https://github.com/tstellar updated
https://github.com/llvm/llvm-project/pull/91425
>From 2fc32a278e4fd46c6dd085845e69e84c321a3f75 Mon Sep 17 00:00:00 2001
From: Phoebe Wang <phoebe.w...@intel.com>
Date: Mon, 6 May 2024 10:59:44 +0800
Subject: [PATCH 1/2] [X86][FP16] Do not create VBROADCAST_LOAD for f16 without
AVX2 (#91125)
AVX doesn't provide 16-bit BROADCAST instruction.
Fixes #91005
---
llvm/lib/Target/X86/X86ISelLowering.cpp | 2 +-
llvm/test/CodeGen/X86/pr91005.ll | 39 +++++++++++++++++++++++++
2 files changed, 40 insertions(+), 1 deletion(-)
create mode 100644 llvm/test/CodeGen/X86/pr91005.ll
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp
b/llvm/lib/Target/X86/X86ISelLowering.cpp
index c572b27fe401e..3e4ecab8443a9 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -7295,7 +7295,7 @@ static SDValue
lowerBuildVectorAsBroadcast(BuildVectorSDNode *BVOp,
// With pattern matching, the VBROADCAST node may become a VMOVDDUP.
if (ScalarSize == 32 ||
(ScalarSize == 64 && (IsGE256 || Subtarget.hasVLX())) ||
- CVT == MVT::f16 ||
+ (CVT == MVT::f16 && Subtarget.hasAVX2()) ||
(OptForSize && (ScalarSize == 64 || Subtarget.hasAVX2()))) {
const Constant *C = nullptr;
if (ConstantSDNode *CI = dyn_cast<ConstantSDNode>(Ld))
diff --git a/llvm/test/CodeGen/X86/pr91005.ll b/llvm/test/CodeGen/X86/pr91005.ll
new file mode 100644
index 0000000000000..97fd1ce456882
--- /dev/null
+++ b/llvm/test/CodeGen/X86/pr91005.ll
@@ -0,0 +1,39 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
UTC_ARGS: --version 4
+; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=+f16c < %s | FileCheck %s
+
+define void @PR91005(ptr %0) minsize {
+; CHECK-LABEL: PR91005:
+; CHECK: # %bb.0:
+; CHECK-NEXT: xorl %eax, %eax
+; CHECK-NEXT: testb %al, %al
+; CHECK-NEXT: je .LBB0_2
+; CHECK-NEXT: # %bb.1:
+; CHECK-NEXT: vbroadcastss {{.*#+}} xmm0 = [31744,31744,31744,31744]
+; CHECK-NEXT: vpcmpeqw %xmm0, %xmm0, %xmm0
+; CHECK-NEXT: vpinsrw $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm1
+; CHECK-NEXT: vpand %xmm1, %xmm0, %xmm0
+; CHECK-NEXT: vcvtph2ps %xmm0, %xmm0
+; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
+; CHECK-NEXT: vmulss %xmm1, %xmm0, %xmm0
+; CHECK-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; CHECK-NEXT: vmovd %xmm0, %eax
+; CHECK-NEXT: movw %ax, (%rdi)
+; CHECK-NEXT: .LBB0_2: # %common.ret
+; CHECK-NEXT: retq
+ %2 = bitcast <2 x half> poison to <2 x i16>
+ %3 = icmp eq <2 x i16> %2, <i16 31744, i16 31744>
+ br i1 poison, label %4, label %common.ret
+
+common.ret: ; preds = %4, %1
+ ret void
+
+4: ; preds = %1
+ %5 = select <2 x i1> %3, <2 x half> <half 0xH3C00, half 0xH3C00>, <2 x half>
zeroinitializer
+ %6 = fmul <2 x half> %5, zeroinitializer
+ %7 = fsub <2 x half> %6, zeroinitializer
+ %8 = extractelement <2 x half> %7, i64 0
+ store half %8, ptr %0, align 2
+ br label %common.ret
+}
+
+declare <2 x half> @llvm.fabs.v2f16(<2 x half>)
>From 4d284b853f26a6cb848028720163561cabf63d95 Mon Sep 17 00:00:00 2001
From: Phoebe Wang <phoebe.w...@intel.com>
Date: Wed, 8 May 2024 10:59:31 +0800
Subject: [PATCH 2/2] Fix difference with LLVM 18 release
---
llvm/test/CodeGen/X86/pr91005.ll | 11 ++++++-----
1 file changed, 6 insertions(+), 5 deletions(-)
diff --git a/llvm/test/CodeGen/X86/pr91005.ll b/llvm/test/CodeGen/X86/pr91005.ll
index 97fd1ce456882..16b78bf1e7e17 100644
--- a/llvm/test/CodeGen/X86/pr91005.ll
+++ b/llvm/test/CodeGen/X86/pr91005.ll
@@ -8,12 +8,13 @@ define void @PR91005(ptr %0) minsize {
; CHECK-NEXT: testb %al, %al
; CHECK-NEXT: je .LBB0_2
; CHECK-NEXT: # %bb.1:
-; CHECK-NEXT: vbroadcastss {{.*#+}} xmm0 = [31744,31744,31744,31744]
-; CHECK-NEXT: vpcmpeqw %xmm0, %xmm0, %xmm0
-; CHECK-NEXT: vpinsrw $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm1
-; CHECK-NEXT: vpand %xmm1, %xmm0, %xmm0
+; CHECK-NEXT: vpcmpeqw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; CHECK-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; CHECK-NEXT: vpextrw $0, %xmm0, %eax
+; CHECK-NEXT: movzwl %ax, %eax
+; CHECK-NEXT: vmovd %eax, %xmm0
; CHECK-NEXT: vcvtph2ps %xmm0, %xmm0
-; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
+; CHECK-NEXT: vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vmulss %xmm1, %xmm0, %xmm0
; CHECK-NEXT: vcvtps2ph $4, %xmm0, %xmm0
; CHECK-NEXT: vmovd %xmm0, %eax
_______________________________________________
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
