[llvm] Make extract vector for bf16 types legal (PR #222913)

[via llvm-commits]

https://github.com/tfzee created https://github.com/llvm/llvm-project/pull/222913

Fixes #222585

Makes EXTRACT_SUBVECTOR legal for bf16 types leading to this new codegen
```asm
# $ ./build/bin/clang test.c -O3 -march=novalake -mprefer-vector-width=512 -ffast-math -S -o -
foo:
        vmovups (%rsi), %zmm0
        vmulph  (%rdi), %zmm0, %zmm0
        vextractf64x4   $1, %zmm0, %ymm1
        vaddph  %zmm1, %zmm0, %zmm0
        vextractf128    $1, %ymm0, %xmm1
        vaddph  %xmm1, %xmm0, %xmm0
        vshufpd $1, %xmm0, %xmm0, %xmm1         # xmm1 = xmm0[1,0]
        vaddph  %xmm1, %xmm0, %xmm0
        vmovshdup       %xmm0, %xmm1            # xmm1 = xmm0[1,1,3,3]
        vaddph  %xmm1, %xmm0, %xmm0
        vpsrld  $16, %xmm0, %xmm1
        vaddsh  %xmm1, %xmm0, %xmm0
        vzeroupper
        retq

bar:
        vmovups (%rsi), %zmm0
        vmulbf16        (%rdi), %zmm0, %zmm0
        vextractf64x4   $1, %zmm0, %ymm1
        vaddbf16        %zmm1, %zmm0, %zmm0
        vextractf32x4   $1, %zmm0, %xmm1
        vaddbf16        %xmm1, %xmm0, %xmm0
        vshufpd $1, %xmm0, %xmm0, %xmm1         # xmm1 = xmm0[1,0]
        vaddbf16        %xmm1, %xmm0, %xmm0
        vmovshdup       %xmm0, %xmm1            # xmm1 = xmm0[1,1,3,3]
        vaddbf16        %xmm1, %xmm0, %xmm0
        vpsrld  $16, %xmm0, %xmm1
        vaddbf16        %xmm1, %xmm0, %xmm0
        vzeroupper
        retq
```

>From 75afeef38c475492bdba26baa5c8ba45a10c40b9 Mon Sep 17 00:00:00 2001
From: "Ziegler, Tim" <tim.ziegler at intel.com>
Date: Fri, 11 Sep 2026 12:33:58 +0200
Subject: [PATCH] Make extract vector for bf16 types legal

---
 llvm/lib/Target/X86/X86ISelLowering.cpp       |  1 +
 .../avx512-shuffles/shuffle-chained-bf16.ll   | 24 ++-----------------
 2 files changed, 3 insertions(+), 22 deletions(-)

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index a41cb2914660c..fd22ada6109c1 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -2590,6 +2590,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
         setOperationAction(ISD::VSELECT, VT, Custom);
       setOperationAction(ISD::BUILD_VECTOR, VT, Custom);
       setOperationAction(ISD::VECTOR_SHUFFLE, VT, Custom);
+      setOperationAction(ISD::EXTRACT_SUBVECTOR, VT, Legal);
       setOperationAction(ISD::INSERT_SUBVECTOR, VT, Legal);
       setOperationAction(ISD::CONCAT_VECTORS, VT, Custom);
     }
diff --git a/llvm/test/CodeGen/X86/avx512-shuffles/shuffle-chained-bf16.ll b/llvm/test/CodeGen/X86/avx512-shuffles/shuffle-chained-bf16.ll
index a7c79016334fa..85e5a7b6c39be 100644
--- a/llvm/test/CodeGen/X86/avx512-shuffles/shuffle-chained-bf16.ll
+++ b/llvm/test/CodeGen/X86/avx512-shuffles/shuffle-chained-bf16.ll
@@ -6,19 +6,9 @@ target triple = "x86_64-unknown-linux-gnu"
 define <2 x bfloat> @shuffle_chained_v32bf16_v2bf16(<32 x bfloat> %a) {
 ; CHECK-LABEL: shuffle_chained_v32bf16_v2bf16:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    pushq %rbp
-; CHECK-NEXT:    .cfi_def_cfa_offset 16
-; CHECK-NEXT:    .cfi_offset %rbp, -16
-; CHECK-NEXT:    movq %rsp, %rbp
-; CHECK-NEXT:    .cfi_def_cfa_register %rbp
-; CHECK-NEXT:    andq $-64, %rsp
-; CHECK-NEXT:    addq $-128, %rsp
 ; CHECK-NEXT:    vmovd {{.*#+}} xmm1 = [0,16,0,0,0,0,0,0]
 ; CHECK-NEXT:    vpermw %zmm0, %zmm1, %zmm0
 ; CHECK-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
-; CHECK-NEXT:    movq %rbp, %rsp
-; CHECK-NEXT:    popq %rbp
-; CHECK-NEXT:    .cfi_def_cfa %rsp, 8
 ; CHECK-NEXT:    vzeroupper
 ; CHECK-NEXT:    retq
   %s = shufflevector <32 x bfloat> %a, <32 x bfloat> zeroinitializer, <32 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23, i32 8, i32 24, i32 9, i32 25, i32 10, i32 26, i32 11, i32 27, i32 12, i32 28, i32 13, i32 29, i32 14, i32 30, i32 15, i32 31>
@@ -29,18 +19,8 @@ define <2 x bfloat> @shuffle_chained_v32bf16_v2bf16(<32 x bfloat> %a) {
 define <2 x bfloat> @shuffle_chained_v16bf16(<16 x bfloat> %a) {
 ; CHECK-LABEL: shuffle_chained_v16bf16:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    pushq %rbp
-; CHECK-NEXT:    .cfi_def_cfa_offset 16
-; CHECK-NEXT:    .cfi_offset %rbp, -16
-; CHECK-NEXT:    movq %rsp, %rbp
-; CHECK-NEXT:    .cfi_def_cfa_register %rbp
-; CHECK-NEXT:    andq $-32, %rsp
-; CHECK-NEXT:    subq $96, %rsp
-; CHECK-NEXT:    vmovdqa %ymm0, (%rsp)
-; CHECK-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; CHECK-NEXT:    movq %rbp, %rsp
-; CHECK-NEXT:    popq %rbp
-; CHECK-NEXT:    .cfi_def_cfa %rsp, 8
+; CHECK-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; CHECK-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
 ; CHECK-NEXT:    vzeroupper
 ; CHECK-NEXT:    retq
   %s = shufflevector <16 x bfloat> %a, <16 x bfloat> zeroinitializer, <16 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11, i32 4, i32 12, i32 5, i32 13, i32 6, i32 14, i32 7, i32 15>