diff options
Diffstat (limited to 'lib/Target/X86/X86InstrSSE.td')
| -rw-r--r-- | lib/Target/X86/X86InstrSSE.td | 99 |
1 files changed, 36 insertions, 63 deletions
diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td index 91c84dd6a1..10f527cded 100644 --- a/lib/Target/X86/X86InstrSSE.td +++ b/lib/Target/X86/X86InstrSSE.td @@ -7655,7 +7655,6 @@ defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq", // Variable Bit Shifts // multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, - PatFrag pf128, PatFrag pf256, Intrinsic Int128, Intrinsic Int256> { def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2), @@ -7664,7 +7663,8 @@ multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - [(set VR128:$dst, (Int128 VR128:$src1, (pf128 addr:$src2)))]>, + [(set VR128:$dst, + (Int128 VR128:$src1, (bitconvert (memopv2i64 addr:$src2))))]>, VEX_4V; def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src1, VR256:$src2), @@ -7673,70 +7673,43 @@ multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - [(set VR256:$dst, (Int256 VR256:$src1, (pf256 addr:$src2)))]>, + [(set VR256:$dst, + (Int256 VR256:$src1, (bitconvert (memopv4i64 addr:$src2))))]>, VEX_4V; } -defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", memopv4i32, memopv8i32, - int_x86_avx2_psllv_d, int_x86_avx2_psllv_d_256>; -defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", memopv2i64, memopv4i64, - int_x86_avx2_psllv_q, int_x86_avx2_psllv_q_256>, - VEX_W; -defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", memopv4i32, memopv8i32, - int_x86_avx2_psrlv_d, int_x86_avx2_psrlv_d_256>; -defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", memopv2i64, memopv4i64, - int_x86_avx2_psrlv_q, int_x86_avx2_psrlv_q_256>, - VEX_W; -defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", memopv4i32, memopv8i32, - int_x86_avx2_psrav_d, int_x86_avx2_psrav_d_256>; - - -let Predicates = [HasAVX2] in { - - def : Pat<(v4i32 (shl (v4i32 VR128:$src1), (v4i32 VR128:$src2))), - (VPSLLVDrr VR128:$src1, VR128:$src2)>; - def : Pat<(v2i64 (shl (v2i64 VR128:$src1), (v2i64 VR128:$src2))), - (VPSLLVQrr VR128:$src1, VR128:$src2)>; - def : Pat<(v4i32 (srl (v4i32 VR128:$src1), (v4i32 VR128:$src2))), - (VPSRLVDrr VR128:$src1, VR128:$src2)>; - def : Pat<(v2i64 (srl (v2i64 VR128:$src1), (v2i64 VR128:$src2))), - (VPSRLVQrr VR128:$src1, VR128:$src2)>; - def : Pat<(v4i32 (sra (v4i32 VR128:$src1), (v4i32 VR128:$src2))), - (VPSRAVDrr VR128:$src1, VR128:$src2)>; - def : Pat<(v8i32 (shl (v8i32 VR256:$src1), (v8i32 VR256:$src2))), - (VPSLLVDYrr VR256:$src1, VR256:$src2)>; - def : Pat<(v4i64 (shl (v4i64 VR256:$src1), (v4i64 VR256:$src2))), - (VPSLLVQYrr VR256:$src1, VR256:$src2)>; - def : Pat<(v8i32 (srl (v8i32 VR256:$src1), (v8i32 VR256:$src2))), - (VPSRLVDYrr VR256:$src1, VR256:$src2)>; - def : Pat<(v4i64 (srl (v4i64 VR256:$src1), (v4i64 VR256:$src2))), - (VPSRLVQYrr VR256:$src1, VR256:$src2)>; - def : Pat<(v8i32 (sra (v8i32 VR256:$src1), (v8i32 VR256:$src2))), - (VPSRAVDYrr VR256:$src1, VR256:$src2)>; - - def : Pat<(v4i32 (shl (v4i32 VR128:$src1),(loadv4i32 addr:$src2))), - (VPSLLVDrm VR128:$src1, addr:$src2)>; - def : Pat<(v4i32 (shl (v4i32 VR128:$src1),(loadv2i64 addr:$src2))), - (VPSLLVDrm VR128:$src1, addr:$src2)>; - def : Pat<(v2i64 (shl (v2i64 VR128:$src1),(loadv2i64 addr:$src2))), - (VPSLLVQrm VR128:$src1, addr:$src2)>; - def : Pat<(v4i32 (srl (v4i32 VR128:$src1),(loadv4i32 addr:$src2))), - (VPSRLVDrm VR128:$src1, addr:$src2)>; - def : Pat<(v2i64 (srl (v2i64 VR128:$src1),(loadv2i64 addr:$src2))), - (VPSRLVQrm VR128:$src1, addr:$src2)>; - def : Pat<(v4i32 (sra (v4i32 VR128:$src1),(loadv4i32 addr:$src2))), - (VPSRAVDrm VR128:$src1, addr:$src2)>; - def : Pat<(v8i32 (shl (v8i32 VR256:$src1),(loadv8i32 addr:$src2))), - (VPSLLVDYrm VR256:$src1, addr:$src2)>; - def : Pat<(v4i64 (shl (v4i64 VR256:$src1),(loadv4i64 addr:$src2))), - (VPSLLVQYrm VR256:$src1, addr:$src2)>; - def : Pat<(v8i32 (srl (v8i32 VR256:$src1),(loadv8i32 addr:$src2))), - (VPSRLVDYrm VR256:$src1, addr:$src2)>; - def : Pat<(v4i64 (srl (v4i64 VR256:$src1),(loadv4i64 addr:$src2))), - (VPSRLVQYrm VR256:$src1, addr:$src2)>; - def : Pat<(v8i32 (sra (v8i32 VR256:$src1),(loadv8i32 addr:$src2))), - (VPSRAVDYrm VR256:$src1, addr:$src2)>; +multiclass avx2_var_shift_i64<bits<8> opc, string OpcodeStr, + Intrinsic Int128, Intrinsic Int256> { + def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), + (ins VR128:$src1, VR128:$src2), + !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set VR128:$dst, (Int128 VR128:$src1, VR128:$src2))]>, VEX_4V; + def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), + (ins VR128:$src1, i128mem:$src2), + !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set VR128:$dst, + (Int128 VR128:$src1, (memopv2i64 addr:$src2)))]>, + VEX_4V; + def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), + (ins VR256:$src1, VR256:$src2), + !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set VR256:$dst, (Int256 VR256:$src1, VR256:$src2))]>, VEX_4V; + def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), + (ins VR256:$src1, i256mem:$src2), + !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set VR256:$dst, + (Int256 VR256:$src1, (memopv4i64 addr:$src2)))]>, + VEX_4V; } - +defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", int_x86_avx2_psllv_d, + int_x86_avx2_psllv_d_256>; +defm VPSLLVQ : avx2_var_shift_i64<0x47, "vpsllvq", int_x86_avx2_psllv_q, + int_x86_avx2_psllv_q_256>, VEX_W; +defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", int_x86_avx2_psrlv_d, + int_x86_avx2_psrlv_d_256>; +defm VPSRLVQ : avx2_var_shift_i64<0x45, "vpsrlvq", int_x86_avx2_psrlv_q, + int_x86_avx2_psrlv_q_256>, VEX_W; +defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", int_x86_avx2_psrav_d, + int_x86_avx2_psrav_d_256>; |