RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-23 11:13:28 +01:00
Code Issues Projects Releases Wiki Activity

[X86][SSE] Add support for X86ISD::PACKSS to ComputeNumSignBitsForTargetNode

Browse Source 
Helps improve combineLogicBlendIntoPBLENDV support by allowing us to peek into through PACKSS truncations of vector comparison results.

Differential Revision: https://reviews.llvm.org/D37680

llvm-svn: 312916
This commit is contained in:
Simon Pilgrim 2017-09-11 14:03:47 +00:00
parent 0ce7d42fef
commit 8beb4573f2
3 changed files with 1731 additions and 1723 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
lib/Target/X86/X86ISelLowering.cpp
View File

@ -27091,6 +27091,18 @@ unsigned X86TargetLowering::ComputeNumSignBitsForTargetNode(
return Tmp;
}
case X86ISD::PACKSS: {
// PACKSS is just a truncation if the sign bits extend to the packed size.
// TODO: Add DemandedElts support.
unsigned SrcBits = Op.getOperand(0).getScalarValueSizeInBits();
unsigned Tmp0 = DAG.ComputeNumSignBits(Op.getOperand(0), Depth + 1);
unsigned Tmp1 = DAG.ComputeNumSignBits(Op.getOperand(1), Depth + 1);
unsigned Tmp = std::min(Tmp0, Tmp1);
if (Tmp > (SrcBits - VTBits))
return Tmp - (SrcBits - VTBits);
return 1;
}
case X86ISD::VSHLI: {
SDValue Src = Op.getOperand(0);
unsigned Tmp = DAG.ComputeNumSignBits(Src, Depth + 1);

3368
test/CodeGen/X86/bitcast-and-setcc-512.ll
View File

File diff suppressed because it is too large Load Diff

74
test/CodeGen/X86/vselect-packss.ll
View File

@ -10,15 +10,24 @@
;
define <16 x i8> @vselect_packss_v16i16(<16 x i16> %a0, <16 x i16> %a1, <16 x i8> %a2, <16 x i8> %a3) {
; SSE-LABEL: vselect_packss_v16i16:
; SSE: # BB#0:
; SSE-NEXT: pcmpeqw %xmm3, %xmm1
; SSE-NEXT: pcmpeqw %xmm2, %xmm0
; SSE-NEXT: packsswb %xmm1, %xmm0
; SSE-NEXT: pand %xmm0, %xmm4
; SSE-NEXT: pandn %xmm5, %xmm0
; SSE-NEXT: por %xmm4, %xmm0
; SSE-NEXT: retq
; SSE2-LABEL: vselect_packss_v16i16:
; SSE2: # BB#0:
; SSE2-NEXT: pcmpeqw %xmm3, %xmm1
; SSE2-NEXT: pcmpeqw %xmm2, %xmm0
; SSE2-NEXT: packsswb %xmm1, %xmm0
; SSE2-NEXT: pand %xmm0, %xmm4
; SSE2-NEXT: pandn %xmm5, %xmm0
; SSE2-NEXT: por %xmm4, %xmm0
; SSE2-NEXT: retq
;
; SSE42-LABEL: vselect_packss_v16i16:
; SSE42: # BB#0:
; SSE42-NEXT: pcmpeqw %xmm3, %xmm1
; SSE42-NEXT: pcmpeqw %xmm2, %xmm0
; SSE42-NEXT: packsswb %xmm1, %xmm0
; SSE42-NEXT: pblendvb %xmm0, %xmm4, %xmm5
; SSE42-NEXT: movdqa %xmm5, %xmm0
; SSE42-NEXT: retq
;
; AVX1-LABEL: vselect_packss_v16i16:
; AVX1: # BB#0:
@ -27,9 +36,7 @@ define <16 x i8> @vselect_packss_v16i16(<16 x i16> %a0, <16 x i16> %a1, <16 x i8
; AVX1-NEXT: vpcmpeqw %xmm4, %xmm5, %xmm4
; AVX1-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vpacksswb %xmm4, %xmm0, %xmm0
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm1
; AVX1-NEXT: vpandn %xmm3, %xmm0, %xmm0
; AVX1-NEXT: vpor %xmm0, %xmm1, %xmm0
; AVX1-NEXT: vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq
;
@ -38,9 +45,7 @@ define <16 x i8> @vselect_packss_v16i16(<16 x i16> %a0, <16 x i16> %a1, <16 x i8
; AVX2-NEXT: vpcmpeqw %ymm1, %ymm0, %ymm0
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX2-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpand %xmm2, %xmm0, %xmm1
; AVX2-NEXT: vpandn %xmm3, %xmm0, %xmm0
; AVX2-NEXT: vpor %xmm0, %xmm1, %xmm0
; AVX2-NEXT: vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
@ -394,15 +399,24 @@ define <16 x i8> @vselect_packss_v16i64(<16 x i64> %a0, <16 x i64> %a1, <16 x i8
;
define <16 x i8> @vselect_packss(<16 x i16> %a0, <16 x i16> %a1, <16 x i8> %a2, <16 x i8> %a3) {
; SSE-LABEL: vselect_packss:
; SSE: # BB#0:
; SSE-NEXT: pcmpeqw %xmm3, %xmm1
; SSE-NEXT: pcmpeqw %xmm2, %xmm0
; SSE-NEXT: packsswb %xmm1, %xmm0
; SSE-NEXT: pand %xmm0, %xmm4
; SSE-NEXT: pandn %xmm5, %xmm0
; SSE-NEXT: por %xmm4, %xmm0
; SSE-NEXT: retq
; SSE2-LABEL: vselect_packss:
; SSE2: # BB#0:
; SSE2-NEXT: pcmpeqw %xmm3, %xmm1
; SSE2-NEXT: pcmpeqw %xmm2, %xmm0
; SSE2-NEXT: packsswb %xmm1, %xmm0
; SSE2-NEXT: pand %xmm0, %xmm4
; SSE2-NEXT: pandn %xmm5, %xmm0
; SSE2-NEXT: por %xmm4, %xmm0
; SSE2-NEXT: retq
;
; SSE42-LABEL: vselect_packss:
; SSE42: # BB#0:
; SSE42-NEXT: pcmpeqw %xmm3, %xmm1
; SSE42-NEXT: pcmpeqw %xmm2, %xmm0
; SSE42-NEXT: packsswb %xmm1, %xmm0
; SSE42-NEXT: pblendvb %xmm0, %xmm4, %xmm5
; SSE42-NEXT: movdqa %xmm5, %xmm0
; SSE42-NEXT: retq
;
; AVX1-LABEL: vselect_packss:
; AVX1: # BB#0:
@ -411,9 +425,7 @@ define <16 x i8> @vselect_packss(<16 x i16> %a0, <16 x i16> %a1, <16 x i8> %a2,
; AVX1-NEXT: vpcmpeqw %xmm4, %xmm5, %xmm4
; AVX1-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vpacksswb %xmm4, %xmm0, %xmm0
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm1
; AVX1-NEXT: vpandn %xmm3, %xmm0, %xmm0
; AVX1-NEXT: vpor %xmm0, %xmm1, %xmm0
; AVX1-NEXT: vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq
;
@ -422,9 +434,7 @@ define <16 x i8> @vselect_packss(<16 x i16> %a0, <16 x i16> %a1, <16 x i8> %a2,
; AVX2-NEXT: vpcmpeqw %ymm1, %ymm0, %ymm0
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX2-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpand %xmm2, %xmm0, %xmm1
; AVX2-NEXT: vpandn %xmm3, %xmm0, %xmm0
; AVX2-NEXT: vpor %xmm0, %xmm1, %xmm0
; AVX2-NEXT: vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
@ -433,9 +443,7 @@ define <16 x i8> @vselect_packss(<16 x i16> %a0, <16 x i16> %a1, <16 x i8> %a2,
; AVX512-NEXT: vpcmpeqw %ymm1, %ymm0, %ymm0
; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX512-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
; AVX512-NEXT: vpand %xmm2, %xmm0, %xmm1
; AVX512-NEXT: vpandn %xmm3, %xmm0, %xmm0
; AVX512-NEXT: vpor %xmm0, %xmm1, %xmm0
; AVX512-NEXT: vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
; AVX512-NEXT: vzeroupper
; AVX512-NEXT: retq
%1 = icmp eq <16 x i16> %a0, %a1