diff options
Diffstat (limited to 'lib/Target/X86/X86InstrSSE.td')
| -rw-r--r-- | lib/Target/X86/X86InstrSSE.td | 153 |
1 files changed, 82 insertions, 71 deletions
diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td index cde3f6b..b64c03a 100644 --- a/lib/Target/X86/X86InstrSSE.td +++ b/lib/Target/X86/X86InstrSSE.td @@ -1056,13 +1056,37 @@ let neverHasSideEffects = 1 in { XD, VEX_4V; } +let Constraints = "$src1 = $dst" in { +def CMPSSrr : SIi8<0xC2, MRMSrcReg, + (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, SSECC:$cc), + "cmp${cc}ss\t{$src2, $dst|$dst, $src2}", + [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), FR32:$src2, imm:$cc))]>, XS; +def CMPSSrm : SIi8<0xC2, MRMSrcMem, + (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2, SSECC:$cc), + "cmp${cc}ss\t{$src2, $dst|$dst, $src2}", + [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), (loadf32 addr:$src2), imm:$cc))]>, XS; +def CMPSDrr : SIi8<0xC2, MRMSrcReg, + (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, SSECC:$cc), + "cmp${cc}sd\t{$src2, $dst|$dst, $src2}", + [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), FR64:$src2, imm:$cc))]>, XD; +def CMPSDrm : SIi8<0xC2, MRMSrcMem, + (outs FR64:$dst), (ins FR64:$src1, f64mem:$src2, SSECC:$cc), + "cmp${cc}sd\t{$src2, $dst|$dst, $src2}", + [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), (loadf64 addr:$src2), imm:$cc))]>, XD; +} let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in { - defm CMPSS : sse12_cmp_scalar<FR32, f32mem, - "cmp${cc}ss\t{$src, $dst|$dst, $src}", - "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}">, XS; - defm CMPSD : sse12_cmp_scalar<FR64, f64mem, - "cmp${cc}sd\t{$src, $dst|$dst, $src}", - "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}">, XD; +def CMPSSrr_alt : SIi8<0xC2, MRMSrcReg, + (outs FR32:$dst), (ins FR32:$src1, FR32:$src, i8imm:$src2), + "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS; +def CMPSSrm_alt : SIi8<0xC2, MRMSrcMem, + (outs FR32:$dst), (ins FR32:$src1, f32mem:$src, i8imm:$src2), + "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS; +def CMPSDrr_alt : SIi8<0xC2, MRMSrcReg, + (outs FR64:$dst), (ins FR64:$src1, FR64:$src, i8imm:$src2), + "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD; +def CMPSDrm_alt : SIi8<0xC2, MRMSrcMem, + (outs FR64:$dst), (ins FR64:$src1, f64mem:$src, i8imm:$src2), + "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD; } multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop, @@ -1327,11 +1351,6 @@ multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm, } // Mask creation -defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps", - SSEPackedSingle>, TB; -defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd", - SSEPackedDouble>, TB, OpSize; - defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps", SSEPackedSingle>, VEX; defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, @@ -1342,6 +1361,24 @@ defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256, defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256, "movmskpd", SSEPackedDouble>, OpSize, VEX; +defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps", + SSEPackedSingle>, TB; +defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd", + SSEPackedDouble>, TB, OpSize; + +// X86fgetsign +def MOVMSKPDrr32_alt : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins FR64:$src), + "movmskpd\t{$src, $dst|$dst, $src}", + [(set GR32:$dst, (X86fgetsign FR64:$src))], SSEPackedDouble>, TB, OpSize; +def MOVMSKPDrr64_alt : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins FR64:$src), + "movmskpd\t{$src, $dst|$dst, $src}", + [(set GR64:$dst, (X86fgetsign FR64:$src))], SSEPackedDouble>, TB, OpSize; +def MOVMSKPSrr32_alt : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins FR32:$src), + "movmskps\t{$src, $dst|$dst, $src}", + [(set GR32:$dst, (X86fgetsign FR32:$src))], SSEPackedSingle>, TB; +def MOVMSKPSrr64_alt : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins FR32:$src), + "movmskps\t{$src, $dst|$dst, $src}", + [(set GR64:$dst, (X86fgetsign FR32:$src))], SSEPackedSingle>, TB; // Assembler Only def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src), @@ -1875,21 +1912,6 @@ defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>, // SSE 1 & 2 - Non-temporal stores //===----------------------------------------------------------------------===// -def VMOVNTPSmr_Int : VPSI<0x2B, MRMDestMem, (outs), - (ins i128mem:$dst, VR128:$src), - "movntps\t{$src, $dst|$dst, $src}", - [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>, VEX; -def VMOVNTPDmr_Int : VPDI<0x2B, MRMDestMem, (outs), - (ins i128mem:$dst, VR128:$src), - "movntpd\t{$src, $dst|$dst, $src}", - [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>, VEX; - -let ExeDomain = SSEPackedInt in - def VMOVNTDQmr_Int : VPDI<0xE7, MRMDestMem, (outs), - (ins f128mem:$dst, VR128:$src), - "movntdq\t{$src, $dst|$dst, $src}", - [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>, VEX; - let AddedComplexity = 400 in { // Prefer non-temporal versions def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src), @@ -1906,12 +1928,16 @@ let AddedComplexity = 400 in { // Prefer non-temporal versions "movntdq\t{$src, $dst|$dst, $src}", [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>, VEX; + let ExeDomain = SSEPackedInt in - def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs), - (ins f128mem:$dst, VR128:$src), - "movntdq\t{$src, $dst|$dst, $src}", - [(alignednontemporalstore (v4f32 VR128:$src), - addr:$dst)]>, VEX; + def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs), + (ins f128mem:$dst, VR128:$src), + "movntdq\t{$src, $dst|$dst, $src}", + [(alignednontemporalstore (v4f32 VR128:$src), + addr:$dst)]>, VEX; + + def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst), + (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>; def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src), @@ -1943,18 +1969,6 @@ def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src), def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src), (VMOVNTPSYmr addr:$dst, VR256:$src)>; -def MOVNTPSmr_Int : PSI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src), - "movntps\t{$src, $dst|$dst, $src}", - [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>; -def MOVNTPDmr_Int : PDI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src), - "movntpd\t{$src, $dst|$dst, $src}", - [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>; - -let ExeDomain = SSEPackedInt in -def MOVNTDQmr_Int : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src), - "movntdq\t{$src, $dst|$dst, $src}", - [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>; - let AddedComplexity = 400 in { // Prefer non-temporal versions def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src), "movntps\t{$src, $dst|$dst, $src}", @@ -1972,22 +1986,19 @@ def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src), "movntdq\t{$src, $dst|$dst, $src}", [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>; +def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst), + (MOVNTDQmr addr:$dst, VR128:$src)>; + // There is no AVX form for instructions below this point def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src), "movnti\t{$src, $dst|$dst, $src}", [(nontemporalstore (i32 GR32:$src), addr:$dst)]>, TB, Requires<[HasSSE2]>; - def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src), "movnti\t{$src, $dst|$dst, $src}", [(nontemporalstore (i64 GR64:$src), addr:$dst)]>, TB, Requires<[HasSSE2]>; - } -def MOVNTImr_Int : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src), - "movnti\t{$src, $dst|$dst, $src}", - [(int_x86_sse2_movnt_i addr:$dst, GR32:$src)]>, - TB, Requires<[HasSSE2]>; //===----------------------------------------------------------------------===// // SSE 1 & 2 - Misc Instructions (No AVX form) @@ -4733,14 +4744,14 @@ let Uses = [XMM0], Constraints = "$src1 = $dst" in { def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2), !strconcat(OpcodeStr, - "\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}"), + "\t{$src2, $dst|$dst, $src2}"), [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>, OpSize; def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2), !strconcat(OpcodeStr, - "\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}"), + "\t{$src2, $dst|$dst, $src2}"), [(set VR128:$dst, (IntId VR128:$src1, (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize; @@ -4961,66 +4972,66 @@ defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>; // This set of instructions are only rm, the only difference is the size // of r and m. let Constraints = "$src1 = $dst" in { - def CRC32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst), + def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst), (ins GR32:$src1, i8mem:$src2), "crc32{b} \t{$src2, $src1|$src1, $src2}", [(set GR32:$dst, - (int_x86_sse42_crc32_8 GR32:$src1, + (int_x86_sse42_crc32_32_8 GR32:$src1, (load addr:$src2)))]>; - def CRC32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst), + def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src1, GR8:$src2), "crc32{b} \t{$src2, $src1|$src1, $src2}", [(set GR32:$dst, - (int_x86_sse42_crc32_8 GR32:$src1, GR8:$src2))]>; - def CRC32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst), + (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>; + def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst), (ins GR32:$src1, i16mem:$src2), "crc32{w} \t{$src2, $src1|$src1, $src2}", [(set GR32:$dst, - (int_x86_sse42_crc32_16 GR32:$src1, + (int_x86_sse42_crc32_32_16 GR32:$src1, (load addr:$src2)))]>, OpSize; - def CRC32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst), + def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src1, GR16:$src2), "crc32{w} \t{$src2, $src1|$src1, $src2}", [(set GR32:$dst, - (int_x86_sse42_crc32_16 GR32:$src1, GR16:$src2))]>, + (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>, OpSize; - def CRC32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst), + def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst), (ins GR32:$src1, i32mem:$src2), "crc32{l} \t{$src2, $src1|$src1, $src2}", [(set GR32:$dst, - (int_x86_sse42_crc32_32 GR32:$src1, + (int_x86_sse42_crc32_32_32 GR32:$src1, (load addr:$src2)))]>; - def CRC32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst), + def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src1, GR32:$src2), "crc32{l} \t{$src2, $src1|$src1, $src2}", [(set GR32:$dst, - (int_x86_sse42_crc32_32 GR32:$src1, GR32:$src2))]>; - def CRC64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst), + (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>; + def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst), (ins GR64:$src1, i8mem:$src2), "crc32{b} \t{$src2, $src1|$src1, $src2}", [(set GR64:$dst, - (int_x86_sse42_crc64_8 GR64:$src1, + (int_x86_sse42_crc32_64_8 GR64:$src1, (load addr:$src2)))]>, REX_W; - def CRC64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst), + def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src1, GR8:$src2), "crc32{b} \t{$src2, $src1|$src1, $src2}", [(set GR64:$dst, - (int_x86_sse42_crc64_8 GR64:$src1, GR8:$src2))]>, + (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>, REX_W; - def CRC64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst), + def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2), "crc32{q} \t{$src2, $src1|$src1, $src2}", [(set GR64:$dst, - (int_x86_sse42_crc64_64 GR64:$src1, + (int_x86_sse42_crc32_64_64 GR64:$src1, (load addr:$src2)))]>, REX_W; - def CRC64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst), + def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src1, GR64:$src2), "crc32{q} \t{$src2, $src1|$src1, $src2}", [(set GR64:$dst, - (int_x86_sse42_crc64_64 GR64:$src1, GR64:$src2))]>, + (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>, REX_W; } |
