diff options
| -rw-r--r-- | lib/Target/X86/X86InstrSSE.td | 123 |
1 files changed, 54 insertions, 69 deletions
diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td index 5546ea94ce..e080c2310e 100644 --- a/lib/Target/X86/X86InstrSSE.td +++ b/lib/Target/X86/X86InstrSSE.td @@ -1135,6 +1135,16 @@ let Constraints = "$src1 = $dst" in { "cmp${cc}ps\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse_cmp_ps VR128:$src1, (memop addr:$src), imm:$cc))]>; + def CMPPDrri : PDIi8<0xC2, MRMSrcReg, + (outs VR128:$dst), (ins VR128:$src1, VR128:$src, SSECC:$cc), + "cmp${cc}pd\t{$src, $dst|$dst, $src}", + [(set VR128:$dst, (int_x86_sse2_cmp_pd VR128:$src1, + VR128:$src, imm:$cc))]>; + def CMPPDrmi : PDIi8<0xC2, MRMSrcMem, + (outs VR128:$dst), (ins VR128:$src1, f128mem:$src, SSECC:$cc), + "cmp${cc}pd\t{$src, $dst|$dst, $src}", + [(set VR128:$dst, (int_x86_sse2_cmp_pd VR128:$src1, + (memop addr:$src), imm:$cc))]>; // Accept explicit immediate argument form instead of comparison code. let isAsmParserOnly = 1 in { @@ -1144,12 +1154,22 @@ let isAsmParserOnly = 1 in { def CMPPSrmi_alt : PSIi8<0xC2, MRMSrcMem, (outs VR128:$dst), (ins VR128:$src1, f128mem:$src, i8imm:$src2), "cmpps\t{$src2, $src, $dst|$dst, $src, $src}", []>; + def CMPPDrri_alt : PDIi8<0xC2, MRMSrcReg, + (outs VR128:$dst), (ins VR128:$src1, VR128:$src, i8imm:$src2), + "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}", []>; + def CMPPDrmi_alt : PDIi8<0xC2, MRMSrcMem, + (outs VR128:$dst), (ins VR128:$src1, f128mem:$src, i8imm:$src2), + "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}", []>; } } def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)), (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>; def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)), (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>; +def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)), + (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>; +def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)), + (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>; // Shuffle and unpack instructions let Constraints = "$src1 = $dst" in { @@ -1167,6 +1187,18 @@ let Constraints = "$src1 = $dst" in { [(set VR128:$dst, (v4f32 (shufp:$src3 VR128:$src1, (memopv4f32 addr:$src2))))]>; + def SHUFPDrri : PDIi8<0xC6, MRMSrcReg, + (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i8imm:$src3), + "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}", + [(set VR128:$dst, + (v2f64 (shufp:$src3 VR128:$src1, VR128:$src2)))]>; + def SHUFPDrmi : PDIi8<0xC6, MRMSrcMem, + (outs VR128:$dst), (ins VR128:$src1, + f128mem:$src2, i8imm:$src3), + "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}", + [(set VR128:$dst, + (v2f64 (shufp:$src3 + VR128:$src1, (memopv2f64 addr:$src2))))]>; let AddedComplexity = 10 in { def UNPCKHPSrr : PSI<0x15, MRMSrcReg, @@ -1191,6 +1223,28 @@ let Constraints = "$src1 = $dst" in { "unpcklps\t{$src2, $dst|$dst, $src2}", [(set VR128:$dst, (unpckl VR128:$src1, (memopv4f32 addr:$src2)))]>; + def UNPCKHPDrr : PDI<0x15, MRMSrcReg, + (outs VR128:$dst), (ins VR128:$src1, VR128:$src2), + "unpckhpd\t{$src2, $dst|$dst, $src2}", + [(set VR128:$dst, + (v2f64 (unpckh VR128:$src1, VR128:$src2)))]>; + def UNPCKHPDrm : PDI<0x15, MRMSrcMem, + (outs VR128:$dst), (ins VR128:$src1, f128mem:$src2), + "unpckhpd\t{$src2, $dst|$dst, $src2}", + [(set VR128:$dst, + (v2f64 (unpckh VR128:$src1, + (memopv2f64 addr:$src2))))]>; + + def UNPCKLPDrr : PDI<0x14, MRMSrcReg, + (outs VR128:$dst), (ins VR128:$src1, VR128:$src2), + "unpcklpd\t{$src2, $dst|$dst, $src2}", + [(set VR128:$dst, + (v2f64 (unpckl VR128:$src1, VR128:$src2)))]>; + def UNPCKLPDrm : PDI<0x14, MRMSrcMem, + (outs VR128:$dst), (ins VR128:$src1, f128mem:$src2), + "unpcklpd\t{$src2, $dst|$dst, $src2}", + [(set VR128:$dst, + (unpckl VR128:$src1, (memopv2f64 addr:$src2)))]>; } // AddedComplexity } // Constraints = "$src1 = $dst" @@ -1786,75 +1840,6 @@ defm SQRT : sse2_fp_unop_rm<0x51, "sqrt", fsqrt, // There is no f64 version of the reciprocal approximation instructions. -let Constraints = "$src1 = $dst" in { - def CMPPDrri : PDIi8<0xC2, MRMSrcReg, - (outs VR128:$dst), (ins VR128:$src1, VR128:$src, SSECC:$cc), - "cmp${cc}pd\t{$src, $dst|$dst, $src}", - [(set VR128:$dst, (int_x86_sse2_cmp_pd VR128:$src1, - VR128:$src, imm:$cc))]>; - def CMPPDrmi : PDIi8<0xC2, MRMSrcMem, - (outs VR128:$dst), (ins VR128:$src1, f128mem:$src, SSECC:$cc), - "cmp${cc}pd\t{$src, $dst|$dst, $src}", - [(set VR128:$dst, (int_x86_sse2_cmp_pd VR128:$src1, - (memop addr:$src), imm:$cc))]>; - - // Accept explicit immediate argument form instead of comparison code. -let isAsmParserOnly = 1 in { - def CMPPDrri_alt : PDIi8<0xC2, MRMSrcReg, - (outs VR128:$dst), (ins VR128:$src1, VR128:$src, i8imm:$src2), - "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}", []>; - def CMPPDrmi_alt : PDIi8<0xC2, MRMSrcMem, - (outs VR128:$dst), (ins VR128:$src1, f128mem:$src, i8imm:$src2), - "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}", []>; -} -} -def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)), - (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>; -def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)), - (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>; - -// Shuffle and unpack instructions -let Constraints = "$src1 = $dst" in { - def SHUFPDrri : PDIi8<0xC6, MRMSrcReg, - (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i8imm:$src3), - "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}", - [(set VR128:$dst, - (v2f64 (shufp:$src3 VR128:$src1, VR128:$src2)))]>; - def SHUFPDrmi : PDIi8<0xC6, MRMSrcMem, - (outs VR128:$dst), (ins VR128:$src1, - f128mem:$src2, i8imm:$src3), - "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}", - [(set VR128:$dst, - (v2f64 (shufp:$src3 - VR128:$src1, (memopv2f64 addr:$src2))))]>; - - let AddedComplexity = 10 in { - def UNPCKHPDrr : PDI<0x15, MRMSrcReg, - (outs VR128:$dst), (ins VR128:$src1, VR128:$src2), - "unpckhpd\t{$src2, $dst|$dst, $src2}", - [(set VR128:$dst, - (v2f64 (unpckh VR128:$src1, VR128:$src2)))]>; - def UNPCKHPDrm : PDI<0x15, MRMSrcMem, - (outs VR128:$dst), (ins VR128:$src1, f128mem:$src2), - "unpckhpd\t{$src2, $dst|$dst, $src2}", - [(set VR128:$dst, - (v2f64 (unpckh VR128:$src1, - (memopv2f64 addr:$src2))))]>; - - def UNPCKLPDrr : PDI<0x14, MRMSrcReg, - (outs VR128:$dst), (ins VR128:$src1, VR128:$src2), - "unpcklpd\t{$src2, $dst|$dst, $src2}", - [(set VR128:$dst, - (v2f64 (unpckl VR128:$src1, VR128:$src2)))]>; - def UNPCKLPDrm : PDI<0x14, MRMSrcMem, - (outs VR128:$dst), (ins VR128:$src1, f128mem:$src2), - "unpcklpd\t{$src2, $dst|$dst, $src2}", - [(set VR128:$dst, - (unpckl VR128:$src1, (memopv2f64 addr:$src2)))]>; - } // AddedComplexity -} // Constraints = "$src1 = $dst" - - //===---------------------------------------------------------------------===// // SSE integer instructions let ExeDomain = SSEPackedInt in { |