[llvm] [X86] Add assembly comment for VMOVSH instructions (PR #156848)
via llvm-commits
llvm-commits at lists.llvm.org
Thu Sep 4 03:32:37 PDT 2025
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-x86
Author: Simon Pilgrim (RKSimon)
<details>
<summary>Changes</summary>
Matches existing MOVSD/S implementations
---
Full diff: https://github.com/llvm/llvm-project/pull/156848.diff
5 Files Affected:
- (modified) llvm/lib/Target/X86/MCTargetDesc/X86InstComments.cpp (+7)
- (modified) llvm/test/CodeGen/X86/avx512fp16-fold-load-binops.ll (+2-2)
- (modified) llvm/test/CodeGen/X86/avx512fp16-intrinsics.ll (+1-1)
- (modified) llvm/test/CodeGen/X86/avx512fp16-mov.ll (+12-12)
- (modified) llvm/test/CodeGen/X86/vec-strict-128-fp16.ll (+2-2)
``````````diff
diff --git a/llvm/lib/Target/X86/MCTargetDesc/X86InstComments.cpp b/llvm/lib/Target/X86/MCTargetDesc/X86InstComments.cpp
index 547745fdba9d6..76731437931a7 100644
--- a/llvm/lib/Target/X86/MCTargetDesc/X86InstComments.cpp
+++ b/llvm/lib/Target/X86/MCTargetDesc/X86InstComments.cpp
@@ -1668,6 +1668,13 @@ bool llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS,
DestName = getRegName(MI->getOperand(0).getReg());
break;
+ case X86::VMOVSHZrr:
+ Src2Name = getRegName(MI->getOperand(2).getReg());
+ Src1Name = getRegName(MI->getOperand(1).getReg());
+ DecodeScalarMoveMask(8, false, ShuffleMask);
+ DestName = getRegName(MI->getOperand(0).getReg());
+ break;
+
case X86::MOVPQI2QIrr:
case X86::MOVZPQILo2PQIrr:
case X86::VMOVPQI2QIrr:
diff --git a/llvm/test/CodeGen/X86/avx512fp16-fold-load-binops.ll b/llvm/test/CodeGen/X86/avx512fp16-fold-load-binops.ll
index 56d923d7c4cf1..4a5c1fe5a2a04 100644
--- a/llvm/test/CodeGen/X86/avx512fp16-fold-load-binops.ll
+++ b/llvm/test/CodeGen/X86/avx512fp16-fold-load-binops.ll
@@ -57,7 +57,7 @@ define <8 x half> @minsh(<8 x half> %va, ptr %pb) {
; CHECK-LABEL: minsh:
; CHECK: # %bb.0:
; CHECK-NEXT: vminsh (%rdi), %xmm0, %xmm1
-; CHECK-NEXT: vmovsh %xmm1, %xmm0, %xmm0
+; CHECK-NEXT: vmovsh {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3,4,5,6,7]
; CHECK-NEXT: retq
%a = extractelement <8 x half> %va, i32 0
%b = load half, ptr %pb
@@ -70,7 +70,7 @@ define <8 x half> @maxsh(<8 x half> %va, ptr %pb) {
; CHECK-LABEL: maxsh:
; CHECK: # %bb.0:
; CHECK-NEXT: vminsh (%rdi), %xmm0, %xmm1
-; CHECK-NEXT: vmovsh %xmm1, %xmm0, %xmm0
+; CHECK-NEXT: vmovsh {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3,4,5,6,7]
; CHECK-NEXT: retq
%a = extractelement <8 x half> %va, i32 0
%b = load half, ptr %pb
diff --git a/llvm/test/CodeGen/X86/avx512fp16-intrinsics.ll b/llvm/test/CodeGen/X86/avx512fp16-intrinsics.ll
index b1bacd92f073b..44ea3ce64ccf4 100644
--- a/llvm/test/CodeGen/X86/avx512fp16-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512fp16-intrinsics.ll
@@ -1369,7 +1369,7 @@ define <8 x half> @PR153570(ptr %p) {
; CHECK-NEXT: vpbroadcastw {{.*#+}} xmm1 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; CHECK-NEXT: vmulsh {rn-sae}, %xmm0, %xmm1, %xmm0
; CHECK-NEXT: vpbroadcastw {{.*#+}} xmm2 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0]
-; CHECK-NEXT: vmovsh %xmm2, %xmm1, %xmm1
+; CHECK-NEXT: vmovsh {{.*#+}} xmm1 = xmm2[0],xmm1[1,2,3,4,5,6,7]
; CHECK-NEXT: vmovaps %xmm1, (%rdi)
; CHECK-NEXT: retq
%r = tail call <8 x half> @llvm.x86.avx512fp16.mask.mul.sh.round(<8 x half> <half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00>, <8 x half> <half 0xH4000, half 0xH4000, half 0xH4000, half 0xH4000, half 0xH4000, half 0xH4000, half 0xH4000, half 0xH4000>, <8 x half> <half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000, half 0xH8000>, i8 0, i32 8)
diff --git a/llvm/test/CodeGen/X86/avx512fp16-mov.ll b/llvm/test/CodeGen/X86/avx512fp16-mov.ll
index 526511c850451..316e3f27a0a1f 100644
--- a/llvm/test/CodeGen/X86/avx512fp16-mov.ll
+++ b/llvm/test/CodeGen/X86/avx512fp16-mov.ll
@@ -303,7 +303,7 @@ define <8 x half> @test14(half %x) {
; X64-LABEL: test14:
; X64: # %bb.0:
; X64-NEXT: vxorps %xmm1, %xmm1, %xmm1
-; X64-NEXT: vmovsh %xmm0, %xmm1, %xmm0
+; X64-NEXT: vmovsh {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
; X64-NEXT: retq
;
; X86-LABEL: test14:
@@ -318,7 +318,7 @@ define <16 x half> @test14b(half %x) {
; X64VL-LABEL: test14b:
; X64VL: # %bb.0:
; X64VL-NEXT: vxorps %xmm1, %xmm1, %xmm1
-; X64VL-NEXT: vmovsh %xmm0, %xmm1, %xmm0
+; X64VL-NEXT: vmovsh {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
; X64VL-NEXT: retq
;
; X86-LABEL: test14b:
@@ -329,7 +329,7 @@ define <16 x half> @test14b(half %x) {
; X64-NOVL-LABEL: test14b:
; X64-NOVL: # %bb.0:
; X64-NOVL-NEXT: vxorps %xmm1, %xmm1, %xmm1
-; X64-NOVL-NEXT: vmovsh %xmm0, %xmm1, %xmm0
+; X64-NOVL-NEXT: vmovsh {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
; X64-NOVL-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X64-NOVL-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
; X64-NOVL-NEXT: retq
@@ -341,7 +341,7 @@ define <32 x half> @test14c(half %x) {
; X64VL-LABEL: test14c:
; X64VL: # %bb.0:
; X64VL-NEXT: vxorps %xmm1, %xmm1, %xmm1
-; X64VL-NEXT: vmovsh %xmm0, %xmm1, %xmm0
+; X64VL-NEXT: vmovsh {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
; X64VL-NEXT: retq
;
; X86-LABEL: test14c:
@@ -352,7 +352,7 @@ define <32 x half> @test14c(half %x) {
; X64-NOVL-LABEL: test14c:
; X64-NOVL: # %bb.0:
; X64-NOVL-NEXT: vxorps %xmm1, %xmm1, %xmm1
-; X64-NOVL-NEXT: vmovsh %xmm0, %xmm1, %xmm0
+; X64-NOVL-NEXT: vmovsh {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
; X64-NOVL-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X64-NOVL-NEXT: vinsertf32x4 $0, %xmm0, %zmm1, %zmm0
; X64-NOVL-NEXT: retq
@@ -1464,21 +1464,21 @@ define <8 x half> @movsh(<8 x half> %a, <8 x half> %b) {
; X64VL-LABEL: movsh:
; X64VL: # %bb.0:
; X64VL-NEXT: vpshufb {{.*#+}} xmm2 = xmm0[0,1,14,15,0,1,2,3,4,5,6,7,14,15,10,11]
-; X64VL-NEXT: vmovsh %xmm0, %xmm1, %xmm0
+; X64VL-NEXT: vmovsh {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
; X64VL-NEXT: vaddph %xmm0, %xmm2, %xmm0
; X64VL-NEXT: retq
;
; X86-LABEL: movsh:
; X86: # %bb.0:
; X86-NEXT: vpshufb {{.*#+}} xmm2 = xmm0[0,1,14,15,0,1,2,3,4,5,6,7,14,15,10,11]
-; X86-NEXT: vmovsh %xmm0, %xmm1, %xmm0
+; X86-NEXT: vmovsh {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
; X86-NEXT: vaddph %xmm0, %xmm2, %xmm0
; X86-NEXT: retl
;
; X64-NOVL-LABEL: movsh:
; X64-NOVL: # %bb.0:
; X64-NOVL-NEXT: vpshufb {{.*#+}} xmm2 = xmm0[0,1,14,15,0,1,2,3,4,5,6,7,14,15,10,11]
-; X64-NOVL-NEXT: vmovsh %xmm0, %xmm1, %xmm3
+; X64-NOVL-NEXT: vmovsh {{.*#+}} xmm3 = xmm0[0],xmm1[1,2,3,4,5,6,7]
; X64-NOVL-NEXT: vpsrldq {{.*#+}} xmm4 = xmm3[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
; X64-NOVL-NEXT: vpsrldq {{.*#+}} xmm5 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
; X64-NOVL-NEXT: vaddsh %xmm4, %xmm5, %xmm4
@@ -2311,7 +2311,7 @@ define <8 x half> @test21(half %a, half %b, half %c) nounwind {
; X64-LABEL: test21:
; X64: # %bb.0:
; X64-NEXT: vxorps %xmm3, %xmm3, %xmm3
-; X64-NEXT: vmovsh %xmm2, %xmm3, %xmm2
+; X64-NEXT: vmovsh {{.*#+}} xmm2 = xmm2[0],xmm3[1,2,3,4,5,6,7]
; X64-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; X64-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0],xmm2[0],zero,zero
; X64-NEXT: retq
@@ -2427,7 +2427,7 @@ define <16 x i32> @pr52561(<16 x i32> %a, <16 x i32> %b) "min-legal-vector-width
; X64VL-NEXT: vpaddd %ymm2, %ymm1, %ymm1
; X64VL-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
; X64VL-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; X64VL-NEXT: vmovsh %xmm0, %xmm2, %xmm0
+; X64VL-NEXT: vmovsh {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3,4,5,6,7]
; X64VL-NEXT: retq
;
; X86-LABEL: pr52561:
@@ -2443,7 +2443,7 @@ define <16 x i32> @pr52561(<16 x i32> %a, <16 x i32> %b) "min-legal-vector-width
; X86-NEXT: vpaddd %ymm2, %ymm1, %ymm1
; X86-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}, %ymm1, %ymm1
; X86-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; X86-NEXT: vmovsh %xmm0, %xmm2, %xmm0
+; X86-NEXT: vmovsh {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3,4,5,6,7]
; X86-NEXT: movl %ebp, %esp
; X86-NEXT: popl %ebp
; X86-NEXT: retl
@@ -2474,7 +2474,7 @@ define <8 x i16> @pr59628_xmm(i16 %arg) {
; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; X86-NEXT: vxorps %xmm0, %xmm0, %xmm0
; X86-NEXT: vpbroadcastw %eax, %xmm1
-; X86-NEXT: vmovsh %xmm1, %xmm0, %xmm0
+; X86-NEXT: vmovsh {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3,4,5,6,7]
; X86-NEXT: vpcmpneqw {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1, %k1
; X86-NEXT: vmovdqu16 %xmm0, %xmm0 {%k1} {z}
; X86-NEXT: retl
diff --git a/llvm/test/CodeGen/X86/vec-strict-128-fp16.ll b/llvm/test/CodeGen/X86/vec-strict-128-fp16.ll
index 35688e59fc9f4..766ccdbada539 100644
--- a/llvm/test/CodeGen/X86/vec-strict-128-fp16.ll
+++ b/llvm/test/CodeGen/X86/vec-strict-128-fp16.ll
@@ -79,7 +79,7 @@ define <8 x half> @f11(<2 x double> %a0, <8 x half> %a1) #0 {
; CHECK-LABEL: f11:
; CHECK: # %bb.0:
; CHECK-NEXT: vcvtsd2sh %xmm0, %xmm0, %xmm0
-; CHECK-NEXT: vmovsh %xmm0, %xmm1, %xmm0
+; CHECK-NEXT: vmovsh {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
; CHECK-NEXT: ret{{[l|q]}}
%ext = extractelement <2 x double> %a0, i32 0
%cvt = call half @llvm.experimental.constrained.fptrunc.f16.f64(double %ext,
@@ -140,7 +140,7 @@ define <8 x half> @f17(<4 x float> %a0, <8 x half> %a1) #0 {
; CHECK-LABEL: f17:
; CHECK: # %bb.0:
; CHECK-NEXT: vcvtss2sh %xmm0, %xmm0, %xmm0
-; CHECK-NEXT: vmovsh %xmm0, %xmm1, %xmm0
+; CHECK-NEXT: vmovsh {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
; CHECK-NEXT: ret{{[l|q]}}
%ext = extractelement <4 x float> %a0, i32 0
%cvt = call half @llvm.experimental.constrained.fptrunc.f16.f32(float %ext,
``````````
</details>
https://github.com/llvm/llvm-project/pull/156848
More information about the llvm-commits
mailing list