[llvm-branch-commits] [llvm] [SelectionDAG] Split vector types for atomic store (PR #197860)

[Matt Arsenault via llvm-branch-commits]

================
@@ -751,6 +751,456 @@ define <2 x float> @atomic_vec2_float_align(ptr %x) {
   ret <2 x float> %ret
 }
 
+define void @store_atomic_vec2_half(ptr %x, <2 x half> %v) {
+; CHECK-SSE-O3-LABEL: store_atomic_vec2_half:
+; CHECK-SSE-O3:       # %bb.0:
+; CHECK-SSE-O3-NEXT:    pextrw $0, %xmm0, %eax
+; CHECK-SSE-O3-NEXT:    psrld $16, %xmm0
+; CHECK-SSE-O3-NEXT:    pextrw $0, %xmm0, %ecx
+; CHECK-SSE-O3-NEXT:    shll $16, %ecx
+; CHECK-SSE-O3-NEXT:    movzwl %ax, %eax
+; CHECK-SSE-O3-NEXT:    orl %ecx, %eax
+; CHECK-SSE-O3-NEXT:    movl %eax, (%rdi)
+; CHECK-SSE-O3-NEXT:    retq
+;
+; CHECK-AVX-O3-LABEL: store_atomic_vec2_half:
+; CHECK-AVX-O3:       # %bb.0:
+; CHECK-AVX-O3-NEXT:    vmovd %xmm0, %eax
+; CHECK-AVX-O3-NEXT:    movl %eax, (%rdi)
+; CHECK-AVX-O3-NEXT:    retq
+;
+; CHECK-SSE-O0-LABEL: store_atomic_vec2_half:
+; CHECK-SSE-O0:       # %bb.0:
+; CHECK-SSE-O0-NEXT:    movaps %xmm0, %xmm1
+; CHECK-SSE-O0-NEXT:    movaps %xmm1, %xmm0
+; CHECK-SSE-O0-NEXT:    psrld $16, %xmm1
+; CHECK-SSE-O0-NEXT:    pextrw $0, %xmm1, %eax
+; CHECK-SSE-O0-NEXT:    # kill: def $ax killed $ax killed $eax
+; CHECK-SSE-O0-NEXT:    # implicit-def: $ecx
+; CHECK-SSE-O0-NEXT:    movw %ax, %cx
+; CHECK-SSE-O0-NEXT:    shll $16, %ecx
+; CHECK-SSE-O0-NEXT:    pextrw $0, %xmm0, %eax
+; CHECK-SSE-O0-NEXT:    # kill: def $ax killed $ax killed $eax
+; CHECK-SSE-O0-NEXT:    movzwl %ax, %eax
+; CHECK-SSE-O0-NEXT:    orl %ecx, %eax
+; CHECK-SSE-O0-NEXT:    movl %eax, (%rdi)
+; CHECK-SSE-O0-NEXT:    retq
+;
+; CHECK-AVX-O0-LABEL: store_atomic_vec2_half:
+; CHECK-AVX-O0:       # %bb.0:
+; CHECK-AVX-O0-NEXT:    vmovd %xmm0, %eax
+; CHECK-AVX-O0-NEXT:    movl %eax, (%rdi)
+; CHECK-AVX-O0-NEXT:    retq
+  store atomic <2 x half> %v, ptr %x release, align 4
+  ret void
+}
+
+define void @store_atomic_vec2_bfloat(ptr %x, <2 x bfloat> %v) {
+; CHECK-SSE-O3-LABEL: store_atomic_vec2_bfloat:
+; CHECK-SSE-O3:       # %bb.0:
+; CHECK-SSE-O3-NEXT:    pextrw $0, %xmm0, %eax
+; CHECK-SSE-O3-NEXT:    psrld $16, %xmm0
+; CHECK-SSE-O3-NEXT:    pextrw $0, %xmm0, %ecx
+; CHECK-SSE-O3-NEXT:    shll $16, %ecx
+; CHECK-SSE-O3-NEXT:    movzwl %ax, %eax
+; CHECK-SSE-O3-NEXT:    orl %ecx, %eax
+; CHECK-SSE-O3-NEXT:    movl %eax, (%rdi)
+; CHECK-SSE-O3-NEXT:    retq
+;
+; CHECK-AVX-O3-LABEL: store_atomic_vec2_bfloat:
+; CHECK-AVX-O3:       # %bb.0:
+; CHECK-AVX-O3-NEXT:    vmovd %xmm0, %eax
+; CHECK-AVX-O3-NEXT:    movl %eax, (%rdi)
+; CHECK-AVX-O3-NEXT:    retq
+;
+; CHECK-SSE-O0-LABEL: store_atomic_vec2_bfloat:
+; CHECK-SSE-O0:       # %bb.0:
+; CHECK-SSE-O0-NEXT:    subq $24, %rsp
+; CHECK-SSE-O0-NEXT:    .cfi_def_cfa_offset 32
----------------
arsenm wrote:

Add nounwind to functions 

https://github.com/llvm/llvm-project/pull/197860