diff options
| -rw-r--r-- | lib/Target/X86/X86InstrSSE.td | 47 |
1 files changed, 24 insertions, 23 deletions
diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td index 23073a9..b3d6066 100644 --- a/lib/Target/X86/X86InstrSSE.td +++ b/lib/Target/X86/X86InstrSSE.td @@ -1110,34 +1110,41 @@ def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src), // SSE 1 & 2 - Move Low packed FP Instructions //===----------------------------------------------------------------------===// -multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC, - SDNode psnode, SDNode pdnode, string base_opc, - string asm_opr, InstrItinClass itin> { +multiclass sse12_mov_hilo_packed_base<bits<8>opc, SDNode psnode, SDNode pdnode, + string base_opc, string asm_opr, + InstrItinClass itin> { def PSrm : PI<opc, MRMSrcMem, (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2), !strconcat(base_opc, "s", asm_opr), - [(set RC:$dst, - (psnode RC:$src1, + [(set VR128:$dst, + (psnode VR128:$src1, (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))], itin, SSEPackedSingle>, TB; def PDrm : PI<opc, MRMSrcMem, - (outs RC:$dst), (ins RC:$src1, f64mem:$src2), + (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2), !strconcat(base_opc, "d", asm_opr), - [(set RC:$dst, (v2f64 (pdnode RC:$src1, + [(set VR128:$dst, (v2f64 (pdnode VR128:$src1, (scalar_to_vector (loadf64 addr:$src2)))))], itin, SSEPackedDouble>, TB, OpSize; + } -let AddedComplexity = 20 in { - defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp", - "\t{$src2, $src1, $dst|$dst, $src1, $src2}", - IIC_SSE_MOV_LH>, VEX_4V; +multiclass sse12_mov_hilo_packed<bits<8>opc, SDNode psnode, SDNode pdnode, + string base_opc, InstrItinClass itin> { + defm V#NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}", + itin>, VEX_4V; + +let Constraints = "$src1 = $dst" in + defm NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc, + "\t{$src2, $dst|$dst, $src2}", + itin>; } -let Constraints = "$src1 = $dst", AddedComplexity = 20 in { - defm MOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp", - "\t{$src2, $dst|$dst, $src2}", - IIC_SSE_MOV_LH>; + +let AddedComplexity = 20 in { + defm MOVL : sse12_mov_hilo_packed<0x12, X86Movlps, X86Movlpd, "movlp", + IIC_SSE_MOV_LH>; } def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src), @@ -1235,14 +1242,8 @@ let Predicates = [UseSSE2] in { //===----------------------------------------------------------------------===// let AddedComplexity = 20 in { - defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp", - "\t{$src2, $src1, $dst|$dst, $src1, $src2}", - IIC_SSE_MOV_LH>, VEX_4V; -} -let Constraints = "$src1 = $dst", AddedComplexity = 20 in { - defm MOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp", - "\t{$src2, $dst|$dst, $src2}", - IIC_SSE_MOV_LH>; + defm MOVH : sse12_mov_hilo_packed<0x16, X86Movlhps, X86Movlhpd, "movhp", + IIC_SSE_MOV_LH>; } // v2f64 extract element 1 is always custom lowered to unpack high to low |