[X86] Allow more cmpps/pd immediate encodings to be commuted during isel.

The code that checks the immediate wasn't masking to the lower 3-bits like the code in X86InstrInfo.cpp that's used by the peephole pass does.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@322060 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 189fb46..30fca8e 100644 (file)
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -2398,8 +2398,8 @@ let Constraints = "$src1 = $dst" in {
 }
 
 def CommutableCMPCC : PatLeaf<(imm), [{
-  return (N->getZExtValue() == 0x00 || N->getZExtValue() == 0x03 ||
-          N->getZExtValue() == 0x04 || N->getZExtValue() == 0x07);
+  uint64_t Imm = N->getZExtValue() & 0x7;
+  return (Imm == 0x00 || Imm == 0x03 || Imm == 0x04 || Imm == 0x07);
 }]>;
 
 // Patterns to select compares with loads in first operand.

diff --git a/test/CodeGen/X86/commute-fcmp.ll b/test/CodeGen/X86/commute-fcmp.ll
index f5b7030..c6573b4 100644 (file)
--- a/test/CodeGen/X86/commute-fcmp.ll
+++ b/test/CodeGen/X86/commute-fcmp.ll
@@ -112,14 +112,12 @@ define <4 x i32> @commute_cmpps_ueq(<4 x float>* %a0, <4 x float> %a1) {
 ;
 ; AVX-LABEL: commute_cmpps_ueq:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovaps (%rdi), %xmm1
-; AVX-NEXT:    vcmpeq_uqps %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    vcmpeq_uqps (%rdi), %xmm0, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: commute_cmpps_ueq:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vmovaps (%rdi), %xmm1
-; AVX512-NEXT:    vcmpeq_uqps %xmm0, %xmm1, %k1
+; AVX512-NEXT:    vcmpeq_uqps (%rdi), %xmm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
 ; AVX512-NEXT:    retq
@@ -141,14 +139,12 @@ define <4 x i32> @commute_cmpps_one(<4 x float>* %a0, <4 x float> %a1) {
 ;
 ; AVX-LABEL: commute_cmpps_one:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovaps (%rdi), %xmm1
-; AVX-NEXT:    vcmpneq_oqps %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    vcmpneq_oqps (%rdi), %xmm0, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: commute_cmpps_one:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vmovaps (%rdi), %xmm1
-; AVX512-NEXT:    vcmpneq_oqps %xmm0, %xmm1, %k1
+; AVX512-NEXT:    vcmpneq_oqps (%rdi), %xmm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
 ; AVX512-NEXT:    retq
@@ -325,14 +321,12 @@ define <8 x i32> @commute_cmpps_ueq_ymm(<8 x float>* %a0, <8 x float> %a1) {
 ;
 ; AVX-LABEL: commute_cmpps_ueq_ymm:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovaps (%rdi), %ymm1
-; AVX-NEXT:    vcmpeq_uqps %ymm0, %ymm1, %ymm0
+; AVX-NEXT:    vcmpeq_uqps (%rdi), %ymm0, %ymm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: commute_cmpps_ueq_ymm:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vmovaps (%rdi), %ymm1
-; AVX512-NEXT:    vcmpeq_uqps %ymm0, %ymm1, %k1
+; AVX512-NEXT:    vcmpeq_uqps (%rdi), %ymm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
 ; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
 ; AVX512-NEXT:    retq
@@ -359,14 +353,12 @@ define <8 x i32> @commute_cmpps_one_ymm(<8 x float>* %a0, <8 x float> %a1) {
 ;
 ; AVX-LABEL: commute_cmpps_one_ymm:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovaps (%rdi), %ymm1
-; AVX-NEXT:    vcmpneq_oqps %ymm0, %ymm1, %ymm0
+; AVX-NEXT:    vcmpneq_oqps (%rdi), %ymm0, %ymm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: commute_cmpps_one_ymm:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vmovaps (%rdi), %ymm1
-; AVX512-NEXT:    vcmpneq_oqps %ymm0, %ymm1, %k1
+; AVX512-NEXT:    vcmpneq_oqps (%rdi), %ymm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
 ; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
 ; AVX512-NEXT:    retq
@@ -522,14 +514,12 @@ define <2 x i64> @commute_cmppd_ueq(<2 x double>* %a0, <2 x double> %a1) {
 ;
 ; AVX-LABEL: commute_cmppd_ueq:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovapd (%rdi), %xmm1
-; AVX-NEXT:    vcmpeq_uqpd %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    vcmpeq_uqpd (%rdi), %xmm0, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: commute_cmppd_ueq:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vmovapd (%rdi), %xmm1
-; AVX512-NEXT:    vcmpeq_uqpd %xmm0, %xmm1, %k1
+; AVX512-NEXT:    vcmpeq_uqpd (%rdi), %xmm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z}
 ; AVX512-NEXT:    retq
@@ -551,14 +541,12 @@ define <2 x i64> @commute_cmppd_one(<2 x double>* %a0, <2 x double> %a1) {
 ;
 ; AVX-LABEL: commute_cmppd_one:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovapd (%rdi), %xmm1
-; AVX-NEXT:    vcmpneq_oqpd %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    vcmpneq_oqpd (%rdi), %xmm0, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: commute_cmppd_one:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vmovapd (%rdi), %xmm1
-; AVX512-NEXT:    vcmpneq_oqpd %xmm0, %xmm1, %k1
+; AVX512-NEXT:    vcmpneq_oqpd (%rdi), %xmm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z}
 ; AVX512-NEXT:    retq
@@ -758,14 +746,12 @@ define <4 x i64> @commute_cmppd_ueq_ymmm(<4 x double>* %a0, <4 x double> %a1) {
 ;
 ; AVX-LABEL: commute_cmppd_ueq_ymmm:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovapd (%rdi), %ymm1
-; AVX-NEXT:    vcmpeq_uqpd %ymm0, %ymm1, %ymm0
+; AVX-NEXT:    vcmpeq_uqpd (%rdi), %ymm0, %ymm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: commute_cmppd_ueq_ymmm:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vmovapd (%rdi), %ymm1
-; AVX512-NEXT:    vcmpeq_uqpd %ymm0, %ymm1, %k1
+; AVX512-NEXT:    vcmpeq_uqpd (%rdi), %ymm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
 ; AVX512-NEXT:    vmovdqa64 %ymm0, %ymm0 {%k1} {z}
 ; AVX512-NEXT:    retq
@@ -792,14 +778,12 @@ define <4 x i64> @commute_cmppd_one_ymmm(<4 x double>* %a0, <4 x double> %a1) {
 ;
 ; AVX-LABEL: commute_cmppd_one_ymmm:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovapd (%rdi), %ymm1
-; AVX-NEXT:    vcmpneq_oqpd %ymm0, %ymm1, %ymm0
+; AVX-NEXT:    vcmpneq_oqpd (%rdi), %ymm0, %ymm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: commute_cmppd_one_ymmm:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vmovapd (%rdi), %ymm1
-; AVX512-NEXT:    vcmpneq_oqpd %ymm0, %ymm1, %k1
+; AVX512-NEXT:    vcmpneq_oqpd (%rdi), %ymm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
 ; AVX512-NEXT:    vmovdqa64 %ymm0, %ymm0 {%k1} {z}
 ; AVX512-NEXT:    retq