xm << 14); |
||
| 1428 | Result[1] = 0x7f800000 | (pSource->ym << 14); |
||
| 1429 | Result[2] = 0x7f800000 | (pSource->zm << 14); |
||
| 1430 | } |
||
| 1431 | else if ( pSource->e != 0 ) // The values are all normalized |
||
| 1432 | { |
||
| 1433 | Exponent = pSource->e; |
||
| 1434 | |||
| 1435 | ExpBits = (Exponent + 112) << 23; |
||
| 1436 | |||
| 1437 | Mantissa = pSource->xm; |
||
| 1438 | Result[0] = ExpBits | (Mantissa << 14); |
||
| 1439 | |||
| 1440 | Mantissa = pSource->ym; |
||
| 1441 | Result[1] = ExpBits | (Mantissa << 14); |
||
| 1442 | |||
| 1443 | Mantissa = pSource->zm; |
||
| 1444 | Result[2] = ExpBits | (Mantissa << 14); |
||
| 1445 | } |
||
| 1446 | else |
||
| 1447 | { |
||
| 1448 | // X Channel |
||
| 1449 | Mantissa = pSource->xm; |
||
| 1450 | |||
| 1451 | if (Mantissa != 0) // The value is denormalized |
||
| 1452 | { |
||
| 1453 | // Normalize the value in the resulting float |
||
| 1454 | Exponent = 1; |
||
| 1455 | |||
| 1456 | do |
||
| 1457 | { |
||
| 1458 | Exponent--; |
||
| 1459 | Mantissa <<= 1; |
||
| 1460 | } while ((Mantissa & 0x200) == 0); |
||
| 1461 | |||
| 1462 | Mantissa &= 0x1FF; |
||
| 1463 | } |
||
| 1464 | else // The value is zero |
||
| 1465 | { |
||
| 1466 | Exponent = (UINT)-112; |
||
| 1467 | } |
||
| 1468 | |||
| 1469 | Result[0] = ((Exponent + 112) << 23) | (Mantissa << 14); |
||
| 1470 | |||
| 1471 | // Y Channel |
||
| 1472 | Mantissa = pSource->ym; |
||
| 1473 | |||
| 1474 | if (Mantissa != 0) // The value is denormalized |
||
| 1475 | { |
||
| 1476 | // Normalize the value in the resulting float |
||
| 1477 | Exponent = 1; |
||
| 1478 | |||
| 1479 | do |
||
| 1480 | { |
||
| 1481 | Exponent--; |
||
| 1482 | Mantissa <<= 1; |
||
| 1483 | } while ((Mantissa & 0x200) == 0); |
||
| 1484 | |||
| 1485 | Mantissa &= 0x1FF; |
||
| 1486 | } |
||
| 1487 | else // The value is zero |
||
| 1488 | { |
||
| 1489 | Exponent = (UINT)-112; |
||
| 1490 | } |
||
| 1491 | |||
| 1492 | Result[1] = ((Exponent + 112) << 23) | (Mantissa << 14); |
||
| 1493 | |||
| 1494 | // Z Channel |
||
| 1495 | Mantissa = pSource->zm; |
||
| 1496 | |||
| 1497 | if (Mantissa != 0) // The value is denormalized |
||
| 1498 | { |
||
| 1499 | // Normalize the value in the resulting float |
||
| 1500 | Exponent = 1; |
||
| 1501 | |||
| 1502 | do |
||
| 1503 | { |
||
| 1504 | Exponent--; |
||
| 1505 | Mantissa <<= 1; |
||
| 1506 | } while ((Mantissa & 0x200) == 0); |
||
| 1507 | |||
| 1508 | Mantissa &= 0x1FF; |
||
| 1509 | } |
||
| 1510 | else // The value is zero |
||
| 1511 | { |
||
| 1512 | Exponent = (UINT)-112; |
||
| 1513 | } |
||
| 1514 | |||
| 1515 | Result[2] = ((Exponent + 112) << 23) | (Mantissa << 14); |
||
| 1516 | } |
||
| 1517 | |||
| 1518 | return XMLoadFloat3( (XMFLOAT3*)&Result ); |
||
| 1519 | } |
||
| 1520 | |||
| 1521 | //------------------------------------------------------------------------------ |
||
| 1522 | |||
| 1523 | XMFINLINE XMVECTOR XMLoadInt4 |
||
| 1524 | ( |
||
| 1525 | CONST UINT* pSource |
||
| 1526 | ) |
||
| 1527 | { |
||
| 1528 | #if defined(_XM_NO_INTRINSICS_) |
||
| 1529 | |||
| 1530 | XMVECTOR V; |
||
| 1531 | |||
| 1532 | XMASSERT(pSource); |
||
| 1533 | |||
| 1534 | V.vector4_u32[0] = pSource[0]; |
||
| 1535 | V.vector4_u32[1] = pSource[1]; |
||
| 1536 | V.vector4_u32[2] = pSource[2]; |
||
| 1537 | V.vector4_u32[3] = pSource[3]; |
||
| 1538 | |||
| 1539 | return V; |
||
| 1540 | |||
| 1541 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 1542 | |||
| 1543 | XMASSERT(pSource); |
||
| 1544 | __m128i V = _mm_loadu_si128( (const __m128i*)pSource ); |
||
| 1545 | return reinterpret_cast<__m128 *>(&V)[0]; |
||
| 1546 | |||
| 1547 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 1548 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 1549 | } |
||
| 1550 | |||
| 1551 | //------------------------------------------------------------------------------ |
||
| 1552 | |||
| 1553 | XMFINLINE XMVECTOR XMLoadInt4A |
||
| 1554 | ( |
||
| 1555 | CONST UINT* pSource |
||
| 1556 | ) |
||
| 1557 | { |
||
| 1558 | #if defined(_XM_NO_INTRINSICS_) |
||
| 1559 | |||
| 1560 | XMVECTOR V; |
||
| 1561 | |||
| 1562 | XMASSERT(pSource); |
||
| 1563 | XMASSERT(((UINT_PTR)pSource & 0xF) == 0); |
||
| 1564 | |||
| 1565 | V.vector4_u32[0] = pSource[0]; |
||
| 1566 | V.vector4_u32[1] = pSource[1]; |
||
| 1567 | V.vector4_u32[2] = pSource[2]; |
||
| 1568 | V.vector4_u32[3] = pSource[3]; |
||
| 1569 | |||
| 1570 | return V; |
||
| 1571 | |||
| 1572 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 1573 | |||
| 1574 | XMASSERT(pSource); |
||
| 1575 | XMASSERT(((UINT_PTR)pSource & 0xF) == 0); |
||
| 1576 | |||
| 1577 | __m128i V = _mm_load_si128( (const __m128i*)pSource ); |
||
| 1578 | return reinterpret_cast<__m128 *>(&V)[0]; |
||
| 1579 | |||
| 1580 | |||
| 1581 | #else // _XM_VMX128_INTRINSICS_ |
||
| 1582 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 1583 | } |
||
| 1584 | |||
| 1585 | //------------------------------------------------------------------------------ |
||
| 1586 | |||
| 1587 | XMFINLINE XMVECTOR XMLoadFloat4 |
||
| 1588 | ( |
||
| 1589 | CONST XMFLOAT4* pSource |
||
| 1590 | ) |
||
| 1591 | { |
||
| 1592 | #if defined(_XM_NO_INTRINSICS_) |
||
| 1593 | XMVECTOR V; |
||
| 1594 | XMASSERT(pSource); |
||
| 1595 | ((UINT *)(&V.vector4_f32[0]))[0] = ((const UINT *)(&pSource->x))[0]; |
||
| 1596 | ((UINT *)(&V.vector4_f32[1]))[0] = ((const UINT *)(&pSource->y))[0]; |
||
| 1597 | ((UINT *)(&V.vector4_f32[2]))[0] = ((const UINT *)(&pSource->z))[0]; |
||
| 1598 | ((UINT *)(&V.vector4_f32[3]))[0] = ((const UINT *)(&pSource->w))[0]; |
||
| 1599 | return V; |
||
| 1600 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 1601 | XMASSERT(pSource); |
||
| 1602 | return _mm_loadu_ps( &pSource->x ); |
||
| 1603 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 1604 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 1605 | } |
||
| 1606 | |||
| 1607 | //------------------------------------------------------------------------------ |
||
| 1608 | |||
| 1609 | XMFINLINE XMVECTOR XMLoadFloat4A |
||
| 1610 | ( |
||
| 1611 | CONST XMFLOAT4A* pSource |
||
| 1612 | ) |
||
| 1613 | { |
||
| 1614 | #if defined(_XM_NO_INTRINSICS_) |
||
| 1615 | |||
| 1616 | XMVECTOR V; |
||
| 1617 | |||
| 1618 | XMASSERT(pSource); |
||
| 1619 | XMASSERT(((UINT_PTR)pSource & 0xF) == 0); |
||
| 1620 | |||
| 1621 | V.vector4_f32[0] = pSource->x; |
||
| 1622 | V.vector4_f32[1] = pSource->y; |
||
| 1623 | V.vector4_f32[2] = pSource->z; |
||
| 1624 | V.vector4_f32[3] = pSource->w; |
||
| 1625 | |||
| 1626 | return V; |
||
| 1627 | |||
| 1628 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 1629 | |||
| 1630 | XMASSERT(pSource); |
||
| 1631 | XMASSERT(((UINT_PTR)pSource & 0xF) == 0); |
||
| 1632 | |||
| 1633 | return _mm_load_ps( &pSource->x ); |
||
| 1634 | |||
| 1635 | #else // _XM_VMX128_INTRINSICS_ |
||
| 1636 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 1637 | } |
||
| 1638 | |||
| 1639 | //------------------------------------------------------------------------------ |
||
| 1640 | |||
| 1641 | XMFINLINE XMVECTOR XMLoadHalf4 |
||
| 1642 | ( |
||
| 1643 | CONST XMHALF4* pSource |
||
| 1644 | ) |
||
| 1645 | { |
||
| 1646 | #if defined(_XM_NO_INTRINSICS_) |
||
| 1647 | XMASSERT(pSource); |
||
| 1648 | { |
||
| 1649 | XMVECTOR vResult = { |
||
| 1650 | XMConvertHalfToFloat(pSource->x), |
||
| 1651 | XMConvertHalfToFloat(pSource->y), |
||
| 1652 | XMConvertHalfToFloat(pSource->z), |
||
| 1653 | XMConvertHalfToFloat(pSource->w) |
||
| 1654 | }; |
||
| 1655 | return vResult; |
||
| 1656 | } |
||
| 1657 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 1658 | XMASSERT(pSource); |
||
| 1659 | XMVECTOR vResult = { |
||
| 1660 | XMConvertHalfToFloat(pSource->x), |
||
| 1661 | XMConvertHalfToFloat(pSource->y), |
||
| 1662 | XMConvertHalfToFloat(pSource->z), |
||
| 1663 | XMConvertHalfToFloat(pSource->w) |
||
| 1664 | }; |
||
| 1665 | return vResult; |
||
| 1666 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 1667 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 1668 | } |
||
| 1669 | |||
| 1670 | //------------------------------------------------------------------------------ |
||
| 1671 | |||
| 1672 | XMFINLINE XMVECTOR XMLoadShortN4 |
||
| 1673 | ( |
||
| 1674 | CONST XMSHORTN4* pSource |
||
| 1675 | ) |
||
| 1676 | { |
||
| 1677 | #if defined(_XM_NO_INTRINSICS_) |
||
| 1678 | XMASSERT(pSource); |
||
| 1679 | XMASSERT(pSource->x != -32768); |
||
| 1680 | XMASSERT(pSource->y != -32768); |
||
| 1681 | XMASSERT(pSource->z != -32768); |
||
| 1682 | XMASSERT(pSource->w != -32768); |
||
| 1683 | { |
||
| 1684 | XMVECTOR vResult = { |
||
| 1685 | (FLOAT)pSource->x * (1.0f/32767.0f), |
||
| 1686 | (FLOAT)pSource->y * (1.0f/32767.0f), |
||
| 1687 | (FLOAT)pSource->z * (1.0f/32767.0f), |
||
| 1688 | (FLOAT)pSource->w * (1.0f/32767.0f) |
||
| 1689 | }; |
||
| 1690 | return vResult; |
||
| 1691 | } |
||
| 1692 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 1693 | XMASSERT(pSource); |
||
| 1694 | XMASSERT(pSource->x != -32768); |
||
| 1695 | XMASSERT(pSource->y != -32768); |
||
| 1696 | XMASSERT(pSource->z != -32768); |
||
| 1697 | XMASSERT(pSource->w != -32768); |
||
| 1698 | // Splat the color in all four entries (x,z,y,w) |
||
| 1699 | __m128d vIntd = _mm_load1_pd(reinterpret_cast<const double *>(&pSource->x)); |
||
| 1700 | // Shift x&0ffff,z&0xffff,y&0xffff0000,w&0xffff0000 |
||
| 1701 | __m128 vTemp = _mm_and_ps(reinterpret_cast<const __m128 *>(&vIntd)[0],g_XMMaskX16Y16Z16W16); |
||
| 1702 | // x and z are unsigned! Flip the bits to convert the order to signed |
||
| 1703 | vTemp = _mm_xor_ps(vTemp,g_XMFlipX16Y16Z16W16); |
||
| 1704 | // Convert to floating point numbers |
||
| 1705 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 1706 | // x and z - 0x8000 to complete the conversion |
||
| 1707 | vTemp = _mm_add_ps(vTemp,g_XMFixX16Y16Z16W16); |
||
| 1708 | // Convert -32767-32767 to -1.0f-1.0f |
||
| 1709 | vTemp = _mm_mul_ps(vTemp,g_XMNormalizeX16Y16Z16W16); |
||
| 1710 | // Very important! The entries are x,z,y,w, flip it to x,y,z,w |
||
| 1711 | vTemp = _mm_shuffle_ps(vTemp,vTemp,_MM_SHUFFLE(3,1,2,0)); |
||
| 1712 | return vTemp; |
||
| 1713 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 1714 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 1715 | } |
||
| 1716 | |||
| 1717 | //------------------------------------------------------------------------------ |
||
| 1718 | |||
| 1719 | XMFINLINE XMVECTOR XMLoadShort4 |
||
| 1720 | ( |
||
| 1721 | CONST XMSHORT4* pSource |
||
| 1722 | ) |
||
| 1723 | { |
||
| 1724 | #if defined(_XM_NO_INTRINSICS_) |
||
| 1725 | |||
| 1726 | XMVECTOR V; |
||
| 1727 | |||
| 1728 | XMASSERT(pSource); |
||
| 1729 | XMASSERT(pSource->x != -32768); |
||
| 1730 | XMASSERT(pSource->y != -32768); |
||
| 1731 | XMASSERT(pSource->z != -32768); |
||
| 1732 | XMASSERT(pSource->w != -32768); |
||
| 1733 | |||
| 1734 | V.vector4_f32[0] = (FLOAT)pSource->x; |
||
| 1735 | V.vector4_f32[1] = (FLOAT)pSource->y; |
||
| 1736 | V.vector4_f32[2] = (FLOAT)pSource->z; |
||
| 1737 | V.vector4_f32[3] = (FLOAT)pSource->w; |
||
| 1738 | |||
| 1739 | return V; |
||
| 1740 | |||
| 1741 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 1742 | XMASSERT(pSource); |
||
| 1743 | XMASSERT(pSource->x != -32768); |
||
| 1744 | XMASSERT(pSource->y != -32768); |
||
| 1745 | XMASSERT(pSource->z != -32768); |
||
| 1746 | XMASSERT(pSource->w != -32768); |
||
| 1747 | // Splat the color in all four entries (x,z,y,w) |
||
| 1748 | __m128d vIntd = _mm_load1_pd(reinterpret_cast<const double *>(&pSource->x)); |
||
| 1749 | // Shift x&0ffff,z&0xffff,y&0xffff0000,w&0xffff0000 |
||
| 1750 | __m128 vTemp = _mm_and_ps(reinterpret_cast<const __m128 *>(&vIntd)[0],g_XMMaskX16Y16Z16W16); |
||
| 1751 | // x and z are unsigned! Flip the bits to convert the order to signed |
||
| 1752 | vTemp = _mm_xor_ps(vTemp,g_XMFlipX16Y16Z16W16); |
||
| 1753 | // Convert to floating point numbers |
||
| 1754 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 1755 | // x and z - 0x8000 to complete the conversion |
||
| 1756 | vTemp = _mm_add_ps(vTemp,g_XMFixX16Y16Z16W16); |
||
| 1757 | // Fix y and w because they are 65536 too large |
||
| 1758 | vTemp = _mm_mul_ps(vTemp,g_XMFixupY16W16); |
||
| 1759 | // Very important! The entries are x,z,y,w, flip it to x,y,z,w |
||
| 1760 | return _mm_shuffle_ps(vTemp,vTemp,_MM_SHUFFLE(3,1,2,0)); |
||
| 1761 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 1762 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 1763 | } |
||
| 1764 | |||
| 1765 | //------------------------------------------------------------------------------ |
||
| 1766 | |||
| 1767 | XMFINLINE XMVECTOR XMLoadUShortN4 |
||
| 1768 | ( |
||
| 1769 | CONST XMUSHORTN4* pSource |
||
| 1770 | ) |
||
| 1771 | { |
||
| 1772 | #if defined(_XM_NO_INTRINSICS_) |
||
| 1773 | |||
| 1774 | XMVECTOR V; |
||
| 1775 | |||
| 1776 | XMASSERT(pSource); |
||
| 1777 | |||
| 1778 | V.vector4_f32[0] = (FLOAT)pSource->x / 65535.0f; |
||
| 1779 | V.vector4_f32[1] = (FLOAT)pSource->y / 65535.0f; |
||
| 1780 | V.vector4_f32[2] = (FLOAT)pSource->z / 65535.0f; |
||
| 1781 | V.vector4_f32[3] = (FLOAT)pSource->w / 65535.0f; |
||
| 1782 | |||
| 1783 | return V; |
||
| 1784 | |||
| 1785 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 1786 | XMASSERT(pSource); |
||
| 1787 | static const XMVECTORF32 FixupY16W16 = {1.0f/65535.0f,1.0f/65535.0f,1.0f/(65535.0f*65536.0f),1.0f/(65535.0f*65536.0f)}; |
||
| 1788 | static const XMVECTORF32 FixaddY16W16 = {0,0,32768.0f*65536.0f,32768.0f*65536.0f}; |
||
| 1789 | XMASSERT(pSource); |
||
| 1790 | // Splat the color in all four entries (x,z,y,w) |
||
| 1791 | __m128d vIntd = _mm_load1_pd(reinterpret_cast<const double *>(&pSource->x)); |
||
| 1792 | // Shift x&0ffff,z&0xffff,y&0xffff0000,w&0xffff0000 |
||
| 1793 | __m128 vTemp = _mm_and_ps(reinterpret_cast<const __m128 *>(&vIntd)[0],g_XMMaskX16Y16Z16W16); |
||
| 1794 | // y and w are signed! Flip the bits to convert the order to unsigned |
||
| 1795 | vTemp = _mm_xor_ps(vTemp,g_XMFlipZW); |
||
| 1796 | // Convert to floating point numbers |
||
| 1797 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 1798 | // y and w + 0x8000 to complete the conversion |
||
| 1799 | vTemp = _mm_add_ps(vTemp,FixaddY16W16); |
||
| 1800 | // Fix y and w because they are 65536 too large |
||
| 1801 | vTemp = _mm_mul_ps(vTemp,FixupY16W16); |
||
| 1802 | // Very important! The entries are x,z,y,w, flip it to x,y,z,w |
||
| 1803 | return _mm_shuffle_ps(vTemp,vTemp,_MM_SHUFFLE(3,1,2,0)); |
||
| 1804 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 1805 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 1806 | } |
||
| 1807 | |||
| 1808 | //------------------------------------------------------------------------------ |
||
| 1809 | |||
| 1810 | XMFINLINE XMVECTOR XMLoadUShort4 |
||
| 1811 | ( |
||
| 1812 | CONST XMUSHORT4* pSource |
||
| 1813 | ) |
||
| 1814 | { |
||
| 1815 | #if defined(_XM_NO_INTRINSICS_) |
||
| 1816 | |||
| 1817 | XMVECTOR V; |
||
| 1818 | |||
| 1819 | XMASSERT(pSource); |
||
| 1820 | |||
| 1821 | V.vector4_f32[0] = (FLOAT)pSource->x; |
||
| 1822 | V.vector4_f32[1] = (FLOAT)pSource->y; |
||
| 1823 | V.vector4_f32[2] = (FLOAT)pSource->z; |
||
| 1824 | V.vector4_f32[3] = (FLOAT)pSource->w; |
||
| 1825 | |||
| 1826 | return V; |
||
| 1827 | |||
| 1828 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 1829 | XMASSERT(pSource); |
||
| 1830 | static const XMVECTORF32 FixaddY16W16 = {0,0,32768.0f,32768.0f}; |
||
| 1831 | XMASSERT(pSource); |
||
| 1832 | // Splat the color in all four entries (x,z,y,w) |
||
| 1833 | __m128d vIntd = _mm_load1_pd(reinterpret_cast<const double *>(&pSource->x)); |
||
| 1834 | // Shift x&0ffff,z&0xffff,y&0xffff0000,w&0xffff0000 |
||
| 1835 | __m128 vTemp = _mm_and_ps(reinterpret_cast<const __m128 *>(&vIntd)[0],g_XMMaskX16Y16Z16W16); |
||
| 1836 | // y and w are signed! Flip the bits to convert the order to unsigned |
||
| 1837 | vTemp = _mm_xor_ps(vTemp,g_XMFlipZW); |
||
| 1838 | // Convert to floating point numbers |
||
| 1839 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 1840 | // Fix y and w because they are 65536 too large |
||
| 1841 | vTemp = _mm_mul_ps(vTemp,g_XMFixupY16W16); |
||
| 1842 | // y and w + 0x8000 to complete the conversion |
||
| 1843 | vTemp = _mm_add_ps(vTemp,FixaddY16W16); |
||
| 1844 | // Very important! The entries are x,z,y,w, flip it to x,y,z,w |
||
| 1845 | return _mm_shuffle_ps(vTemp,vTemp,_MM_SHUFFLE(3,1,2,0)); |
||
| 1846 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 1847 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 1848 | } |
||
| 1849 | |||
| 1850 | //------------------------------------------------------------------------------ |
||
| 1851 | |||
| 1852 | XMFINLINE XMVECTOR XMLoadXIcoN4 |
||
| 1853 | ( |
||
| 1854 | CONST XMXICON4* pSource |
||
| 1855 | ) |
||
| 1856 | { |
||
| 1857 | #if defined(_XM_NO_INTRINSICS_) |
||
| 1858 | |||
| 1859 | XMVECTOR V; |
||
| 1860 | UINT Element; |
||
| 1861 | static CONST UINT SignExtend[] = {0x00000000, 0xFFF00000}; |
||
| 1862 | |||
| 1863 | XMASSERT(pSource); |
||
| 1864 | XMASSERT((pSource->v & 0xFFFFFull) != 0x80000ull); |
||
| 1865 | XMASSERT(((pSource->v >> 20) & 0xFFFFFull) != 0x80000ull); |
||
| 1866 | XMASSERT(((pSource->v >> 40) & 0xFFFFFull) != 0x80000ull); |
||
| 1867 | |||
| 1868 | Element = (UINT)pSource->v & 0xFFFFF; |
||
| 1869 | V.vector4_f32[0] = (FLOAT)(INT)(Element | SignExtend[Element >> 19]) / 524287.0f; |
||
| 1870 | Element = (UINT)(pSource->v >> 20) & 0xFFFFF; |
||
| 1871 | V.vector4_f32[1] = (FLOAT)(INT)(Element | SignExtend[Element >> 19]) / 524287.0f; |
||
| 1872 | Element = (UINT)(pSource->v >> 40) & 0xFFFFF; |
||
| 1873 | V.vector4_f32[2] = (FLOAT)(INT)(Element | SignExtend[Element >> 19]) / 524287.0f; |
||
| 1874 | V.vector4_f32[3] = (FLOAT)(pSource->v >> 60) / 15.0f; |
||
| 1875 | |||
| 1876 | return V; |
||
| 1877 | |||
| 1878 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 1879 | XMASSERT((pSource->v & 0xFFFFFull) != 0x80000ull); |
||
| 1880 | XMASSERT(((pSource->v >> 20) & 0xFFFFFull) != 0x80000ull); |
||
| 1881 | XMASSERT(((pSource->v >> 40) & 0xFFFFFull) != 0x80000ull); |
||
| 1882 | static const XMVECTORF32 LoadXIcoN4Mul = {1.0f/524287.0f,1.0f/(524287.0f*4096.0f),1.0f/524287.0f,1.0f/(15.0f*4096.0f*65536.0f)}; |
||
| 1883 | XMASSERT(pSource); |
||
| 1884 | // Grab the 64 bit structure |
||
| 1885 | __m128d vResultd = _mm_load_sd(reinterpret_cast<const double *>(&pSource->v)); |
||
| 1886 | // By shifting down 8 bits, y and z are in seperate 32 bit elements |
||
| 1887 | __m128i vResulti = _mm_srli_si128(reinterpret_cast<const __m128i *>(&vResultd)[0],8/8); |
||
| 1888 | // vResultd has x and w, vResulti has y and z, merge into one as x,w,y,z |
||
| 1889 | XMVECTOR vTemp = _mm_shuffle_ps(reinterpret_cast<const __m128 *>(&vResultd)[0],reinterpret_cast<const __m128 *>(&vResulti)[0],_MM_SHUFFLE(1,0,1,0)); |
||
| 1890 | // Fix the entries to x,y,z,w |
||
| 1891 | vTemp = _mm_shuffle_ps(vTemp,vTemp,_MM_SHUFFLE(1,3,2,0)); |
||
| 1892 | // Mask x,y,z and w |
||
| 1893 | vTemp = _mm_and_ps(vTemp,g_XMMaskIco4); |
||
| 1894 | // x and z are unsigned! Flip the bits to convert the order to signed |
||
| 1895 | vTemp = _mm_xor_ps(vTemp,g_XMXorXIco4); |
||
| 1896 | // Convert to floating point numbers |
||
| 1897 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 1898 | // x and z - 0x80 to complete the conversion |
||
| 1899 | vTemp = _mm_add_ps(vTemp,g_XMAddXIco4); |
||
| 1900 | // Fix y and w because they are too large |
||
| 1901 | vTemp = _mm_mul_ps(vTemp,LoadXIcoN4Mul); |
||
| 1902 | return vTemp; |
||
| 1903 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 1904 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 1905 | } |
||
| 1906 | |||
| 1907 | //------------------------------------------------------------------------------ |
||
| 1908 | |||
| 1909 | XMFINLINE XMVECTOR XMLoadXIco4 |
||
| 1910 | ( |
||
| 1911 | CONST XMXICO4* pSource |
||
| 1912 | ) |
||
| 1913 | { |
||
| 1914 | #if defined(_XM_NO_INTRINSICS_) |
||
| 1915 | |||
| 1916 | XMVECTOR V; |
||
| 1917 | UINT Element; |
||
| 1918 | static CONST UINT SignExtend[] = {0x00000000, 0xFFF00000}; |
||
| 1919 | |||
| 1920 | XMASSERT(pSource); |
||
| 1921 | XMASSERT((pSource->v & 0xFFFFFull) != 0x80000ull); |
||
| 1922 | XMASSERT(((pSource->v >> 20) & 0xFFFFFull) != 0x80000ull); |
||
| 1923 | XMASSERT(((pSource->v >> 40) & 0xFFFFFull) != 0x80000ull); |
||
| 1924 | |||
| 1925 | Element = (UINT)pSource->v & 0xFFFFF; |
||
| 1926 | V.vector4_f32[0] = (FLOAT)(INT)(Element | SignExtend[Element >> 19]); |
||
| 1927 | Element = (UINT)(pSource->v >> 20) & 0xFFFFF; |
||
| 1928 | V.vector4_f32[1] = (FLOAT)(INT)(Element | SignExtend[Element >> 19]); |
||
| 1929 | Element = (UINT)(pSource->v >> 40) & 0xFFFFF; |
||
| 1930 | V.vector4_f32[2] = (FLOAT)(INT)(Element | SignExtend[Element >> 19]); |
||
| 1931 | V.vector4_f32[3] = (FLOAT)(pSource->v >> 60); |
||
| 1932 | |||
| 1933 | return V; |
||
| 1934 | |||
| 1935 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 1936 | XMASSERT((pSource->v & 0xFFFFFull) != 0x80000ull); |
||
| 1937 | XMASSERT(((pSource->v >> 20) & 0xFFFFFull) != 0x80000ull); |
||
| 1938 | XMASSERT(((pSource->v >> 40) & 0xFFFFFull) != 0x80000ull); |
||
| 1939 | XMASSERT(pSource); |
||
| 1940 | // Grab the 64 bit structure |
||
| 1941 | __m128d vResultd = _mm_load_sd(reinterpret_cast<const double *>(&pSource->v)); |
||
| 1942 | // By shifting down 8 bits, y and z are in seperate 32 bit elements |
||
| 1943 | __m128i vResulti = _mm_srli_si128(reinterpret_cast<const __m128i *>(&vResultd)[0],8/8); |
||
| 1944 | // vResultd has x and w, vResulti has y and z, merge into one as x,w,y,z |
||
| 1945 | XMVECTOR vTemp = _mm_shuffle_ps(reinterpret_cast<const __m128 *>(&vResultd)[0],reinterpret_cast<const __m128 *>(&vResulti)[0],_MM_SHUFFLE(1,0,1,0)); |
||
| 1946 | // Fix the entries to x,y,z,w |
||
| 1947 | vTemp = _mm_shuffle_ps(vTemp,vTemp,_MM_SHUFFLE(1,3,2,0)); |
||
| 1948 | // Mask x,y,z and w |
||
| 1949 | vTemp = _mm_and_ps(vTemp,g_XMMaskIco4); |
||
| 1950 | // x and z are unsigned! Flip the bits to convert the order to signed |
||
| 1951 | vTemp = _mm_xor_ps(vTemp,g_XMXorXIco4); |
||
| 1952 | // Convert to floating point numbers |
||
| 1953 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 1954 | // x and z - 0x80 to complete the conversion |
||
| 1955 | vTemp = _mm_add_ps(vTemp,g_XMAddXIco4); |
||
| 1956 | // Fix y and w because they are too large |
||
| 1957 | vTemp = _mm_mul_ps(vTemp,g_XMMulIco4); |
||
| 1958 | return vTemp; |
||
| 1959 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 1960 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 1961 | } |
||
| 1962 | |||
| 1963 | //------------------------------------------------------------------------------ |
||
| 1964 | |||
| 1965 | XMFINLINE XMVECTOR XMLoadUIcoN4 |
||
| 1966 | ( |
||
| 1967 | CONST XMUICON4* pSource |
||
| 1968 | ) |
||
| 1969 | { |
||
| 1970 | #if defined(_XM_NO_INTRINSICS_) |
||
| 1971 | |||
| 1972 | XMVECTOR V; |
||
| 1973 | |||
| 1974 | XMASSERT(pSource); |
||
| 1975 | |||
| 1976 | V.vector4_f32[0] = (FLOAT)(pSource->v & 0xFFFFF) / 1048575.0f; |
||
| 1977 | V.vector4_f32[1] = (FLOAT)((pSource->v >> 20) & 0xFFFFF) / 1048575.0f; |
||
| 1978 | V.vector4_f32[2] = (FLOAT)((pSource->v >> 40) & 0xFFFFF) / 1048575.0f; |
||
| 1979 | V.vector4_f32[3] = (FLOAT)(pSource->v >> 60) / 15.0f; |
||
| 1980 | |||
| 1981 | return V; |
||
| 1982 | |||
| 1983 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 1984 | static const XMVECTORF32 LoadUIcoN4Mul = {1.0f/1048575.0f,1.0f/(1048575.0f*4096.0f),1.0f/1048575.0f,1.0f/(15.0f*4096.0f*65536.0f)}; |
||
| 1985 | XMASSERT(pSource); |
||
| 1986 | // Grab the 64 bit structure |
||
| 1987 | __m128d vResultd = _mm_load_sd(reinterpret_cast<const double *>(&pSource->v)); |
||
| 1988 | // By shifting down 8 bits, y and z are in seperate 32 bit elements |
||
| 1989 | __m128i vResulti = _mm_srli_si128(reinterpret_cast<const __m128i *>(&vResultd)[0],8/8); |
||
| 1990 | // vResultd has x and w, vResulti has y and z, merge into one as x,w,y,z |
||
| 1991 | XMVECTOR vTemp = _mm_shuffle_ps(reinterpret_cast<const __m128 *>(&vResultd)[0],reinterpret_cast<const __m128 *>(&vResulti)[0],_MM_SHUFFLE(1,0,1,0)); |
||
| 1992 | // Fix the entries to x,y,z,w |
||
| 1993 | vTemp = _mm_shuffle_ps(vTemp,vTemp,_MM_SHUFFLE(1,3,2,0)); |
||
| 1994 | // Mask x,y,z and w |
||
| 1995 | vTemp = _mm_and_ps(vTemp,g_XMMaskIco4); |
||
| 1996 | // x and z are unsigned! Flip the bits to convert the order to signed |
||
| 1997 | vTemp = _mm_xor_ps(vTemp,g_XMFlipYW); |
||
| 1998 | // Convert to floating point numbers |
||
| 1999 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 2000 | // x and z - 0x80 to complete the conversion |
||
| 2001 | vTemp = _mm_add_ps(vTemp,g_XMAddUIco4); |
||
| 2002 | // Fix y and w because they are too large |
||
| 2003 | vTemp = _mm_mul_ps(vTemp,LoadUIcoN4Mul); |
||
| 2004 | return vTemp; |
||
| 2005 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 2006 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 2007 | } |
||
| 2008 | |||
| 2009 | //------------------------------------------------------------------------------ |
||
| 2010 | |||
| 2011 | XMFINLINE XMVECTOR XMLoadUIco4 |
||
| 2012 | ( |
||
| 2013 | CONST XMUICO4* pSource |
||
| 2014 | ) |
||
| 2015 | { |
||
| 2016 | #if defined(_XM_NO_INTRINSICS_) |
||
| 2017 | |||
| 2018 | XMVECTOR V; |
||
| 2019 | |||
| 2020 | XMASSERT(pSource); |
||
| 2021 | |||
| 2022 | V.vector4_f32[0] = (FLOAT)(pSource->v & 0xFFFFF); |
||
| 2023 | V.vector4_f32[1] = (FLOAT)((pSource->v >> 20) & 0xFFFFF); |
||
| 2024 | V.vector4_f32[2] = (FLOAT)((pSource->v >> 40) & 0xFFFFF); |
||
| 2025 | V.vector4_f32[3] = (FLOAT)(pSource->v >> 60); |
||
| 2026 | |||
| 2027 | return V; |
||
| 2028 | |||
| 2029 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 2030 | XMASSERT(pSource); |
||
| 2031 | // Grab the 64 bit structure |
||
| 2032 | __m128d vResultd = _mm_load_sd(reinterpret_cast<const double *>(&pSource->v)); |
||
| 2033 | // By shifting down 8 bits, y and z are in seperate 32 bit elements |
||
| 2034 | __m128i vResulti = _mm_srli_si128(reinterpret_cast<const __m128i *>(&vResultd)[0],8/8); |
||
| 2035 | // vResultd has x and w, vResulti has y and z, merge into one as x,w,y,z |
||
| 2036 | XMVECTOR vTemp = _mm_shuffle_ps(reinterpret_cast<const __m128 *>(&vResultd)[0],reinterpret_cast<const __m128 *>(&vResulti)[0],_MM_SHUFFLE(1,0,1,0)); |
||
| 2037 | // Fix the entries to x,y,z,w |
||
| 2038 | vTemp = _mm_shuffle_ps(vTemp,vTemp,_MM_SHUFFLE(1,3,2,0)); |
||
| 2039 | // Mask x,y,z and w |
||
| 2040 | vTemp = _mm_and_ps(vTemp,g_XMMaskIco4); |
||
| 2041 | // x and z are unsigned! Flip the bits to convert the order to signed |
||
| 2042 | vTemp = _mm_xor_ps(vTemp,g_XMFlipYW); |
||
| 2043 | // Convert to floating point numbers |
||
| 2044 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 2045 | // x and z - 0x80 to complete the conversion |
||
| 2046 | vTemp = _mm_add_ps(vTemp,g_XMAddUIco4); |
||
| 2047 | // Fix y and w because they are too large |
||
| 2048 | vTemp = _mm_mul_ps(vTemp,g_XMMulIco4); |
||
| 2049 | return vTemp; |
||
| 2050 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 2051 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 2052 | } |
||
| 2053 | |||
| 2054 | //------------------------------------------------------------------------------ |
||
| 2055 | |||
| 2056 | XMFINLINE XMVECTOR XMLoadIcoN4 |
||
| 2057 | ( |
||
| 2058 | CONST XMICON4* pSource |
||
| 2059 | ) |
||
| 2060 | { |
||
| 2061 | #if defined(_XM_NO_INTRINSICS_) |
||
| 2062 | |||
| 2063 | XMVECTOR V; |
||
| 2064 | UINT Element; |
||
| 2065 | static CONST UINT SignExtend[] = {0x00000000, 0xFFF00000}; |
||
| 2066 | static CONST UINT SignExtendW[] = {0x00000000, 0xFFFFFFF0}; |
||
| 2067 | |||
| 2068 | XMASSERT(pSource); |
||
| 2069 | |||
| 2070 | Element = (UINT)pSource->v & 0xFFFFF; |
||
| 2071 | V.vector4_f32[0] = (FLOAT)(INT)(Element | SignExtend[Element >> 19]) / 524287.0f; |
||
| 2072 | Element = (UINT)(pSource->v >> 20) & 0xFFFFF; |
||
| 2073 | V.vector4_f32[1] = (FLOAT)(INT)(Element | SignExtend[Element >> 19]) / 524287.0f; |
||
| 2074 | Element = (UINT)(pSource->v >> 40) & 0xFFFFF; |
||
| 2075 | V.vector4_f32[2] = (FLOAT)(INT)(Element | SignExtend[Element >> 19]) / 524287.0f; |
||
| 2076 | Element = (UINT)(pSource->v >> 60); |
||
| 2077 | V.vector4_f32[3] = (FLOAT)(INT)(Element | SignExtendW[Element >> 3]) / 7.0f; |
||
| 2078 | |||
| 2079 | return V; |
||
| 2080 | |||
| 2081 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 2082 | static const XMVECTORF32 LoadIcoN4Mul = {1.0f/524287.0f,1.0f/(524287.0f*4096.0f),1.0f/524287.0f,1.0f/(7.0f*4096.0f*65536.0f)}; |
||
| 2083 | XMASSERT(pSource); |
||
| 2084 | // Grab the 64 bit structure |
||
| 2085 | __m128d vResultd = _mm_load_sd(reinterpret_cast<const double *>(&pSource->v)); |
||
| 2086 | // By shifting down 8 bits, y and z are in seperate 32 bit elements |
||
| 2087 | __m128i vResulti = _mm_srli_si128(reinterpret_cast<const __m128i *>(&vResultd)[0],8/8); |
||
| 2088 | // vResultd has x and w, vResulti has y and z, merge into one as x,w,y,z |
||
| 2089 | XMVECTOR vTemp = _mm_shuffle_ps(reinterpret_cast<const __m128 *>(&vResultd)[0],reinterpret_cast<const __m128 *>(&vResulti)[0],_MM_SHUFFLE(1,0,1,0)); |
||
| 2090 | // Fix the entries to x,y,z,w |
||
| 2091 | vTemp = _mm_shuffle_ps(vTemp,vTemp,_MM_SHUFFLE(1,3,2,0)); |
||
| 2092 | // Mask x,y,z and w |
||
| 2093 | vTemp = _mm_and_ps(vTemp,g_XMMaskIco4); |
||
| 2094 | // x and z are unsigned! Flip the bits to convert the order to signed |
||
| 2095 | vTemp = _mm_xor_ps(vTemp,g_XMXorIco4); |
||
| 2096 | // Convert to floating point numbers |
||
| 2097 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 2098 | // x and z - 0x80 to complete the conversion |
||
| 2099 | vTemp = _mm_add_ps(vTemp,g_XMAddIco4); |
||
| 2100 | // Fix y and w because they are too large |
||
| 2101 | vTemp = _mm_mul_ps(vTemp,LoadIcoN4Mul); |
||
| 2102 | return vTemp; |
||
| 2103 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 2104 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 2105 | } |
||
| 2106 | |||
| 2107 | //------------------------------------------------------------------------------ |
||
| 2108 | |||
| 2109 | XMFINLINE XMVECTOR XMLoadIco4 |
||
| 2110 | ( |
||
| 2111 | CONST XMICO4* pSource |
||
| 2112 | ) |
||
| 2113 | { |
||
| 2114 | #if defined(_XM_NO_INTRINSICS_) |
||
| 2115 | |||
| 2116 | XMVECTOR V; |
||
| 2117 | UINT Element; |
||
| 2118 | static CONST UINT SignExtend[] = {0x00000000, 0xFFF00000}; |
||
| 2119 | static CONST UINT SignExtendW[] = {0x00000000, 0xFFFFFFF0}; |
||
| 2120 | |||
| 2121 | XMASSERT(pSource); |
||
| 2122 | |||
| 2123 | Element = (UINT)pSource->v & 0xFFFFF; |
||
| 2124 | V.vector4_f32[0] = (FLOAT)(INT)(Element | SignExtend[Element >> 19]); |
||
| 2125 | Element = (UINT)(pSource->v >> 20) & 0xFFFFF; |
||
| 2126 | V.vector4_f32[1] = (FLOAT)(INT)(Element | SignExtend[Element >> 19]); |
||
| 2127 | Element = (UINT)(pSource->v >> 40) & 0xFFFFF; |
||
| 2128 | V.vector4_f32[2] = (FLOAT)(INT)(Element | SignExtend[Element >> 19]); |
||
| 2129 | Element = (UINT)(pSource->v >> 60); |
||
| 2130 | V.vector4_f32[3] = (FLOAT)(INT)(Element | SignExtendW[Element >> 3]); |
||
| 2131 | |||
| 2132 | return V; |
||
| 2133 | |||
| 2134 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 2135 | XMASSERT(pSource); |
||
| 2136 | // Grab the 64 bit structure |
||
| 2137 | __m128d vResultd = _mm_load_sd(reinterpret_cast<const double *>(&pSource->v)); |
||
| 2138 | // By shifting down 8 bits, y and z are in seperate 32 bit elements |
||
| 2139 | __m128i vResulti = _mm_srli_si128(reinterpret_cast<const __m128i *>(&vResultd)[0],8/8); |
||
| 2140 | // vResultd has x and w, vResulti has y and z, merge into one as x,w,y,z |
||
| 2141 | XMVECTOR vTemp = _mm_shuffle_ps(reinterpret_cast<const __m128 *>(&vResultd)[0],reinterpret_cast<const __m128 *>(&vResulti)[0],_MM_SHUFFLE(1,0,1,0)); |
||
| 2142 | // Fix the entries to x,y,z,w |
||
| 2143 | vTemp = _mm_shuffle_ps(vTemp,vTemp,_MM_SHUFFLE(1,3,2,0)); |
||
| 2144 | // Mask x,y,z and w |
||
| 2145 | vTemp = _mm_and_ps(vTemp,g_XMMaskIco4); |
||
| 2146 | // x and z are unsigned! Flip the bits to convert the order to signed |
||
| 2147 | vTemp = _mm_xor_ps(vTemp,g_XMXorIco4); |
||
| 2148 | // Convert to floating point numbers |
||
| 2149 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 2150 | // x and z - 0x80 to complete the conversion |
||
| 2151 | vTemp = _mm_add_ps(vTemp,g_XMAddIco4); |
||
| 2152 | // Fix y and w because they are too large |
||
| 2153 | vTemp = _mm_mul_ps(vTemp,g_XMMulIco4); |
||
| 2154 | return vTemp; |
||
| 2155 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 2156 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 2157 | } |
||
| 2158 | |||
| 2159 | |||
| 2160 | //------------------------------------------------------------------------------ |
||
| 2161 | |||
| 2162 | XMFINLINE XMVECTOR XMLoadXDecN4 |
||
| 2163 | ( |
||
| 2164 | CONST XMXDECN4* pSource |
||
| 2165 | ) |
||
| 2166 | { |
||
| 2167 | #if defined(_XM_NO_INTRINSICS_) |
||
| 2168 | XMVECTOR V; |
||
| 2169 | UINT Element; |
||
| 2170 | static CONST UINT SignExtend[] = {0x00000000, 0xFFFFFC00}; |
||
| 2171 | |||
| 2172 | XMASSERT(pSource); |
||
| 2173 | XMASSERT((pSource->v & 0x3FF) != 0x200); |
||
| 2174 | XMASSERT(((pSource->v >> 10) & 0x3FF) != 0x200); |
||
| 2175 | XMASSERT(((pSource->v >> 20) & 0x3FF) != 0x200); |
||
| 2176 | |||
| 2177 | Element = pSource->v & 0x3FF; |
||
| 2178 | V.vector4_f32[0] = (FLOAT)(SHORT)(Element | SignExtend[Element >> 9]) / 511.0f; |
||
| 2179 | Element = (pSource->v >> 10) & 0x3FF; |
||
| 2180 | V.vector4_f32[1] = (FLOAT)(SHORT)(Element | SignExtend[Element >> 9]) / 511.0f; |
||
| 2181 | Element = (pSource->v >> 20) & 0x3FF; |
||
| 2182 | V.vector4_f32[2] = (FLOAT)(SHORT)(Element | SignExtend[Element >> 9]) / 511.0f; |
||
| 2183 | V.vector4_f32[3] = (FLOAT)(pSource->v >> 30) / 3.0f; |
||
| 2184 | |||
| 2185 | return V; |
||
| 2186 | |||
| 2187 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 2188 | XMASSERT(pSource); |
||
| 2189 | // Splat the color in all four entries |
||
| 2190 | __m128 vTemp = _mm_load_ps1(reinterpret_cast<const float *>(&pSource->v)); |
||
| 2191 | // Shift R&0xFF0000, G&0xFF00, B&0xFF, A&0xFF000000 |
||
| 2192 | vTemp = _mm_and_ps(vTemp,g_XMMaskA2B10G10R10); |
||
| 2193 | // a is unsigned! Flip the bit to convert the order to signed |
||
| 2194 | vTemp = _mm_xor_ps(vTemp,g_XMFlipA2B10G10R10); |
||
| 2195 | // Convert to floating point numbers |
||
| 2196 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 2197 | // RGB + 0, A + 0x80000000.f to undo the signed order. |
||
| 2198 | vTemp = _mm_add_ps(vTemp,g_XMFixAA2B10G10R10); |
||
| 2199 | // Convert 0-255 to 0.0f-1.0f |
||
| 2200 | return _mm_mul_ps(vTemp,g_XMNormalizeA2B10G10R10); |
||
| 2201 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 2202 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 2203 | } |
||
| 2204 | |||
| 2205 | //------------------------------------------------------------------------------ |
||
| 2206 | |||
| 2207 | XMFINLINE XMVECTOR XMLoadXDec4 |
||
| 2208 | ( |
||
| 2209 | CONST XMXDEC4* pSource |
||
| 2210 | ) |
||
| 2211 | { |
||
| 2212 | #if defined(_XM_NO_INTRINSICS_) |
||
| 2213 | |||
| 2214 | XMVECTOR V; |
||
| 2215 | UINT Element; |
||
| 2216 | static CONST UINT SignExtend[] = {0x00000000, 0xFFFFFC00}; |
||
| 2217 | |||
| 2218 | XMASSERT(pSource); |
||
| 2219 | XMASSERT((pSource->v & 0x3FF) != 0x200); |
||
| 2220 | XMASSERT(((pSource->v >> 10) & 0x3FF) != 0x200); |
||
| 2221 | XMASSERT(((pSource->v >> 20) & 0x3FF) != 0x200); |
||
| 2222 | |||
| 2223 | Element = pSource->v & 0x3FF; |
||
| 2224 | V.vector4_f32[0] = (FLOAT)(SHORT)(Element | SignExtend[Element >> 9]); |
||
| 2225 | Element = (pSource->v >> 10) & 0x3FF; |
||
| 2226 | V.vector4_f32[1] = (FLOAT)(SHORT)(Element | SignExtend[Element >> 9]); |
||
| 2227 | Element = (pSource->v >> 20) & 0x3FF; |
||
| 2228 | V.vector4_f32[2] = (FLOAT)(SHORT)(Element | SignExtend[Element >> 9]); |
||
| 2229 | V.vector4_f32[3] = (FLOAT)(pSource->v >> 30); |
||
| 2230 | |||
| 2231 | return V; |
||
| 2232 | |||
| 2233 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 2234 | XMASSERT((pSource->v & 0x3FF) != 0x200); |
||
| 2235 | XMASSERT(((pSource->v >> 10) & 0x3FF) != 0x200); |
||
| 2236 | XMASSERT(((pSource->v >> 20) & 0x3FF) != 0x200); |
||
| 2237 | static const XMVECTORI32 XDec4Xor = {0x200, 0x200<<10, 0x200<<20, 0x80000000}; |
||
| 2238 | static const XMVECTORF32 XDec4Add = {-512.0f,-512.0f*1024.0f,-512.0f*1024.0f*1024.0f,32768*65536.0f}; |
||
| 2239 | XMASSERT(pSource); |
||
| 2240 | // Splat the color in all four entries |
||
| 2241 | XMVECTOR vTemp = _mm_load_ps1(reinterpret_cast<const float *>(&pSource->v)); |
||
| 2242 | // Shift R&0xFF0000, G&0xFF00, B&0xFF, A&0xFF000000 |
||
| 2243 | vTemp = _mm_and_ps(vTemp,g_XMMaskDec4); |
||
| 2244 | // a is unsigned! Flip the bit to convert the order to signed |
||
| 2245 | vTemp = _mm_xor_ps(vTemp,XDec4Xor); |
||
| 2246 | // Convert to floating point numbers |
||
| 2247 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 2248 | // RGB + 0, A + 0x80000000.f to undo the signed order. |
||
| 2249 | vTemp = _mm_add_ps(vTemp,XDec4Add); |
||
| 2250 | // Convert 0-255 to 0.0f-1.0f |
||
| 2251 | vTemp = _mm_mul_ps(vTemp,g_XMMulDec4); |
||
| 2252 | return vTemp; |
||
| 2253 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 2254 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 2255 | } |
||
| 2256 | |||
| 2257 | //------------------------------------------------------------------------------ |
||
| 2258 | |||
| 2259 | XMFINLINE XMVECTOR XMLoadUDecN4 |
||
| 2260 | ( |
||
| 2261 | CONST XMUDECN4* pSource |
||
| 2262 | ) |
||
| 2263 | { |
||
| 2264 | #if defined(_XM_NO_INTRINSICS_) |
||
| 2265 | |||
| 2266 | XMVECTOR V; |
||
| 2267 | UINT Element; |
||
| 2268 | |||
| 2269 | XMASSERT(pSource); |
||
| 2270 | |||
| 2271 | Element = pSource->v & 0x3FF; |
||
| 2272 | V.vector4_f32[0] = (FLOAT)Element / 1023.0f; |
||
| 2273 | Element = (pSource->v >> 10) & 0x3FF; |
||
| 2274 | V.vector4_f32[1] = (FLOAT)Element / 1023.0f; |
||
| 2275 | Element = (pSource->v >> 20) & 0x3FF; |
||
| 2276 | V.vector4_f32[2] = (FLOAT)Element / 1023.0f; |
||
| 2277 | V.vector4_f32[3] = (FLOAT)(pSource->v >> 30) / 3.0f; |
||
| 2278 | |||
| 2279 | return V; |
||
| 2280 | |||
| 2281 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 2282 | XMASSERT(pSource); |
||
| 2283 | static const XMVECTORF32 UDecN4Mul = {1.0f/1023.0f,1.0f/(1023.0f*1024.0f),1.0f/(1023.0f*1024.0f*1024.0f),1.0f/(3.0f*1024.0f*1024.0f*1024.0f)}; |
||
| 2284 | // Splat the color in all four entries |
||
| 2285 | XMVECTOR vTemp = _mm_load_ps1(reinterpret_cast<const float *>(&pSource->v)); |
||
| 2286 | // Shift R&0xFF0000, G&0xFF00, B&0xFF, A&0xFF000000 |
||
| 2287 | vTemp = _mm_and_ps(vTemp,g_XMMaskDec4); |
||
| 2288 | // a is unsigned! Flip the bit to convert the order to signed |
||
| 2289 | vTemp = _mm_xor_ps(vTemp,g_XMFlipW); |
||
| 2290 | // Convert to floating point numbers |
||
| 2291 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 2292 | // RGB + 0, A + 0x80000000.f to undo the signed order. |
||
| 2293 | vTemp = _mm_add_ps(vTemp,g_XMAddUDec4); |
||
| 2294 | // Convert 0-255 to 0.0f-1.0f |
||
| 2295 | vTemp = _mm_mul_ps(vTemp,UDecN4Mul); |
||
| 2296 | return vTemp; |
||
| 2297 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 2298 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 2299 | } |
||
| 2300 | |||
| 2301 | //------------------------------------------------------------------------------ |
||
| 2302 | |||
| 2303 | XMFINLINE XMVECTOR XMLoadUDec4 |
||
| 2304 | ( |
||
| 2305 | CONST XMUDEC4* pSource |
||
| 2306 | ) |
||
| 2307 | { |
||
| 2308 | #if defined(_XM_NO_INTRINSICS_) |
||
| 2309 | |||
| 2310 | XMVECTOR V; |
||
| 2311 | UINT Element; |
||
| 2312 | |||
| 2313 | XMASSERT(pSource); |
||
| 2314 | |||
| 2315 | Element = pSource->v & 0x3FF; |
||
| 2316 | V.vector4_f32[0] = (FLOAT)Element; |
||
| 2317 | Element = (pSource->v >> 10) & 0x3FF; |
||
| 2318 | V.vector4_f32[1] = (FLOAT)Element; |
||
| 2319 | Element = (pSource->v >> 20) & 0x3FF; |
||
| 2320 | V.vector4_f32[2] = (FLOAT)Element; |
||
| 2321 | V.vector4_f32[3] = (FLOAT)(pSource->v >> 30); |
||
| 2322 | |||
| 2323 | return V; |
||
| 2324 | |||
| 2325 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 2326 | XMASSERT(pSource); |
||
| 2327 | // Splat the color in all four entries |
||
| 2328 | XMVECTOR vTemp = _mm_load_ps1(reinterpret_cast<const float *>(&pSource->v)); |
||
| 2329 | // Shift R&0xFF0000, G&0xFF00, B&0xFF, A&0xFF000000 |
||
| 2330 | vTemp = _mm_and_ps(vTemp,g_XMMaskDec4); |
||
| 2331 | // a is unsigned! Flip the bit to convert the order to signed |
||
| 2332 | vTemp = _mm_xor_ps(vTemp,g_XMFlipW); |
||
| 2333 | // Convert to floating point numbers |
||
| 2334 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 2335 | // RGB + 0, A + 0x80000000.f to undo the signed order. |
||
| 2336 | vTemp = _mm_add_ps(vTemp,g_XMAddUDec4); |
||
| 2337 | // Convert 0-255 to 0.0f-1.0f |
||
| 2338 | vTemp = _mm_mul_ps(vTemp,g_XMMulDec4); |
||
| 2339 | return vTemp; |
||
| 2340 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 2341 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 2342 | } |
||
| 2343 | |||
| 2344 | //------------------------------------------------------------------------------ |
||
| 2345 | |||
| 2346 | XMFINLINE XMVECTOR XMLoadDecN4 |
||
| 2347 | ( |
||
| 2348 | CONST XMDECN4* pSource |
||
| 2349 | ) |
||
| 2350 | { |
||
| 2351 | #if defined(_XM_NO_INTRINSICS_) |
||
| 2352 | |||
| 2353 | XMVECTOR V; |
||
| 2354 | UINT Element; |
||
| 2355 | static CONST UINT SignExtend[] = {0x00000000, 0xFFFFFC00}; |
||
| 2356 | static CONST UINT SignExtendW[] = {0x00000000, 0xFFFFFFFC}; |
||
| 2357 | |||
| 2358 | XMASSERT(pSource); |
||
| 2359 | XMASSERT((pSource->v & 0x3FF) != 0x200); |
||
| 2360 | XMASSERT(((pSource->v >> 10) & 0x3FF) != 0x200); |
||
| 2361 | XMASSERT(((pSource->v >> 20) & 0x3FF) != 0x200); |
||
| 2362 | XMASSERT(((pSource->v >> 30) & 0x3) != 0x2); |
||
| 2363 | |||
| 2364 | Element = pSource->v & 0x3FF; |
||
| 2365 | V.vector4_f32[0] = (FLOAT)(SHORT)(Element | SignExtend[Element >> 9]) / 511.0f; |
||
| 2366 | Element = (pSource->v >> 10) & 0x3FF; |
||
| 2367 | V.vector4_f32[1] = (FLOAT)(SHORT)(Element | SignExtend[Element >> 9]) / 511.0f; |
||
| 2368 | Element = (pSource->v >> 20) & 0x3FF; |
||
| 2369 | V.vector4_f32[2] = (FLOAT)(SHORT)(Element | SignExtend[Element >> 9]) / 511.0f; |
||
| 2370 | Element = pSource->v >> 30; |
||
| 2371 | V.vector4_f32[3] = (FLOAT)(SHORT)(Element | SignExtendW[Element >> 1]); |
||
| 2372 | |||
| 2373 | return V; |
||
| 2374 | |||
| 2375 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 2376 | XMASSERT(pSource); |
||
| 2377 | XMASSERT((pSource->v & 0x3FF) != 0x200); |
||
| 2378 | XMASSERT(((pSource->v >> 10) & 0x3FF) != 0x200); |
||
| 2379 | XMASSERT(((pSource->v >> 20) & 0x3FF) != 0x200); |
||
| 2380 | XMASSERT(((pSource->v >> 30) & 0x3) != 0x2); |
||
| 2381 | static const XMVECTORF32 DecN4Mul = {1.0f/511.0f,1.0f/(511.0f*1024.0f),1.0f/(511.0f*1024.0f*1024.0f),1.0f/(1024.0f*1024.0f*1024.0f)}; |
||
| 2382 | // Splat the color in all four entries |
||
| 2383 | XMVECTOR vTemp = _mm_load_ps1(reinterpret_cast<const float *>(&pSource->v)); |
||
| 2384 | // Shift R&0xFF0000, G&0xFF00, B&0xFF, A&0xFF000000 |
||
| 2385 | vTemp = _mm_and_ps(vTemp,g_XMMaskDec4); |
||
| 2386 | // a is unsigned! Flip the bit to convert the order to signed |
||
| 2387 | vTemp = _mm_xor_ps(vTemp,g_XMXorDec4); |
||
| 2388 | // Convert to floating point numbers |
||
| 2389 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 2390 | // RGB + 0, A + 0x80000000.f to undo the signed order. |
||
| 2391 | vTemp = _mm_add_ps(vTemp,g_XMAddDec4); |
||
| 2392 | // Convert 0-255 to 0.0f-1.0f |
||
| 2393 | vTemp = _mm_mul_ps(vTemp,DecN4Mul); |
||
| 2394 | return vTemp; |
||
| 2395 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 2396 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 2397 | } |
||
| 2398 | |||
| 2399 | //------------------------------------------------------------------------------ |
||
| 2400 | |||
| 2401 | XMFINLINE XMVECTOR XMLoadDec4 |
||
| 2402 | ( |
||
| 2403 | CONST XMDEC4* pSource |
||
| 2404 | ) |
||
| 2405 | { |
||
| 2406 | #if defined(_XM_NO_INTRINSICS_) |
||
| 2407 | |||
| 2408 | XMVECTOR V; |
||
| 2409 | UINT Element; |
||
| 2410 | static CONST UINT SignExtend[] = {0x00000000, 0xFFFFFC00}; |
||
| 2411 | static CONST UINT SignExtendW[] = {0x00000000, 0xFFFFFFFC}; |
||
| 2412 | |||
| 2413 | XMASSERT(pSource); |
||
| 2414 | XMASSERT((pSource->v & 0x3FF) != 0x200); |
||
| 2415 | XMASSERT(((pSource->v >> 10) & 0x3FF) != 0x200); |
||
| 2416 | XMASSERT(((pSource->v >> 20) & 0x3FF) != 0x200); |
||
| 2417 | XMASSERT(((pSource->v >> 30) & 0x3) != 0x2); |
||
| 2418 | |||
| 2419 | Element = pSource->v & 0x3FF; |
||
| 2420 | V.vector4_f32[0] = (FLOAT)(SHORT)(Element | SignExtend[Element >> 9]); |
||
| 2421 | Element = (pSource->v >> 10) & 0x3FF; |
||
| 2422 | V.vector4_f32[1] = (FLOAT)(SHORT)(Element | SignExtend[Element >> 9]); |
||
| 2423 | Element = (pSource->v >> 20) & 0x3FF; |
||
| 2424 | V.vector4_f32[2] = (FLOAT)(SHORT)(Element | SignExtend[Element >> 9]); |
||
| 2425 | Element = pSource->v >> 30; |
||
| 2426 | V.vector4_f32[3] = (FLOAT)(SHORT)(Element | SignExtendW[Element >> 1]); |
||
| 2427 | |||
| 2428 | return V; |
||
| 2429 | |||
| 2430 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 2431 | XMASSERT((pSource->v & 0x3FF) != 0x200); |
||
| 2432 | XMASSERT(((pSource->v >> 10) & 0x3FF) != 0x200); |
||
| 2433 | XMASSERT(((pSource->v >> 20) & 0x3FF) != 0x200); |
||
| 2434 | XMASSERT(((pSource->v >> 30) & 0x3) != 0x2); |
||
| 2435 | XMASSERT(pSource); |
||
| 2436 | // Splat the color in all four entries |
||
| 2437 | XMVECTOR vTemp = _mm_load_ps1(reinterpret_cast<const float *>(&pSource->v)); |
||
| 2438 | // Shift R&0xFF0000, G&0xFF00, B&0xFF, A&0xFF000000 |
||
| 2439 | vTemp = _mm_and_ps(vTemp,g_XMMaskDec4); |
||
| 2440 | // a is unsigned! Flip the bit to convert the order to signed |
||
| 2441 | vTemp = _mm_xor_ps(vTemp,g_XMXorDec4); |
||
| 2442 | // Convert to floating point numbers |
||
| 2443 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 2444 | // RGB + 0, A + 0x80000000.f to undo the signed order. |
||
| 2445 | vTemp = _mm_add_ps(vTemp,g_XMAddDec4); |
||
| 2446 | // Convert 0-255 to 0.0f-1.0f |
||
| 2447 | vTemp = _mm_mul_ps(vTemp,g_XMMulDec4); |
||
| 2448 | return vTemp; |
||
| 2449 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 2450 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 2451 | } |
||
| 2452 | |||
| 2453 | //------------------------------------------------------------------------------ |
||
| 2454 | |||
| 2455 | XMFINLINE XMVECTOR XMLoadUByteN4 |
||
| 2456 | ( |
||
| 2457 | CONST XMUBYTEN4* pSource |
||
| 2458 | ) |
||
| 2459 | { |
||
| 2460 | #if defined(_XM_NO_INTRINSICS_) |
||
| 2461 | |||
| 2462 | XMVECTOR V; |
||
| 2463 | |||
| 2464 | XMASSERT(pSource); |
||
| 2465 | |||
| 2466 | V.vector4_f32[0] = (FLOAT)pSource->x / 255.0f; |
||
| 2467 | V.vector4_f32[1] = (FLOAT)pSource->y / 255.0f; |
||
| 2468 | V.vector4_f32[2] = (FLOAT)pSource->z / 255.0f; |
||
| 2469 | V.vector4_f32[3] = (FLOAT)pSource->w / 255.0f; |
||
| 2470 | |||
| 2471 | return V; |
||
| 2472 | |||
| 2473 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 2474 | static const XMVECTORF32 LoadUByteN4Mul = {1.0f/255.0f,1.0f/(255.0f*256.0f),1.0f/(255.0f*65536.0f),1.0f/(255.0f*65536.0f*256.0f)}; |
||
| 2475 | XMASSERT(pSource); |
||
| 2476 | // Splat the color in all four entries (x,z,y,w) |
||
| 2477 | XMVECTOR vTemp = _mm_load1_ps(reinterpret_cast<const float *>(&pSource->x)); |
||
| 2478 | // Mask x&0ff,y&0xff00,z&0xff0000,w&0xff000000 |
||
| 2479 | vTemp = _mm_and_ps(vTemp,g_XMMaskByte4); |
||
| 2480 | // w is signed! Flip the bits to convert the order to unsigned |
||
| 2481 | vTemp = _mm_xor_ps(vTemp,g_XMFlipW); |
||
| 2482 | // Convert to floating point numbers |
||
| 2483 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 2484 | // w + 0x80 to complete the conversion |
||
| 2485 | vTemp = _mm_add_ps(vTemp,g_XMAddUDec4); |
||
| 2486 | // Fix y, z and w because they are too large |
||
| 2487 | vTemp = _mm_mul_ps(vTemp,LoadUByteN4Mul); |
||
| 2488 | return vTemp; |
||
| 2489 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 2490 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 2491 | } |
||
| 2492 | |||
| 2493 | //------------------------------------------------------------------------------ |
||
| 2494 | |||
| 2495 | XMFINLINE XMVECTOR XMLoadUByte4 |
||
| 2496 | ( |
||
| 2497 | CONST XMUBYTE4* pSource |
||
| 2498 | ) |
||
| 2499 | { |
||
| 2500 | #if defined(_XM_NO_INTRINSICS_) |
||
| 2501 | |||
| 2502 | XMVECTOR V; |
||
| 2503 | |||
| 2504 | XMASSERT(pSource); |
||
| 2505 | |||
| 2506 | V.vector4_f32[0] = (FLOAT)pSource->x; |
||
| 2507 | V.vector4_f32[1] = (FLOAT)pSource->y; |
||
| 2508 | V.vector4_f32[2] = (FLOAT)pSource->z; |
||
| 2509 | V.vector4_f32[3] = (FLOAT)pSource->w; |
||
| 2510 | |||
| 2511 | return V; |
||
| 2512 | |||
| 2513 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 2514 | static const XMVECTORF32 LoadUByte4Mul = {1.0f,1.0f/256.0f,1.0f/65536.0f,1.0f/(65536.0f*256.0f)}; |
||
| 2515 | XMASSERT(pSource); |
||
| 2516 | // Splat the color in all four entries (x,z,y,w) |
||
| 2517 | XMVECTOR vTemp = _mm_load1_ps(reinterpret_cast<const float *>(&pSource->x)); |
||
| 2518 | // Mask x&0ff,y&0xff00,z&0xff0000,w&0xff000000 |
||
| 2519 | vTemp = _mm_and_ps(vTemp,g_XMMaskByte4); |
||
| 2520 | // w is signed! Flip the bits to convert the order to unsigned |
||
| 2521 | vTemp = _mm_xor_ps(vTemp,g_XMFlipW); |
||
| 2522 | // Convert to floating point numbers |
||
| 2523 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 2524 | // w + 0x80 to complete the conversion |
||
| 2525 | vTemp = _mm_add_ps(vTemp,g_XMAddUDec4); |
||
| 2526 | // Fix y, z and w because they are too large |
||
| 2527 | vTemp = _mm_mul_ps(vTemp,LoadUByte4Mul); |
||
| 2528 | return vTemp; |
||
| 2529 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 2530 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 2531 | } |
||
| 2532 | |||
| 2533 | //------------------------------------------------------------------------------ |
||
| 2534 | |||
| 2535 | XMFINLINE XMVECTOR XMLoadByteN4 |
||
| 2536 | ( |
||
| 2537 | CONST XMBYTEN4* pSource |
||
| 2538 | ) |
||
| 2539 | { |
||
| 2540 | #if defined(_XM_NO_INTRINSICS_) |
||
| 2541 | |||
| 2542 | XMVECTOR V; |
||
| 2543 | |||
| 2544 | XMASSERT(pSource); |
||
| 2545 | XMASSERT(pSource->x != -128); |
||
| 2546 | XMASSERT(pSource->y != -128); |
||
| 2547 | XMASSERT(pSource->z != -128); |
||
| 2548 | XMASSERT(pSource->w != -128); |
||
| 2549 | |||
| 2550 | V.vector4_f32[0] = (FLOAT)pSource->x / 127.0f; |
||
| 2551 | V.vector4_f32[1] = (FLOAT)pSource->y / 127.0f; |
||
| 2552 | V.vector4_f32[2] = (FLOAT)pSource->z / 127.0f; |
||
| 2553 | V.vector4_f32[3] = (FLOAT)pSource->w / 127.0f; |
||
| 2554 | |||
| 2555 | return V; |
||
| 2556 | |||
| 2557 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 2558 | static const XMVECTORF32 LoadByteN4Mul = {1.0f/127.0f,1.0f/(127.0f*256.0f),1.0f/(127.0f*65536.0f),1.0f/(127.0f*65536.0f*256.0f)}; |
||
| 2559 | XMASSERT(pSource); |
||
| 2560 | XMASSERT(pSource->x != -128); |
||
| 2561 | XMASSERT(pSource->y != -128); |
||
| 2562 | XMASSERT(pSource->z != -128); |
||
| 2563 | XMASSERT(pSource->w != -128); |
||
| 2564 | // Splat the color in all four entries (x,z,y,w) |
||
| 2565 | XMVECTOR vTemp = _mm_load1_ps(reinterpret_cast<const float *>(&pSource->x)); |
||
| 2566 | // Mask x&0ff,y&0xff00,z&0xff0000,w&0xff000000 |
||
| 2567 | vTemp = _mm_and_ps(vTemp,g_XMMaskByte4); |
||
| 2568 | // x,y and z are unsigned! Flip the bits to convert the order to signed |
||
| 2569 | vTemp = _mm_xor_ps(vTemp,g_XMXorByte4); |
||
| 2570 | // Convert to floating point numbers |
||
| 2571 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 2572 | // x, y and z - 0x80 to complete the conversion |
||
| 2573 | vTemp = _mm_add_ps(vTemp,g_XMAddByte4); |
||
| 2574 | // Fix y, z and w because they are too large |
||
| 2575 | vTemp = _mm_mul_ps(vTemp,LoadByteN4Mul); |
||
| 2576 | return vTemp; |
||
| 2577 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 2578 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 2579 | } |
||
| 2580 | |||
| 2581 | //------------------------------------------------------------------------------ |
||
| 2582 | |||
| 2583 | XMFINLINE XMVECTOR XMLoadByte4 |
||
| 2584 | ( |
||
| 2585 | CONST XMBYTE4* pSource |
||
| 2586 | ) |
||
| 2587 | { |
||
| 2588 | #if defined(_XM_NO_INTRINSICS_) |
||
| 2589 | |||
| 2590 | XMVECTOR V; |
||
| 2591 | |||
| 2592 | XMASSERT(pSource); |
||
| 2593 | XMASSERT(pSource->x != -128); |
||
| 2594 | XMASSERT(pSource->y != -128); |
||
| 2595 | XMASSERT(pSource->z != -128); |
||
| 2596 | XMASSERT(pSource->w != -128); |
||
| 2597 | |||
| 2598 | V.vector4_f32[0] = (FLOAT)pSource->x; |
||
| 2599 | V.vector4_f32[1] = (FLOAT)pSource->y; |
||
| 2600 | V.vector4_f32[2] = (FLOAT)pSource->z; |
||
| 2601 | V.vector4_f32[3] = (FLOAT)pSource->w; |
||
| 2602 | |||
| 2603 | return V; |
||
| 2604 | |||
| 2605 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 2606 | static const XMVECTORF32 LoadByte4Mul = {1.0f,1.0f/256.0f,1.0f/65536.0f,1.0f/(65536.0f*256.0f)}; |
||
| 2607 | XMASSERT(pSource); |
||
| 2608 | XMASSERT(pSource->x != -128); |
||
| 2609 | XMASSERT(pSource->y != -128); |
||
| 2610 | XMASSERT(pSource->z != -128); |
||
| 2611 | XMASSERT(pSource->w != -128); |
||
| 2612 | // Splat the color in all four entries (x,z,y,w) |
||
| 2613 | XMVECTOR vTemp = _mm_load1_ps(reinterpret_cast<const float *>(&pSource->x)); |
||
| 2614 | // Mask x&0ff,y&0xff00,z&0xff0000,w&0xff000000 |
||
| 2615 | vTemp = _mm_and_ps(vTemp,g_XMMaskByte4); |
||
| 2616 | // x,y and z are unsigned! Flip the bits to convert the order to signed |
||
| 2617 | vTemp = _mm_xor_ps(vTemp,g_XMXorByte4); |
||
| 2618 | // Convert to floating point numbers |
||
| 2619 | vTemp = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vTemp)[0]); |
||
| 2620 | // x, y and z - 0x80 to complete the conversion |
||
| 2621 | vTemp = _mm_add_ps(vTemp,g_XMAddByte4); |
||
| 2622 | // Fix y, z and w because they are too large |
||
| 2623 | vTemp = _mm_mul_ps(vTemp,LoadByte4Mul); |
||
| 2624 | return vTemp; |
||
| 2625 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 2626 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 2627 | } |
||
| 2628 | |||
| 2629 | //------------------------------------------------------------------------------ |
||
| 2630 | |||
| 2631 | XMFINLINE XMVECTOR XMLoadUNibble4 |
||
| 2632 | ( |
||
| 2633 | CONST XMUNIBBLE4* pSource |
||
| 2634 | ) |
||
| 2635 | { |
||
| 2636 | #if defined(_XM_SSE_INTRINSICS_) && !defined(_XM_NO_INTRINSICS_) |
||
| 2637 | static const XMVECTORI32 UNibble4And = {0xF,0xF0,0xF00,0xF000}; |
||
| 2638 | static const XMVECTORF32 UNibble4Mul = {1.0f,1.0f/16.f,1.0f/256.f,1.0f/4096.f}; |
||
| 2639 | XMASSERT(pSource); |
||
| 2640 | // Get the 32 bit value and splat it |
||
| 2641 | XMVECTOR vResult = _mm_load_ps1(reinterpret_cast<const float *>(&pSource->v)); |
||
| 2642 | // Mask off x, y and z |
||
| 2643 | vResult = _mm_and_ps(vResult,UNibble4And); |
||
| 2644 | // Convert to float |
||
| 2645 | vResult = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vResult)[0]); |
||
| 2646 | // Normalize x, y, and z |
||
| 2647 | vResult = _mm_mul_ps(vResult,UNibble4Mul); |
||
| 2648 | return vResult; |
||
| 2649 | #else |
||
| 2650 | XMVECTOR V; |
||
| 2651 | UINT Element; |
||
| 2652 | |||
| 2653 | XMASSERT(pSource); |
||
| 2654 | |||
| 2655 | Element = pSource->v & 0xF; |
||
| 2656 | V.vector4_f32[0] = (FLOAT)Element; |
||
| 2657 | Element = (pSource->v >> 4) & 0xF; |
||
| 2658 | V.vector4_f32[1] = (FLOAT)Element; |
||
| 2659 | Element = (pSource->v >> 8) & 0xF; |
||
| 2660 | V.vector4_f32[2] = (FLOAT)Element; |
||
| 2661 | Element = (pSource->v >> 12) & 0xF; |
||
| 2662 | V.vector4_f32[3] = (FLOAT)Element; |
||
| 2663 | |||
| 2664 | return V; |
||
| 2665 | #endif // !_XM_SSE_INTRISICS_ |
||
| 2666 | } |
||
| 2667 | |||
| 2668 | //------------------------------------------------------------------------------ |
||
| 2669 | |||
| 2670 | XMFINLINE XMVECTOR XMLoadU555 |
||
| 2671 | ( |
||
| 2672 | CONST XMU555* pSource |
||
| 2673 | ) |
||
| 2674 | { |
||
| 2675 | #if defined(_XM_SSE_INTRINSICS_) && !defined(_XM_NO_INTRINSICS_) |
||
| 2676 | static const XMVECTORI32 U555And = {0x1F,0x1F<<5,0x1F<<10,0x8000}; |
||
| 2677 | static const XMVECTORF32 U555Mul = {1.0f,1.0f/32.f,1.0f/1024.f,1.0f/32768.f}; |
||
| 2678 | XMASSERT(pSource); |
||
| 2679 | // Get the 32 bit value and splat it |
||
| 2680 | XMVECTOR vResult = _mm_load_ps1(reinterpret_cast<const float *>(&pSource->v)); |
||
| 2681 | // Mask off x, y and z |
||
| 2682 | vResult = _mm_and_ps(vResult,U555And); |
||
| 2683 | // Convert to float |
||
| 2684 | vResult = _mm_cvtepi32_ps(reinterpret_cast<const __m128i *>(&vResult)[0]); |
||
| 2685 | // Normalize x, y, and z |
||
| 2686 | vResult = _mm_mul_ps(vResult,U555Mul); |
||
| 2687 | return vResult; |
||
| 2688 | #else |
||
| 2689 | XMVECTOR V; |
||
| 2690 | UINT Element; |
||
| 2691 | |||
| 2692 | XMASSERT(pSource); |
||
| 2693 | |||
| 2694 | Element = pSource->v & 0x1F; |
||
| 2695 | V.vector4_f32[0] = (FLOAT)Element; |
||
| 2696 | Element = (pSource->v >> 5) & 0x1F; |
||
| 2697 | V.vector4_f32[1] = (FLOAT)Element; |
||
| 2698 | Element = (pSource->v >> 10) & 0x1F; |
||
| 2699 | V.vector4_f32[2] = (FLOAT)Element; |
||
| 2700 | Element = (pSource->v >> 15) & 0x1; |
||
| 2701 | V.vector4_f32[3] = (FLOAT)Element; |
||
| 2702 | |||
| 2703 | return V; |
||
| 2704 | #endif // !_XM_SSE_INTRISICS_ |
||
| 2705 | } |
||
| 2706 | |||
| 2707 | //------------------------------------------------------------------------------ |
||
| 2708 | |||
| 2709 | XMFINLINE XMVECTOR XMLoadColor |
||
| 2710 | ( |
||
| 2711 | CONST XMCOLOR* pSource |
||
| 2712 | ) |
||
| 2713 | { |
||
| 2714 | #if defined(_XM_NO_INTRINSICS_) |
||
| 2715 | XMASSERT(pSource); |
||
| 2716 | { |
||
| 2717 | // INT -> Float conversions are done in one instruction. |
||
| 2718 | // UINT -> Float calls a runtime function. Keep in INT |
||
| 2719 | INT iColor = (INT)(pSource->c); |
||
| 2720 | XMVECTOR vColor = { |
||
| 2721 | (FLOAT)((iColor >> 16) & 0xFF) * (1.0f/255.0f), |
||
| 2722 | (FLOAT)((iColor >> 8) & 0xFF) * (1.0f/255.0f), |
||
| 2723 | (FLOAT)(iColor & 0xFF) * (1.0f/255.0f), |
||
| 2724 | (FLOAT)((iColor >> 24) & 0xFF) * (1.0f/255.0f) |
||
| 2725 | }; |
||
| 2726 | return vColor; |
||
| 2727 | } |
||
| 2728 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 2729 | XMASSERT(pSource); |
||
| 2730 | // Splat the color in all four entries |
||
| 2731 | __m128i vInt = _mm_set1_epi32(pSource->c); |
||
| 2732 | // Shift R&0xFF0000, G&0xFF00, B&0xFF, A&0xFF000000 |
||
| 2733 | vInt = _mm_and_si128(vInt,g_XMMaskA8R8G8B8); |
||
| 2734 | // a is unsigned! Flip the bit to convert the order to signed |
||
| 2735 | vInt = _mm_xor_si128(vInt,g_XMFlipA8R8G8B8); |
||
| 2736 | // Convert to floating point numbers |
||
| 2737 | XMVECTOR vTemp = _mm_cvtepi32_ps(vInt); |
||
| 2738 | // RGB + 0, A + 0x80000000.f to undo the signed order. |
||
| 2739 | vTemp = _mm_add_ps(vTemp,g_XMFixAA8R8G8B8); |
||
| 2740 | // Convert 0-255 to 0.0f-1.0f |
||
| 2741 | return _mm_mul_ps(vTemp,g_XMNormalizeA8R8G8B8); |
||
| 2742 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 2743 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 2744 | } |
||
| 2745 | |||
| 2746 | //------------------------------------------------------------------------------ |
||
| 2747 | |||
| 2748 | XMFINLINE XMMATRIX XMLoadFloat3x3 |
||
| 2749 | ( |
||
| 2750 | CONST XMFLOAT3X3* pSource |
||
| 2751 | ) |
||
| 2752 | { |
||
| 2753 | #if defined(_XM_NO_INTRINSICS_) |
||
| 2754 | |||
| 2755 | XMMATRIX M; |
||
| 2756 | |||
| 2757 | XMASSERT(pSource); |
||
| 2758 | |||
| 2759 | M.r[0].vector4_f32[0] = pSource->m[0][0]; |
||
| 2760 | M.r[0].vector4_f32[1] = pSource->m[0][1]; |
||
| 2761 | M.r[0].vector4_f32[2] = pSource->m[0][2]; |
||
| 2762 | M.r[0].vector4_f32[3] = 0.0f; |
||
| 2763 | |||
| 2764 | M.r[1].vector4_f32[0] = pSource->m[1][0]; |
||
| 2765 | M.r[1].vector4_f32[1] = pSource->m[1][1]; |
||
| 2766 | M.r[1].vector4_f32[2] = pSource->m[1][2]; |
||
| 2767 | M.r[1].vector4_f32[3] = 0.0f; |
||
| 2768 | |||
| 2769 | M.r[2].vector4_f32[0] = pSource->m[2][0]; |
||
| 2770 | M.r[2].vector4_f32[1] = pSource->m[2][1]; |
||
| 2771 | M.r[2].vector4_f32[2] = pSource->m[2][2]; |
||
| 2772 | M.r[2].vector4_f32[3] = 0.0f; |
||
| 2773 | |||
| 2774 | M.r[3].vector4_f32[0] = 0.0f; |
||
| 2775 | M.r[3].vector4_f32[1] = 0.0f; |
||
| 2776 | M.r[3].vector4_f32[2] = 0.0f; |
||
| 2777 | M.r[3].vector4_f32[3] = 1.0f; |
||
| 2778 | |||
| 2779 | return M; |
||
| 2780 | |||
| 2781 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 2782 | XMMATRIX M; |
||
| 2783 | XMVECTOR V1, V2, V3, Z, T1, T2, T3, T4, T5; |
||
| 2784 | |||
| 2785 | Z = _mm_setzero_ps(); |
||
| 2786 | |||
| 2787 | XMASSERT(pSource); |
||
| 2788 | |||
| 2789 | V1 = _mm_loadu_ps( &pSource->m[0][0] ); |
||
| 2790 | V2 = _mm_loadu_ps( &pSource->m[1][1] ); |
||
| 2791 | V3 = _mm_load_ss( &pSource->m[2][2] ); |
||
| 2792 | |||
| 2793 | T1 = _mm_unpackhi_ps( V1, Z ); |
||
| 2794 | T2 = _mm_unpacklo_ps( V2, Z ); |
||
| 2795 | T3 = _mm_shuffle_ps( V3, T2, _MM_SHUFFLE( 0, 1, 0, 0 ) ); |
||
| 2796 | T4 = _mm_movehl_ps( T2, T3 ); |
||
| 2797 | T5 = _mm_movehl_ps( Z, T1 ); |
||
| 2798 | |||
| 2799 | M.r[0] = _mm_movelh_ps( V1, T1 ); |
||
| 2800 | M.r[1] = _mm_add_ps( T4, T5 ); |
||
| 2801 | M.r[2] = _mm_shuffle_ps( V2, V3, _MM_SHUFFLE(1, 0, 3, 2) ); |
||
| 2802 | M.r[3] = g_XMIdentityR3; |
||
| 2803 | |||
| 2804 | return M; |
||
| 2805 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 2806 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 2807 | } |
||
| 2808 | |||
| 2809 | //------------------------------------------------------------------------------ |
||
| 2810 | |||
| 2811 | XMFINLINE XMMATRIX XMLoadFloat4x3 |
||
| 2812 | ( |
||
| 2813 | CONST XMFLOAT4X3* pSource |
||
| 2814 | ) |
||
| 2815 | { |
||
| 2816 | #if defined(_XM_NO_INTRINSICS_) |
||
| 2817 | XMMATRIX M; |
||
| 2818 | XMASSERT(pSource); |
||
| 2819 | |||
| 2820 | ((UINT *)(&M.r[0].vector4_f32[0]))[0] = ((const UINT *)(&pSource->m[0][0]))[0]; |
||
| 2821 | ((UINT *)(&M.r[0].vector4_f32[1]))[0] = ((const UINT *)(&pSource->m[0][1]))[0]; |
||
| 2822 | ((UINT *)(&M.r[0].vector4_f32[2]))[0] = ((const UINT *)(&pSource->m[0][2]))[0]; |
||
| 2823 | M.r[0].vector4_f32[3] = 0.0f; |
||
| 2824 | |||
| 2825 | ((UINT *)(&M.r[1].vector4_f32[0]))[0] = ((const UINT *)(&pSource->m[1][0]))[0]; |
||
| 2826 | ((UINT *)(&M.r[1].vector4_f32[1]))[0] = ((const UINT *)(&pSource->m[1][1]))[0]; |
||
| 2827 | ((UINT *)(&M.r[1].vector4_f32[2]))[0] = ((const UINT *)(&pSource->m[1][2]))[0]; |
||
| 2828 | M.r[1].vector4_f32[3] = 0.0f; |
||
| 2829 | |||
| 2830 | ((UINT *)(&M.r[2].vector4_f32[0]))[0] = ((const UINT *)(&pSource->m[2][0]))[0]; |
||
| 2831 | ((UINT *)(&M.r[2].vector4_f32[1]))[0] = ((const UINT *)(&pSource->m[2][1]))[0]; |
||
| 2832 | ((UINT *)(&M.r[2].vector4_f32[2]))[0] = ((const UINT *)(&pSource->m[2][2]))[0]; |
||
| 2833 | M.r[2].vector4_f32[3] = 0.0f; |
||
| 2834 | |||
| 2835 | ((UINT *)(&M.r[3].vector4_f32[0]))[0] = ((const UINT *)(&pSource->m[3][0]))[0]; |
||
| 2836 | ((UINT *)(&M.r[3].vector4_f32[1]))[0] = ((const UINT *)(&pSource->m[3][1]))[0]; |
||
| 2837 | ((UINT *)(&M.r[3].vector4_f32[2]))[0] = ((const UINT *)(&pSource->m[3][2]))[0]; |
||
| 2838 | M.r[3].vector4_f32[3] = 1.0f; |
||
| 2839 | |||
| 2840 | return M; |
||
| 2841 | |||
| 2842 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 2843 | XMASSERT(pSource); |
||
| 2844 | // Use unaligned load instructions to |
||
| 2845 | // load the 12 floats |
||
| 2846 | // vTemp1 = x1,y1,z1,x2 |
||
| 2847 | XMVECTOR vTemp1 = _mm_loadu_ps(&pSource->m[0][0]); |
||
| 2848 | // vTemp2 = y2,z2,x3,y3 |
||
| 2849 | XMVECTOR vTemp2 = _mm_loadu_ps(&pSource->m[1][1]); |
||
| 2850 | // vTemp4 = z3,x4,y4,z4 |
||
| 2851 | XMVECTOR vTemp4 = _mm_loadu_ps(&pSource->m[2][2]); |
||
| 2852 | // vTemp3 = x3,y3,z3,z3 |
||
| 2853 | XMVECTOR vTemp3 = _mm_shuffle_ps(vTemp2,vTemp4,_MM_SHUFFLE(0,0,3,2)); |
||
| 2854 | // vTemp2 = y2,z2,x2,x2 |
||
| 2855 | vTemp2 = _mm_shuffle_ps(vTemp2,vTemp1,_MM_SHUFFLE(3,3,1,0)); |
||
| 2856 | // vTemp2 = x2,y2,z2,z2 |
||
| 2857 | vTemp2 = _mm_shuffle_ps(vTemp2,vTemp2,_MM_SHUFFLE(1,1,0,2)); |
||
| 2858 | // vTemp1 = x1,y1,z1,0 |
||
| 2859 | vTemp1 = _mm_and_ps(vTemp1,g_XMMask3); |
||
| 2860 | // vTemp2 = x2,y2,z2,0 |
||
| 2861 | vTemp2 = _mm_and_ps(vTemp2,g_XMMask3); |
||
| 2862 | // vTemp3 = x3,y3,z3,0 |
||
| 2863 | vTemp3 = _mm_and_ps(vTemp3,g_XMMask3); |
||
| 2864 | // vTemp4i = x4,y4,z4,0 |
||
| 2865 | __m128i vTemp4i = _mm_srli_si128(reinterpret_cast<const __m128i *>(&vTemp4)[0],32/8); |
||
| 2866 | // vTemp4i = x4,y4,z4,1.0f |
||
| 2867 | vTemp4i = _mm_or_si128(vTemp4i,g_XMIdentityR3); |
||
| 2868 | XMMATRIX M(vTemp1, |
||
| 2869 | vTemp2, |
||
| 2870 | vTemp3, |
||
| 2871 | reinterpret_cast<const __m128 *>(&vTemp4i)[0]); |
||
| 2872 | return M; |
||
| 2873 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 2874 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 2875 | } |
||
| 2876 | |||
| 2877 | //------------------------------------------------------------------------------ |
||
| 2878 | |||
| 2879 | XMFINLINE XMMATRIX XMLoadFloat4x3A |
||
| 2880 | ( |
||
| 2881 | CONST XMFLOAT4X3A* pSource |
||
| 2882 | ) |
||
| 2883 | { |
||
| 2884 | #if defined(_XM_NO_INTRINSICS_) |
||
| 2885 | |||
| 2886 | XMMATRIX M; |
||
| 2887 | |||
| 2888 | XMASSERT(pSource); |
||
| 2889 | XMASSERT(((UINT_PTR)pSource & 0xF) == 0); |
||
| 2890 | |||
| 2891 | M.r[0].vector4_f32[0] = pSource->m[0][0]; |
||
| 2892 | M.r[0].vector4_f32[1] = pSource->m[0][1]; |
||
| 2893 | M.r[0].vector4_f32[2] = pSource->m[0][2]; |
||
| 2894 | M.r[0].vector4_f32[3] = 0.0f; |
||
| 2895 | |||
| 2896 | M.r[1].vector4_f32[0] = pSource->m[1][0]; |
||
| 2897 | M.r[1].vector4_f32[1] = pSource->m[1][1]; |
||
| 2898 | M.r[1].vector4_f32[2] = pSource->m[1][2]; |
||
| 2899 | M.r[1].vector4_f32[3] = 0.0f; |
||
| 2900 | |||
| 2901 | M.r[2].vector4_f32[0] = pSource->m[2][0]; |
||
| 2902 | M.r[2].vector4_f32[1] = pSource->m[2][1]; |
||
| 2903 | M.r[2].vector4_f32[2] = pSource->m[2][2]; |
||
| 2904 | M.r[2].vector4_f32[3] = 0.0f; |
||
| 2905 | |||
| 2906 | M.r[3].vector4_f32[0] = pSource->m[3][0]; |
||
| 2907 | M.r[3].vector4_f32[1] = pSource->m[3][1]; |
||
| 2908 | M.r[3].vector4_f32[2] = pSource->m[3][2]; |
||
| 2909 | M.r[3].vector4_f32[3] = 1.0f; |
||
| 2910 | |||
| 2911 | return M; |
||
| 2912 | |||
| 2913 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 2914 | XMASSERT(pSource); |
||
| 2915 | // Use aligned load instructions to |
||
| 2916 | // load the 12 floats |
||
| 2917 | // vTemp1 = x1,y1,z1,x2 |
||
| 2918 | XMVECTOR vTemp1 = _mm_load_ps(&pSource->m[0][0]); |
||
| 2919 | // vTemp2 = y2,z2,x3,y3 |
||
| 2920 | XMVECTOR vTemp2 = _mm_load_ps(&pSource->m[1][1]); |
||
| 2921 | // vTemp4 = z3,x4,y4,z4 |
||
| 2922 | XMVECTOR vTemp4 = _mm_load_ps(&pSource->m[2][2]); |
||
| 2923 | // vTemp3 = x3,y3,z3,z3 |
||
| 2924 | XMVECTOR vTemp3 = _mm_shuffle_ps(vTemp2,vTemp4,_MM_SHUFFLE(0,0,3,2)); |
||
| 2925 | // vTemp2 = y2,z2,x2,x2 |
||
| 2926 | vTemp2 = _mm_shuffle_ps(vTemp2,vTemp1,_MM_SHUFFLE(3,3,1,0)); |
||
| 2927 | // vTemp2 = x2,y2,z2,z2 |
||
| 2928 | vTemp2 = _mm_shuffle_ps(vTemp2,vTemp2,_MM_SHUFFLE(1,1,0,2)); |
||
| 2929 | // vTemp1 = x1,y1,z1,0 |
||
| 2930 | vTemp1 = _mm_and_ps(vTemp1,g_XMMask3); |
||
| 2931 | // vTemp2 = x2,y2,z2,0 |
||
| 2932 | vTemp2 = _mm_and_ps(vTemp2,g_XMMask3); |
||
| 2933 | // vTemp3 = x3,y3,z3,0 |
||
| 2934 | vTemp3 = _mm_and_ps(vTemp3,g_XMMask3); |
||
| 2935 | // vTemp4i = x4,y4,z4,0 |
||
| 2936 | __m128i vTemp4i = _mm_srli_si128(reinterpret_cast<const __m128i *>(&vTemp4)[0],32/8); |
||
| 2937 | // vTemp4i = x4,y4,z4,1.0f |
||
| 2938 | vTemp4i = _mm_or_si128(vTemp4i,g_XMIdentityR3); |
||
| 2939 | XMMATRIX M(vTemp1, |
||
| 2940 | vTemp2, |
||
| 2941 | vTemp3, |
||
| 2942 | reinterpret_cast<const __m128 *>(&vTemp4i)[0]); |
||
| 2943 | return M; |
||
| 2944 | #else // _XM_VMX128_INTRINSICS_ |
||
| 2945 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 2946 | } |
||
| 2947 | |||
| 2948 | //------------------------------------------------------------------------------ |
||
| 2949 | |||
| 2950 | XMFINLINE XMMATRIX XMLoadFloat4x4 |
||
| 2951 | ( |
||
| 2952 | CONST XMFLOAT4X4* pSource |
||
| 2953 | ) |
||
| 2954 | { |
||
| 2955 | #if defined(_XM_NO_INTRINSICS_) |
||
| 2956 | XMMATRIX M; |
||
| 2957 | XMASSERT(pSource); |
||
| 2958 | |||
| 2959 | ((UINT *)(&M.r[0].vector4_f32[0]))[0] = ((const UINT *)(&pSource->m[0][0]))[0]; |
||
| 2960 | ((UINT *)(&M.r[0].vector4_f32[1]))[0] = ((const UINT *)(&pSource->m[0][1]))[0]; |
||
| 2961 | ((UINT *)(&M.r[0].vector4_f32[2]))[0] = ((const UINT *)(&pSource->m[0][2]))[0]; |
||
| 2962 | ((UINT *)(&M.r[0].vector4_f32[3]))[0] = ((const UINT *)(&pSource->m[0][3]))[0]; |
||
| 2963 | |||
| 2964 | ((UINT *)(&M.r[1].vector4_f32[0]))[0] = ((const UINT *)(&pSource->m[1][0]))[0]; |
||
| 2965 | ((UINT *)(&M.r[1].vector4_f32[1]))[0] = ((const UINT *)(&pSource->m[1][1]))[0]; |
||
| 2966 | ((UINT *)(&M.r[1].vector4_f32[2]))[0] = ((const UINT *)(&pSource->m[1][2]))[0]; |
||
| 2967 | ((UINT *)(&M.r[1].vector4_f32[3]))[0] = ((const UINT *)(&pSource->m[1][3]))[0]; |
||
| 2968 | |||
| 2969 | ((UINT *)(&M.r[2].vector4_f32[0]))[0] = ((const UINT *)(&pSource->m[2][0]))[0]; |
||
| 2970 | ((UINT *)(&M.r[2].vector4_f32[1]))[0] = ((const UINT *)(&pSource->m[2][1]))[0]; |
||
| 2971 | ((UINT *)(&M.r[2].vector4_f32[2]))[0] = ((const UINT *)(&pSource->m[2][2]))[0]; |
||
| 2972 | ((UINT *)(&M.r[2].vector4_f32[3]))[0] = ((const UINT *)(&pSource->m[2][3]))[0]; |
||
| 2973 | |||
| 2974 | ((UINT *)(&M.r[3].vector4_f32[0]))[0] = ((const UINT *)(&pSource->m[3][0]))[0]; |
||
| 2975 | ((UINT *)(&M.r[3].vector4_f32[1]))[0] = ((const UINT *)(&pSource->m[3][1]))[0]; |
||
| 2976 | ((UINT *)(&M.r[3].vector4_f32[2]))[0] = ((const UINT *)(&pSource->m[3][2]))[0]; |
||
| 2977 | ((UINT *)(&M.r[3].vector4_f32[3]))[0] = ((const UINT *)(&pSource->m[3][3]))[0]; |
||
| 2978 | |||
| 2979 | return M; |
||
| 2980 | |||
| 2981 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 2982 | XMASSERT(pSource); |
||
| 2983 | XMMATRIX M; |
||
| 2984 | |||
| 2985 | M.r[0] = _mm_loadu_ps( &pSource->_11 ); |
||
| 2986 | M.r[1] = _mm_loadu_ps( &pSource->_21 ); |
||
| 2987 | M.r[2] = _mm_loadu_ps( &pSource->_31 ); |
||
| 2988 | M.r[3] = _mm_loadu_ps( &pSource->_41 ); |
||
| 2989 | |||
| 2990 | return M; |
||
| 2991 | #elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 2992 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 2993 | } |
||
| 2994 | |||
| 2995 | //------------------------------------------------------------------------------ |
||
| 2996 | |||
| 2997 | XMFINLINE XMMATRIX XMLoadFloat4x4A |
||
| 2998 | ( |
||
| 2999 | CONST XMFLOAT4X4A* pSource |
||
| 3000 | ) |
||
| 3001 | { |
||
| 3002 | #if defined(_XM_NO_INTRINSICS_) |
||
| 3003 | |||
| 3004 | XMMATRIX M; |
||
| 3005 | |||
| 3006 | XMASSERT(pSource); |
||
| 3007 | XMASSERT(((UINT_PTR)pSource & 0xF) == 0); |
||
| 3008 | |||
| 3009 | M.r[0].vector4_f32[0] = pSource->m[0][0]; |
||
| 3010 | M.r[0].vector4_f32[1] = pSource->m[0][1]; |
||
| 3011 | M.r[0].vector4_f32[2] = pSource->m[0][2]; |
||
| 3012 | M.r[0].vector4_f32[3] = pSource->m[0][3]; |
||
| 3013 | |||
| 3014 | M.r[1].vector4_f32[0] = pSource->m[1][0]; |
||
| 3015 | M.r[1].vector4_f32[1] = pSource->m[1][1]; |
||
| 3016 | M.r[1].vector4_f32[2] = pSource->m[1][2]; |
||
| 3017 | M.r[1].vector4_f32[3] = pSource->m[1][3]; |
||
| 3018 | |||
| 3019 | M.r[2].vector4_f32[0] = pSource->m[2][0]; |
||
| 3020 | M.r[2].vector4_f32[1] = pSource->m[2][1]; |
||
| 3021 | M.r[2].vector4_f32[2] = pSource->m[2][2]; |
||
| 3022 | M.r[2].vector4_f32[3] = pSource->m[2][3]; |
||
| 3023 | |||
| 3024 | M.r[3].vector4_f32[0] = pSource->m[3][0]; |
||
| 3025 | M.r[3].vector4_f32[1] = pSource->m[3][1]; |
||
| 3026 | M.r[3].vector4_f32[2] = pSource->m[3][2]; |
||
| 3027 | M.r[3].vector4_f32[3] = pSource->m[3][3]; |
||
| 3028 | |||
| 3029 | return M; |
||
| 3030 | |||
| 3031 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 3032 | XMMATRIX M; |
||
| 3033 | |||
| 3034 | XMASSERT(pSource); |
||
| 3035 | |||
| 3036 | M.r[0] = _mm_load_ps( &pSource->_11 ); |
||
| 3037 | M.r[1] = _mm_load_ps( &pSource->_21 ); |
||
| 3038 | M.r[2] = _mm_load_ps( &pSource->_31 ); |
||
| 3039 | M.r[3] = _mm_load_ps( &pSource->_41 ); |
||
| 3040 | |||
| 3041 | return M; |
||
| 3042 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3043 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3044 | } |
||
| 3045 | |||
| 3046 | /**************************************************************************** |
||
| 3047 | * |
||
| 3048 | * Vector and matrix store operations |
||
| 3049 | * |
||
| 3050 | ****************************************************************************/ |
||
| 3051 | |||
| 3052 | XMFINLINE VOID XMStoreInt |
||
| 3053 | ( |
||
| 3054 | UINT* pDestination, |
||
| 3055 | FXMVECTOR V |
||
| 3056 | ) |
||
| 3057 | { |
||
| 3058 | #if defined(_XM_NO_INTRINSICS_) || defined(_XM_SSE_INTRINSICS_) |
||
| 3059 | |||
| 3060 | XMASSERT(pDestination); |
||
| 3061 | XMASSERT(((UINT_PTR)pDestination & 3) == 0); |
||
| 3062 | |||
| 3063 | *pDestination = XMVectorGetIntX( V ); |
||
| 3064 | |||
| 3065 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3066 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3067 | } |
||
| 3068 | |||
| 3069 | //------------------------------------------------------------------------------ |
||
| 3070 | |||
| 3071 | XMFINLINE VOID XMStoreFloat |
||
| 3072 | ( |
||
| 3073 | FLOAT* pDestination, |
||
| 3074 | FXMVECTOR V |
||
| 3075 | ) |
||
| 3076 | { |
||
| 3077 | #if defined(_XM_NO_INTRINSICS_) || defined(_XM_SSE_INTRINSICS_) |
||
| 3078 | |||
| 3079 | XMASSERT(pDestination); |
||
| 3080 | XMASSERT(((UINT_PTR)pDestination & 3) == 0); |
||
| 3081 | |||
| 3082 | *pDestination = XMVectorGetX( V ); |
||
| 3083 | |||
| 3084 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3085 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3086 | } |
||
| 3087 | |||
| 3088 | //------------------------------------------------------------------------------ |
||
| 3089 | |||
| 3090 | XMFINLINE VOID XMStoreInt2 |
||
| 3091 | ( |
||
| 3092 | UINT* pDestination, |
||
| 3093 | FXMVECTOR V |
||
| 3094 | ) |
||
| 3095 | { |
||
| 3096 | #if defined(_XM_NO_INTRINSICS_) |
||
| 3097 | |||
| 3098 | XMASSERT(pDestination); |
||
| 3099 | XMASSERT(((UINT_PTR)pDestination & 3) == 0); |
||
| 3100 | |||
| 3101 | pDestination[0] = V.vector4_u32[0]; |
||
| 3102 | pDestination[1] = V.vector4_u32[1]; |
||
| 3103 | |||
| 3104 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 3105 | |||
| 3106 | XMASSERT(pDestination); |
||
| 3107 | XMASSERT(((UINT_PTR)pDestination & 3) == 0); |
||
| 3108 | pDestination[0] = XMVectorGetIntX( V ); |
||
| 3109 | pDestination[1] = XMVectorGetIntY( V ); |
||
| 3110 | |||
| 3111 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3112 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3113 | } |
||
| 3114 | |||
| 3115 | //------------------------------------------------------------------------------ |
||
| 3116 | |||
| 3117 | XMFINLINE VOID XMStoreInt2A |
||
| 3118 | ( |
||
| 3119 | UINT* pDestination, |
||
| 3120 | FXMVECTOR V |
||
| 3121 | ) |
||
| 3122 | { |
||
| 3123 | #if defined(_XM_NO_INTRINSICS_) |
||
| 3124 | |||
| 3125 | XMASSERT(pDestination); |
||
| 3126 | XMASSERT(((UINT_PTR)pDestination & 0xF) == 0); |
||
| 3127 | |||
| 3128 | pDestination[0] = V.vector4_u32[0]; |
||
| 3129 | pDestination[1] = V.vector4_u32[1]; |
||
| 3130 | |||
| 3131 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 3132 | |||
| 3133 | XMASSERT(pDestination); |
||
| 3134 | XMASSERT(((UINT_PTR)pDestination & 3) == 0); |
||
| 3135 | |||
| 3136 | _mm_storel_epi64( (__m128i*)pDestination, reinterpret_cast<const __m128i *>(&V)[0] ); |
||
| 3137 | |||
| 3138 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3139 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3140 | } |
||
| 3141 | |||
| 3142 | //------------------------------------------------------------------------------ |
||
| 3143 | |||
| 3144 | XMFINLINE VOID XMStoreFloat2 |
||
| 3145 | ( |
||
| 3146 | XMFLOAT2* pDestination, |
||
| 3147 | FXMVECTOR V |
||
| 3148 | ) |
||
| 3149 | { |
||
| 3150 | #if defined(_XM_NO_INTRINSICS_) |
||
| 3151 | |||
| 3152 | XMASSERT(pDestination); |
||
| 3153 | XMASSERT(((UINT_PTR)pDestination & 3) == 0); |
||
| 3154 | |||
| 3155 | pDestination->x = V.vector4_f32[0]; |
||
| 3156 | pDestination->y = V.vector4_f32[1]; |
||
| 3157 | |||
| 3158 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 3159 | |||
| 3160 | XMASSERT(pDestination); |
||
| 3161 | XMASSERT(((UINT_PTR)pDestination & 3) == 0); |
||
| 3162 | |||
| 3163 | XMVECTOR T = _mm_shuffle_ps( V, V, _MM_SHUFFLE( 1, 1, 1, 1 ) ); |
||
| 3164 | _mm_store_ss( &pDestination->x, V ); |
||
| 3165 | _mm_store_ss( &pDestination->y, T ); |
||
| 3166 | |||
| 3167 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3168 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3169 | } |
||
| 3170 | |||
| 3171 | //------------------------------------------------------------------------------ |
||
| 3172 | |||
| 3173 | XMFINLINE VOID XMStoreFloat2A |
||
| 3174 | ( |
||
| 3175 | XMFLOAT2A* pDestination, |
||
| 3176 | FXMVECTOR V |
||
| 3177 | ) |
||
| 3178 | { |
||
| 3179 | #if defined(_XM_NO_INTRINSICS_) |
||
| 3180 | |||
| 3181 | XMASSERT(pDestination); |
||
| 3182 | XMASSERT(((UINT_PTR)pDestination & 0xF) == 0); |
||
| 3183 | |||
| 3184 | pDestination->x = V.vector4_f32[0]; |
||
| 3185 | pDestination->y = V.vector4_f32[1]; |
||
| 3186 | |||
| 3187 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 3188 | |||
| 3189 | XMASSERT(pDestination); |
||
| 3190 | XMASSERT(((UINT_PTR)pDestination & 3) == 0); |
||
| 3191 | |||
| 3192 | XMVECTOR T = _mm_shuffle_ps( V, V, _MM_SHUFFLE( 1, 1, 1, 1 ) ); |
||
| 3193 | _mm_store_ss( &pDestination->x, V ); |
||
| 3194 | _mm_store_ss( &pDestination->y, T ); |
||
| 3195 | |||
| 3196 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3197 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3198 | } |
||
| 3199 | |||
| 3200 | //------------------------------------------------------------------------------ |
||
| 3201 | |||
| 3202 | XMFINLINE VOID XMStoreHalf2 |
||
| 3203 | ( |
||
| 3204 | XMHALF2* pDestination, |
||
| 3205 | FXMVECTOR V |
||
| 3206 | ) |
||
| 3207 | { |
||
| 3208 | #if defined(_XM_NO_INTRINSICS_) |
||
| 3209 | |||
| 3210 | XMASSERT(pDestination); |
||
| 3211 | |||
| 3212 | pDestination->x = XMConvertFloatToHalf(V.vector4_f32[0]); |
||
| 3213 | pDestination->y = XMConvertFloatToHalf(V.vector4_f32[1]); |
||
| 3214 | |||
| 3215 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 3216 | XMASSERT(pDestination); |
||
| 3217 | pDestination->x = XMConvertFloatToHalf(XMVectorGetX(V)); |
||
| 3218 | pDestination->y = XMConvertFloatToHalf(XMVectorGetY(V)); |
||
| 3219 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3220 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3221 | } |
||
| 3222 | |||
| 3223 | //------------------------------------------------------------------------------ |
||
| 3224 | |||
| 3225 | XMFINLINE VOID XMStoreShortN2 |
||
| 3226 | ( |
||
| 3227 | XMSHORTN2* pDestination, |
||
| 3228 | FXMVECTOR V |
||
| 3229 | ) |
||
| 3230 | { |
||
| 3231 | #if defined(_XM_NO_INTRINSICS_) |
||
| 3232 | |||
| 3233 | XMVECTOR N; |
||
| 3234 | static CONST XMVECTORF32 Scale = {32767.0f, 32767.0f, 32767.0f, 32767.0f}; |
||
| 3235 | |||
| 3236 | XMASSERT(pDestination); |
||
| 3237 | |||
| 3238 | N = XMVectorClamp(V, g_XMNegativeOne.v, g_XMOne.v); |
||
| 3239 | N = XMVectorMultiply(N, Scale.v); |
||
| 3240 | N = XMVectorRound(N); |
||
| 3241 | |||
| 3242 | pDestination->x = (SHORT)N.vector4_f32[0]; |
||
| 3243 | pDestination->y = (SHORT)N.vector4_f32[1]; |
||
| 3244 | |||
| 3245 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 3246 | XMASSERT(pDestination); |
||
| 3247 | static CONST XMVECTORF32 Scale = {32767.0f, 32767.0f, 32767.0f, 32767.0f}; |
||
| 3248 | |||
| 3249 | XMVECTOR vResult = _mm_max_ps(V,g_XMNegativeOne); |
||
| 3250 | vResult = _mm_min_ps(vResult,g_XMOne); |
||
| 3251 | vResult = _mm_mul_ps(vResult,Scale); |
||
| 3252 | __m128i vResulti = _mm_cvtps_epi32(vResult); |
||
| 3253 | vResulti = _mm_packs_epi32(vResulti,vResulti); |
||
| 3254 | _mm_store_ss(reinterpret_cast<float *>(&pDestination->x),reinterpret_cast<const __m128 *>(&vResulti)[0]); |
||
| 3255 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3256 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3257 | } |
||
| 3258 | |||
| 3259 | //------------------------------------------------------------------------------ |
||
| 3260 | |||
| 3261 | XMFINLINE VOID XMStoreShort2 |
||
| 3262 | ( |
||
| 3263 | XMSHORT2* pDestination, |
||
| 3264 | FXMVECTOR V |
||
| 3265 | ) |
||
| 3266 | { |
||
| 3267 | #if defined(_XM_NO_INTRINSICS_) |
||
| 3268 | |||
| 3269 | XMVECTOR N; |
||
| 3270 | static CONST XMVECTOR Min = {-32767.0f, -32767.0f, -32767.0f, -32767.0f}; |
||
| 3271 | static CONST XMVECTOR Max = {32767.0f, 32767.0f, 32767.0f, 32767.0f}; |
||
| 3272 | |||
| 3273 | XMASSERT(pDestination); |
||
| 3274 | |||
| 3275 | N = XMVectorClamp(V, Min, Max); |
||
| 3276 | N = XMVectorRound(N); |
||
| 3277 | |||
| 3278 | pDestination->x = (SHORT)N.vector4_f32[0]; |
||
| 3279 | pDestination->y = (SHORT)N.vector4_f32[1]; |
||
| 3280 | |||
| 3281 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 3282 | XMASSERT(pDestination); |
||
| 3283 | static CONST XMVECTORF32 Min = {-32767.0f, -32767.0f, -32767.0f, -32767.0f}; |
||
| 3284 | static CONST XMVECTORF32 Max = {32767.0f, 32767.0f, 32767.0f, 32767.0f}; |
||
| 3285 | // Bounds check |
||
| 3286 | XMVECTOR vResult = _mm_max_ps(V,Min); |
||
| 3287 | vResult = _mm_min_ps(vResult,Max); |
||
| 3288 | // Convert to int with rounding |
||
| 3289 | __m128i vInt = _mm_cvtps_epi32(vResult); |
||
| 3290 | // Pack the ints into shorts |
||
| 3291 | vInt = _mm_packs_epi32(vInt,vInt); |
||
| 3292 | _mm_store_ss(reinterpret_cast<float *>(&pDestination->x),reinterpret_cast<const __m128 *>(&vInt)[0]); |
||
| 3293 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3294 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3295 | } |
||
| 3296 | |||
| 3297 | //------------------------------------------------------------------------------ |
||
| 3298 | |||
| 3299 | XMFINLINE VOID XMStoreUShortN2 |
||
| 3300 | ( |
||
| 3301 | XMUSHORTN2* pDestination, |
||
| 3302 | FXMVECTOR V |
||
| 3303 | ) |
||
| 3304 | { |
||
| 3305 | #if defined(_XM_NO_INTRINSICS_) |
||
| 3306 | |||
| 3307 | XMVECTOR N; |
||
| 3308 | static CONST XMVECTORF32 Scale = {65535.0f, 65535.0f, 65535.0f, 65535.0f}; |
||
| 3309 | |||
| 3310 | XMASSERT(pDestination); |
||
| 3311 | |||
| 3312 | N = XMVectorClamp(V, XMVectorZero(), g_XMOne.v); |
||
| 3313 | N = XMVectorMultiplyAdd(N, Scale.v, g_XMOneHalf.v); |
||
| 3314 | N = XMVectorTruncate(N); |
||
| 3315 | |||
| 3316 | pDestination->x = (SHORT)N.vector4_f32[0]; |
||
| 3317 | pDestination->y = (SHORT)N.vector4_f32[1]; |
||
| 3318 | |||
| 3319 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 3320 | XMASSERT(pDestination); |
||
| 3321 | static CONST XMVECTORF32 Scale = {65535.0f, 65535.0f, 65535.0f, 65535.0f}; |
||
| 3322 | // Bounds check |
||
| 3323 | XMVECTOR vResult = _mm_max_ps(V,g_XMZero); |
||
| 3324 | vResult = _mm_min_ps(vResult,g_XMOne); |
||
| 3325 | vResult = _mm_mul_ps(vResult,Scale); |
||
| 3326 | // Convert to int with rounding |
||
| 3327 | __m128i vInt = _mm_cvtps_epi32(vResult); |
||
| 3328 | // Since the SSE pack instruction clamps using signed rules, |
||
| 3329 | // manually extract the values to store them to memory |
||
| 3330 | pDestination->x = static_cast<SHORT>(_mm_extract_epi16(vInt,0)); |
||
| 3331 | pDestination->y = static_cast<SHORT>(_mm_extract_epi16(vInt,2)); |
||
| 3332 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3333 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3334 | } |
||
| 3335 | |||
| 3336 | //------------------------------------------------------------------------------ |
||
| 3337 | |||
| 3338 | XMFINLINE VOID XMStoreUShort2 |
||
| 3339 | ( |
||
| 3340 | XMUSHORT2* pDestination, |
||
| 3341 | FXMVECTOR V |
||
| 3342 | ) |
||
| 3343 | { |
||
| 3344 | #if defined(_XM_NO_INTRINSICS_) |
||
| 3345 | |||
| 3346 | XMVECTOR N; |
||
| 3347 | static CONST XMVECTOR Max = {65535.0f, 65535.0f, 65535.0f, 65535.0f}; |
||
| 3348 | |||
| 3349 | XMASSERT(pDestination); |
||
| 3350 | |||
| 3351 | N = XMVectorClamp(V, XMVectorZero(), Max); |
||
| 3352 | N = XMVectorRound(N); |
||
| 3353 | |||
| 3354 | pDestination->x = (SHORT)N.vector4_f32[0]; |
||
| 3355 | pDestination->y = (SHORT)N.vector4_f32[1]; |
||
| 3356 | |||
| 3357 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 3358 | XMASSERT(pDestination); |
||
| 3359 | static CONST XMVECTORF32 Max = {65535.0f, 65535.0f, 65535.0f, 65535.0f}; |
||
| 3360 | // Bounds check |
||
| 3361 | XMVECTOR vResult = _mm_max_ps(V,g_XMZero); |
||
| 3362 | vResult = _mm_min_ps(vResult,Max); |
||
| 3363 | // Convert to int with rounding |
||
| 3364 | __m128i vInt = _mm_cvtps_epi32(vResult); |
||
| 3365 | // Since the SSE pack instruction clamps using signed rules, |
||
| 3366 | // manually extract the values to store them to memory |
||
| 3367 | pDestination->x = static_cast<SHORT>(_mm_extract_epi16(vInt,0)); |
||
| 3368 | pDestination->y = static_cast<SHORT>(_mm_extract_epi16(vInt,2)); |
||
| 3369 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3370 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3371 | } |
||
| 3372 | |||
| 3373 | //------------------------------------------------------------------------------ |
||
| 3374 | |||
| 3375 | XMFINLINE VOID XMStoreInt3 |
||
| 3376 | ( |
||
| 3377 | UINT* pDestination, |
||
| 3378 | FXMVECTOR V |
||
| 3379 | ) |
||
| 3380 | { |
||
| 3381 | #if defined(_XM_NO_INTRINSICS_) |
||
| 3382 | |||
| 3383 | XMASSERT(pDestination); |
||
| 3384 | XMASSERT(((UINT_PTR)pDestination & 3) == 0); |
||
| 3385 | |||
| 3386 | pDestination[0] = V.vector4_u32[0]; |
||
| 3387 | pDestination[1] = V.vector4_u32[1]; |
||
| 3388 | pDestination[2] = V.vector4_u32[2]; |
||
| 3389 | |||
| 3390 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 3391 | |||
| 3392 | XMASSERT(pDestination); |
||
| 3393 | XMASSERT(((UINT_PTR)pDestination & 3) == 0); |
||
| 3394 | pDestination[0] = XMVectorGetIntX( V ); |
||
| 3395 | pDestination[1] = XMVectorGetIntY( V ); |
||
| 3396 | pDestination[2] = XMVectorGetIntZ( V ); |
||
| 3397 | |||
| 3398 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3399 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3400 | } |
||
| 3401 | |||
| 3402 | //------------------------------------------------------------------------------ |
||
| 3403 | |||
| 3404 | XMFINLINE VOID XMStoreInt3A |
||
| 3405 | ( |
||
| 3406 | UINT* pDestination, |
||
| 3407 | FXMVECTOR V |
||
| 3408 | ) |
||
| 3409 | { |
||
| 3410 | #if defined(_XM_NO_INTRINSICS_) |
||
| 3411 | |||
| 3412 | XMASSERT(pDestination); |
||
| 3413 | XMASSERT(((UINT_PTR)pDestination & 0xF) == 0); |
||
| 3414 | |||
| 3415 | pDestination[0] = V.vector4_u32[0]; |
||
| 3416 | pDestination[1] = V.vector4_u32[1]; |
||
| 3417 | pDestination[2] = V.vector4_u32[2]; |
||
| 3418 | |||
| 3419 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 3420 | |||
| 3421 | XMASSERT(pDestination); |
||
| 3422 | XMASSERT(((UINT_PTR)pDestination & 3) == 0); |
||
| 3423 | pDestination[0] = XMVectorGetIntX( V ); |
||
| 3424 | pDestination[1] = XMVectorGetIntY( V ); |
||
| 3425 | pDestination[2] = XMVectorGetIntZ( V ); |
||
| 3426 | |||
| 3427 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3428 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3429 | } |
||
| 3430 | |||
| 3431 | //------------------------------------------------------------------------------ |
||
| 3432 | |||
| 3433 | XMFINLINE VOID XMStoreFloat3 |
||
| 3434 | ( |
||
| 3435 | XMFLOAT3* pDestination, |
||
| 3436 | FXMVECTOR V |
||
| 3437 | ) |
||
| 3438 | { |
||
| 3439 | #if defined(_XM_NO_INTRINSICS_) |
||
| 3440 | |||
| 3441 | XMASSERT(pDestination); |
||
| 3442 | XMASSERT(((UINT_PTR)pDestination & 3) == 0); |
||
| 3443 | |||
| 3444 | pDestination->x = V.vector4_f32[0]; |
||
| 3445 | pDestination->y = V.vector4_f32[1]; |
||
| 3446 | pDestination->z = V.vector4_f32[2]; |
||
| 3447 | |||
| 3448 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 3449 | |||
| 3450 | XMASSERT(pDestination); |
||
| 3451 | XMASSERT(((UINT_PTR)pDestination & 3) == 0); |
||
| 3452 | |||
| 3453 | XMVECTOR T1 = _mm_shuffle_ps(V,V,_MM_SHUFFLE(1,1,1,1)); |
||
| 3454 | XMVECTOR T2 = _mm_shuffle_ps(V,V,_MM_SHUFFLE(2,2,2,2)); |
||
| 3455 | _mm_store_ss( &pDestination->x, V ); |
||
| 3456 | _mm_store_ss( &pDestination->y, T1 ); |
||
| 3457 | _mm_store_ss( &pDestination->z, T2 ); |
||
| 3458 | |||
| 3459 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3460 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3461 | } |
||
| 3462 | |||
| 3463 | //------------------------------------------------------------------------------ |
||
| 3464 | |||
| 3465 | XMFINLINE VOID XMStoreFloat3A |
||
| 3466 | ( |
||
| 3467 | XMFLOAT3A* pDestination, |
||
| 3468 | FXMVECTOR V |
||
| 3469 | ) |
||
| 3470 | { |
||
| 3471 | #if defined(_XM_NO_INTRINSICS_) |
||
| 3472 | |||
| 3473 | XMASSERT(pDestination); |
||
| 3474 | XMASSERT(((UINT_PTR)pDestination & 0xF) == 0); |
||
| 3475 | |||
| 3476 | pDestination->x = V.vector4_f32[0]; |
||
| 3477 | pDestination->y = V.vector4_f32[1]; |
||
| 3478 | pDestination->z = V.vector4_f32[2]; |
||
| 3479 | |||
| 3480 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 3481 | |||
| 3482 | XMASSERT(pDestination); |
||
| 3483 | XMASSERT(((UINT_PTR)pDestination & 0xF) == 0); |
||
| 3484 | |||
| 3485 | XMVECTOR T1 = _mm_shuffle_ps( V, V, _MM_SHUFFLE( 1, 1, 1, 1 ) ); |
||
| 3486 | XMVECTOR T2 = _mm_unpackhi_ps( V, V ); |
||
| 3487 | _mm_store_ss( &pDestination->x, V ); |
||
| 3488 | _mm_store_ss( &pDestination->y, T1 ); |
||
| 3489 | _mm_store_ss( &pDestination->z, T2 ); |
||
| 3490 | |||
| 3491 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3492 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3493 | } |
||
| 3494 | |||
| 3495 | //------------------------------------------------------------------------------ |
||
| 3496 | |||
| 3497 | XMFINLINE VOID XMStoreUHenDN3 |
||
| 3498 | ( |
||
| 3499 | XMUHENDN3* pDestination, |
||
| 3500 | FXMVECTOR V |
||
| 3501 | ) |
||
| 3502 | { |
||
| 3503 | #if defined(_XM_NO_INTRINSICS_) |
||
| 3504 | |||
| 3505 | XMVECTOR N; |
||
| 3506 | static CONST XMVECTORF32 Scale = {2047.0f, 2047.0f, 1023.0f, 0.0f}; |
||
| 3507 | |||
| 3508 | XMASSERT(pDestination); |
||
| 3509 | |||
| 3510 | N = XMVectorClamp(V, XMVectorZero(), g_XMOne.v); |
||
| 3511 | N = XMVectorMultiply(N, Scale.v); |
||
| 3512 | |||
| 3513 | pDestination->v = (((UINT)N.vector4_f32[2] & 0x3FF) << 22) | |
||
| 3514 | (((UINT)N.vector4_f32[1] & 0x7FF) << 11) | |
||
| 3515 | (((UINT)N.vector4_f32[0] & 0x7FF)); |
||
| 3516 | |||
| 3517 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 3518 | XMASSERT(pDestination); |
||
| 3519 | static const XMVECTORF32 ScaleUHenDN3 = {2047.0f, 2047.0f*2048.0f,1023.0f*(2048.0f*2048.0f)/2.0f,1.0f}; |
||
| 3520 | static const XMVECTORI32 MaskUHenDN3 = {0x7FF,0x7FF<<11,0x3FF<<(22-1),0}; |
||
| 3521 | // Clamp to bounds |
||
| 3522 | XMVECTOR vResult = _mm_max_ps(V,g_XMZero); |
||
| 3523 | vResult = _mm_min_ps(vResult,g_XMOne); |
||
| 3524 | // Scale by multiplication |
||
| 3525 | vResult = _mm_mul_ps(vResult,ScaleUHenDN3); |
||
| 3526 | // Convert to int |
||
| 3527 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 3528 | // Mask off any fraction |
||
| 3529 | vResulti = _mm_and_si128(vResulti,MaskUHenDN3); |
||
| 3530 | // Do a horizontal or of 3 entries |
||
| 3531 | __m128i vResulti2 = _mm_shuffle_epi32(vResulti,_MM_SHUFFLE(0,3,2,1)); |
||
| 3532 | // i = x|y |
||
| 3533 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 3534 | // Move Z to the x position |
||
| 3535 | vResulti2 = _mm_shuffle_epi32(vResulti2,_MM_SHUFFLE(0,3,2,1)); |
||
| 3536 | // Add Z to itself to perform a single bit left shift |
||
| 3537 | vResulti2 = _mm_add_epi32(vResulti2,vResulti2); |
||
| 3538 | // i = x|y|z |
||
| 3539 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 3540 | _mm_store_ss(reinterpret_cast<float *>(&pDestination->v),reinterpret_cast<const __m128 *>(&vResulti)[0]); |
||
| 3541 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3542 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3543 | } |
||
| 3544 | |||
| 3545 | //------------------------------------------------------------------------------ |
||
| 3546 | |||
| 3547 | XMFINLINE VOID XMStoreUHenD3 |
||
| 3548 | ( |
||
| 3549 | XMUHEND3* pDestination, |
||
| 3550 | FXMVECTOR V |
||
| 3551 | ) |
||
| 3552 | { |
||
| 3553 | #if defined(_XM_NO_INTRINSICS_) |
||
| 3554 | |||
| 3555 | XMVECTOR N; |
||
| 3556 | static CONST XMVECTOR Max = {2047.0f, 2047.0f, 1023.0f, 0.0f}; |
||
| 3557 | |||
| 3558 | XMASSERT(pDestination); |
||
| 3559 | |||
| 3560 | N = XMVectorClamp(V, XMVectorZero(), Max); |
||
| 3561 | |||
| 3562 | pDestination->v = (((UINT)N.vector4_f32[2] & 0x3FF) << 22) | |
||
| 3563 | (((UINT)N.vector4_f32[1] & 0x7FF) << 11) | |
||
| 3564 | (((UINT)N.vector4_f32[0] & 0x7FF)); |
||
| 3565 | |||
| 3566 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 3567 | XMASSERT(pDestination); |
||
| 3568 | static const XMVECTORF32 MaxUHenD3 = { 2047.0f, 2047.0f, 1023.0f, 1.0f}; |
||
| 3569 | static const XMVECTORF32 ScaleUHenD3 = {1.0f, 2048.0f,(2048.0f*2048.0f)/2.0f,1.0f}; |
||
| 3570 | static const XMVECTORI32 MaskUHenD3 = {0x7FF,0x7FF<<11,0x3FF<<(22-1),0}; |
||
| 3571 | // Clamp to bounds |
||
| 3572 | XMVECTOR vResult = _mm_max_ps(V,g_XMZero); |
||
| 3573 | vResult = _mm_min_ps(vResult,MaxUHenD3); |
||
| 3574 | // Scale by multiplication |
||
| 3575 | vResult = _mm_mul_ps(vResult,ScaleUHenD3); |
||
| 3576 | // Convert to int |
||
| 3577 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 3578 | // Mask off any fraction |
||
| 3579 | vResulti = _mm_and_si128(vResulti,MaskUHenD3); |
||
| 3580 | // Do a horizontal or of 3 entries |
||
| 3581 | __m128i vResulti2 = _mm_shuffle_epi32(vResulti,_MM_SHUFFLE(0,3,2,1)); |
||
| 3582 | // i = x|y |
||
| 3583 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 3584 | // Move Z to the x position |
||
| 3585 | vResulti2 = _mm_shuffle_epi32(vResulti2,_MM_SHUFFLE(0,3,2,1)); |
||
| 3586 | // Add Z to itself to perform a single bit left shift |
||
| 3587 | vResulti2 = _mm_add_epi32(vResulti2,vResulti2); |
||
| 3588 | // i = x|y|z |
||
| 3589 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 3590 | _mm_store_ss(reinterpret_cast<float *>(&pDestination->v),reinterpret_cast<const __m128 *>(&vResulti)[0]); |
||
| 3591 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3592 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3593 | } |
||
| 3594 | |||
| 3595 | //------------------------------------------------------------------------------ |
||
| 3596 | |||
| 3597 | XMFINLINE VOID XMStoreHenDN3 |
||
| 3598 | ( |
||
| 3599 | XMHENDN3* pDestination, |
||
| 3600 | FXMVECTOR V |
||
| 3601 | ) |
||
| 3602 | { |
||
| 3603 | #if defined(_XM_NO_INTRINSICS_) |
||
| 3604 | |||
| 3605 | XMVECTOR N; |
||
| 3606 | static CONST XMVECTORF32 Scale = {1023.0f, 1023.0f, 511.0f, 1.0f}; |
||
| 3607 | |||
| 3608 | XMASSERT(pDestination); |
||
| 3609 | |||
| 3610 | N = XMVectorClamp(V, g_XMNegativeOne.v, g_XMOne.v); |
||
| 3611 | N = XMVectorMultiply(N, Scale.v); |
||
| 3612 | |||
| 3613 | pDestination->v = (((INT)N.vector4_f32[2] & 0x3FF) << 22) | |
||
| 3614 | (((INT)N.vector4_f32[1] & 0x7FF) << 11) | |
||
| 3615 | (((INT)N.vector4_f32[0] & 0x7FF)); |
||
| 3616 | |||
| 3617 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 3618 | XMASSERT(pDestination); |
||
| 3619 | static const XMVECTORF32 ScaleHenDN3 = {1023.0f, 1023.0f*2048.0f,511.0f*(2048.0f*2048.0f),1.0f}; |
||
| 3620 | // Clamp to bounds |
||
| 3621 | XMVECTOR vResult = _mm_max_ps(V,g_XMNegativeOne); |
||
| 3622 | vResult = _mm_min_ps(vResult,g_XMOne); |
||
| 3623 | // Scale by multiplication |
||
| 3624 | vResult = _mm_mul_ps(vResult,ScaleHenDN3); |
||
| 3625 | // Convert to int |
||
| 3626 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 3627 | // Mask off any fraction |
||
| 3628 | vResulti = _mm_and_si128(vResulti,g_XMMaskHenD3); |
||
| 3629 | // Do a horizontal or of all 4 entries |
||
| 3630 | vResult = _mm_shuffle_ps(reinterpret_cast<const __m128 *>(&vResulti)[0],reinterpret_cast<const __m128 *>(&vResulti)[0],_MM_SHUFFLE(0,3,2,1)); |
||
| 3631 | vResulti = _mm_or_si128(vResulti,reinterpret_cast<const __m128i *>(&vResult)[0]); |
||
| 3632 | vResult = _mm_shuffle_ps(vResult,vResult,_MM_SHUFFLE(0,3,2,1)); |
||
| 3633 | vResulti = _mm_or_si128(vResulti,reinterpret_cast<const __m128i *>(&vResult)[0]); |
||
| 3634 | _mm_store_ss(reinterpret_cast<float *>(&pDestination->v),reinterpret_cast<const __m128 *>(&vResulti)[0]); |
||
| 3635 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3636 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3637 | } |
||
| 3638 | |||
| 3639 | //------------------------------------------------------------------------------ |
||
| 3640 | |||
| 3641 | XMFINLINE VOID XMStoreHenD3 |
||
| 3642 | ( |
||
| 3643 | XMHEND3* pDestination, |
||
| 3644 | FXMVECTOR V |
||
| 3645 | ) |
||
| 3646 | { |
||
| 3647 | #if defined(_XM_NO_INTRINSICS_) |
||
| 3648 | |||
| 3649 | XMVECTOR N; |
||
| 3650 | static CONST XMVECTOR Min = {-1023.0f, -1023.0f, -511.0f, -1.0f}; |
||
| 3651 | static CONST XMVECTOR Max = {1023.0f, 1023.0f, 511.0f, 1.0f}; |
||
| 3652 | |||
| 3653 | XMASSERT(pDestination); |
||
| 3654 | |||
| 3655 | N = XMVectorClamp(V, Min, Max); |
||
| 3656 | |||
| 3657 | pDestination->v = (((INT)N.vector4_f32[2] & 0x3FF) << 22) | |
||
| 3658 | (((INT)N.vector4_f32[1] & 0x7FF) << 11) | |
||
| 3659 | (((INT)N.vector4_f32[0] & 0x7FF)); |
||
| 3660 | |||
| 3661 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 3662 | XMASSERT(pDestination); |
||
| 3663 | static const XMVECTORF32 MinHenD3 = {-1023.0f,-1023.0f,-511.0f,-1.0f}; |
||
| 3664 | static const XMVECTORF32 MaxHenD3 = { 1023.0f, 1023.0f, 511.0f, 1.0f}; |
||
| 3665 | static const XMVECTORF32 ScaleHenD3 = {1.0f, 2048.0f,(2048.0f*2048.0f),1.0f}; |
||
| 3666 | // Clamp to bounds |
||
| 3667 | XMVECTOR vResult = _mm_max_ps(V,MinHenD3); |
||
| 3668 | vResult = _mm_min_ps(vResult,MaxHenD3); |
||
| 3669 | // Scale by multiplication |
||
| 3670 | vResult = _mm_mul_ps(vResult,ScaleHenD3); |
||
| 3671 | // Convert to int |
||
| 3672 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 3673 | // Mask off any fraction |
||
| 3674 | vResulti = _mm_and_si128(vResulti,g_XMMaskHenD3); |
||
| 3675 | // Do a horizontal or of all 4 entries |
||
| 3676 | vResult = _mm_shuffle_ps(reinterpret_cast<const __m128 *>(&vResulti)[0],reinterpret_cast<const __m128 *>(&vResulti)[0],_MM_SHUFFLE(0,3,2,1)); |
||
| 3677 | vResulti = _mm_or_si128(vResulti,reinterpret_cast<const __m128i *>(&vResult)[0]); |
||
| 3678 | vResult = _mm_shuffle_ps(vResult,vResult,_MM_SHUFFLE(0,3,2,1)); |
||
| 3679 | vResulti = _mm_or_si128(vResulti,reinterpret_cast<const __m128i *>(&vResult)[0]); |
||
| 3680 | _mm_store_ss(reinterpret_cast<float *>(&pDestination->v),reinterpret_cast<const __m128 *>(&vResulti)[0]); |
||
| 3681 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3682 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3683 | } |
||
| 3684 | |||
| 3685 | //------------------------------------------------------------------------------ |
||
| 3686 | |||
| 3687 | XMFINLINE VOID XMStoreUDHenN3 |
||
| 3688 | ( |
||
| 3689 | XMUDHENN3* pDestination, |
||
| 3690 | FXMVECTOR V |
||
| 3691 | ) |
||
| 3692 | { |
||
| 3693 | #if defined(_XM_NO_INTRINSICS_) |
||
| 3694 | |||
| 3695 | XMVECTOR N; |
||
| 3696 | static CONST XMVECTORF32 Scale = {1023.0f, 2047.0f, 2047.0f, 0.0f}; |
||
| 3697 | |||
| 3698 | XMASSERT(pDestination); |
||
| 3699 | |||
| 3700 | N = XMVectorClamp(V, XMVectorZero(), g_XMOne.v); |
||
| 3701 | N = XMVectorMultiply(N, Scale.v); |
||
| 3702 | |||
| 3703 | pDestination->v = (((UINT)N.vector4_f32[2] & 0x7FF) << 21) | |
||
| 3704 | (((UINT)N.vector4_f32[1] & 0x7FF) << 10) | |
||
| 3705 | (((UINT)N.vector4_f32[0] & 0x3FF)); |
||
| 3706 | |||
| 3707 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 3708 | XMASSERT(pDestination); |
||
| 3709 | static const XMVECTORF32 ScaleUDHenN3 = {1023.0f,2047.0f*1024.0f,2047.0f*(1024.0f*2048.0f)/2.0f,1.0f}; |
||
| 3710 | static const XMVECTORI32 MaskUDHenN3 = {0x3FF,0x7FF<<10,0x7FF<<(21-1),0}; |
||
| 3711 | // Clamp to bounds |
||
| 3712 | XMVECTOR vResult = _mm_max_ps(V,g_XMZero); |
||
| 3713 | vResult = _mm_min_ps(vResult,g_XMOne); |
||
| 3714 | // Scale by multiplication |
||
| 3715 | vResult = _mm_mul_ps(vResult,ScaleUDHenN3); |
||
| 3716 | // Convert to int |
||
| 3717 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 3718 | // Mask off any fraction |
||
| 3719 | vResulti = _mm_and_si128(vResulti,MaskUDHenN3); |
||
| 3720 | // Do a horizontal or of 3 entries |
||
| 3721 | __m128i vResulti2 = _mm_shuffle_epi32(vResulti,_MM_SHUFFLE(0,3,2,1)); |
||
| 3722 | // i = x|y |
||
| 3723 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 3724 | // Move Z to the x position |
||
| 3725 | vResulti2 = _mm_shuffle_epi32(vResulti2,_MM_SHUFFLE(0,3,2,1)); |
||
| 3726 | // Add Z to itself to perform a single bit left shift |
||
| 3727 | vResulti2 = _mm_add_epi32(vResulti2,vResulti2); |
||
| 3728 | // i = x|y|z |
||
| 3729 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 3730 | _mm_store_ss(reinterpret_cast<float *>(&pDestination->v),reinterpret_cast<const __m128 *>(&vResulti)[0]); |
||
| 3731 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3732 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3733 | } |
||
| 3734 | |||
| 3735 | //------------------------------------------------------------------------------ |
||
| 3736 | |||
| 3737 | XMFINLINE VOID XMStoreUDHen3 |
||
| 3738 | ( |
||
| 3739 | XMUDHEN3* pDestination, |
||
| 3740 | FXMVECTOR V |
||
| 3741 | ) |
||
| 3742 | { |
||
| 3743 | #if defined(_XM_NO_INTRINSICS_) |
||
| 3744 | |||
| 3745 | XMVECTOR N; |
||
| 3746 | static CONST XMVECTOR Max = {1023.0f, 2047.0f, 2047.0f, 0.0f}; |
||
| 3747 | |||
| 3748 | XMASSERT(pDestination); |
||
| 3749 | |||
| 3750 | N = XMVectorClamp(V, XMVectorZero(), Max); |
||
| 3751 | |||
| 3752 | pDestination->v = (((UINT)N.vector4_f32[2] & 0x7FF) << 21) | |
||
| 3753 | (((UINT)N.vector4_f32[1] & 0x7FF) << 10) | |
||
| 3754 | (((UINT)N.vector4_f32[0] & 0x3FF)); |
||
| 3755 | |||
| 3756 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 3757 | XMASSERT(pDestination); |
||
| 3758 | static const XMVECTORF32 MaxUDHen3 = { 1023.0f, 2047.0f, 2047.0f, 1.0f}; |
||
| 3759 | static const XMVECTORF32 ScaleUDHen3 = {1.0f, 1024.0f,(1024.0f*2048.0f)/2.0f,1.0f}; |
||
| 3760 | static const XMVECTORI32 MaskUDHen3 = {0x3FF,0x7FF<<10,0x7FF<<(21-1),0}; |
||
| 3761 | // Clamp to bounds |
||
| 3762 | XMVECTOR vResult = _mm_max_ps(V,g_XMZero); |
||
| 3763 | vResult = _mm_min_ps(vResult,MaxUDHen3); |
||
| 3764 | // Scale by multiplication |
||
| 3765 | vResult = _mm_mul_ps(vResult,ScaleUDHen3); |
||
| 3766 | // Convert to int |
||
| 3767 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 3768 | // Mask off any fraction |
||
| 3769 | vResulti = _mm_and_si128(vResulti,MaskUDHen3); |
||
| 3770 | // Do a horizontal or of 3 entries |
||
| 3771 | __m128i vResulti2 = _mm_shuffle_epi32(vResulti,_MM_SHUFFLE(0,3,2,1)); |
||
| 3772 | // i = x|y |
||
| 3773 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 3774 | // Move Z to the x position |
||
| 3775 | vResulti2 = _mm_shuffle_epi32(vResulti2,_MM_SHUFFLE(0,3,2,1)); |
||
| 3776 | // Add Z to itself to perform a single bit left shift |
||
| 3777 | vResulti2 = _mm_add_epi32(vResulti2,vResulti2); |
||
| 3778 | // i = x|y|z |
||
| 3779 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 3780 | _mm_store_ss(reinterpret_cast<float *>(&pDestination->v),reinterpret_cast<const __m128 *>(&vResulti)[0]); |
||
| 3781 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3782 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3783 | } |
||
| 3784 | |||
| 3785 | //------------------------------------------------------------------------------ |
||
| 3786 | |||
| 3787 | XMFINLINE VOID XMStoreDHenN3 |
||
| 3788 | ( |
||
| 3789 | XMDHENN3* pDestination, |
||
| 3790 | FXMVECTOR V |
||
| 3791 | ) |
||
| 3792 | { |
||
| 3793 | #if defined(_XM_NO_INTRINSICS_) |
||
| 3794 | |||
| 3795 | XMVECTOR N; |
||
| 3796 | static CONST XMVECTORF32 Scale = {511.0f, 1023.0f, 1023.0f, 1.0f}; |
||
| 3797 | |||
| 3798 | XMASSERT(pDestination); |
||
| 3799 | |||
| 3800 | N = XMVectorClamp(V, g_XMNegativeOne.v, g_XMOne.v); |
||
| 3801 | N = XMVectorMultiply(N, Scale.v); |
||
| 3802 | |||
| 3803 | pDestination->v = (((INT)N.vector4_f32[2] & 0x7FF) << 21) | |
||
| 3804 | (((INT)N.vector4_f32[1] & 0x7FF) << 10) | |
||
| 3805 | (((INT)N.vector4_f32[0] & 0x3FF)); |
||
| 3806 | |||
| 3807 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 3808 | XMASSERT(pDestination); |
||
| 3809 | static const XMVECTORF32 ScaleDHenN3 = {511.0f, 1023.0f*1024.0f,1023.0f*(1024.0f*2048.0f),1.0f}; |
||
| 3810 | // Clamp to bounds |
||
| 3811 | XMVECTOR vResult = _mm_max_ps(V,g_XMNegativeOne); |
||
| 3812 | vResult = _mm_min_ps(vResult,g_XMOne); |
||
| 3813 | // Scale by multiplication |
||
| 3814 | vResult = _mm_mul_ps(vResult,ScaleDHenN3); |
||
| 3815 | // Convert to int |
||
| 3816 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 3817 | // Mask off any fraction |
||
| 3818 | vResulti = _mm_and_si128(vResulti,g_XMMaskDHen3); |
||
| 3819 | // Do a horizontal or of all 4 entries |
||
| 3820 | vResult = _mm_shuffle_ps(reinterpret_cast<const __m128 *>(&vResulti)[0],reinterpret_cast<const __m128 *>(&vResulti)[0],_MM_SHUFFLE(0,3,2,1)); |
||
| 3821 | vResulti = _mm_or_si128(vResulti,reinterpret_cast<const __m128i *>(&vResult)[0]); |
||
| 3822 | vResult = _mm_shuffle_ps(vResult,vResult,_MM_SHUFFLE(0,3,2,1)); |
||
| 3823 | vResulti = _mm_or_si128(vResulti,reinterpret_cast<const __m128i *>(&vResult)[0]); |
||
| 3824 | _mm_store_ss(reinterpret_cast<float *>(&pDestination->v),reinterpret_cast<const __m128 *>(&vResulti)[0]); |
||
| 3825 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3826 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3827 | } |
||
| 3828 | |||
| 3829 | //------------------------------------------------------------------------------ |
||
| 3830 | |||
| 3831 | XMFINLINE VOID XMStoreDHen3 |
||
| 3832 | ( |
||
| 3833 | XMDHEN3* pDestination, |
||
| 3834 | FXMVECTOR V |
||
| 3835 | ) |
||
| 3836 | { |
||
| 3837 | #if defined(_XM_NO_INTRINSICS_) |
||
| 3838 | |||
| 3839 | XMVECTOR N; |
||
| 3840 | static CONST XMVECTOR Min = {-511.0f, -1023.0f, -1023.0f, -1.0f}; |
||
| 3841 | static CONST XMVECTOR Max = {511.0f, 1023.0f, 1023.0f, 1.0f}; |
||
| 3842 | |||
| 3843 | XMASSERT(pDestination); |
||
| 3844 | |||
| 3845 | N = XMVectorClamp(V, Min, Max); |
||
| 3846 | |||
| 3847 | pDestination->v = (((INT)N.vector4_f32[2] & 0x7FF) << 21) | |
||
| 3848 | (((INT)N.vector4_f32[1] & 0x7FF) << 10) | |
||
| 3849 | (((INT)N.vector4_f32[0] & 0x3FF)); |
||
| 3850 | |||
| 3851 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 3852 | XMASSERT(pDestination); |
||
| 3853 | static const XMVECTORF32 MinDHen3 = {-511.0f,-1023.0f,-1023.0f,-1.0f}; |
||
| 3854 | static const XMVECTORF32 MaxDHen3 = { 511.0f, 1023.0f, 1023.0f, 1.0f}; |
||
| 3855 | static const XMVECTORF32 ScaleDHen3 = {1.0f, 1024.0f,(1024.0f*2048.0f),1.0f}; |
||
| 3856 | // Clamp to bounds |
||
| 3857 | XMVECTOR vResult = _mm_max_ps(V,MinDHen3); |
||
| 3858 | vResult = _mm_min_ps(vResult,MaxDHen3); |
||
| 3859 | // Scale by multiplication |
||
| 3860 | vResult = _mm_mul_ps(vResult,ScaleDHen3); |
||
| 3861 | // Convert to int |
||
| 3862 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 3863 | // Mask off any fraction |
||
| 3864 | vResulti = _mm_and_si128(vResulti,g_XMMaskDHen3); |
||
| 3865 | // Do a horizontal or of all 4 entries |
||
| 3866 | vResult = _mm_shuffle_ps(reinterpret_cast<const __m128 *>(&vResulti)[0],reinterpret_cast<const __m128 *>(&vResulti)[0],_MM_SHUFFLE(0,3,2,1)); |
||
| 3867 | vResulti = _mm_or_si128(vResulti,reinterpret_cast<const __m128i *>(&vResult)[0]); |
||
| 3868 | vResult = _mm_shuffle_ps(vResult,vResult,_MM_SHUFFLE(0,3,2,1)); |
||
| 3869 | vResulti = _mm_or_si128(vResulti,reinterpret_cast<const __m128i *>(&vResult)[0]); |
||
| 3870 | _mm_store_ss(reinterpret_cast<float *>(&pDestination->v),reinterpret_cast<const __m128 *>(&vResulti)[0]); |
||
| 3871 | #else // _XM_VMX128_INTRINSICS_ |
||
| 3872 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 3873 | } |
||
| 3874 | |||
| 3875 | //------------------------------------------------------------------------------ |
||
| 3876 | |||
| 3877 | XMFINLINE VOID XMStoreU565 |
||
| 3878 | ( |
||
| 3879 | XMU565* pDestination, |
||
| 3880 | FXMVECTOR V |
||
| 3881 | ) |
||
| 3882 | { |
||
| 3883 | #if defined(_XM_SSE_INTRINSICS_) && !defined(_XM_NO_INTRINSICS_) |
||
| 3884 | XMASSERT(pDestination); |
||
| 3885 | static CONST XMVECTORF32 Max = {31.0f, 63.0f, 31.0f, 0.0f}; |
||
| 3886 | // Bounds check |
||
| 3887 | XMVECTOR vResult = _mm_max_ps(V,g_XMZero); |
||
| 3888 | vResult = _mm_min_ps(vResult,Max); |
||
| 3889 | // Convert to int with rounding |
||
| 3890 | __m128i vInt = _mm_cvtps_epi32(vResult); |
||
| 3891 | // No SSE operations will write to 16-bit values, so we have to extract them manually |
||
| 3892 | USHORT x = static_cast<USHORT>(_mm_extract_epi16(vInt,0)); |
||
| 3893 | USHORT y = static_cast<USHORT>(_mm_extract_epi16(vInt,2)); |
||
| 3894 | USHORT z = static_cast<USHORT>(_mm_extract_epi16(vInt,4)); |
||
| 3895 | pDestination->v = ((z & 0x1F) << 11) | |
||
| 3896 | ((y & 0x3F) << 5) | |
||
| 3897 | ((x & 0x1F)); |
||
| 3898 | #else |
||
| 3899 | XMVECTOR N; |
||
| 3900 | static CONST XMVECTORF32 Max = {31.0f, 63.0f, 31.0f, 0.0f}; |
||
| 3901 | |||
| 3902 | XMASSERT(pDestination); |
||
| 3903 | |||
| 3904 | N = XMVectorClamp(V, XMVectorZero(), Max.v); |
||
| 3905 | N = XMVectorRound(N); |
||
| 3906 | |||
| 3907 | pDestination->v = (((USHORT)N.vector4_f32[2] & 0x1F) << 11) | |
||
| 3908 | (((USHORT)N.vector4_f32[1] & 0x3F) << 5) | |
||
| 3909 | (((USHORT)N.vector4_f32[0] & 0x1F)); |
||
| 3910 | #endif !_XM_SSE_INTRINSICS_ |
||
| 3911 | } |
||
| 3912 | |||
| 3913 | //------------------------------------------------------------------------------ |
||
| 3914 | |||
| 3915 | XMFINLINE VOID XMStoreFloat3PK |
||
| 3916 | ( |
||
| 3917 | XMFLOAT3PK* pDestination, |
||
| 3918 | FXMVECTOR V |
||
| 3919 | ) |
||
| 3920 | { |
||
| 3921 | UINT I, Sign, j; |
||
| 3922 | UINT IValue[3]; |
||
| 3923 | UINT Result[3]; |
||
| 3924 | |||
| 3925 | XMASSERT(pDestination); |
||
| 3926 | |||
| 3927 | XMStoreFloat3( (XMFLOAT3*)&IValue, V ); |
||
| 3928 | |||
| 3929 | // X & Y Channels (5-bit exponent, 6-bit mantissa) |
||
| 3930 | for(j=0; j < 2; ++j) |
||
| 3931 | { |
||
| 3932 | Sign = IValue[j] & 0x80000000; |
||
| 3933 | I = IValue[j] & 0x7FFFFFFF; |
||
| 3934 | |||
| 3935 | if ((I & 0x7F800000) == 0x7F800000) |
||
| 3936 | { |
||
| 3937 | // INF or NAN |
||
| 3938 | Result[j] = 0x7c0; |
||
| 3939 | if (( I & 0x7FFFFF ) != 0) |
||
| 3940 | { |
||
| 3941 | Result[j] = 0x7c0 | (((I>>17)|(I>11)|(I>>6)|(I))&0x3f); |
||
| 3942 | } |
||
| 3943 | else if ( Sign ) |
||
| 3944 | { |
||
| 3945 | // -INF is clamped to 0 since 3PK is positive only |
||
| 3946 | Result[j] = 0; |
||
| 3947 | } |
||
| 3948 | } |
||
| 3949 | else if ( Sign ) |
||
| 3950 | { |
||
| 3951 | // 3PK is positive only, so clamp to zero |
||
| 3952 | Result[j] = 0; |
||
| 3953 | } |
||
| 3954 | else if (I > 0x477E0000U) |
||
| 3955 | { |
||
| 3956 | // The number is too large to be represented as a float11, set to max |
||
| 3957 | Result[j] = 0x7BF; |
||
| 3958 | } |
||
| 3959 | else |
||
| 3960 | { |
||
| 3961 | if (I < 0x38800000U) |
||
| 3962 | { |
||
| 3963 | // The number is too small to be represented as a normalized float11 |
||
| 3964 | // Convert it to a denormalized value. |
||
| 3965 | UINT Shift = 113U - (I >> 23U); |
||
| 3966 | I = (0x800000U | (I & 0x7FFFFFU)) >> Shift; |
||
| 3967 | } |
||
| 3968 | else |
||
| 3969 | { |
||
| 3970 | // Rebias the exponent to represent the value as a normalized float11 |
||
| 3971 | I += 0xC8000000U; |
||
| 3972 | } |
||
| 3973 | |||
| 3974 | Result[j] = ((I + 0xFFFFU + ((I >> 17U) & 1U)) >> 17U)&0x7ffU; |
||
| 3975 | } |
||
| 3976 | } |
||
| 3977 | |||
| 3978 | // Z Channel (5-bit exponent, 5-bit mantissa) |
||
| 3979 | Sign = IValue[2] & 0x80000000; |
||
| 3980 | I = IValue[2] & 0x7FFFFFFF; |
||
| 3981 | |||
| 3982 | if ((I & 0x7F800000) == 0x7F800000) |
||
| 3983 | { |
||
| 3984 | // INF or NAN |
||
| 3985 | Result[2] = 0x3e0; |
||
| 3986 | if ( I & 0x7FFFFF ) |
||
| 3987 | { |
||
| 3988 | Result[2] = 0x3e0 | (((I>>18)|(I>13)|(I>>3)|(I))&0x1f); |
||
| 3989 | } |
||
| 3990 | else if ( Sign ) |
||
| 3991 | { |
||
| 3992 | // -INF is clamped to 0 since 3PK is positive only |
||
| 3993 | Result[2] = 0; |
||
| 3994 | } |
||
| 3995 | } |
||
| 3996 | else if ( Sign ) |
||
| 3997 | { |
||
| 3998 | // 3PK is positive only, so clamp to zero |
||
| 3999 | Result[2] = 0; |
||
| 4000 | } |
||
| 4001 | else if (I > 0x477C0000U) |
||
| 4002 | { |
||
| 4003 | // The number is too large to be represented as a float10, set to max |
||
| 4004 | Result[2] = 0x3df; |
||
| 4005 | } |
||
| 4006 | else |
||
| 4007 | { |
||
| 4008 | if (I < 0x38800000U) |
||
| 4009 | { |
||
| 4010 | // The number is too small to be represented as a normalized float10 |
||
| 4011 | // Convert it to a denormalized value. |
||
| 4012 | UINT Shift = 113U - (I >> 23U); |
||
| 4013 | I = (0x800000U | (I & 0x7FFFFFU)) >> Shift; |
||
| 4014 | } |
||
| 4015 | else |
||
| 4016 | { |
||
| 4017 | // Rebias the exponent to represent the value as a normalized float10 |
||
| 4018 | I += 0xC8000000U; |
||
| 4019 | } |
||
| 4020 | |||
| 4021 | Result[2] = ((I + 0x1FFFFU + ((I >> 18U) & 1U)) >> 18U)&0x3ffU; |
||
| 4022 | } |
||
| 4023 | |||
| 4024 | // Pack Result into memory |
||
| 4025 | pDestination->v = (Result[0] & 0x7ff) |
||
| 4026 | | ( (Result[1] & 0x7ff) << 11 ) |
||
| 4027 | | ( (Result[2] & 0x3ff) << 22 ); |
||
| 4028 | } |
||
| 4029 | |||
| 4030 | |||
| 4031 | //------------------------------------------------------------------------------ |
||
| 4032 | |||
| 4033 | XMFINLINE VOID XMStoreFloat3SE |
||
| 4034 | ( |
||
| 4035 | XMFLOAT3SE* pDestination, |
||
| 4036 | FXMVECTOR V |
||
| 4037 | ) |
||
| 4038 | { |
||
| 4039 | UINT I, Sign, j, T; |
||
| 4040 | UINT IValue[3]; |
||
| 4041 | UINT Frac[3]; |
||
| 4042 | UINT Exp[3]; |
||
| 4043 | |||
| 4044 | XMASSERT(pDestination); |
||
| 4045 | |||
| 4046 | XMStoreFloat3( (XMFLOAT3*)&IValue, V ); |
||
| 4047 | |||
| 4048 | // X, Y, Z Channels (5-bit exponent, 9-bit mantissa) |
||
| 4049 | for(j=0; j < 3; ++j) |
||
| 4050 | { |
||
| 4051 | Sign = IValue[j] & 0x80000000; |
||
| 4052 | I = IValue[j] & 0x7FFFFFFF; |
||
| 4053 | |||
| 4054 | if ((I & 0x7F800000) == 0x7F800000) |
||
| 4055 | { |
||
| 4056 | // INF or NAN |
||
| 4057 | Exp[j] = 0x1f; |
||
| 4058 | if (( I & 0x7FFFFF ) != 0) |
||
| 4059 | { |
||
| 4060 | Frac[j] = ((I>>14)|(I>5)|(I))&0x1ff; |
||
| 4061 | } |
||
| 4062 | else if ( Sign ) |
||
| 4063 | { |
||
| 4064 | // -INF is clamped to 0 since 3SE is positive only |
||
| 4065 | Exp[j] = Frac[j] = 0; |
||
| 4066 | } |
||
| 4067 | } |
||
| 4068 | else if ( Sign ) |
||
| 4069 | { |
||
| 4070 | // 3SE is positive only, so clamp to zero |
||
| 4071 | Exp[j] = Frac[j] = 0; |
||
| 4072 | } |
||
| 4073 | else if (I > 0x477FC000U) |
||
| 4074 | { |
||
| 4075 | // The number is too large, set to max |
||
| 4076 | Exp[j] = 0x1e; |
||
| 4077 | Frac[j] = 0x1ff; |
||
| 4078 | } |
||
| 4079 | else |
||
| 4080 | { |
||
| 4081 | if (I < 0x38800000U) |
||
| 4082 | { |
||
| 4083 | // The number is too small to be represented as a normalized float11 |
||
| 4084 | // Convert it to a denormalized value. |
||
| 4085 | UINT Shift = 113U - (I >> 23U); |
||
| 4086 | I = (0x800000U | (I & 0x7FFFFFU)) >> Shift; |
||
| 4087 | } |
||
| 4088 | else |
||
| 4089 | { |
||
| 4090 | // Rebias the exponent to represent the value as a normalized float11 |
||
| 4091 | I += 0xC8000000U; |
||
| 4092 | } |
||
| 4093 | |||
| 4094 | T = ((I + 0x1FFFU + ((I >> 14U) & 1U)) >> 14U)&0x3fffU; |
||
| 4095 | |||
| 4096 | Exp[j] = (T & 0x3E00) >> 9; |
||
| 4097 | Frac[j] = T & 0x1ff; |
||
| 4098 | } |
||
| 4099 | } |
||
| 4100 | |||
| 4101 | // Adjust to a shared exponent |
||
| 4102 | T = XMMax( Exp[0], XMMax( Exp[1], Exp[2] ) ); |
||
| 4103 | |||
| 4104 | Frac[0] = Frac[0] >> (T - Exp[0]); |
||
| 4105 | Frac[1] = Frac[1] >> (T - Exp[1]); |
||
| 4106 | Frac[2] = Frac[2] >> (T - Exp[2]); |
||
| 4107 | |||
| 4108 | // Store packed into memory |
||
| 4109 | pDestination->xm = Frac[0]; |
||
| 4110 | pDestination->ym = Frac[1]; |
||
| 4111 | pDestination->zm = Frac[2]; |
||
| 4112 | pDestination->e = T; |
||
| 4113 | } |
||
| 4114 | |||
| 4115 | //------------------------------------------------------------------------------ |
||
| 4116 | |||
| 4117 | XMFINLINE VOID XMStoreInt4 |
||
| 4118 | ( |
||
| 4119 | UINT* pDestination, |
||
| 4120 | FXMVECTOR V |
||
| 4121 | ) |
||
| 4122 | { |
||
| 4123 | #if defined(_XM_NO_INTRINSICS_) |
||
| 4124 | |||
| 4125 | XMASSERT(pDestination); |
||
| 4126 | |||
| 4127 | pDestination[0] = V.vector4_u32[0]; |
||
| 4128 | pDestination[1] = V.vector4_u32[1]; |
||
| 4129 | pDestination[2] = V.vector4_u32[2]; |
||
| 4130 | pDestination[3] = V.vector4_u32[3]; |
||
| 4131 | |||
| 4132 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 4133 | XMASSERT(pDestination); |
||
| 4134 | |||
| 4135 | _mm_storeu_si128( (__m128i*)pDestination, reinterpret_cast<const __m128i *>(&V)[0] ); |
||
| 4136 | |||
| 4137 | #else // _XM_VMX128_INTRINSICS_ |
||
| 4138 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 4139 | } |
||
| 4140 | |||
| 4141 | //------------------------------------------------------------------------------ |
||
| 4142 | |||
| 4143 | XMFINLINE VOID XMStoreInt4A |
||
| 4144 | ( |
||
| 4145 | UINT* pDestination, |
||
| 4146 | FXMVECTOR V |
||
| 4147 | ) |
||
| 4148 | { |
||
| 4149 | #if defined(_XM_NO_INTRINSICS_) |
||
| 4150 | |||
| 4151 | XMASSERT(pDestination); |
||
| 4152 | XMASSERT(((UINT_PTR)pDestination & 0xF) == 0); |
||
| 4153 | |||
| 4154 | pDestination[0] = V.vector4_u32[0]; |
||
| 4155 | pDestination[1] = V.vector4_u32[1]; |
||
| 4156 | pDestination[2] = V.vector4_u32[2]; |
||
| 4157 | pDestination[3] = V.vector4_u32[3]; |
||
| 4158 | |||
| 4159 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 4160 | XMASSERT(pDestination); |
||
| 4161 | |||
| 4162 | _mm_store_si128( (__m128i*)pDestination, reinterpret_cast<const __m128i *>(&V)[0] ); |
||
| 4163 | |||
| 4164 | #else // _XM_VMX128_INTRINSICS_ |
||
| 4165 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 4166 | } |
||
| 4167 | |||
| 4168 | //------------------------------------------------------------------------------ |
||
| 4169 | |||
| 4170 | XMFINLINE VOID XMStoreInt4NC |
||
| 4171 | ( |
||
| 4172 | UINT* pDestination, |
||
| 4173 | FXMVECTOR V |
||
| 4174 | ) |
||
| 4175 | { |
||
| 4176 | #if defined(_XM_NO_INTRINSICS_) |
||
| 4177 | |||
| 4178 | XMASSERT(pDestination); |
||
| 4179 | |||
| 4180 | pDestination[0] = V.vector4_u32[0]; |
||
| 4181 | pDestination[1] = V.vector4_u32[1]; |
||
| 4182 | pDestination[2] = V.vector4_u32[2]; |
||
| 4183 | pDestination[3] = V.vector4_u32[3]; |
||
| 4184 | |||
| 4185 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 4186 | XMASSERT(pDestination); |
||
| 4187 | |||
| 4188 | _mm_storeu_si128( (__m128i*)pDestination, reinterpret_cast<const __m128i *>(&V)[0] ); |
||
| 4189 | |||
| 4190 | #else // _XM_VMX128_INTRINSICS_ |
||
| 4191 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 4192 | } |
||
| 4193 | |||
| 4194 | //------------------------------------------------------------------------------ |
||
| 4195 | |||
| 4196 | XMFINLINE VOID XMStoreFloat4 |
||
| 4197 | ( |
||
| 4198 | XMFLOAT4* pDestination, |
||
| 4199 | FXMVECTOR V |
||
| 4200 | ) |
||
| 4201 | { |
||
| 4202 | #if defined(_XM_NO_INTRINSICS_) |
||
| 4203 | |||
| 4204 | XMASSERT(pDestination); |
||
| 4205 | |||
| 4206 | pDestination->x = V.vector4_f32[0]; |
||
| 4207 | pDestination->y = V.vector4_f32[1]; |
||
| 4208 | pDestination->z = V.vector4_f32[2]; |
||
| 4209 | pDestination->w = V.vector4_f32[3]; |
||
| 4210 | |||
| 4211 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 4212 | XMASSERT(pDestination); |
||
| 4213 | |||
| 4214 | _mm_storeu_ps( &pDestination->x, V ); |
||
| 4215 | |||
| 4216 | #else // _XM_VMX128_INTRINSICS_ |
||
| 4217 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 4218 | } |
||
| 4219 | |||
| 4220 | //------------------------------------------------------------------------------ |
||
| 4221 | |||
| 4222 | XMFINLINE VOID XMStoreFloat4A |
||
| 4223 | ( |
||
| 4224 | XMFLOAT4A* pDestination, |
||
| 4225 | FXMVECTOR V |
||
| 4226 | ) |
||
| 4227 | { |
||
| 4228 | #if defined(_XM_NO_INTRINSICS_) |
||
| 4229 | |||
| 4230 | XMASSERT(pDestination); |
||
| 4231 | XMASSERT(((UINT_PTR)pDestination & 0xF) == 0); |
||
| 4232 | |||
| 4233 | pDestination->x = V.vector4_f32[0]; |
||
| 4234 | pDestination->y = V.vector4_f32[1]; |
||
| 4235 | pDestination->z = V.vector4_f32[2]; |
||
| 4236 | pDestination->w = V.vector4_f32[3]; |
||
| 4237 | |||
| 4238 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 4239 | XMASSERT(pDestination); |
||
| 4240 | XMASSERT(((UINT_PTR)pDestination & 0xF) == 0); |
||
| 4241 | |||
| 4242 | _mm_store_ps( &pDestination->x, V ); |
||
| 4243 | #else // _XM_VMX128_INTRINSICS_ |
||
| 4244 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 4245 | } |
||
| 4246 | |||
| 4247 | //------------------------------------------------------------------------------ |
||
| 4248 | |||
| 4249 | XMFINLINE VOID XMStoreFloat4NC |
||
| 4250 | ( |
||
| 4251 | XMFLOAT4* pDestination, |
||
| 4252 | FXMVECTOR V |
||
| 4253 | ) |
||
| 4254 | { |
||
| 4255 | #if defined(_XM_NO_INTRINSICS_) |
||
| 4256 | |||
| 4257 | XMASSERT(pDestination); |
||
| 4258 | |||
| 4259 | pDestination->x = V.vector4_f32[0]; |
||
| 4260 | pDestination->y = V.vector4_f32[1]; |
||
| 4261 | pDestination->z = V.vector4_f32[2]; |
||
| 4262 | pDestination->w = V.vector4_f32[3]; |
||
| 4263 | |||
| 4264 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 4265 | XMASSERT(pDestination); |
||
| 4266 | |||
| 4267 | _mm_storeu_ps( &pDestination->x, V ); |
||
| 4268 | |||
| 4269 | #else // _XM_VMX128_INTRINSICS_ |
||
| 4270 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 4271 | } |
||
| 4272 | |||
| 4273 | //------------------------------------------------------------------------------ |
||
| 4274 | |||
| 4275 | XMFINLINE VOID XMStoreHalf4 |
||
| 4276 | ( |
||
| 4277 | XMHALF4* pDestination, |
||
| 4278 | FXMVECTOR V |
||
| 4279 | ) |
||
| 4280 | { |
||
| 4281 | #if defined(_XM_NO_INTRINSICS_) |
||
| 4282 | |||
| 4283 | XMASSERT(pDestination); |
||
| 4284 | |||
| 4285 | pDestination->x = XMConvertFloatToHalf(V.vector4_f32[0]); |
||
| 4286 | pDestination->y = XMConvertFloatToHalf(V.vector4_f32[1]); |
||
| 4287 | pDestination->z = XMConvertFloatToHalf(V.vector4_f32[2]); |
||
| 4288 | pDestination->w = XMConvertFloatToHalf(V.vector4_f32[3]); |
||
| 4289 | |||
| 4290 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 4291 | XMASSERT(pDestination); |
||
| 4292 | pDestination->x = XMConvertFloatToHalf(XMVectorGetX(V)); |
||
| 4293 | pDestination->y = XMConvertFloatToHalf(XMVectorGetY(V)); |
||
| 4294 | pDestination->z = XMConvertFloatToHalf(XMVectorGetZ(V)); |
||
| 4295 | pDestination->w = XMConvertFloatToHalf(XMVectorGetW(V)); |
||
| 4296 | #else // _XM_VMX128_INTRINSICS_ |
||
| 4297 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 4298 | } |
||
| 4299 | |||
| 4300 | //------------------------------------------------------------------------------ |
||
| 4301 | |||
| 4302 | XMFINLINE VOID XMStoreShortN4 |
||
| 4303 | ( |
||
| 4304 | XMSHORTN4* pDestination, |
||
| 4305 | FXMVECTOR V |
||
| 4306 | ) |
||
| 4307 | { |
||
| 4308 | #if defined(_XM_NO_INTRINSICS_) |
||
| 4309 | |||
| 4310 | XMVECTOR N; |
||
| 4311 | static CONST XMVECTORF32 Scale = {32767.0f, 32767.0f, 32767.0f, 32767.0f}; |
||
| 4312 | |||
| 4313 | XMASSERT(pDestination); |
||
| 4314 | |||
| 4315 | N = XMVectorClamp(V, g_XMNegativeOne.v, g_XMOne.v); |
||
| 4316 | N = XMVectorMultiply(N, Scale.v); |
||
| 4317 | N = XMVectorRound(N); |
||
| 4318 | |||
| 4319 | pDestination->x = (SHORT)N.vector4_f32[0]; |
||
| 4320 | pDestination->y = (SHORT)N.vector4_f32[1]; |
||
| 4321 | pDestination->z = (SHORT)N.vector4_f32[2]; |
||
| 4322 | pDestination->w = (SHORT)N.vector4_f32[3]; |
||
| 4323 | |||
| 4324 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 4325 | XMASSERT(pDestination); |
||
| 4326 | static CONST XMVECTORF32 Scale = {32767.0f, 32767.0f, 32767.0f, 32767.0f}; |
||
| 4327 | |||
| 4328 | XMVECTOR vResult = _mm_max_ps(V,g_XMNegativeOne); |
||
| 4329 | vResult = _mm_min_ps(vResult,g_XMOne); |
||
| 4330 | vResult = _mm_mul_ps(vResult,Scale); |
||
| 4331 | __m128i vResulti = _mm_cvtps_epi32(vResult); |
||
| 4332 | vResulti = _mm_packs_epi32(vResulti,vResulti); |
||
| 4333 | _mm_store_sd(reinterpret_cast<double *>(&pDestination->x),reinterpret_cast<const __m128d *>(&vResulti)[0]); |
||
| 4334 | #else // _XM_VMX128_INTRINSICS_ |
||
| 4335 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 4336 | } |
||
| 4337 | |||
| 4338 | //------------------------------------------------------------------------------ |
||
| 4339 | |||
| 4340 | XMFINLINE VOID XMStoreShort4 |
||
| 4341 | ( |
||
| 4342 | XMSHORT4* pDestination, |
||
| 4343 | FXMVECTOR V |
||
| 4344 | ) |
||
| 4345 | { |
||
| 4346 | #if defined(_XM_NO_INTRINSICS_) |
||
| 4347 | |||
| 4348 | XMVECTOR N; |
||
| 4349 | static CONST XMVECTOR Min = {-32767.0f, -32767.0f, -32767.0f, -32767.0f}; |
||
| 4350 | static CONST XMVECTOR Max = {32767.0f, 32767.0f, 32767.0f, 32767.0f}; |
||
| 4351 | |||
| 4352 | XMASSERT(pDestination); |
||
| 4353 | |||
| 4354 | N = XMVectorClamp(V, Min, Max); |
||
| 4355 | N = XMVectorRound(N); |
||
| 4356 | |||
| 4357 | pDestination->x = (SHORT)N.vector4_f32[0]; |
||
| 4358 | pDestination->y = (SHORT)N.vector4_f32[1]; |
||
| 4359 | pDestination->z = (SHORT)N.vector4_f32[2]; |
||
| 4360 | pDestination->w = (SHORT)N.vector4_f32[3]; |
||
| 4361 | |||
| 4362 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 4363 | XMASSERT(pDestination); |
||
| 4364 | static CONST XMVECTORF32 Min = {-32767.0f, -32767.0f, -32767.0f, -32767.0f}; |
||
| 4365 | static CONST XMVECTORF32 Max = {32767.0f, 32767.0f, 32767.0f, 32767.0f}; |
||
| 4366 | // Bounds check |
||
| 4367 | XMVECTOR vResult = _mm_max_ps(V,Min); |
||
| 4368 | vResult = _mm_min_ps(vResult,Max); |
||
| 4369 | // Convert to int with rounding |
||
| 4370 | __m128i vInt = _mm_cvtps_epi32(vResult); |
||
| 4371 | // Pack the ints into shorts |
||
| 4372 | vInt = _mm_packs_epi32(vInt,vInt); |
||
| 4373 | _mm_store_sd(reinterpret_cast<double *>(&pDestination->x),reinterpret_cast<const __m128d *>(&vInt)[0]); |
||
| 4374 | #else // _XM_VMX128_INTRINSICS_ |
||
| 4375 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 4376 | } |
||
| 4377 | |||
| 4378 | //------------------------------------------------------------------------------ |
||
| 4379 | |||
| 4380 | XMFINLINE VOID XMStoreUShortN4 |
||
| 4381 | ( |
||
| 4382 | XMUSHORTN4* pDestination, |
||
| 4383 | FXMVECTOR V |
||
| 4384 | ) |
||
| 4385 | { |
||
| 4386 | #if defined(_XM_NO_INTRINSICS_) |
||
| 4387 | |||
| 4388 | XMVECTOR N; |
||
| 4389 | static CONST XMVECTORF32 Scale = {65535.0f, 65535.0f, 65535.0f, 65535.0f}; |
||
| 4390 | |||
| 4391 | XMASSERT(pDestination); |
||
| 4392 | |||
| 4393 | N = XMVectorClamp(V, XMVectorZero(), g_XMOne.v); |
||
| 4394 | N = XMVectorMultiplyAdd(N, Scale.v, g_XMOneHalf.v); |
||
| 4395 | N = XMVectorTruncate(N); |
||
| 4396 | |||
| 4397 | pDestination->x = (SHORT)N.vector4_f32[0]; |
||
| 4398 | pDestination->y = (SHORT)N.vector4_f32[1]; |
||
| 4399 | pDestination->z = (SHORT)N.vector4_f32[2]; |
||
| 4400 | pDestination->w = (SHORT)N.vector4_f32[3]; |
||
| 4401 | |||
| 4402 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 4403 | XMASSERT(pDestination); |
||
| 4404 | static CONST XMVECTORF32 Scale = {65535.0f, 65535.0f, 65535.0f, 65535.0f}; |
||
| 4405 | // Bounds check |
||
| 4406 | XMVECTOR vResult = _mm_max_ps(V,g_XMZero); |
||
| 4407 | vResult = _mm_min_ps(vResult,g_XMOne); |
||
| 4408 | vResult = _mm_mul_ps(vResult,Scale); |
||
| 4409 | // Convert to int with rounding |
||
| 4410 | __m128i vInt = _mm_cvtps_epi32(vResult); |
||
| 4411 | // Since the SSE pack instruction clamps using signed rules, |
||
| 4412 | // manually extract the values to store them to memory |
||
| 4413 | pDestination->x = static_cast<SHORT>(_mm_extract_epi16(vInt,0)); |
||
| 4414 | pDestination->y = static_cast<SHORT>(_mm_extract_epi16(vInt,2)); |
||
| 4415 | pDestination->z = static_cast<SHORT>(_mm_extract_epi16(vInt,4)); |
||
| 4416 | pDestination->w = static_cast<SHORT>(_mm_extract_epi16(vInt,6)); |
||
| 4417 | #else // _XM_VMX128_INTRINSICS_ |
||
| 4418 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 4419 | } |
||
| 4420 | |||
| 4421 | //------------------------------------------------------------------------------ |
||
| 4422 | |||
| 4423 | XMFINLINE VOID XMStoreUShort4 |
||
| 4424 | ( |
||
| 4425 | XMUSHORT4* pDestination, |
||
| 4426 | FXMVECTOR V |
||
| 4427 | ) |
||
| 4428 | { |
||
| 4429 | #if defined(_XM_NO_INTRINSICS_) |
||
| 4430 | |||
| 4431 | XMVECTOR N; |
||
| 4432 | static CONST XMVECTOR Max = {65535.0f, 65535.0f, 65535.0f, 65535.0f}; |
||
| 4433 | |||
| 4434 | XMASSERT(pDestination); |
||
| 4435 | |||
| 4436 | N = XMVectorClamp(V, XMVectorZero(), Max); |
||
| 4437 | N = XMVectorRound(N); |
||
| 4438 | |||
| 4439 | pDestination->x = (SHORT)N.vector4_f32[0]; |
||
| 4440 | pDestination->y = (SHORT)N.vector4_f32[1]; |
||
| 4441 | pDestination->z = (SHORT)N.vector4_f32[2]; |
||
| 4442 | pDestination->w = (SHORT)N.vector4_f32[3]; |
||
| 4443 | |||
| 4444 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 4445 | XMASSERT(pDestination); |
||
| 4446 | static CONST XMVECTORF32 Max = {65535.0f, 65535.0f, 65535.0f, 65535.0f}; |
||
| 4447 | // Bounds check |
||
| 4448 | XMVECTOR vResult = _mm_max_ps(V,g_XMZero); |
||
| 4449 | vResult = _mm_min_ps(vResult,Max); |
||
| 4450 | // Convert to int with rounding |
||
| 4451 | __m128i vInt = _mm_cvtps_epi32(vResult); |
||
| 4452 | // Since the SSE pack instruction clamps using signed rules, |
||
| 4453 | // manually extract the values to store them to memory |
||
| 4454 | pDestination->x = static_cast<SHORT>(_mm_extract_epi16(vInt,0)); |
||
| 4455 | pDestination->y = static_cast<SHORT>(_mm_extract_epi16(vInt,2)); |
||
| 4456 | pDestination->z = static_cast<SHORT>(_mm_extract_epi16(vInt,4)); |
||
| 4457 | pDestination->w = static_cast<SHORT>(_mm_extract_epi16(vInt,6)); |
||
| 4458 | #else // _XM_VMX128_INTRINSICS_ |
||
| 4459 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 4460 | } |
||
| 4461 | |||
| 4462 | //------------------------------------------------------------------------------ |
||
| 4463 | |||
| 4464 | XMFINLINE VOID XMStoreXIcoN4 |
||
| 4465 | ( |
||
| 4466 | XMXICON4* pDestination, |
||
| 4467 | FXMVECTOR V |
||
| 4468 | ) |
||
| 4469 | { |
||
| 4470 | #if defined(_XM_NO_INTRINSICS_) |
||
| 4471 | |||
| 4472 | XMVECTOR N; |
||
| 4473 | static CONST XMVECTORF32 Min = {-1.0f, -1.0f, -1.0f, 0.0f}; |
||
| 4474 | static CONST XMVECTORF32 Scale = {524287.0f, 524287.0f, 524287.0f, 15.0f}; |
||
| 4475 | |||
| 4476 | XMASSERT(pDestination); |
||
| 4477 | |||
| 4478 | N = XMVectorClamp(V, Min.v, g_XMOne.v); |
||
| 4479 | N = XMVectorMultiply(N, Scale.v); |
||
| 4480 | N = XMVectorRound(N); |
||
| 4481 | |||
| 4482 | pDestination->v = ((UINT64)N.vector4_f32[3] << 60) | |
||
| 4483 | (((INT64)N.vector4_f32[2] & 0xFFFFF) << 40) | |
||
| 4484 | (((INT64)N.vector4_f32[1] & 0xFFFFF) << 20) | |
||
| 4485 | (((INT64)N.vector4_f32[0] & 0xFFFFF)); |
||
| 4486 | |||
| 4487 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 4488 | XMASSERT(pDestination); |
||
| 4489 | // Note: Masks are x,w,y and z |
||
| 4490 | static const XMVECTORF32 MinXIcoN4 = {-1.0f, 0.0f,-1.0f,-1.0f}; |
||
| 4491 | static const XMVECTORF32 ScaleXIcoN4 = {524287.0f,15.0f*4096.0f*65536.0f*0.5f,524287.0f*4096.0f,524287.0f}; |
||
| 4492 | static const XMVECTORI32 MaskXIcoN4 = {0xFFFFF,0xF<<((60-32)-1),0xFFFFF000,0xFFFFF}; |
||
| 4493 | |||
| 4494 | // Clamp to bounds |
||
| 4495 | XMVECTOR vResult = _mm_shuffle_ps(V,V,_MM_SHUFFLE(2,1,3,0)); |
||
| 4496 | vResult = _mm_max_ps(vResult,MinXIcoN4); |
||
| 4497 | vResult = _mm_min_ps(vResult,g_XMOne); |
||
| 4498 | // Scale by multiplication |
||
| 4499 | vResult = _mm_mul_ps(vResult,ScaleXIcoN4); |
||
| 4500 | // Convert to integer (w is unsigned) |
||
| 4501 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 4502 | // Mask off unused bits |
||
| 4503 | vResulti = _mm_and_si128(vResulti,MaskXIcoN4); |
||
| 4504 | // Isolate Y |
||
| 4505 | __m128i vResulti2 = _mm_and_si128(vResulti,g_XMMaskY); |
||
| 4506 | // Double Y (Really W) to fixup for unsigned conversion |
||
| 4507 | vResulti = _mm_add_epi32(vResulti,vResulti2); |
||
| 4508 | // Shift y and z to straddle the 32-bit boundary |
||
| 4509 | vResulti2 = _mm_srli_si128(vResulti,(64+12)/8); |
||
| 4510 | // Shift it into place |
||
| 4511 | vResulti2 = _mm_slli_si128(vResulti2,20/8); |
||
| 4512 | // i = x|y<<20|z<<40|w<<60 |
||
| 4513 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 4514 | _mm_store_sd(reinterpret_cast<double *>(&pDestination->v),reinterpret_cast<const __m128d *>(&vResulti)[0]); |
||
| 4515 | #else // _XM_VMX128_INTRINSICS_ |
||
| 4516 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 4517 | } |
||
| 4518 | |||
| 4519 | //------------------------------------------------------------------------------ |
||
| 4520 | |||
| 4521 | XMFINLINE VOID XMStoreXIco4 |
||
| 4522 | ( |
||
| 4523 | XMXICO4* pDestination, |
||
| 4524 | FXMVECTOR V |
||
| 4525 | ) |
||
| 4526 | { |
||
| 4527 | #if defined(_XM_NO_INTRINSICS_) |
||
| 4528 | |||
| 4529 | XMVECTOR N; |
||
| 4530 | static CONST XMVECTORF32 Min = {-524287.0f, -524287.0f, -524287.0f, 0.0f}; |
||
| 4531 | static CONST XMVECTORF32 Max = {524287.0f, 524287.0f, 524287.0f, 15.0f}; |
||
| 4532 | |||
| 4533 | XMASSERT(pDestination); |
||
| 4534 | N = XMVectorClamp(V, Min.v, Max.v); |
||
| 4535 | pDestination->v = ((UINT64)N.vector4_f32[3] << 60) | |
||
| 4536 | (((INT64)N.vector4_f32[2] & 0xFFFFF) << 40) | |
||
| 4537 | (((INT64)N.vector4_f32[1] & 0xFFFFF) << 20) | |
||
| 4538 | (((INT64)N.vector4_f32[0] & 0xFFFFF)); |
||
| 4539 | |||
| 4540 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 4541 | XMASSERT(pDestination); |
||
| 4542 | // Note: Masks are x,w,y and z |
||
| 4543 | static const XMVECTORF32 MinXIco4 = {-524287.0f, 0.0f,-524287.0f,-524287.0f}; |
||
| 4544 | static const XMVECTORF32 MaxXIco4 = { 524287.0f,15.0f, 524287.0f, 524287.0f}; |
||
| 4545 | static const XMVECTORF32 ScaleXIco4 = {1.0f,4096.0f*65536.0f*0.5f,4096.0f,1.0f}; |
||
| 4546 | static const XMVECTORI32 MaskXIco4 = {0xFFFFF,0xF<<((60-1)-32),0xFFFFF000,0xFFFFF}; |
||
| 4547 | // Clamp to bounds |
||
| 4548 | XMVECTOR vResult = _mm_shuffle_ps(V,V,_MM_SHUFFLE(2,1,3,0)); |
||
| 4549 | vResult = _mm_max_ps(vResult,MinXIco4); |
||
| 4550 | vResult = _mm_min_ps(vResult,MaxXIco4); |
||
| 4551 | // Scale by multiplication |
||
| 4552 | vResult = _mm_mul_ps(vResult,ScaleXIco4); |
||
| 4553 | // Convert to int |
||
| 4554 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 4555 | // Mask off any fraction |
||
| 4556 | vResulti = _mm_and_si128(vResulti,MaskXIco4); |
||
| 4557 | // Isolate Y |
||
| 4558 | __m128i vResulti2 = _mm_and_si128(vResulti,g_XMMaskY); |
||
| 4559 | // Double Y (Really W) to fixup for unsigned conversion |
||
| 4560 | vResulti = _mm_add_epi32(vResulti,vResulti2); |
||
| 4561 | // Shift y and z to straddle the 32-bit boundary |
||
| 4562 | vResulti2 = _mm_srli_si128(vResulti,(64+12)/8); |
||
| 4563 | // Shift it into place |
||
| 4564 | vResulti2 = _mm_slli_si128(vResulti2,20/8); |
||
| 4565 | // i = x|y<<20|z<<40|w<<60 |
||
| 4566 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 4567 | _mm_store_sd(reinterpret_cast<double *>(&pDestination->v),reinterpret_cast<const __m128d *>(&vResulti)[0]); |
||
| 4568 | #else // _XM_VMX128_INTRINSICS_ |
||
| 4569 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 4570 | } |
||
| 4571 | |||
| 4572 | //------------------------------------------------------------------------------ |
||
| 4573 | |||
| 4574 | XMFINLINE VOID XMStoreUIcoN4 |
||
| 4575 | ( |
||
| 4576 | XMUICON4* pDestination, |
||
| 4577 | FXMVECTOR V |
||
| 4578 | ) |
||
| 4579 | { |
||
| 4580 | #define XM_URange ((FLOAT)(1 << 20)) |
||
| 4581 | #define XM_URangeDiv2 ((FLOAT)(1 << 19)) |
||
| 4582 | #define XM_UMaxXYZ ((FLOAT)((1 << 20) - 1)) |
||
| 4583 | #define XM_UMaxW ((FLOAT)((1 << 4) - 1)) |
||
| 4584 | #define XM_ScaleXYZ (-(FLOAT)((1 << 20) - 1) / XM_PACK_FACTOR) |
||
| 4585 | #define XM_ScaleW (-(FLOAT)((1 << 4) - 1) / XM_PACK_FACTOR) |
||
| 4586 | #define XM_Scale (-1.0f / XM_PACK_FACTOR) |
||
| 4587 | #define XM_Offset (3.0f) |
||
| 4588 | |||
| 4589 | #if defined(_XM_NO_INTRINSICS_) |
||
| 4590 | |||
| 4591 | XMVECTOR N; |
||
| 4592 | static CONST XMVECTORF32 Scale = {1048575.0f, 1048575.0f, 1048575.0f, 15.0f}; |
||
| 4593 | |||
| 4594 | XMASSERT(pDestination); |
||
| 4595 | |||
| 4596 | N = XMVectorClamp(V, XMVectorZero(), g_XMOne.v); |
||
| 4597 | N = XMVectorMultiplyAdd(N, Scale.v, g_XMOneHalf.v); |
||
| 4598 | |||
| 4599 | pDestination->v = ((UINT64)N.vector4_f32[3] << 60) | |
||
| 4600 | (((UINT64)N.vector4_f32[2] & 0xFFFFF) << 40) | |
||
| 4601 | (((UINT64)N.vector4_f32[1] & 0xFFFFF) << 20) | |
||
| 4602 | (((UINT64)N.vector4_f32[0] & 0xFFFFF)); |
||
| 4603 | |||
| 4604 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 4605 | XMASSERT(pDestination); |
||
| 4606 | // Note: Masks are x,w,y and z |
||
| 4607 | static const XMVECTORF32 ScaleUIcoN4 = {1048575.0f,15.0f*4096.0f*65536.0f,1048575.0f*4096.0f,1048575.0f}; |
||
| 4608 | static const XMVECTORI32 MaskUIcoN4 = {0xFFFFF,0xF<<(60-32),0xFFFFF000,0xFFFFF}; |
||
| 4609 | static const XMVECTORF32 AddUIcoN4 = {0.0f,-32768.0f*65536.0f,-32768.0f*65536.0f,0.0f}; |
||
| 4610 | // Clamp to bounds |
||
| 4611 | XMVECTOR vResult = _mm_shuffle_ps(V,V,_MM_SHUFFLE(2,1,3,0)); |
||
| 4612 | vResult = _mm_max_ps(vResult,g_XMZero); |
||
| 4613 | vResult = _mm_min_ps(vResult,g_XMOne); |
||
| 4614 | // Scale by multiplication |
||
| 4615 | vResult = _mm_mul_ps(vResult,ScaleUIcoN4); |
||
| 4616 | // Adjust for unsigned entries |
||
| 4617 | vResult = _mm_add_ps(vResult,AddUIcoN4); |
||
| 4618 | // Convert to int |
||
| 4619 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 4620 | // Fix the signs on the unsigned entries |
||
| 4621 | vResulti = _mm_xor_si128(vResulti,g_XMFlipYZ); |
||
| 4622 | // Mask off any fraction |
||
| 4623 | vResulti = _mm_and_si128(vResulti,MaskUIcoN4); |
||
| 4624 | // Shift y and z to straddle the 32-bit boundary |
||
| 4625 | __m128i vResulti2 = _mm_srli_si128(vResulti,(64+12)/8); |
||
| 4626 | // Shift it into place |
||
| 4627 | vResulti2 = _mm_slli_si128(vResulti2,20/8); |
||
| 4628 | // i = x|y<<20|z<<40|w<<60 |
||
| 4629 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 4630 | _mm_store_sd(reinterpret_cast<double *>(&pDestination->v),reinterpret_cast<const __m128d *>(&vResulti)[0]); |
||
| 4631 | #else // _XM_VMX128_INTRINSICS_ |
||
| 4632 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 4633 | |||
| 4634 | #undef XM_URange |
||
| 4635 | #undef XM_URangeDiv2 |
||
| 4636 | #undef XM_UMaxXYZ |
||
| 4637 | #undef XM_UMaxW |
||
| 4638 | #undef XM_ScaleXYZ |
||
| 4639 | #undef XM_ScaleW |
||
| 4640 | #undef XM_Scale |
||
| 4641 | #undef XM_Offset |
||
| 4642 | } |
||
| 4643 | |||
| 4644 | //------------------------------------------------------------------------------ |
||
| 4645 | |||
| 4646 | XMFINLINE VOID XMStoreUIco4 |
||
| 4647 | ( |
||
| 4648 | XMUICO4* pDestination, |
||
| 4649 | FXMVECTOR V |
||
| 4650 | ) |
||
| 4651 | { |
||
| 4652 | #define XM_Scale (-1.0f / XM_PACK_FACTOR) |
||
| 4653 | #define XM_URange ((FLOAT)(1 << 20)) |
||
| 4654 | #define XM_URangeDiv2 ((FLOAT)(1 << 19)) |
||
| 4655 | |||
| 4656 | #if defined(_XM_NO_INTRINSICS_) |
||
| 4657 | |||
| 4658 | XMVECTOR N; |
||
| 4659 | static CONST XMVECTOR Max = {1048575.0f, 1048575.0f, 1048575.0f, 15.0f}; |
||
| 4660 | |||
| 4661 | XMASSERT(pDestination); |
||
| 4662 | |||
| 4663 | N = XMVectorClamp(V, XMVectorZero(), Max); |
||
| 4664 | N = XMVectorRound(N); |
||
| 4665 | |||
| 4666 | pDestination->v = ((UINT64)N.vector4_f32[3] << 60) | |
||
| 4667 | (((UINT64)N.vector4_f32[2] & 0xFFFFF) << 40) | |
||
| 4668 | (((UINT64)N.vector4_f32[1] & 0xFFFFF) << 20) | |
||
| 4669 | (((UINT64)N.vector4_f32[0] & 0xFFFFF)); |
||
| 4670 | |||
| 4671 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 4672 | XMASSERT(pDestination); |
||
| 4673 | // Note: Masks are x,w,y and z |
||
| 4674 | static const XMVECTORF32 MaxUIco4 = { 1048575.0f, 15.0f, 1048575.0f, 1048575.0f}; |
||
| 4675 | static const XMVECTORF32 ScaleUIco4 = {1.0f,4096.0f*65536.0f,4096.0f,1.0f}; |
||
| 4676 | static const XMVECTORI32 MaskUIco4 = {0xFFFFF,0xF<<(60-32),0xFFFFF000,0xFFFFF}; |
||
| 4677 | static const XMVECTORF32 AddUIco4 = {0.0f,-32768.0f*65536.0f,-32768.0f*65536.0f,0.0f}; |
||
| 4678 | // Clamp to bounds |
||
| 4679 | XMVECTOR vResult = _mm_shuffle_ps(V,V,_MM_SHUFFLE(2,1,3,0)); |
||
| 4680 | vResult = _mm_max_ps(vResult,g_XMZero); |
||
| 4681 | vResult = _mm_min_ps(vResult,MaxUIco4); |
||
| 4682 | // Scale by multiplication |
||
| 4683 | vResult = _mm_mul_ps(vResult,ScaleUIco4); |
||
| 4684 | vResult = _mm_add_ps(vResult,AddUIco4); |
||
| 4685 | // Convert to int |
||
| 4686 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 4687 | vResulti = _mm_xor_si128(vResulti,g_XMFlipYZ); |
||
| 4688 | // Mask off any fraction |
||
| 4689 | vResulti = _mm_and_si128(vResulti,MaskUIco4); |
||
| 4690 | // Shift y and z to straddle the 32-bit boundary |
||
| 4691 | __m128i vResulti2 = _mm_srli_si128(vResulti,(64+12)/8); |
||
| 4692 | // Shift it into place |
||
| 4693 | vResulti2 = _mm_slli_si128(vResulti2,20/8); |
||
| 4694 | // i = x|y<<20|z<<40|w<<60 |
||
| 4695 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 4696 | _mm_store_sd(reinterpret_cast<double *>(&pDestination->v),reinterpret_cast<const __m128d *>(&vResulti)[0]); |
||
| 4697 | #else // _XM_VMX128_INTRINSICS_ |
||
| 4698 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 4699 | |||
| 4700 | #undef XM_Scale |
||
| 4701 | #undef XM_URange |
||
| 4702 | #undef XM_URangeDiv2 |
||
| 4703 | } |
||
| 4704 | |||
| 4705 | //------------------------------------------------------------------------------ |
||
| 4706 | |||
| 4707 | XMFINLINE VOID XMStoreIcoN4 |
||
| 4708 | ( |
||
| 4709 | XMICON4* pDestination, |
||
| 4710 | FXMVECTOR V |
||
| 4711 | ) |
||
| 4712 | { |
||
| 4713 | #define XM_Scale (-1.0f / XM_PACK_FACTOR) |
||
| 4714 | #define XM_URange ((FLOAT)(1 << 4)) |
||
| 4715 | #define XM_Offset (3.0f) |
||
| 4716 | #define XM_UMaxXYZ ((FLOAT)((1 << (20 - 1)) - 1)) |
||
| 4717 | #define XM_UMaxW ((FLOAT)((1 << (4 - 1)) - 1)) |
||
| 4718 | |||
| 4719 | #if defined(_XM_NO_INTRINSICS_) |
||
| 4720 | |||
| 4721 | XMVECTOR N; |
||
| 4722 | static CONST XMVECTORF32 Scale = {524287.0f, 524287.0f, 524287.0f, 7.0f}; |
||
| 4723 | |||
| 4724 | XMASSERT(pDestination); |
||
| 4725 | |||
| 4726 | N = XMVectorClamp(V, g_XMNegativeOne.v, g_XMOne.v); |
||
| 4727 | N = XMVectorMultiplyAdd(N, Scale.v, g_XMNegativeZero.v); |
||
| 4728 | N = XMVectorRound(N); |
||
| 4729 | |||
| 4730 | pDestination->v = ((UINT64)N.vector4_f32[3] << 60) | |
||
| 4731 | (((UINT64)N.vector4_f32[2] & 0xFFFFF) << 40) | |
||
| 4732 | (((UINT64)N.vector4_f32[1] & 0xFFFFF) << 20) | |
||
| 4733 | (((UINT64)N.vector4_f32[0] & 0xFFFFF)); |
||
| 4734 | |||
| 4735 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 4736 | XMASSERT(pDestination); |
||
| 4737 | // Note: Masks are x,w,y and z |
||
| 4738 | static const XMVECTORF32 ScaleIcoN4 = {524287.0f,7.0f*4096.0f*65536.0f,524287.0f*4096.0f,524287.0f}; |
||
| 4739 | static const XMVECTORI32 MaskIcoN4 = {0xFFFFF,0xF<<(60-32),0xFFFFF000,0xFFFFF}; |
||
| 4740 | // Clamp to bounds |
||
| 4741 | XMVECTOR vResult = _mm_shuffle_ps(V,V,_MM_SHUFFLE(2,1,3,0)); |
||
| 4742 | vResult = _mm_max_ps(vResult,g_XMNegativeOne); |
||
| 4743 | vResult = _mm_min_ps(vResult,g_XMOne); |
||
| 4744 | // Scale by multiplication |
||
| 4745 | vResult = _mm_mul_ps(vResult,ScaleIcoN4); |
||
| 4746 | // Convert to int |
||
| 4747 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 4748 | // Mask off any fraction |
||
| 4749 | vResulti = _mm_and_si128(vResulti,MaskIcoN4); |
||
| 4750 | // Shift y and z to straddle the 32-bit boundary |
||
| 4751 | __m128i vResulti2 = _mm_srli_si128(vResulti,(64+12)/8); |
||
| 4752 | // Shift it into place |
||
| 4753 | vResulti2 = _mm_slli_si128(vResulti2,20/8); |
||
| 4754 | // i = x|y<<20|z<<40|w<<60 |
||
| 4755 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 4756 | _mm_store_sd(reinterpret_cast<double *>(&pDestination->v),reinterpret_cast<const __m128d *>(&vResulti)[0]); |
||
| 4757 | #else // _XM_VMX128_INTRINSICS_ |
||
| 4758 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 4759 | |||
| 4760 | #undef XM_Scale |
||
| 4761 | #undef XM_URange |
||
| 4762 | #undef XM_Offset |
||
| 4763 | #undef XM_UMaxXYZ |
||
| 4764 | #undef XM_UMaxW |
||
| 4765 | } |
||
| 4766 | |||
| 4767 | //------------------------------------------------------------------------------ |
||
| 4768 | |||
| 4769 | XMFINLINE VOID XMStoreIco4 |
||
| 4770 | ( |
||
| 4771 | XMICO4* pDestination, |
||
| 4772 | FXMVECTOR V |
||
| 4773 | ) |
||
| 4774 | { |
||
| 4775 | #define XM_Scale (-1.0f / XM_PACK_FACTOR) |
||
| 4776 | #define XM_URange ((FLOAT)(1 << 4)) |
||
| 4777 | #define XM_Offset (3.0f) |
||
| 4778 | |||
| 4779 | #if defined(_XM_NO_INTRINSICS_) |
||
| 4780 | |||
| 4781 | XMVECTOR N; |
||
| 4782 | static CONST XMVECTOR Min = {-524287.0f, -524287.0f, -524287.0f, -7.0f}; |
||
| 4783 | static CONST XMVECTOR Max = {524287.0f, 524287.0f, 524287.0f, 7.0f}; |
||
| 4784 | |||
| 4785 | XMASSERT(pDestination); |
||
| 4786 | |||
| 4787 | N = XMVectorClamp(V, Min, Max); |
||
| 4788 | N = XMVectorRound(N); |
||
| 4789 | |||
| 4790 | pDestination->v = ((INT64)N.vector4_f32[3] << 60) | |
||
| 4791 | (((INT64)N.vector4_f32[2] & 0xFFFFF) << 40) | |
||
| 4792 | (((INT64)N.vector4_f32[1] & 0xFFFFF) << 20) | |
||
| 4793 | (((INT64)N.vector4_f32[0] & 0xFFFFF)); |
||
| 4794 | |||
| 4795 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 4796 | XMASSERT(pDestination); |
||
| 4797 | // Note: Masks are x,w,y and z |
||
| 4798 | static const XMVECTORF32 MinIco4 = {-524287.0f,-7.0f,-524287.0f,-524287.0f}; |
||
| 4799 | static const XMVECTORF32 MaxIco4 = { 524287.0f, 7.0f, 524287.0f, 524287.0f}; |
||
| 4800 | static const XMVECTORF32 ScaleIco4 = {1.0f,4096.0f*65536.0f,4096.0f,1.0f}; |
||
| 4801 | static const XMVECTORI32 MaskIco4 = {0xFFFFF,0xF<<(60-32),0xFFFFF000,0xFFFFF}; |
||
| 4802 | // Clamp to bounds |
||
| 4803 | XMVECTOR vResult = _mm_shuffle_ps(V,V,_MM_SHUFFLE(2,1,3,0)); |
||
| 4804 | vResult = _mm_max_ps(vResult,MinIco4); |
||
| 4805 | vResult = _mm_min_ps(vResult,MaxIco4); |
||
| 4806 | // Scale by multiplication |
||
| 4807 | vResult = _mm_mul_ps(vResult,ScaleIco4); |
||
| 4808 | // Convert to int |
||
| 4809 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 4810 | // Mask off any fraction |
||
| 4811 | vResulti = _mm_and_si128(vResulti,MaskIco4); |
||
| 4812 | // Shift y and z to straddle the 32-bit boundary |
||
| 4813 | __m128i vResulti2 = _mm_srli_si128(vResulti,(64+12)/8); |
||
| 4814 | // Shift it into place |
||
| 4815 | vResulti2 = _mm_slli_si128(vResulti2,20/8); |
||
| 4816 | // i = x|y<<20|z<<40|w<<60 |
||
| 4817 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 4818 | _mm_store_sd(reinterpret_cast<double *>(&pDestination->v),reinterpret_cast<const __m128d *>(&vResulti)[0]); |
||
| 4819 | #else // _XM_VMX128_INTRINSICS_ |
||
| 4820 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 4821 | |||
| 4822 | #undef XM_Scale |
||
| 4823 | #undef XM_URange |
||
| 4824 | #undef XM_Offset |
||
| 4825 | } |
||
| 4826 | |||
| 4827 | //------------------------------------------------------------------------------ |
||
| 4828 | |||
| 4829 | XMFINLINE VOID XMStoreXDecN4 |
||
| 4830 | ( |
||
| 4831 | XMXDECN4* pDestination, |
||
| 4832 | FXMVECTOR V |
||
| 4833 | ) |
||
| 4834 | { |
||
| 4835 | #if defined(_XM_NO_INTRINSICS_) |
||
| 4836 | |||
| 4837 | XMVECTOR N; |
||
| 4838 | static CONST XMVECTORF32 Min = {-1.0f, -1.0f, -1.0f, 0.0f}; |
||
| 4839 | static CONST XMVECTORF32 Scale = {511.0f, 511.0f, 511.0f, 3.0f}; |
||
| 4840 | |||
| 4841 | XMASSERT(pDestination); |
||
| 4842 | |||
| 4843 | N = XMVectorClamp(V, Min.v, g_XMOne.v); |
||
| 4844 | N = XMVectorMultiply(N, Scale.v); |
||
| 4845 | N = XMVectorRound(N); |
||
| 4846 | |||
| 4847 | pDestination->v = ((UINT)N.vector4_f32[3] << 30) | |
||
| 4848 | (((INT)N.vector4_f32[2] & 0x3FF) << 20) | |
||
| 4849 | (((INT)N.vector4_f32[1] & 0x3FF) << 10) | |
||
| 4850 | (((INT)N.vector4_f32[0] & 0x3FF)); |
||
| 4851 | |||
| 4852 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 4853 | static const XMVECTORF32 Min = {-1.0f, -1.0f, -1.0f, 0.0f}; |
||
| 4854 | static const XMVECTORF32 Scale = {511.0f, 511.0f*1024.0f, 511.0f*1048576.0f,3.0f*536870912.0f}; |
||
| 4855 | static const XMVECTORI32 ScaleMask = {0x3FF,0x3FF<<10,0x3FF<<20,0x3<<29}; |
||
| 4856 | XMASSERT(pDestination); |
||
| 4857 | XMVECTOR vResult = _mm_max_ps(V,Min); |
||
| 4858 | vResult = _mm_min_ps(vResult,g_XMOne); |
||
| 4859 | // Scale by multiplication |
||
| 4860 | vResult = _mm_mul_ps(vResult,Scale); |
||
| 4861 | // Convert to int (W is unsigned) |
||
| 4862 | __m128i vResulti = _mm_cvtps_epi32(vResult); |
||
| 4863 | // Mask off any fraction |
||
| 4864 | vResulti = _mm_and_si128(vResulti,ScaleMask); |
||
| 4865 | // To fix W, add itself to shift it up to <<30 instead of <<29 |
||
| 4866 | __m128i vResultw = _mm_and_si128(vResulti,g_XMMaskW); |
||
| 4867 | vResulti = _mm_add_epi32(vResulti,vResultw); |
||
| 4868 | // Do a horizontal or of all 4 entries |
||
| 4869 | vResult = _mm_shuffle_ps(reinterpret_cast<const __m128 *>(&vResulti)[0],reinterpret_cast<const __m128 *>(&vResulti)[0],_MM_SHUFFLE(0,3,2,1)); |
||
| 4870 | vResulti = _mm_or_si128(vResulti,reinterpret_cast<const __m128i *>(&vResult)[0]); |
||
| 4871 | vResult = _mm_shuffle_ps(vResult,vResult,_MM_SHUFFLE(0,3,2,1)); |
||
| 4872 | vResulti = _mm_or_si128(vResulti,reinterpret_cast<const __m128i *>(&vResult)[0]); |
||
| 4873 | vResult = _mm_shuffle_ps(vResult,vResult,_MM_SHUFFLE(0,3,2,1)); |
||
| 4874 | vResulti = _mm_or_si128(vResulti,reinterpret_cast<const __m128i *>(&vResult)[0]); |
||
| 4875 | _mm_store_ss(reinterpret_cast<float *>(&pDestination->v),reinterpret_cast<const __m128 *>(&vResulti)[0]); |
||
| 4876 | #else // _XM_VMX128_INTRINSICS_ |
||
| 4877 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 4878 | } |
||
| 4879 | |||
| 4880 | //------------------------------------------------------------------------------ |
||
| 4881 | |||
| 4882 | XMFINLINE VOID XMStoreXDec4 |
||
| 4883 | ( |
||
| 4884 | XMXDEC4* pDestination, |
||
| 4885 | FXMVECTOR V |
||
| 4886 | ) |
||
| 4887 | { |
||
| 4888 | #if defined(_XM_NO_INTRINSICS_) |
||
| 4889 | |||
| 4890 | XMVECTOR N; |
||
| 4891 | static CONST XMVECTOR Min = {-511.0f, -511.0f, -511.0f, 0.0f}; |
||
| 4892 | static CONST XMVECTOR Max = {511.0f, 511.0f, 511.0f, 3.0f}; |
||
| 4893 | |||
| 4894 | XMASSERT(pDestination); |
||
| 4895 | |||
| 4896 | N = XMVectorClamp(V, Min, Max); |
||
| 4897 | |||
| 4898 | pDestination->v = ((UINT)N.vector4_f32[3] << 30) | |
||
| 4899 | (((INT)N.vector4_f32[2] & 0x3FF) << 20) | |
||
| 4900 | (((INT)N.vector4_f32[1] & 0x3FF) << 10) | |
||
| 4901 | (((INT)N.vector4_f32[0] & 0x3FF)); |
||
| 4902 | |||
| 4903 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 4904 | XMASSERT(pDestination); |
||
| 4905 | static const XMVECTORF32 MinXDec4 = {-511.0f,-511.0f,-511.0f, 0.0f}; |
||
| 4906 | static const XMVECTORF32 MaxXDec4 = { 511.0f, 511.0f, 511.0f, 3.0f}; |
||
| 4907 | static const XMVECTORF32 ScaleXDec4 = {1.0f,1024.0f/2.0f,1024.0f*1024.0f,1024.0f*1024.0f*1024.0f/2.0f}; |
||
| 4908 | static const XMVECTORI32 MaskXDec4= {0x3FF,0x3FF<<(10-1),0x3FF<<20,0x3<<(30-1)}; |
||
| 4909 | // Clamp to bounds |
||
| 4910 | XMVECTOR vResult = _mm_max_ps(V,MinXDec4); |
||
| 4911 | vResult = _mm_min_ps(vResult,MaxXDec4); |
||
| 4912 | // Scale by multiplication |
||
| 4913 | vResult = _mm_mul_ps(vResult,ScaleXDec4); |
||
| 4914 | // Convert to int |
||
| 4915 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 4916 | // Mask off any fraction |
||
| 4917 | vResulti = _mm_and_si128(vResulti,MaskXDec4); |
||
| 4918 | // Do a horizontal or of 4 entries |
||
| 4919 | __m128i vResulti2 = _mm_shuffle_epi32(vResulti,_MM_SHUFFLE(3,2,3,2)); |
||
| 4920 | // x = x|z, y = y|w |
||
| 4921 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 4922 | // Move Z to the x position |
||
| 4923 | vResulti2 = _mm_shuffle_epi32(vResulti,_MM_SHUFFLE(1,1,1,1)); |
||
| 4924 | // Perform a single bit left shift on y|w |
||
| 4925 | vResulti2 = _mm_add_epi32(vResulti2,vResulti2); |
||
| 4926 | // i = x|y|z|w |
||
| 4927 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 4928 | _mm_store_ss(reinterpret_cast<float *>(&pDestination->v),reinterpret_cast<const __m128 *>(&vResulti)[0]); |
||
| 4929 | #else // _XM_VMX128_INTRINSICS_ |
||
| 4930 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 4931 | } |
||
| 4932 | |||
| 4933 | //------------------------------------------------------------------------------ |
||
| 4934 | |||
| 4935 | XMFINLINE VOID XMStoreUDecN4 |
||
| 4936 | ( |
||
| 4937 | XMUDECN4* pDestination, |
||
| 4938 | FXMVECTOR V |
||
| 4939 | ) |
||
| 4940 | { |
||
| 4941 | #if defined(_XM_NO_INTRINSICS_) |
||
| 4942 | |||
| 4943 | XMVECTOR N; |
||
| 4944 | static CONST XMVECTORF32 Scale = {1023.0f, 1023.0f, 1023.0f, 3.0f}; |
||
| 4945 | |||
| 4946 | XMASSERT(pDestination); |
||
| 4947 | |||
| 4948 | N = XMVectorClamp(V, XMVectorZero(), g_XMOne.v); |
||
| 4949 | N = XMVectorMultiply(N, Scale.v); |
||
| 4950 | |||
| 4951 | pDestination->v = ((UINT)N.vector4_f32[3] << 30) | |
||
| 4952 | (((UINT)N.vector4_f32[2] & 0x3FF) << 20) | |
||
| 4953 | (((UINT)N.vector4_f32[1] & 0x3FF) << 10) | |
||
| 4954 | (((UINT)N.vector4_f32[0] & 0x3FF)); |
||
| 4955 | |||
| 4956 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 4957 | XMASSERT(pDestination); |
||
| 4958 | static const XMVECTORF32 ScaleUDecN4 = {1023.0f,1023.0f*1024.0f*0.5f,1023.0f*1024.0f*1024.0f,3.0f*1024.0f*1024.0f*1024.0f*0.5f}; |
||
| 4959 | static const XMVECTORI32 MaskUDecN4= {0x3FF,0x3FF<<(10-1),0x3FF<<20,0x3<<(30-1)}; |
||
| 4960 | // Clamp to bounds |
||
| 4961 | XMVECTOR vResult = _mm_max_ps(V,g_XMZero); |
||
| 4962 | vResult = _mm_min_ps(vResult,g_XMOne); |
||
| 4963 | // Scale by multiplication |
||
| 4964 | vResult = _mm_mul_ps(vResult,ScaleUDecN4); |
||
| 4965 | // Convert to int |
||
| 4966 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 4967 | // Mask off any fraction |
||
| 4968 | vResulti = _mm_and_si128(vResulti,MaskUDecN4); |
||
| 4969 | // Do a horizontal or of 4 entries |
||
| 4970 | __m128i vResulti2 = _mm_shuffle_epi32(vResulti,_MM_SHUFFLE(3,2,3,2)); |
||
| 4971 | // x = x|z, y = y|w |
||
| 4972 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 4973 | // Move Z to the x position |
||
| 4974 | vResulti2 = _mm_shuffle_epi32(vResulti,_MM_SHUFFLE(1,1,1,1)); |
||
| 4975 | // Perform a left shift by one bit on y|w |
||
| 4976 | vResulti2 = _mm_add_epi32(vResulti2,vResulti2); |
||
| 4977 | // i = x|y|z|w |
||
| 4978 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 4979 | _mm_store_ss(reinterpret_cast<float *>(&pDestination->v),reinterpret_cast<const __m128 *>(&vResulti)[0]); |
||
| 4980 | #else // _XM_VMX128_INTRINSICS_ |
||
| 4981 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 4982 | } |
||
| 4983 | |||
| 4984 | //------------------------------------------------------------------------------ |
||
| 4985 | |||
| 4986 | XMFINLINE VOID XMStoreUDec4 |
||
| 4987 | ( |
||
| 4988 | XMUDEC4* pDestination, |
||
| 4989 | FXMVECTOR V |
||
| 4990 | ) |
||
| 4991 | { |
||
| 4992 | #if defined(_XM_NO_INTRINSICS_) |
||
| 4993 | |||
| 4994 | XMVECTOR N; |
||
| 4995 | static CONST XMVECTOR Max = {1023.0f, 1023.0f, 1023.0f, 3.0f}; |
||
| 4996 | |||
| 4997 | XMASSERT(pDestination); |
||
| 4998 | |||
| 4999 | N = XMVectorClamp(V, XMVectorZero(), Max); |
||
| 5000 | |||
| 5001 | pDestination->v = ((UINT)N.vector4_f32[3] << 30) | |
||
| 5002 | (((UINT)N.vector4_f32[2] & 0x3FF) << 20) | |
||
| 5003 | (((UINT)N.vector4_f32[1] & 0x3FF) << 10) | |
||
| 5004 | (((UINT)N.vector4_f32[0] & 0x3FF)); |
||
| 5005 | |||
| 5006 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 5007 | XMASSERT(pDestination); |
||
| 5008 | static const XMVECTORF32 MaxUDec4 = { 1023.0f, 1023.0f, 1023.0f, 3.0f}; |
||
| 5009 | static const XMVECTORF32 ScaleUDec4 = {1.0f,1024.0f/2.0f,1024.0f*1024.0f,1024.0f*1024.0f*1024.0f/2.0f}; |
||
| 5010 | static const XMVECTORI32 MaskUDec4= {0x3FF,0x3FF<<(10-1),0x3FF<<20,0x3<<(30-1)}; |
||
| 5011 | // Clamp to bounds |
||
| 5012 | XMVECTOR vResult = _mm_max_ps(V,g_XMZero); |
||
| 5013 | vResult = _mm_min_ps(vResult,MaxUDec4); |
||
| 5014 | // Scale by multiplication |
||
| 5015 | vResult = _mm_mul_ps(vResult,ScaleUDec4); |
||
| 5016 | // Convert to int |
||
| 5017 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 5018 | // Mask off any fraction |
||
| 5019 | vResulti = _mm_and_si128(vResulti,MaskUDec4); |
||
| 5020 | // Do a horizontal or of 4 entries |
||
| 5021 | __m128i vResulti2 = _mm_shuffle_epi32(vResulti,_MM_SHUFFLE(3,2,3,2)); |
||
| 5022 | // x = x|z, y = y|w |
||
| 5023 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 5024 | // Move Z to the x position |
||
| 5025 | vResulti2 = _mm_shuffle_epi32(vResulti,_MM_SHUFFLE(1,1,1,1)); |
||
| 5026 | // Perform a left shift by one bit on y|w |
||
| 5027 | vResulti2 = _mm_add_epi32(vResulti2,vResulti2); |
||
| 5028 | // i = x|y|z|w |
||
| 5029 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 5030 | _mm_store_ss(reinterpret_cast<float *>(&pDestination->v),reinterpret_cast<const __m128 *>(&vResulti)[0]); |
||
| 5031 | #else // _XM_VMX128_INTRINSICS_ |
||
| 5032 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 5033 | } |
||
| 5034 | |||
| 5035 | //------------------------------------------------------------------------------ |
||
| 5036 | |||
| 5037 | XMFINLINE VOID XMStoreDecN4 |
||
| 5038 | ( |
||
| 5039 | XMDECN4* pDestination, |
||
| 5040 | FXMVECTOR V |
||
| 5041 | ) |
||
| 5042 | { |
||
| 5043 | #if defined(_XM_NO_INTRINSICS_) |
||
| 5044 | |||
| 5045 | XMVECTOR N; |
||
| 5046 | static CONST XMVECTORF32 Scale = {511.0f, 511.0f, 511.0f, 1.0f}; |
||
| 5047 | |||
| 5048 | XMASSERT(pDestination); |
||
| 5049 | |||
| 5050 | N = XMVectorClamp(V, g_XMNegativeOne.v, g_XMOne.v); |
||
| 5051 | N = XMVectorMultiply(N, Scale.v); |
||
| 5052 | |||
| 5053 | pDestination->v = ((INT)N.vector4_f32[3] << 30) | |
||
| 5054 | (((INT)N.vector4_f32[2] & 0x3FF) << 20) | |
||
| 5055 | (((INT)N.vector4_f32[1] & 0x3FF) << 10) | |
||
| 5056 | (((INT)N.vector4_f32[0] & 0x3FF)); |
||
| 5057 | |||
| 5058 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 5059 | XMASSERT(pDestination); |
||
| 5060 | static const XMVECTORF32 ScaleDecN4 = {511.0f,511.0f*1024.0f,511.0f*1024.0f*1024.0f,1.0f*1024.0f*1024.0f*1024.0f}; |
||
| 5061 | static const XMVECTORI32 MaskDecN4= {0x3FF,0x3FF<<10,0x3FF<<20,0x3<<30}; |
||
| 5062 | // Clamp to bounds |
||
| 5063 | XMVECTOR vResult = _mm_max_ps(V,g_XMNegativeOne); |
||
| 5064 | vResult = _mm_min_ps(vResult,g_XMOne); |
||
| 5065 | // Scale by multiplication |
||
| 5066 | vResult = _mm_mul_ps(vResult,ScaleDecN4); |
||
| 5067 | // Convert to int |
||
| 5068 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 5069 | // Mask off any fraction |
||
| 5070 | vResulti = _mm_and_si128(vResulti,MaskDecN4); |
||
| 5071 | // Do a horizontal or of 4 entries |
||
| 5072 | __m128i vResulti2 = _mm_shuffle_epi32(vResulti,_MM_SHUFFLE(3,2,3,2)); |
||
| 5073 | // x = x|z, y = y|w |
||
| 5074 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 5075 | // Move Z to the x position |
||
| 5076 | vResulti2 = _mm_shuffle_epi32(vResulti,_MM_SHUFFLE(1,1,1,1)); |
||
| 5077 | // i = x|y|z|w |
||
| 5078 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 5079 | _mm_store_ss(reinterpret_cast<float *>(&pDestination->v),reinterpret_cast<const __m128 *>(&vResulti)[0]); |
||
| 5080 | #else // _XM_VMX128_INTRINSICS_ |
||
| 5081 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 5082 | } |
||
| 5083 | |||
| 5084 | //------------------------------------------------------------------------------ |
||
| 5085 | |||
| 5086 | XMFINLINE VOID XMStoreDec4 |
||
| 5087 | ( |
||
| 5088 | XMDEC4* pDestination, |
||
| 5089 | FXMVECTOR V |
||
| 5090 | ) |
||
| 5091 | { |
||
| 5092 | #if defined(_XM_NO_INTRINSICS_) |
||
| 5093 | |||
| 5094 | XMVECTOR N; |
||
| 5095 | static CONST XMVECTOR Min = {-511.0f, -511.0f, -511.0f, -1.0f}; |
||
| 5096 | static CONST XMVECTOR Max = {511.0f, 511.0f, 511.0f, 1.0f}; |
||
| 5097 | |||
| 5098 | XMASSERT(pDestination); |
||
| 5099 | |||
| 5100 | N = XMVectorClamp(V, Min, Max); |
||
| 5101 | |||
| 5102 | pDestination->v = ((INT)N.vector4_f32[3] << 30) | |
||
| 5103 | (((INT)N.vector4_f32[2] & 0x3FF) << 20) | |
||
| 5104 | (((INT)N.vector4_f32[1] & 0x3FF) << 10) | |
||
| 5105 | (((INT)N.vector4_f32[0] & 0x3FF)); |
||
| 5106 | |||
| 5107 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 5108 | XMASSERT(pDestination); |
||
| 5109 | static const XMVECTORF32 MinDec4 = {-511.0f,-511.0f,-511.0f,-1.0f}; |
||
| 5110 | static const XMVECTORF32 MaxDec4 = { 511.0f, 511.0f, 511.0f, 1.0f}; |
||
| 5111 | static const XMVECTORF32 ScaleDec4 = {1.0f,1024.0f,1024.0f*1024.0f,1024.0f*1024.0f*1024.0f}; |
||
| 5112 | static const XMVECTORI32 MaskDec4= {0x3FF,0x3FF<<10,0x3FF<<20,0x3<<30}; |
||
| 5113 | // Clamp to bounds |
||
| 5114 | XMVECTOR vResult = _mm_max_ps(V,MinDec4); |
||
| 5115 | vResult = _mm_min_ps(vResult,MaxDec4); |
||
| 5116 | // Scale by multiplication |
||
| 5117 | vResult = _mm_mul_ps(vResult,ScaleDec4); |
||
| 5118 | // Convert to int |
||
| 5119 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 5120 | // Mask off any fraction |
||
| 5121 | vResulti = _mm_and_si128(vResulti,MaskDec4); |
||
| 5122 | // Do a horizontal or of 4 entries |
||
| 5123 | __m128i vResulti2 = _mm_shuffle_epi32(vResulti,_MM_SHUFFLE(3,2,3,2)); |
||
| 5124 | // x = x|z, y = y|w |
||
| 5125 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 5126 | // Move Z to the x position |
||
| 5127 | vResulti2 = _mm_shuffle_epi32(vResulti,_MM_SHUFFLE(1,1,1,1)); |
||
| 5128 | // i = x|y|z|w |
||
| 5129 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 5130 | _mm_store_ss(reinterpret_cast<float *>(&pDestination->v),reinterpret_cast<const __m128 *>(&vResulti)[0]); |
||
| 5131 | #else // _XM_VMX128_INTRINSICS_ |
||
| 5132 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 5133 | } |
||
| 5134 | |||
| 5135 | //------------------------------------------------------------------------------ |
||
| 5136 | |||
| 5137 | XMFINLINE VOID XMStoreUByteN4 |
||
| 5138 | ( |
||
| 5139 | XMUBYTEN4* pDestination, |
||
| 5140 | FXMVECTOR V |
||
| 5141 | ) |
||
| 5142 | { |
||
| 5143 | #if defined(_XM_NO_INTRINSICS_) |
||
| 5144 | |||
| 5145 | XMVECTOR N; |
||
| 5146 | static CONST XMVECTORF32 Scale = {255.0f, 255.0f, 255.0f, 255.0f}; |
||
| 5147 | |||
| 5148 | XMASSERT(pDestination); |
||
| 5149 | |||
| 5150 | N = XMVectorSaturate(V); |
||
| 5151 | N = XMVectorMultiply(N, Scale.v); |
||
| 5152 | N = XMVectorRound(N); |
||
| 5153 | |||
| 5154 | pDestination->x = (BYTE)N.vector4_f32[0]; |
||
| 5155 | pDestination->y = (BYTE)N.vector4_f32[1]; |
||
| 5156 | pDestination->z = (BYTE)N.vector4_f32[2]; |
||
| 5157 | pDestination->w = (BYTE)N.vector4_f32[3]; |
||
| 5158 | |||
| 5159 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 5160 | XMASSERT(pDestination); |
||
| 5161 | static const XMVECTORF32 ScaleUByteN4 = {255.0f,255.0f*256.0f*0.5f,255.0f*256.0f*256.0f,255.0f*256.0f*256.0f*256.0f*0.5f}; |
||
| 5162 | static const XMVECTORI32 MaskUByteN4 = {0xFF,0xFF<<(8-1),0xFF<<16,0xFF<<(24-1)}; |
||
| 5163 | // Clamp to bounds |
||
| 5164 | XMVECTOR vResult = _mm_max_ps(V,g_XMZero); |
||
| 5165 | vResult = _mm_min_ps(vResult,g_XMOne); |
||
| 5166 | // Scale by multiplication |
||
| 5167 | vResult = _mm_mul_ps(vResult,ScaleUByteN4); |
||
| 5168 | // Convert to int |
||
| 5169 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 5170 | // Mask off any fraction |
||
| 5171 | vResulti = _mm_and_si128(vResulti,MaskUByteN4); |
||
| 5172 | // Do a horizontal or of 4 entries |
||
| 5173 | __m128i vResulti2 = _mm_shuffle_epi32(vResulti,_MM_SHUFFLE(3,2,3,2)); |
||
| 5174 | // x = x|z, y = y|w |
||
| 5175 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 5176 | // Move Z to the x position |
||
| 5177 | vResulti2 = _mm_shuffle_epi32(vResulti,_MM_SHUFFLE(1,1,1,1)); |
||
| 5178 | // Perform a single bit left shift to fix y|w |
||
| 5179 | vResulti2 = _mm_add_epi32(vResulti2,vResulti2); |
||
| 5180 | // i = x|y|z|w |
||
| 5181 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 5182 | _mm_store_ss(reinterpret_cast<float *>(&pDestination->v),reinterpret_cast<const __m128 *>(&vResulti)[0]); |
||
| 5183 | #else // _XM_VMX128_INTRINSICS_ |
||
| 5184 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 5185 | } |
||
| 5186 | |||
| 5187 | //------------------------------------------------------------------------------ |
||
| 5188 | |||
| 5189 | XMFINLINE VOID XMStoreUByte4 |
||
| 5190 | ( |
||
| 5191 | XMUBYTE4* pDestination, |
||
| 5192 | FXMVECTOR V |
||
| 5193 | ) |
||
| 5194 | { |
||
| 5195 | #if defined(_XM_NO_INTRINSICS_) |
||
| 5196 | |||
| 5197 | XMVECTOR N; |
||
| 5198 | static CONST XMVECTOR Max = {255.0f, 255.0f, 255.0f, 255.0f}; |
||
| 5199 | |||
| 5200 | XMASSERT(pDestination); |
||
| 5201 | |||
| 5202 | N = XMVectorClamp(V, XMVectorZero(), Max); |
||
| 5203 | N = XMVectorRound(N); |
||
| 5204 | |||
| 5205 | pDestination->x = (BYTE)N.vector4_f32[0]; |
||
| 5206 | pDestination->y = (BYTE)N.vector4_f32[1]; |
||
| 5207 | pDestination->z = (BYTE)N.vector4_f32[2]; |
||
| 5208 | pDestination->w = (BYTE)N.vector4_f32[3]; |
||
| 5209 | |||
| 5210 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 5211 | XMASSERT(pDestination); |
||
| 5212 | static const XMVECTORF32 MaxUByte4 = { 255.0f, 255.0f, 255.0f, 255.0f}; |
||
| 5213 | static const XMVECTORF32 ScaleUByte4 = {1.0f,256.0f*0.5f,256.0f*256.0f,256.0f*256.0f*256.0f*0.5f}; |
||
| 5214 | static const XMVECTORI32 MaskUByte4 = {0xFF,0xFF<<(8-1),0xFF<<16,0xFF<<(24-1)}; |
||
| 5215 | // Clamp to bounds |
||
| 5216 | XMVECTOR vResult = _mm_max_ps(V,g_XMZero); |
||
| 5217 | vResult = _mm_min_ps(vResult,MaxUByte4); |
||
| 5218 | // Scale by multiplication |
||
| 5219 | vResult = _mm_mul_ps(vResult,ScaleUByte4); |
||
| 5220 | // Convert to int |
||
| 5221 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 5222 | // Mask off any fraction |
||
| 5223 | vResulti = _mm_and_si128(vResulti,MaskUByte4); |
||
| 5224 | // Do a horizontal or of 4 entries |
||
| 5225 | __m128i vResulti2 = _mm_shuffle_epi32(vResulti,_MM_SHUFFLE(3,2,3,2)); |
||
| 5226 | // x = x|z, y = y|w |
||
| 5227 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 5228 | // Move Z to the x position |
||
| 5229 | vResulti2 = _mm_shuffle_epi32(vResulti,_MM_SHUFFLE(1,1,1,1)); |
||
| 5230 | // Perform a single bit left shift to fix y|w |
||
| 5231 | vResulti2 = _mm_add_epi32(vResulti2,vResulti2); |
||
| 5232 | // i = x|y|z|w |
||
| 5233 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 5234 | _mm_store_ss(reinterpret_cast<float *>(&pDestination->v),reinterpret_cast<const __m128 *>(&vResulti)[0]); |
||
| 5235 | #else // _XM_VMX128_INTRINSICS_ |
||
| 5236 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 5237 | } |
||
| 5238 | |||
| 5239 | //------------------------------------------------------------------------------ |
||
| 5240 | |||
| 5241 | XMFINLINE VOID XMStoreByteN4 |
||
| 5242 | ( |
||
| 5243 | XMBYTEN4* pDestination, |
||
| 5244 | FXMVECTOR V |
||
| 5245 | ) |
||
| 5246 | { |
||
| 5247 | #if defined(_XM_NO_INTRINSICS_) |
||
| 5248 | |||
| 5249 | XMVECTOR N; |
||
| 5250 | static CONST XMVECTORF32 Scale = {127.0f, 127.0f, 127.0f, 127.0f}; |
||
| 5251 | |||
| 5252 | XMASSERT(pDestination); |
||
| 5253 | |||
| 5254 | N = XMVectorMultiply(V, Scale.v); |
||
| 5255 | N = XMVectorRound(N); |
||
| 5256 | |||
| 5257 | pDestination->x = (CHAR)N.vector4_f32[0]; |
||
| 5258 | pDestination->y = (CHAR)N.vector4_f32[1]; |
||
| 5259 | pDestination->z = (CHAR)N.vector4_f32[2]; |
||
| 5260 | pDestination->w = (CHAR)N.vector4_f32[3]; |
||
| 5261 | |||
| 5262 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 5263 | XMASSERT(pDestination); |
||
| 5264 | static const XMVECTORF32 ScaleByteN4 = {127.0f,127.0f*256.0f,127.0f*256.0f*256.0f,127.0f*256.0f*256.0f*256.0f}; |
||
| 5265 | static const XMVECTORI32 MaskByteN4 = {0xFF,0xFF<<8,0xFF<<16,0xFF<<24}; |
||
| 5266 | // Clamp to bounds |
||
| 5267 | XMVECTOR vResult = _mm_max_ps(V,g_XMNegativeOne); |
||
| 5268 | vResult = _mm_min_ps(vResult,g_XMOne); |
||
| 5269 | // Scale by multiplication |
||
| 5270 | vResult = _mm_mul_ps(vResult,ScaleByteN4); |
||
| 5271 | // Convert to int |
||
| 5272 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 5273 | // Mask off any fraction |
||
| 5274 | vResulti = _mm_and_si128(vResulti,MaskByteN4); |
||
| 5275 | // Do a horizontal or of 4 entries |
||
| 5276 | __m128i vResulti2 = _mm_shuffle_epi32(vResulti,_MM_SHUFFLE(3,2,3,2)); |
||
| 5277 | // x = x|z, y = y|w |
||
| 5278 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 5279 | // Move Z to the x position |
||
| 5280 | vResulti2 = _mm_shuffle_epi32(vResulti,_MM_SHUFFLE(1,1,1,1)); |
||
| 5281 | // i = x|y|z|w |
||
| 5282 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 5283 | _mm_store_ss(reinterpret_cast<float *>(&pDestination->v),reinterpret_cast<const __m128 *>(&vResulti)[0]); |
||
| 5284 | #else // _XM_VMX128_INTRINSICS_ |
||
| 5285 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 5286 | } |
||
| 5287 | |||
| 5288 | //------------------------------------------------------------------------------ |
||
| 5289 | |||
| 5290 | XMFINLINE VOID XMStoreByte4 |
||
| 5291 | ( |
||
| 5292 | XMBYTE4* pDestination, |
||
| 5293 | FXMVECTOR V |
||
| 5294 | ) |
||
| 5295 | { |
||
| 5296 | #if defined(_XM_NO_INTRINSICS_) |
||
| 5297 | |||
| 5298 | XMVECTOR N; |
||
| 5299 | static CONST XMVECTOR Min = {-127.0f, -127.0f, -127.0f, -127.0f}; |
||
| 5300 | static CONST XMVECTOR Max = {127.0f, 127.0f, 127.0f, 127.0f}; |
||
| 5301 | |||
| 5302 | XMASSERT(pDestination); |
||
| 5303 | |||
| 5304 | N = XMVectorClamp(V, Min, Max); |
||
| 5305 | N = XMVectorRound(N); |
||
| 5306 | |||
| 5307 | pDestination->x = (CHAR)N.vector4_f32[0]; |
||
| 5308 | pDestination->y = (CHAR)N.vector4_f32[1]; |
||
| 5309 | pDestination->z = (CHAR)N.vector4_f32[2]; |
||
| 5310 | pDestination->w = (CHAR)N.vector4_f32[3]; |
||
| 5311 | |||
| 5312 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 5313 | XMASSERT(pDestination); |
||
| 5314 | static const XMVECTORF32 MinByte4 = {-127.0f,-127.0f,-127.0f,-127.0f}; |
||
| 5315 | static const XMVECTORF32 MaxByte4 = { 127.0f, 127.0f, 127.0f, 127.0f}; |
||
| 5316 | static const XMVECTORF32 ScaleByte4 = {1.0f,256.0f,256.0f*256.0f,256.0f*256.0f*256.0f}; |
||
| 5317 | static const XMVECTORI32 MaskByte4 = {0xFF,0xFF<<8,0xFF<<16,0xFF<<24}; |
||
| 5318 | // Clamp to bounds |
||
| 5319 | XMVECTOR vResult = _mm_max_ps(V,MinByte4); |
||
| 5320 | vResult = _mm_min_ps(vResult,MaxByte4); |
||
| 5321 | // Scale by multiplication |
||
| 5322 | vResult = _mm_mul_ps(vResult,ScaleByte4); |
||
| 5323 | // Convert to int |
||
| 5324 | __m128i vResulti = _mm_cvttps_epi32(vResult); |
||
| 5325 | // Mask off any fraction |
||
| 5326 | vResulti = _mm_and_si128(vResulti,MaskByte4); |
||
| 5327 | // Do a horizontal or of 4 entries |
||
| 5328 | __m128i vResulti2 = _mm_shuffle_epi32(vResulti,_MM_SHUFFLE(3,2,3,2)); |
||
| 5329 | // x = x|z, y = y|w |
||
| 5330 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 5331 | // Move Z to the x position |
||
| 5332 | vResulti2 = _mm_shuffle_epi32(vResulti,_MM_SHUFFLE(1,1,1,1)); |
||
| 5333 | // i = x|y|z|w |
||
| 5334 | vResulti = _mm_or_si128(vResulti,vResulti2); |
||
| 5335 | _mm_store_ss(reinterpret_cast<float *>(&pDestination->v),reinterpret_cast<const __m128 *>(&vResulti)[0]); |
||
| 5336 | #else // _XM_VMX128_INTRINSICS_ |
||
| 5337 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 5338 | } |
||
| 5339 | |||
| 5340 | //------------------------------------------------------------------------------ |
||
| 5341 | |||
| 5342 | XMFINLINE VOID XMStoreUNibble4 |
||
| 5343 | ( |
||
| 5344 | XMUNIBBLE4* pDestination, |
||
| 5345 | FXMVECTOR V |
||
| 5346 | ) |
||
| 5347 | { |
||
| 5348 | #if defined(_XM_SSE_INTRINSICS_) && !defined(_XM_NO_INTRINSICS_) |
||
| 5349 | XMASSERT(pDestination); |
||
| 5350 | static CONST XMVECTORF32 Max = {15.0f,15.0f,15.0f,15.0f}; |
||
| 5351 | // Bounds check |
||
| 5352 | XMVECTOR vResult = _mm_max_ps(V,g_XMZero); |
||
| 5353 | vResult = _mm_min_ps(vResult,Max); |
||
| 5354 | // Convert to int with rounding |
||
| 5355 | __m128i vInt = _mm_cvtps_epi32(vResult); |
||
| 5356 | // No SSE operations will write to 16-bit values, so we have to extract them manually |
||
| 5357 | USHORT x = static_cast<USHORT>(_mm_extract_epi16(vInt,0)); |
||
| 5358 | USHORT y = static_cast<USHORT>(_mm_extract_epi16(vInt,2)); |
||
| 5359 | USHORT z = static_cast<USHORT>(_mm_extract_epi16(vInt,4)); |
||
| 5360 | USHORT w = static_cast<USHORT>(_mm_extract_epi16(vInt,6)); |
||
| 5361 | pDestination->v = ((w & 0xF) << 12) | |
||
| 5362 | ((z & 0xF) << 8) | |
||
| 5363 | ((y & 0xF) << 4) | |
||
| 5364 | ((x & 0xF)); |
||
| 5365 | #else |
||
| 5366 | XMVECTOR N; |
||
| 5367 | static CONST XMVECTORF32 Max = {15.0f,15.0f,15.0f,15.0f}; |
||
| 5368 | |||
| 5369 | XMASSERT(pDestination); |
||
| 5370 | |||
| 5371 | N = XMVectorClamp(V, XMVectorZero(), Max.v); |
||
| 5372 | N = XMVectorRound(N); |
||
| 5373 | |||
| 5374 | pDestination->v = (((USHORT)N.vector4_f32[3] & 0xF) << 12) | |
||
| 5375 | (((USHORT)N.vector4_f32[2] & 0xF) << 8) | |
||
| 5376 | (((USHORT)N.vector4_f32[1] & 0xF) << 4) | |
||
| 5377 | (((USHORT)N.vector4_f32[0] & 0xF)); |
||
| 5378 | #endif !_XM_SSE_INTRINSICS_ |
||
| 5379 | } |
||
| 5380 | |||
| 5381 | //------------------------------------------------------------------------------ |
||
| 5382 | |||
| 5383 | XMFINLINE VOID XMStoreU555( |
||
| 5384 | XMU555* pDestination, |
||
| 5385 | FXMVECTOR V |
||
| 5386 | ) |
||
| 5387 | { |
||
| 5388 | #if defined(_XM_SSE_INTRINSICS_) && !defined(_XM_NO_INTRINSICS_) |
||
| 5389 | XMASSERT(pDestination); |
||
| 5390 | static CONST XMVECTORF32 Max = {31.0f, 31.0f, 31.0f, 1.0f}; |
||
| 5391 | // Bounds check |
||
| 5392 | XMVECTOR vResult = _mm_max_ps(V,g_XMZero); |
||
| 5393 | vResult = _mm_min_ps(vResult,Max); |
||
| 5394 | // Convert to int with rounding |
||
| 5395 | __m128i vInt = _mm_cvtps_epi32(vResult); |
||
| 5396 | // No SSE operations will write to 16-bit values, so we have to extract them manually |
||
| 5397 | USHORT x = static_cast<USHORT>(_mm_extract_epi16(vInt,0)); |
||
| 5398 | USHORT y = static_cast<USHORT>(_mm_extract_epi16(vInt,2)); |
||
| 5399 | USHORT z = static_cast<USHORT>(_mm_extract_epi16(vInt,4)); |
||
| 5400 | USHORT w = static_cast<USHORT>(_mm_extract_epi16(vInt,6)); |
||
| 5401 | pDestination->v = ((w) ? 0x8000 : 0) | |
||
| 5402 | ((z & 0x1F) << 10) | |
||
| 5403 | ((y & 0x1F) << 5) | |
||
| 5404 | ((x & 0x1F)); |
||
| 5405 | #else |
||
| 5406 | XMVECTOR N; |
||
| 5407 | static CONST XMVECTORF32 Max = {31.0f, 31.0f, 31.0f, 1.0f}; |
||
| 5408 | |||
| 5409 | XMASSERT(pDestination); |
||
| 5410 | |||
| 5411 | N = XMVectorClamp(V, XMVectorZero(), Max.v); |
||
| 5412 | N = XMVectorRound(N); |
||
| 5413 | |||
| 5414 | pDestination->v = ((N.vector4_f32[3] > 0.f) ? 0x8000 : 0) | |
||
| 5415 | (((USHORT)N.vector4_f32[2] & 0x1F) << 10) | |
||
| 5416 | (((USHORT)N.vector4_f32[1] & 0x1F) << 5) | |
||
| 5417 | (((USHORT)N.vector4_f32[0] & 0x1F)); |
||
| 5418 | #endif !_XM_SSE_INTRINSICS_ |
||
| 5419 | } |
||
| 5420 | |||
| 5421 | //------------------------------------------------------------------------------ |
||
| 5422 | |||
| 5423 | XMFINLINE VOID XMStoreColor |
||
| 5424 | ( |
||
| 5425 | XMCOLOR* pDestination, |
||
| 5426 | FXMVECTOR V |
||
| 5427 | ) |
||
| 5428 | { |
||
| 5429 | #if defined(_XM_NO_INTRINSICS_) |
||
| 5430 | |||
| 5431 | XMVECTOR N; |
||
| 5432 | static CONST XMVECTORF32 Scale = {255.0f, 255.0f, 255.0f, 255.0f}; |
||
| 5433 | |||
| 5434 | XMASSERT(pDestination); |
||
| 5435 | |||
| 5436 | N = XMVectorSaturate(V); |
||
| 5437 | N = XMVectorMultiply(N, Scale.v); |
||
| 5438 | N = XMVectorRound(N); |
||
| 5439 | |||
| 5440 | pDestination->c = ((UINT)N.vector4_f32[3] << 24) | |
||
| 5441 | ((UINT)N.vector4_f32[0] << 16) | |
||
| 5442 | ((UINT)N.vector4_f32[1] << 8) | |
||
| 5443 | ((UINT)N.vector4_f32[2]); |
||
| 5444 | |||
| 5445 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 5446 | XMASSERT(pDestination); |
||
| 5447 | static CONST XMVECTORF32 Scale = {255.0f,255.0f,255.0f,255.0f}; |
||
| 5448 | // Set <0 to 0 |
||
| 5449 | XMVECTOR vResult = _mm_max_ps(V,g_XMZero); |
||
| 5450 | // Set>1 to 1 |
||
| 5451 | vResult = _mm_min_ps(vResult,g_XMOne); |
||
| 5452 | // Convert to 0-255 |
||
| 5453 | vResult = _mm_mul_ps(vResult,Scale); |
||
| 5454 | // Shuffle RGBA to ARGB |
||
| 5455 | vResult = _mm_shuffle_ps(vResult,vResult,_MM_SHUFFLE(2,1,0,3)); |
||
| 5456 | // Convert to int |
||
| 5457 | __m128i vInt = _mm_cvtps_epi32(vResult); |
||
| 5458 | // Mash to shorts |
||
| 5459 | vInt = _mm_packs_epi32(vInt,vInt); |
||
| 5460 | // Mash to bytes |
||
| 5461 | vInt = _mm_packs_epi16(vInt,vInt); |
||
| 5462 | // Store the color |
||
| 5463 | _mm_store_ss(reinterpret_cast<float *>(&pDestination->c),reinterpret_cast<__m128 *>(&vInt)[0]); |
||
| 5464 | #else // _XM_VMX128_INTRINSICS_ |
||
| 5465 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 5466 | } |
||
| 5467 | |||
| 5468 | //------------------------------------------------------------------------------ |
||
| 5469 | |||
| 5470 | XMFINLINE VOID XMStoreFloat3x3 |
||
| 5471 | ( |
||
| 5472 | XMFLOAT3X3* pDestination, |
||
| 5473 | CXMMATRIX M |
||
| 5474 | ) |
||
| 5475 | { |
||
| 5476 | #if defined(_XM_NO_INTRINSICS_) || defined(XM_NO_MISALIGNED_VECTOR_ACCESS) || defined(_XM_SSE_INTRINSICS_) |
||
| 5477 | |||
| 5478 | XMStoreFloat3x3NC(pDestination, M); |
||
| 5479 | |||
| 5480 | #else // _XM_VMX128_INTRINSICS_ |
||
| 5481 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 5482 | } |
||
| 5483 | |||
| 5484 | //------------------------------------------------------------------------------ |
||
| 5485 | |||
| 5486 | XMFINLINE VOID XMStoreFloat3x3NC |
||
| 5487 | ( |
||
| 5488 | XMFLOAT3X3* pDestination, |
||
| 5489 | CXMMATRIX M |
||
| 5490 | ) |
||
| 5491 | { |
||
| 5492 | #if defined(_XM_NO_INTRINSICS_) |
||
| 5493 | |||
| 5494 | XMASSERT(pDestination); |
||
| 5495 | |||
| 5496 | pDestination->m[0][0] = M.r[0].vector4_f32[0]; |
||
| 5497 | pDestination->m[0][1] = M.r[0].vector4_f32[1]; |
||
| 5498 | pDestination->m[0][2] = M.r[0].vector4_f32[2]; |
||
| 5499 | |||
| 5500 | pDestination->m[1][0] = M.r[1].vector4_f32[0]; |
||
| 5501 | pDestination->m[1][1] = M.r[1].vector4_f32[1]; |
||
| 5502 | pDestination->m[1][2] = M.r[1].vector4_f32[2]; |
||
| 5503 | |||
| 5504 | pDestination->m[2][0] = M.r[2].vector4_f32[0]; |
||
| 5505 | pDestination->m[2][1] = M.r[2].vector4_f32[1]; |
||
| 5506 | pDestination->m[2][2] = M.r[2].vector4_f32[2]; |
||
| 5507 | |||
| 5508 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 5509 | XMASSERT(pDestination); |
||
| 5510 | XMVECTOR vTemp1 = M.r[0]; |
||
| 5511 | XMVECTOR vTemp2 = M.r[1]; |
||
| 5512 | XMVECTOR vTemp3 = M.r[2]; |
||
| 5513 | XMVECTOR vWork = _mm_shuffle_ps(vTemp1,vTemp2,_MM_SHUFFLE(0,0,2,2)); |
||
| 5514 | vTemp1 = _mm_shuffle_ps(vTemp1,vWork,_MM_SHUFFLE(2,0,1,0)); |
||
| 5515 | _mm_storeu_ps(&pDestination->m[0][0],vTemp1); |
||
| 5516 | vTemp2 = _mm_shuffle_ps(vTemp2,vTemp3,_MM_SHUFFLE(1,0,2,1)); |
||
| 5517 | _mm_storeu_ps(&pDestination->m[1][1],vTemp2); |
||
| 5518 | vTemp3 = _mm_shuffle_ps(vTemp3,vTemp3,_MM_SHUFFLE(2,2,2,2)); |
||
| 5519 | _mm_store_ss(&pDestination->m[2][2],vTemp3); |
||
| 5520 | #else // _XM_VMX128_INTRINSICS_ |
||
| 5521 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 5522 | } |
||
| 5523 | |||
| 5524 | //------------------------------------------------------------------------------ |
||
| 5525 | |||
| 5526 | XMFINLINE VOID XMStoreFloat4x3 |
||
| 5527 | ( |
||
| 5528 | XMFLOAT4X3* pDestination, |
||
| 5529 | CXMMATRIX M |
||
| 5530 | ) |
||
| 5531 | { |
||
| 5532 | #if defined(_XM_NO_INTRINSICS_) || defined(XM_NO_MISALIGNED_VECTOR_ACCESS) || defined(_XM_SSE_INTRINSICS_) |
||
| 5533 | |||
| 5534 | XMStoreFloat4x3NC(pDestination, M); |
||
| 5535 | |||
| 5536 | #else // _XM_VMX128_INTRINSICS_ |
||
| 5537 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 5538 | } |
||
| 5539 | |||
| 5540 | //------------------------------------------------------------------------------ |
||
| 5541 | |||
| 5542 | XMFINLINE VOID XMStoreFloat4x3A |
||
| 5543 | ( |
||
| 5544 | XMFLOAT4X3A* pDestination, |
||
| 5545 | CXMMATRIX M |
||
| 5546 | ) |
||
| 5547 | { |
||
| 5548 | #if defined(_XM_NO_INTRINSICS_) |
||
| 5549 | |||
| 5550 | XMASSERT(pDestination); |
||
| 5551 | XMASSERT(((UINT_PTR)pDestination & 0xF) == 0); |
||
| 5552 | |||
| 5553 | pDestination->m[0][0] = M.r[0].vector4_f32[0]; |
||
| 5554 | pDestination->m[0][1] = M.r[0].vector4_f32[1]; |
||
| 5555 | pDestination->m[0][2] = M.r[0].vector4_f32[2]; |
||
| 5556 | |||
| 5557 | pDestination->m[1][0] = M.r[1].vector4_f32[0]; |
||
| 5558 | pDestination->m[1][1] = M.r[1].vector4_f32[1]; |
||
| 5559 | pDestination->m[1][2] = M.r[1].vector4_f32[2]; |
||
| 5560 | |||
| 5561 | pDestination->m[2][0] = M.r[2].vector4_f32[0]; |
||
| 5562 | pDestination->m[2][1] = M.r[2].vector4_f32[1]; |
||
| 5563 | pDestination->m[2][2] = M.r[2].vector4_f32[2]; |
||
| 5564 | |||
| 5565 | pDestination->m[3][0] = M.r[3].vector4_f32[0]; |
||
| 5566 | pDestination->m[3][1] = M.r[3].vector4_f32[1]; |
||
| 5567 | pDestination->m[3][2] = M.r[3].vector4_f32[2]; |
||
| 5568 | |||
| 5569 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 5570 | XMASSERT(pDestination); |
||
| 5571 | XMASSERT(((UINT_PTR)pDestination & 0xF) == 0); |
||
| 5572 | // x1,y1,z1,w1 |
||
| 5573 | XMVECTOR vTemp1 = M.r[0]; |
||
| 5574 | // x2,y2,z2,w2 |
||
| 5575 | XMVECTOR vTemp2 = M.r[1]; |
||
| 5576 | // x3,y3,z3,w3 |
||
| 5577 | XMVECTOR vTemp3 = M.r[2]; |
||
| 5578 | // x4,y4,z4,w4 |
||
| 5579 | XMVECTOR vTemp4 = M.r[3]; |
||
| 5580 | // z1,z1,x2,y2 |
||
| 5581 | XMVECTOR vTemp = _mm_shuffle_ps(vTemp1,vTemp2,_MM_SHUFFLE(1,0,2,2)); |
||
| 5582 | // y2,z2,x3,y3 (Final) |
||
| 5583 | vTemp2 = _mm_shuffle_ps(vTemp2,vTemp3,_MM_SHUFFLE(1,0,2,1)); |
||
| 5584 | // x1,y1,z1,x2 (Final) |
||
| 5585 | vTemp1 = _mm_shuffle_ps(vTemp1,vTemp,_MM_SHUFFLE(2,0,1,0)); |
||
| 5586 | // z3,z3,x4,x4 |
||
| 5587 | vTemp3 = _mm_shuffle_ps(vTemp3,vTemp4,_MM_SHUFFLE(0,0,2,2)); |
||
| 5588 | // z3,x4,y4,z4 (Final) |
||
| 5589 | vTemp3 = _mm_shuffle_ps(vTemp3,vTemp4,_MM_SHUFFLE(2,1,2,0)); |
||
| 5590 | // Store in 3 operations |
||
| 5591 | _mm_store_ps(&pDestination->m[0][0],vTemp1); |
||
| 5592 | _mm_store_ps(&pDestination->m[1][1],vTemp2); |
||
| 5593 | _mm_store_ps(&pDestination->m[2][2],vTemp3); |
||
| 5594 | #else // _XM_VMX128_INTRINSICS_ |
||
| 5595 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 5596 | } |
||
| 5597 | |||
| 5598 | //------------------------------------------------------------------------------ |
||
| 5599 | |||
| 5600 | XMFINLINE VOID XMStoreFloat4x3NC |
||
| 5601 | ( |
||
| 5602 | XMFLOAT4X3* pDestination, |
||
| 5603 | CXMMATRIX M |
||
| 5604 | ) |
||
| 5605 | { |
||
| 5606 | #if defined(_XM_NO_INTRINSICS_) |
||
| 5607 | |||
| 5608 | XMASSERT(pDestination); |
||
| 5609 | |||
| 5610 | pDestination->m[0][0] = M.r[0].vector4_f32[0]; |
||
| 5611 | pDestination->m[0][1] = M.r[0].vector4_f32[1]; |
||
| 5612 | pDestination->m[0][2] = M.r[0].vector4_f32[2]; |
||
| 5613 | |||
| 5614 | pDestination->m[1][0] = M.r[1].vector4_f32[0]; |
||
| 5615 | pDestination->m[1][1] = M.r[1].vector4_f32[1]; |
||
| 5616 | pDestination->m[1][2] = M.r[1].vector4_f32[2]; |
||
| 5617 | |||
| 5618 | pDestination->m[2][0] = M.r[2].vector4_f32[0]; |
||
| 5619 | pDestination->m[2][1] = M.r[2].vector4_f32[1]; |
||
| 5620 | pDestination->m[2][2] = M.r[2].vector4_f32[2]; |
||
| 5621 | |||
| 5622 | pDestination->m[3][0] = M.r[3].vector4_f32[0]; |
||
| 5623 | pDestination->m[3][1] = M.r[3].vector4_f32[1]; |
||
| 5624 | pDestination->m[3][2] = M.r[3].vector4_f32[2]; |
||
| 5625 | |||
| 5626 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 5627 | XMASSERT(pDestination); |
||
| 5628 | XMVECTOR vTemp1 = M.r[0]; |
||
| 5629 | XMVECTOR vTemp2 = M.r[1]; |
||
| 5630 | XMVECTOR vTemp3 = M.r[2]; |
||
| 5631 | XMVECTOR vTemp4 = M.r[3]; |
||
| 5632 | XMVECTOR vTemp2x = _mm_shuffle_ps(vTemp2,vTemp3,_MM_SHUFFLE(1,0,2,1)); |
||
| 5633 | vTemp2 = _mm_shuffle_ps(vTemp2,vTemp1,_MM_SHUFFLE(2,2,0,0)); |
||
| 5634 | vTemp1 = _mm_shuffle_ps(vTemp1,vTemp2,_MM_SHUFFLE(0,2,1,0)); |
||
| 5635 | vTemp3 = _mm_shuffle_ps(vTemp3,vTemp4,_MM_SHUFFLE(0,0,2,2)); |
||
| 5636 | vTemp3 = _mm_shuffle_ps(vTemp3,vTemp4,_MM_SHUFFLE(2,1,2,0)); |
||
| 5637 | _mm_storeu_ps(&pDestination->m[0][0],vTemp1); |
||
| 5638 | _mm_storeu_ps(&pDestination->m[1][1],vTemp2x); |
||
| 5639 | _mm_storeu_ps(&pDestination->m[2][2],vTemp3); |
||
| 5640 | #else // _XM_VMX128_INTRINSICS_ |
||
| 5641 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 5642 | } |
||
| 5643 | |||
| 5644 | //------------------------------------------------------------------------------ |
||
| 5645 | |||
| 5646 | XMFINLINE VOID XMStoreFloat4x4 |
||
| 5647 | ( |
||
| 5648 | XMFLOAT4X4* pDestination, |
||
| 5649 | CXMMATRIX M |
||
| 5650 | ) |
||
| 5651 | { |
||
| 5652 | #if defined(_XM_NO_INTRINSICS_) || defined(XM_NO_MISALIGNED_VECTOR_ACCESS) |
||
| 5653 | |||
| 5654 | XMStoreFloat4x4NC(pDestination, M); |
||
| 5655 | |||
| 5656 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 5657 | XMASSERT(pDestination); |
||
| 5658 | |||
| 5659 | _mm_storeu_ps( &pDestination->_11, M.r[0] ); |
||
| 5660 | _mm_storeu_ps( &pDestination->_21, M.r[1] ); |
||
| 5661 | _mm_storeu_ps( &pDestination->_31, M.r[2] ); |
||
| 5662 | _mm_storeu_ps( &pDestination->_41, M.r[3] ); |
||
| 5663 | #else // _XM_VMX128_INTRINSICS_ |
||
| 5664 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 5665 | } |
||
| 5666 | |||
| 5667 | //------------------------------------------------------------------------------ |
||
| 5668 | |||
| 5669 | XMFINLINE VOID XMStoreFloat4x4A |
||
| 5670 | ( |
||
| 5671 | XMFLOAT4X4A* pDestination, |
||
| 5672 | CXMMATRIX M |
||
| 5673 | ) |
||
| 5674 | { |
||
| 5675 | #if defined(_XM_NO_INTRINSICS_) |
||
| 5676 | |||
| 5677 | XMASSERT(pDestination); |
||
| 5678 | XMASSERT(((UINT_PTR)pDestination & 0xF) == 0); |
||
| 5679 | |||
| 5680 | pDestination->m[0][0] = M.r[0].vector4_f32[0]; |
||
| 5681 | pDestination->m[0][1] = M.r[0].vector4_f32[1]; |
||
| 5682 | pDestination->m[0][2] = M.r[0].vector4_f32[2]; |
||
| 5683 | pDestination->m[0][3] = M.r[0].vector4_f32[3]; |
||
| 5684 | |||
| 5685 | pDestination->m[1][0] = M.r[1].vector4_f32[0]; |
||
| 5686 | pDestination->m[1][1] = M.r[1].vector4_f32[1]; |
||
| 5687 | pDestination->m[1][2] = M.r[1].vector4_f32[2]; |
||
| 5688 | pDestination->m[1][3] = M.r[1].vector4_f32[3]; |
||
| 5689 | |||
| 5690 | pDestination->m[2][0] = M.r[2].vector4_f32[0]; |
||
| 5691 | pDestination->m[2][1] = M.r[2].vector4_f32[1]; |
||
| 5692 | pDestination->m[2][2] = M.r[2].vector4_f32[2]; |
||
| 5693 | pDestination->m[2][3] = M.r[2].vector4_f32[3]; |
||
| 5694 | |||
| 5695 | pDestination->m[3][0] = M.r[3].vector4_f32[0]; |
||
| 5696 | pDestination->m[3][1] = M.r[3].vector4_f32[1]; |
||
| 5697 | pDestination->m[3][2] = M.r[3].vector4_f32[2]; |
||
| 5698 | pDestination->m[3][3] = M.r[3].vector4_f32[3]; |
||
| 5699 | |||
| 5700 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 5701 | XMASSERT(pDestination); |
||
| 5702 | |||
| 5703 | _mm_store_ps( &pDestination->_11, M.r[0] ); |
||
| 5704 | _mm_store_ps( &pDestination->_21, M.r[1] ); |
||
| 5705 | _mm_store_ps( &pDestination->_31, M.r[2] ); |
||
| 5706 | _mm_store_ps( &pDestination->_41, M.r[3] ); |
||
| 5707 | #else // _XM_VMX128_INTRINSICS_ |
||
| 5708 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 5709 | } |
||
| 5710 | |||
| 5711 | //------------------------------------------------------------------------------ |
||
| 5712 | |||
| 5713 | XMFINLINE VOID XMStoreFloat4x4NC |
||
| 5714 | ( |
||
| 5715 | XMFLOAT4X4* pDestination, |
||
| 5716 | CXMMATRIX M |
||
| 5717 | ) |
||
| 5718 | { |
||
| 5719 | #if defined(_XM_NO_INTRINSICS_) |
||
| 5720 | |||
| 5721 | XMASSERT(pDestination); |
||
| 5722 | |||
| 5723 | pDestination->m[0][0] = M.r[0].vector4_f32[0]; |
||
| 5724 | pDestination->m[0][1] = M.r[0].vector4_f32[1]; |
||
| 5725 | pDestination->m[0][2] = M.r[0].vector4_f32[2]; |
||
| 5726 | pDestination->m[0][3] = M.r[0].vector4_f32[3]; |
||
| 5727 | |||
| 5728 | pDestination->m[1][0] = M.r[1].vector4_f32[0]; |
||
| 5729 | pDestination->m[1][1] = M.r[1].vector4_f32[1]; |
||
| 5730 | pDestination->m[1][2] = M.r[1].vector4_f32[2]; |
||
| 5731 | pDestination->m[1][3] = M.r[1].vector4_f32[3]; |
||
| 5732 | |||
| 5733 | pDestination->m[2][0] = M.r[2].vector4_f32[0]; |
||
| 5734 | pDestination->m[2][1] = M.r[2].vector4_f32[1]; |
||
| 5735 | pDestination->m[2][2] = M.r[2].vector4_f32[2]; |
||
| 5736 | pDestination->m[2][3] = M.r[2].vector4_f32[3]; |
||
| 5737 | |||
| 5738 | pDestination->m[3][0] = M.r[3].vector4_f32[0]; |
||
| 5739 | pDestination->m[3][1] = M.r[3].vector4_f32[1]; |
||
| 5740 | pDestination->m[3][2] = M.r[3].vector4_f32[2]; |
||
| 5741 | pDestination->m[3][3] = M.r[3].vector4_f32[3]; |
||
| 5742 | |||
| 5743 | #elif defined(_XM_SSE_INTRINSICS_) |
||
| 5744 | XMASSERT(pDestination); |
||
| 5745 | _mm_storeu_ps(&pDestination->m[0][0],M.r[0]); |
||
| 5746 | _mm_storeu_ps(&pDestination->m[1][0],M.r[1]); |
||
| 5747 | _mm_storeu_ps(&pDestination->m[2][0],M.r[2]); |
||
| 5748 | _mm_storeu_ps(&pDestination->m[3][0],M.r[3]); |
||
| 5749 | #else // _XM_VMX128_INTRINSICS_ |
||
| 5750 | #endif // _XM_VMX128_INTRINSICS_ |
||
| 5751 | } |
||
| 5752 | |||
| 5753 | #endif // __XNAMATHCONVERT_INL__ |
||
| 5754 |