Intel SSE / MMX2 / KNI documentation
转自 http://intel80386.com/simd/mmx2-doc.html
Please note, this is a work-in-progress (ie BETA). Timings are of approximate throughput cycles using average from TSC, the latency and ranges are indicated where known. ADDPS Add Parallel Scalars Opcode Cycles Instruction 0F 58 2 (3) ADDPS xmm reg,xmm reg/mem128 ADDPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op1[0] = op1[0] + op2[0] op1[1] = op1[1] + op2[1] op1[2] = op1[2] + op2[2] op1[3] = op1[3] + op2[3] ADDSS Add Single Scalar Opcode Cycles Instruction F3 0F 58 1 (3) ADDSS xmm reg,xmm reg/mem32 ADDPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 1 single precision 32-bit floating point value op1[0] = op1[0] + op2 op1[1] = op1[1] op1[2] = op1[2] op1[3] = op1[3] ANDNPS And Not Parallel Scalars (bitwise) Opcode Cycles Instruction 0F 55 2 ANDNPS xmm reg,xmm reg/mem128 ANDNPS op1, op2 op1 contains 1 128-bit value op2 contains 1 128-bit value op1 = !op1 & op2 ANDPS And Parallel Scalars (bitwise) Opcode Cycles Instruction 0F 54 2 ANDPS xmm reg,xmm reg/mem128 ANDPS op1, op2 op1 contains 1 128-bit value op2 contains 1 128-bit value op1 = op1 & op2 CMPEQPS Compare Equal Parallel Scalars Opcode Cycles Instruction 0F C2 .. 00 2 (3) CMPEQPS xmm reg,xmm reg/mem128 CMPEQPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op1[0] = op1[0] == op2[0] op1[1] = op1[1] == op2[1] op1[2] = op1[2] == op2[2] op1[3] = op1[3] == op2[3] TRUE = 0xFFFFFFFF FALSE = 0x00000000 CMPEQSS Compare Equal Single Scalar Opcode Cycles Instruction F3 0F C2 .. 00 1 (3) CMPEQSS xmm reg,xmm reg/mem32 CMPEQSS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 1 single precision 32-bit floating point value op1[0] = op1[0] == op2 op1[1] = op1[1] op1[2] = op1[2] op1[3] = op1[3] TRUE = 0xFFFFFFFF FALSE = 0x00000000 CMPLEPS Compare Less than or Equal Parallel Scalars Opcode Cycles Instruction 0F C2 .. 02 2 (3) CMPLEPS xmm reg,xmm reg/mem128 CMPLEPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op1[0] = op1[0] <= op2[0] op1[1] = op1[1] <= op2[1] op1[2] = op1[2] <= op2[2] op1[3] = op1[3] <= op2[3] TRUE = 0xFFFFFFFF FALSE = 0x00000000 CMPLESS Compare Less than or Equal Single Scalar Opcode Cycles Instruction F3 0F C2 .. 02 1 (3) CMPLESS xmm reg,xmm reg/mem32 CMPLESS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 1 single precision 32-bit floating point value op1[0] = op1[0] <= op2 op1[1] = op1[1] op1[2] = op1[2] op1[3] = op1[3] TRUE = 0xFFFFFFFF FALSE = 0x00000000 CMPLTPS Compare Less Than Parallel Scalars Opcode Cycles Instruction 0F C2 .. 01 2 (3) CMPLTPS xmm reg,xmm reg/mem128 CMPLTPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op1[0] = op1[0] < op2[0] op1[1] = op1[1] < op2[1] op1[2] = op1[2] < op2[2] op1[3] = op1[3] < op2[3] TRUE = 0xFFFFFFFF FALSE = 0x00000000 CMPLTSS Compare Less Than Single Scalar Opcode Cycles Instruction F3 0F C2 .. 01 1 (3) CMPLTSS xmm reg,xmm reg/mem32 CMPLTSS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 1 single precision 32-bit floating point value op1[0] = op1[0] < op2 op1[1] = op1[1] op1[2] = op1[2] op1[3] = op1[3] TRUE = 0xFFFFFFFF FALSE = 0x00000000 CMPNEQPS Compare Not Equal Parallel Scalars Opcode Cycles Instruction 0F C2 .. 04 2 (3) CMPNEQPS xmm reg,xmm reg/mem128 CMPNEQPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op1[0] = op1[0] != op2[0] op1[1] = op1[1] != op2[1] op1[2] = op1[2] != op2[2] op1[3] = op1[3] != op2[3] TRUE = 0xFFFFFFFF FALSE = 0x00000000 CMPNEQSS Compare Not Equal Single Scalar Opcode Cycles Instruction F3 0F C2 .. 04 1 (3) CMPNEQSS xmm reg,xmm reg/mem32 CMPNEQSS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 1 single precision 32-bit floating point value op1[0] = op1[0] != op2 op1[1] = op1[1] op1[2] = op1[2] op1[3] = op1[3] TRUE = 0xFFFFFFFF FALSE = 0x00000000 CMPNLEPS Compare Not Less than or Equal Parallel Scalars Opcode Cycles Instruction 0F C2 .. 06 2 (3) CMPNLEPS xmm reg,xmm reg/mem128 CMPNLEPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op1[0] = op1[0] > op2[0] op1[1] = op1[1] > op2[1] op1[2] = op1[2] > op2[2] op1[3] = op1[3] > op2[3] TRUE = 0xFFFFFFFF FALSE = 0x00000000 CMPNLESS Compare Not Less than or Equal Single Scalar Opcode Cycles Instruction F3 0F C2 .. 06 1 (3) CMPNLESS xmm reg,xmm reg/mem32 CMPNLESS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 1 single precision 32-bit floating point value op1[0] = op1[0] > op2 op1[1] = op1[1] op1[2] = op1[2] op1[3] = op1[3] TRUE = 0xFFFFFFFF FALSE = 0x00000000 CMPNLTPS Compare Not Less Than Parallel Scalars Opcode Cycles Instruction 0F C2 .. 05 2 (3) CMPNLTPS xmm reg,xmm reg/mem128 CMPNLTPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op1[0] = op1[0] >= op2[0] op1[1] = op1[1] >= op2[1] op1[2] = op1[2] >= op2[2] op1[3] = op1[3] >= op2[3] TRUE = 0xFFFFFFFF FALSE = 0x00000000 CMPNLTSS Compare Not Less Than Single Scalar Opcode Cycles Instruction F3 0F C2 .. 01 1 (3) CMPNLTSS xmm reg,xmm reg/mem32 CMPNLTSS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 1 single precision 32-bit floating point value op1[0] = op1[0] >= op2 op1[1] = op1[1] op1[2] = op1[2] op1[3] = op1[3] TRUE = 0xFFFFFFFF FALSE = 0x00000000 CMPORDPS Compare Ordered Parallel Scalars Opcode Cycles Instruction 0F C2 .. 07 2 (3) CMPORDPS xmm reg,xmm reg/mem128 CMPORDPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op1[0] = (op1[0] != NaN) && (op2[0] != NaN) op1[1] = (op1[1] != NaN) && (op2[1] != NaN) op1[2] = (op1[2] != NaN) && (op2[2] != NaN) op1[3] = (op1[3] != NaN) && (op2[3] != NaN) TRUE = 0xFFFFFFFF FALSE = 0x00000000 CMPORDSS Compare Ordered Single Scalar Opcode Cycles Instruction F3 0F C2 .. 07 1 (3) CMPORDSS xmm reg,xmm reg/mem32 CMPORDSS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 1 single precision 32-bit floating point value op1[0] = (op1[0] != NaN) && (op2 != NaN) op1[1] = op1[1] op1[2] = op1[2] op1[3] = op1[3] TRUE = 0xFFFFFFFF FALSE = 0x00000000 CMPUNORDPS Compare Unordered Parallel Scalars Opcode Cycles Instruction 0F C2 .. 03 2 (3) CMPUNORDPS xmm reg,xmm reg/mem128 CMPUNORDPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op1[0] = (op1[0] == NaN) || (op2[0] == NaN) op1[1] = (op1[1] == NaN) || (op2[1] == NaN) op1[2] = (op1[2] == NaN) || (op2[2] == NaN) op1[3] = (op1[3] == NaN) || (op2[3] == NaN) TRUE = 0xFFFFFFFF FALSE = 0x00000000 CMPUNORDSS Compare Unordered Single Scalar Opcode Cycles Instruction F3 0F C2 .. 03 1 (3) CMPUNORDSS xmm reg,xmm reg/mem32 CMPUNORDSS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 1 single precision 32-bit floating point value op1[0] = (op1[0] == NaN) || (op2 == NaN) op1[1] = op1[1] op1[2] = op1[2] op1[3] = op1[3] TRUE = 0xFFFFFFFF FALSE = 0x00000000 COMISS Compare Integer Single Scalar Opcode Cycles Instruction 0F 2F COMISS xmm reg,xmm reg/mem32 COMISS op1, op2 CVTPI2PS Convert Parallel Integer to Parallel Scalars Opcode Cycles Instruction 0F 2A CVTPI2PS xmm reg,mm reg/mem64 CVTPI2PS op1, op2 op1 contains 2 single precision 32-bit floating point values op2 contains 2 32-bit integer values op1[0] = (float)op2[0] op1[1] = (float)op2[1] op1[2] = op1[2] op1[3] = op1[3] CVTPS2PI Convert Parallel Scalars to Parallel Integers Opcode Cycles Instruction 0F 2D CVTPS2PI mm reg,xmm reg/mem128 CVTPS2PI op1, op2 op1 contains 2 32-bit integer values op2 contains 2 single precision 32-bit floating point values op1[0] = (long)op2[0] op1[1] = (long)op2[1] CVTSI2SS Convert Parallel Integers to Parallel Scalars Opcode Cycles Instruction F3 0F 2A CVTSI2SS xmm reg,reg32/mem32 CVTSI2SS op1, op2 op1 contains 1 single precision 32-bit floating point value op2 contains 1 32-bit integer value op1[0] = (float)op2 op1[1] = op1[1] op1[2] = op1[2] op1[3] = op1[3] CVTSS2SI Convert Single Scalar to Single Integer Opcode Cycles Instruction F3 0F 2D CVTSS2SI reg32,xmm reg/mem128 CVTPS2PI op1, op2 op1 contains 1 32-bit integer value op2 contains 1 single precision 32-bit floating point value op1 = (long)op2[0] CVTTPS2PI Convert Parallel Scalars to Parallel Integers Opcode Cycles Instruction 0F 2C CVTTPS2PI mm reg,xmm reg/mem128 CVTTPS2PI op1, op2 op1 contains 2 32-bit integer values op2 contains 2 single precision 32-bit floating point values op1[0] = (long)op2[0] op1[1] = (long)op2[1] CVTTSS2SI Convert Single Scalar to Single Integer Opcode Cycles Instruction F3 0F 2C CVTTSS2SI reg32,xmm reg/mem128 CVTTSS2SI op1, op2 op1 contains 1 32-bit integer value op2 contains 1 single precision 32-bit floating point value op1 = (long)op2[0] DIVPS Divide Parallel Scalars Opcode Cycles Instruction 0F 5E 15-115 DIVPS xmm reg,xmm reg/mem128 DIVPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op1[0] = op1[0] / op2[0] op1[1] = op1[1] / op2[1] op1[2] = op1[2] / op2[2] op1[3] = op1[3] / op2[3] DIVSS Divide Single Scalar Opcode Cycles Instruction F3 0F 5E 7-98 DIVSS xmm reg,xmm reg/mem32 DIVSS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 1 single precision 32-bit floating point value op1[0] = op1[0] / op2 op1[1] = op1[1] op1[2] = op1[2] op1[3] = op1[3] FXRSTOR Floating Point Extended Restore Opcode Cycles Instruction 0F AE xx001xxx FXRSTOR mem FXRSTOR op1 op1 contains a 512 byte register context, paragraph aligned FXSAVE Floating Point Extended Save Opcode Cycles Instruction 0F AE xx000xxx FXSAVE mem FXSAVE op1 op1 contains a 512 byte register context, paragraph aligned LDMXCSR Load Multimedia Extended Control Status Register Opcode Cycles Instruction 0F AE xx010xxx LDMXCSR mem32 LDMXCSR op1 op1 contains 1 32-bit register MXCSR = op1 MASKMOVQ Opcode Cycles Instruction 0F F7 MASKMOVQ mm reg,mm reg MASKMOVQ op1, op2 MAXPS Maximum Parallel Scalars Opcode Cycles Instruction 0F 5F 2 (3) MAXPS xmm reg,xmm reg/mem128 MAXPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op1[0] = max(op1[0], op2[0]) op1[1] = max(op1[1], op2[1]) op1[2] = max(op1[2], op2[2]) op1[3] = max(op1[3], op2[3]) MAXSS Maximum Single Scalar Opcode Cycles Instruction F3 0F 5F 1 (3) MAXSS xmm reg,xmm reg/mem32 MAXSS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 1 single precision 32-bit floating point value op1[0] = max(op1[0], op2) op1[1] = op1[1] op1[2] = op1[2] op1[3] = op1[3] MINPS Minimum Parallel Scalars Opcode Cycles Instruction 0F 5D 2 (3) MINPS xmm reg,xmm reg/mem128 MINPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op1[0] = min(op1[0], op2[0]) op1[1] = min(op1[1], op2[1]) op1[2] = min(op1[2], op2[2]) op1[3] = min(op1[3], op2[3]) MINSS Minimum Single Scalar Opcode Cycles Instruction F3 0F 5D 1 (3) MINSS xmm reg,xmm reg/mem32 MINSS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 1 single precision 32-bit floating point value op1[0] = min(op1[0], op2) op1[1] = op1[1] op1[2] = op1[2] op1[3] = op1[3] MOVAPS Aligned Move Parallel Scalars Opcode Cycles Instruction 0F 28 MOVAPS xmm reg,xmm reg/mem128 0F 29 MOVAPS mem128,xmm reg MOVAPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op1[0] = op2[0] op1[1] = op2[1] op1[2] = op2[2] op1[3] = op2[3] * Addresses must be paragraph aligned MOVHPS Move High Pair Parallel Scalars Opcode Cycles Instruction 0F 16 MOVHPS xmm reg,mem64 0F 17 MOVHPS mem64,xmm reg MOVHPS op1, op2 op1 contains 2 single precision 32-bit floating point values op2 contains 2 single precision 32-bit floating point values op1[2] = op2[0] (xmm reg,mem64) op1[3] = op2[1] op1[0] = op2[2] (mem64,xmm reg) op1[1] = op2[3] MOVHLPS Move High to Low Pair Parallel Scalars Opcode Cycles Instruction 0F 12 1 MOVHLPS xmm reg,xmm reg MOVHLPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op1[0] = op2[2] op1[1] = op2[3] op1[2] = op1[2] op1[3] = op1[3] MOVLPS Move Low Pair Parallel Scalars Opcode Cycles Instruction 0F 12 MOVLPS xmm reg,mem64 0F 13 MOVLPS mem64,xmm reg MOVLPS op1, op2 op1 contains 2 single precision 32-bit floating point values op2 contains 2 single precision 32-bit floating point values op1[0] = op2[0] op1[1] = op2[1] MOVLHPS Move Low to High Pair Parallel Scalars Opcode Cycles Instruction 0F 16 1 MOVLHPS xmm reg,xmm reg MOVLHPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op1[0] = op1[0] op1[1] = op1[1] op1[2] = op2[0] op1[3] = op2[1] MOVMSKPS Opcode Cycles Instruction 0F 50 MOVMSKPS reg32,xmm reg MOVMSKPS op1, op2 MOVNTPS Uncached Move Parallel Scalars Opcode Cycles Instruction 0F 2B MOVNTPS mem128,xmm reg MOVNTPS op1, op2 op1 contains 1 128-bit value op2 contains 1 128-bit value op1 = op2 MOVNTQ Uncached Move Quad Word Opcode Cycles Instruction 0F E7 MOVNTQ mem64,mm reg MOVNTQ op1, op2 op1 contains 1 64-bit value op2 contains 1 64-bit value op1 = op2 MOVSS Move Single Scalar Opcode Cycles Instruction F3 0F 10 MOVSS xmm reg,xmm reg/mem32 F3 0F 11 MOVSS mem32,xmm reg MOVSS op1, op2 op1 contains 1 single precision 32-bit floating point value op2 contains 1 single precision 32-bit floating point value op1[0] = op2[0] MOVUPS Unaligned Move Parallel Scalars Opcode Cycles Instruction 0F 10 MOVUPS xmm reg,xmm reg/mem128 0F 11 MOVUPS mem128,xmm reg MOVUPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op1[0] = op2[0] op1[1] = op2[1] op1[2] = op2[2] op1[3] = op2[3] MULPS Multiply Parallel Scalars Opcode Cycles Instruction 0F 59 2 (4) MULPS xmm reg,xmm reg/mem128 MULPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op1[0] = op1[0] * op2[0] op1[1] = op1[1] * op2[1] op1[2] = op1[2] * op2[2] op1[3] = op1[3] * op2[3] MULSS Multiply Single Scalar Opcode Cycles Instruction F3 0F 59 1 (4) MULSS xmm reg,xmm reg/mem32 MULSS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 1 single precision 32-bit floating point value op1[0] = op1[0] * op2 op1[1] = op1[1] op1[2] = op1[2] op1[3] = op1[3] ORPS Or Parallel Scalars (bitwise) Opcode Cycles Instruction 0F 56 2 ORPS xmm reg,xmm reg/mem128 ORPS op1, op2 op1 contains 1 128-bit value op2 contains 1 128-bit value op1 = op1 | op2 PAVGB Parallel Integer Average Byte Opcode Cycles Instruction 0F E0 PAVGB mm reg,mm reg/mem64 PAVGB op1, op2 op1 contains 8 8-bit integer values op2 contains 8 8-bit integer values op1[0] = (op1[0] + op2[0]) / 2 op1[1] = (op1[1] + op2[1]) / 2 op1[2] = (op1[2] + op2[2]) / 2 op1[3] = (op1[3] + op2[3]) / 2 op1[4] = (op1[4] + op2[4]) / 2 op1[5] = (op1[5] + op2[5]) / 2 op1[6] = (op1[6] + op2[6]) / 2 op1[7] = (op1[7] + op2[7]) / 2 PAVGW Parallel Integer Average Word Opcode Cycles Instruction 0F E3 PAVGW mm reg,mm reg/mem64 PAVGW op1, op2 op1 contains 4 16-bit integer values op2 contains 4 16-bit integer values op1[0] = (op1[0] + op2[0]) / 2 op1[1] = (op1[1] + op2[1]) / 2 op1[2] = (op1[2] + op2[2]) / 2 op1[3] = (op1[3] + op2[3]) / 2 PEXTRW Opcode Cycles Instruction 0F C5 PEXTRW reg32,mm reg,imm8 PEXTRW op1, op2, op3 PINSRW Opcode Cycles Instruction 0F C4 PINSRW mm reg,reg32/mem32,imm8 PINSRW op1, op2, op3 PMAXSW Parallel Integer Maximum Signed Word Opcode Cycles Instruction 0F EE PMAXSW mm reg,mm reg/mem64 PMAXSW op1, op2 op1 contains 4 16-bit signed integer values op2 contains 4 16-bit signed integer values op1[0] = max(op1[0], op2[0]) op1[1] = max(op1[1], op2[1]) op1[2] = max(op1[2], op2[2]) op1[3] = max(op1[3], op2[3]) PMAXUB Parallel Integer Maximum Unsigned Byte Opcode Cycles Instruction 0F DE PMAXUB mm reg,mm reg/mem64 PMAXUB op1, op2 op1 contains 8 8-bit unsigned integer values op2 contains 8 8-bit unsigned integer values op1[0] = max(op1[0], op2[0]) op1[1] = max(op1[1], op2[1]) op1[2] = max(op1[2], op2[2]) op1[3] = max(op1[3], op2[3]) op1[4] = max(op1[4], op2[4]) op1[5] = max(op1[5], op2[5]) op1[6] = max(op1[6], op2[6]) op1[7] = max(op1[7], op2[7]) PMINSW Parallel Integer Minimum Signed Word Opcode Cycles Instruction 0F EA PMINSW mm reg,mm reg/mem64 PMINSW op1, op2 op1 contains 4 16-bit signed integer values op2 contains 4 16-bit signed integer values op1[0] = min(op1[0], op2[0]) op1[1] = min(op1[1], op2[1]) op1[2] = min(op1[2], op2[2]) op1[3] = min(op1[3], op2[3]) PMINUB Parallel Integer Minimum Unsigned Byte Opcode Cycles Instruction 0F DA PMINUB mm reg,mm reg/mem64 PMINUB op1, op2 op1 contains 8 8-bit unsigned integer values op2 contains 8 8-bit unsigned integer values op1[0] = min(op1[0], op2[0]) op1[1] = min(op1[1], op2[1]) op1[2] = min(op1[2], op2[2]) op1[3] = min(op1[3], op2[3]) op1[4] = min(op1[4], op2[4]) op1[5] = min(op1[5], op2[5]) op1[6] = min(op1[6], op2[6]) op1[7] = min(op1[7], op2[7]) PMOVMSKB Opcode Cycles Instruction 0F D7 PMOVMSKB reg32,mm reg PMOVMSKB op1, op2 PMULHUW Multiply unsigned word store high Opcode Cycles Instruction 0F E4 PMULHUW mm reg,mm reg/mem64 PMULHUW op1, op2 op1 contains 4 16-bit unsigned integer values op2 contains 4 16-bit unsigned integer values op1[0] = (op1[0] * op2[0]) >> 16 op1[1] = (op1[1] * op2[1]) >> 16 op1[2] = (op1[2] * op2[2]) >> 16 op1[3] = (op1[3] * op2[3]) >> 16 PREFETCHNTA Prefetch Non-caching Aligned ? Opcode Cycles Instruction 0F 18 xx000xxx PREFETCHNTA mem8 PREFETCHNTA op1 PREFETCHT0 Prefetch Task 0 ? Opcode Cycles Instruction 0F 18 xx001xxx PREFETCHT0 mem8 PREFETCHT0 op1 PREFETCHT1 Prefetch Task 1 ? Opcode Cycles Instruction 0F 18 xx010xxx PREFETCHT1 mem8 PREFETCHT1 op1 PREFETCHT2 Prefetch Task 2 ? Opcode Cycles Instruction 0F 18 xx011xxx PREFETCHT2 mem8 PREFETCHT2 op1 PSADBW Opcode Cycles Instruction 0F F6 PSADBW mm reg,mm reg/mem64 PSADBW op1, op2 PSHUFW Shuffle Parallel Words Opcode Cycles Instruction 0F 70 1 (1) PSHUFW mm reg,mm reg/mem64,imm8 PSHUFW op1, op3, op3 op1 contains 4 16-bit integer values op2 contains 4 16-bit integer values op3 contains a bit map dd:cc:bb:aa (MSB to LSB) op1[0] = op2[aa] op1[1] = op2[bb] op1[2] = op2[cc] op1[3] = op2[dd] RCPPS Reciprocal Parallel Scalars Opcode Cycles Instruction 0F 53 2 (2) RCPPS xmm reg,xmm reg/mem128 RCPPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op1[0] = 1 / op2[0] op1[1] = 1 / op2[1] op1[2] = 1 / op2[2] op1[3] = 1 / op2[3] * The results have 12-bit accuracy RCPSS Reciprocal Single Scalar Opcode Cycles Instruction F3 0F 53 1 (1) RCPSS xmm reg,xmm reg/mem32 RCPSS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 1 single precision 32-bit floating point value op1[0] = 1 / op2 op1[1] = op1[1] op1[2] = op1[2] op1[3] = op1[3] * The results have 12-bit accuracy RSQRTPS Reciprocal Square Root Parallel Scalars Opcode Cycles Instruction 0F 52 2 (2) RSQRTPS xmm reg,xmm reg/mem128 RSQRTPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op1[0] = 1 / sqrt(op2[0]) op1[1] = 1 / sqrt(op2[1]) op1[2] = 1 / sqrt(op2[2]) op1[3] = 1 / sqrt(op2[3]) * The results have 12-bit accuracy RSQRTSS Reciprocal Square Root Single Scalar Opcode Cycles Instruction F3 0F 52 1 (1) RSQRTSS xmm reg,xmm reg/mem32 RSQRTSS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 1 single precision 32-bit floating point value op1[0] = 1 / sqrt(op2) op1[1] = op1[1] op1[2] = op1[2] op1[3] = op1[3] * The results have 12-bit accuracy SFENCE Stream Fence Opcode Cycles Instruction 0F AE FF SFENCE SFENCE Provides a demarkation in write combining buffers to force current states to be committed. In other words writes to the same location cannot combine to one if there is a fence placed between them. * Presumed function from AGP definition of fencing SHUFPS Shuffle Parallel Scalars Opcode Cycles Instruction 0F C6 3 SHUFPS xmm reg, xmm reg/mem128, imm8 SHUFPS op1, op3, op3 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op3 contains a bit map dd:cc:bb:aa (MSB to LSB) op1[0] = op1[aa] op1[1] = op1[bb] op1[2] = op2[cc] op1[3] = op2[dd] SQRTPS Square Root Parallel Scalars Opcode Cycles Instruction 0F 51 16-134 SQRTPS xmm reg,xmm reg/mem128 SQRTPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op1[0] = sqrt(op2[0]) op1[1] = sqrt(op2[1]) op1[2] = sqrt(op2[2]) op1[3] = sqrt(op2[3]) SQRTSS Square Root Single Scalar Opcode Cycles Instruction F3 0F 51 8-105 SQRTSS xmm reg,xmm reg/mem32 SQRTSS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 1 single precision 32-bit floating point value op1[0] = sqrt(op2) op1[1] = op1[1] op1[2] = op1[2] op1[3] = op1[3] STMXCSR Store Multimedia Extended Control Status Register Opcode Cycles Instruction 0F AE xx011xxx STMXCSR mem32 STMXCSR op1 op1 contains 1 32-bit register op1 = MXCSR SUBPS Subtract Parallel Scalars Opcode Cycles Instruction 0F 5C 2 (3) SUBPS xmm reg,xmm reg/mem128 SUBPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op1[0] = op1[0] - op2[0] op1[1] = op1[1] - op2[1] op1[2] = op1[2] - op2[2] op1[3] = op1[3] - op2[3] SUBSS Subtract Single Scalar Opcode Cycles Instruction F3 0F 5C 1 (3) SUBSS xmm reg,xmm reg/mem32 SUBSS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 1 single precision 32-bit floating point value op1[0] = op1[0] - op2 op1[1] = op1[1] op1[2] = op1[2] op1[3] = op1[3] UCOMISS Opcode Cycles Instruction 0F 2E UCOMISS xmm reg,xmm reg/mem32 UCOMISS op1, op2 UNPCKHPS Unpack High Parallel Scalars Opcode Cycles Instruction 0F 15 2 UNPCKHPS xmm reg,xmm reg/mem128 UNPCKHPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op1[0] = op1[2] op1[1] = op2[2] op1[2] = op1[3] op1[3] = op2[3] UNPCKLPS Unpack Low Parallel Scalars Opcode Cycles Instruction 0F 14 2 UNPCKLPS xmm reg,xmm reg/mem128 UNPCKLPS op1, op2 op1 contains 4 single precision 32-bit floating point values op2 contains 4 single precision 32-bit floating point values op1[0] = op1[0] op1[1] = op2[0] op1[2] = op1[1] op1[3] = op2[1] XORPS Exclusive-Or Parallel Scalars (bitwise) Opcode Cycles Instruction 0F 57 2 XORPS xmm reg,xmm reg/mem128 XORPS op1, op2 op1 contains 1 128-bit value op2 contains 1 128-bit value op1 = op1 ^ op2