[llvm] [X86] Add test coverage for #214676 (PR #215107)

Simon Pilgrim via llvm-commits llvm-commits at lists.llvm.org
Sun Aug 9 08:31:42 PDT 2026


https://github.com/RKSimon created https://github.com/llvm/llvm-project/pull/215107

None

>From 737de3cb53e3c3e531282e27b550ae8d66915fec Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev at redking.me.uk>
Date: Sun, 9 Aug 2026 16:31:01 +0100
Subject: [PATCH] [X86] Add test coverage for #214676

---
 .../test/CodeGen/X86/avx512bf16-intrinsics.ll | 16 +++++++++
 .../CodeGen/X86/avx512bf16-vl-intrinsics.ll   | 33 +++++++++++++++++++
 2 files changed, 49 insertions(+)

diff --git a/llvm/test/CodeGen/X86/avx512bf16-intrinsics.ll b/llvm/test/CodeGen/X86/avx512bf16-intrinsics.ll
index a5767ecf14a53..ea0be592de710 100644
--- a/llvm/test/CodeGen/X86/avx512bf16-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512bf16-intrinsics.ll
@@ -35,6 +35,22 @@ entry:
   ret <8 x i64> %3
 }
 
+; concatenation test for swapped operands
+define <8 x i64> @test_mm512_cvtne2ps2bf16_512_concat(<16 x float> %A, <16 x float> %B) {
+; CHECK-LABEL: test_mm512_cvtne2ps2bf16_512_concat:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vcvtneps2bf16 %zmm0, %ymm0 # encoding: [0x62,0xf2,0x7e,0x48,0x72,0xc0]
+; CHECK-NEXT:    vcvtneps2bf16 %zmm1, %ymm1 # encoding: [0x62,0xf2,0x7e,0x48,0x72,0xc9]
+; CHECK-NEXT:    vinsertf64x4 $1, %ymm1, %zmm0, %zmm0 # encoding: [0x62,0xf3,0xfd,0x48,0x1a,0xc1,0x01]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+entry:
+  %0 = tail call <16 x bfloat> @llvm.x86.avx512bf16.cvtneps2bf16.512(<16 x float> %A)
+  %1 = tail call <16 x bfloat> @llvm.x86.avx512bf16.cvtneps2bf16.512(<16 x float> %B)
+  %2 = shufflevector <16 x bfloat> %0, <16 x bfloat> %1, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+  %3 = bitcast <32 x bfloat> %2 to <8 x i64>
+  ret <8 x i64> %3
+}
+
 define <8 x i64> @test_mm512_mask_cvtne2ps2bf16_512(<8 x i64> %C, i32 %U, <16 x float> %A, <16 x float> %B) local_unnamed_addr #2 {
 ; X86-LABEL: test_mm512_mask_cvtne2ps2bf16_512:
 ; X86:       # %bb.0: # %entry
diff --git a/llvm/test/CodeGen/X86/avx512bf16-vl-intrinsics.ll b/llvm/test/CodeGen/X86/avx512bf16-vl-intrinsics.ll
index b3c261147c773..936d764a909e0 100644
--- a/llvm/test/CodeGen/X86/avx512bf16-vl-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512bf16-vl-intrinsics.ll
@@ -58,6 +58,23 @@ entry:
   ret <2 x i64> %4
 }
 
+; concatenation test for swapped operands
+define <2 x i64> @test_mm_cvtne2ps2bf16_128_concat(<4 x float> %A, <4 x float> %B) {
+; CHECK-LABEL: test_mm_cvtne2ps2bf16_128_concat:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vcvtneps2bf16 %xmm0, %xmm0 # encoding: [0x62,0xf2,0x7e,0x08,0x72,0xc0]
+; CHECK-NEXT:    vcvtneps2bf16 %xmm1, %xmm1 # encoding: [0x62,0xf2,0x7e,0x08,0x72,0xc9]
+; CHECK-NEXT:    vmovlhps %xmm1, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x16,0xc1]
+; CHECK-NEXT:    # xmm0 = xmm0[0],xmm1[0]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+entry:
+  %0 = tail call <8 x bfloat> @llvm.x86.avx512bf16.mask.cvtneps2bf16.128(<4 x float> %A, <8 x bfloat> poison, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
+  %1 = tail call <8 x bfloat> @llvm.x86.avx512bf16.mask.cvtneps2bf16.128(<4 x float> %B, <8 x bfloat> poison, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
+  %2 = shufflevector <8 x bfloat> %0, <8 x bfloat> %1, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 10, i32 11>
+  %3 = bitcast <8 x bfloat> %2 to <2 x i64>
+  ret <2 x i64> %3
+}
+
 declare <16 x bfloat> @llvm.x86.avx512bf16.cvtne2ps2bf16.256(<8 x float>, <8 x float>) #3
 
 define <4 x i64> @test_mm256_cvtne2ps2bf16_256(<8 x float> %A, <8 x float> %B) local_unnamed_addr #1 {
@@ -112,6 +129,22 @@ entry:
   ret <4 x i64> %4
 }
 
+; concatenation test for swapped operands
+define <4 x i64> @test_mm256_cvtne2ps2bf16_256_concat(<8 x float> %A, <8 x float> %B) {
+; CHECK-LABEL: test_mm256_cvtne2ps2bf16_256_concat:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vcvtneps2bf16 %ymm0, %xmm0 # encoding: [0x62,0xf2,0x7e,0x28,0x72,0xc0]
+; CHECK-NEXT:    vcvtneps2bf16 %ymm1, %xmm1 # encoding: [0x62,0xf2,0x7e,0x28,0x72,0xc9]
+; CHECK-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe3,0x7d,0x18,0xc1,0x01]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+entry:
+  %0 = tail call <8 x bfloat> @llvm.x86.avx512bf16.cvtneps2bf16.256(<8 x float> %A)
+  %1 = tail call <8 x bfloat> @llvm.x86.avx512bf16.cvtneps2bf16.256(<8 x float> %B)
+  %2 = shufflevector <8 x bfloat> %0, <8 x bfloat> %1, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+  %3 = bitcast <16 x bfloat> %2 to <4 x i64>
+  ret <4 x i64> %3
+}
+
 declare <8 x bfloat> @llvm.x86.avx512bf16.cvtneps2bf16.256(<8 x float>) #3
 
 define <2 x i64> @test_mm256_cvtneps2bf16_256(<8 x float> %A) local_unnamed_addr #2 {



More information about the llvm-commits mailing list