8b43c1be23
If we're loading a vector constant that is known to be zero in the upper elements, then attempt to shrink the constant and just scalar load the lower 32/64 bits. Always chose the vzload/broadcast with the smallest constant load, and prefer vzload over broadcasts for same bitwidth to avoid domain flips (mainly a AVX1 issue). Fixes #73783
178 lines
7.2 KiB
LLVM
178 lines
7.2 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
|
|
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefixes=SSE
|
|
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX1
|
|
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX2
|
|
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v4 | FileCheck %s --check-prefixes=AVX512
|
|
|
|
define i32 @PR63108() {
|
|
; SSE-LABEL: PR63108:
|
|
; SSE: # %bb.0: # %entry
|
|
; SSE-NEXT: xorl %eax, %eax
|
|
; SSE-NEXT: testb %al, %al
|
|
; SSE-NEXT: je .LBB0_2
|
|
; SSE-NEXT: # %bb.1:
|
|
; SSE-NEXT: movd {{.*#+}} xmm0 = [57339,0,0,0]
|
|
; SSE-NEXT: jmp .LBB0_5
|
|
; SSE-NEXT: .LBB0_2: # %vector.body.preheader
|
|
; SSE-NEXT: pxor %xmm0, %xmm0
|
|
; SSE-NEXT: movd {{.*#+}} xmm1 = [57339,0,0,0]
|
|
; SSE-NEXT: xorl %eax, %eax
|
|
; SSE-NEXT: .p2align 4, 0x90
|
|
; SSE-NEXT: .LBB0_3: # %vector.body
|
|
; SSE-NEXT: # =>This Inner Loop Header: Depth=1
|
|
; SSE-NEXT: movdqa %xmm1, %xmm2
|
|
; SSE-NEXT: testb %al, %al
|
|
; SSE-NEXT: pxor %xmm1, %xmm1
|
|
; SSE-NEXT: jne .LBB0_3
|
|
; SSE-NEXT: # %bb.4: # %middle.block
|
|
; SSE-NEXT: pxor %xmm2, %xmm0
|
|
; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; SSE-NEXT: pxor %xmm0, %xmm1
|
|
; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm1[1,1,1,1]
|
|
; SSE-NEXT: pxor %xmm1, %xmm2
|
|
; SSE-NEXT: movdqa %xmm2, %xmm0
|
|
; SSE-NEXT: psrld $16, %xmm0
|
|
; SSE-NEXT: pxor %xmm2, %xmm0
|
|
; SSE-NEXT: .LBB0_5: # %for.cond.cleanup
|
|
; SSE-NEXT: movd %xmm0, %eax
|
|
; SSE-NEXT: movsbl %al, %ecx
|
|
; SSE-NEXT: shrl $8, %eax
|
|
; SSE-NEXT: movsbl %al, %eax
|
|
; SSE-NEXT: addl %ecx, %eax
|
|
; SSE-NEXT: retq
|
|
;
|
|
; AVX1-LABEL: PR63108:
|
|
; AVX1: # %bb.0: # %entry
|
|
; AVX1-NEXT: xorl %eax, %eax
|
|
; AVX1-NEXT: testb %al, %al
|
|
; AVX1-NEXT: je .LBB0_2
|
|
; AVX1-NEXT: # %bb.1:
|
|
; AVX1-NEXT: vmovd {{.*#+}} xmm0 = [57339,0,0,0]
|
|
; AVX1-NEXT: jmp .LBB0_5
|
|
; AVX1-NEXT: .LBB0_2: # %vector.body.preheader
|
|
; AVX1-NEXT: vmovss {{.*#+}} xmm0 = [57339,0,0,0]
|
|
; AVX1-NEXT: xorl %eax, %eax
|
|
; AVX1-NEXT: .p2align 4, 0x90
|
|
; AVX1-NEXT: .LBB0_3: # %vector.body
|
|
; AVX1-NEXT: # =>This Inner Loop Header: Depth=1
|
|
; AVX1-NEXT: vmovaps %ymm0, %ymm1
|
|
; AVX1-NEXT: vxorps %xmm0, %xmm0, %xmm0
|
|
; AVX1-NEXT: testb %al, %al
|
|
; AVX1-NEXT: jne .LBB0_3
|
|
; AVX1-NEXT: # %bb.4: # %middle.block
|
|
; AVX1-NEXT: vxorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm0
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; AVX1-NEXT: vxorps %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vshufps {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; AVX1-NEXT: vxorps %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vshufps {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; AVX1-NEXT: vxorps %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpsrld $16, %xmm0, %xmm1
|
|
; AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: .LBB0_5: # %for.cond.cleanup
|
|
; AVX1-NEXT: vmovd %xmm0, %eax
|
|
; AVX1-NEXT: vpextrb $1, %xmm0, %ecx
|
|
; AVX1-NEXT: movsbl %al, %edx
|
|
; AVX1-NEXT: movsbl %cl, %eax
|
|
; AVX1-NEXT: addl %edx, %eax
|
|
; AVX1-NEXT: vzeroupper
|
|
; AVX1-NEXT: retq
|
|
;
|
|
; AVX2-LABEL: PR63108:
|
|
; AVX2: # %bb.0: # %entry
|
|
; AVX2-NEXT: xorl %eax, %eax
|
|
; AVX2-NEXT: testb %al, %al
|
|
; AVX2-NEXT: je .LBB0_2
|
|
; AVX2-NEXT: # %bb.1:
|
|
; AVX2-NEXT: vpbroadcastw {{.*#+}} xmm0 = [251,223,251,223,251,223,251,223,251,223,251,223,251,223,251,223]
|
|
; AVX2-NEXT: jmp .LBB0_5
|
|
; AVX2-NEXT: .LBB0_2: # %vector.body.preheader
|
|
; AVX2-NEXT: vmovd {{.*#+}} xmm0 = [57339,0,0,0]
|
|
; AVX2-NEXT: xorl %eax, %eax
|
|
; AVX2-NEXT: .p2align 4, 0x90
|
|
; AVX2-NEXT: .LBB0_3: # %vector.body
|
|
; AVX2-NEXT: # =>This Inner Loop Header: Depth=1
|
|
; AVX2-NEXT: vmovdqa %ymm0, %ymm1
|
|
; AVX2-NEXT: vpxor %xmm0, %xmm0, %xmm0
|
|
; AVX2-NEXT: testb %al, %al
|
|
; AVX2-NEXT: jne .LBB0_3
|
|
; AVX2-NEXT: # %bb.4: # %middle.block
|
|
; AVX2-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm0
|
|
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
; AVX2-NEXT: vpsrld $16, %xmm0, %xmm1
|
|
; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
; AVX2-NEXT: .LBB0_5: # %for.cond.cleanup
|
|
; AVX2-NEXT: vmovd %xmm0, %eax
|
|
; AVX2-NEXT: vpextrb $1, %xmm0, %ecx
|
|
; AVX2-NEXT: movsbl %al, %edx
|
|
; AVX2-NEXT: movsbl %cl, %eax
|
|
; AVX2-NEXT: addl %edx, %eax
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: retq
|
|
;
|
|
; AVX512-LABEL: PR63108:
|
|
; AVX512: # %bb.0: # %entry
|
|
; AVX512-NEXT: xorl %eax, %eax
|
|
; AVX512-NEXT: testb %al, %al
|
|
; AVX512-NEXT: je .LBB0_2
|
|
; AVX512-NEXT: # %bb.1:
|
|
; AVX512-NEXT: vpbroadcastw {{.*#+}} xmm0 = [251,223,251,223,251,223,251,223,251,223,251,223,251,223,251,223]
|
|
; AVX512-NEXT: jmp .LBB0_5
|
|
; AVX512-NEXT: .LBB0_2: # %vector.body.preheader
|
|
; AVX512-NEXT: vmovd {{.*#+}} xmm0 = [57339,0,0,0]
|
|
; AVX512-NEXT: xorl %eax, %eax
|
|
; AVX512-NEXT: .p2align 4, 0x90
|
|
; AVX512-NEXT: .LBB0_3: # %vector.body
|
|
; AVX512-NEXT: # =>This Inner Loop Header: Depth=1
|
|
; AVX512-NEXT: vmovdqa %ymm0, %ymm1
|
|
; AVX512-NEXT: vpxor %xmm0, %xmm0, %xmm0
|
|
; AVX512-NEXT: testb %al, %al
|
|
; AVX512-NEXT: jne .LBB0_3
|
|
; AVX512-NEXT: # %bb.4: # %middle.block
|
|
; AVX512-NEXT: vpxord {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm0
|
|
; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; AVX512-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; AVX512-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
; AVX512-NEXT: vpsrld $16, %xmm0, %xmm1
|
|
; AVX512-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
; AVX512-NEXT: .LBB0_5: # %for.cond.cleanup
|
|
; AVX512-NEXT: vmovd %xmm0, %eax
|
|
; AVX512-NEXT: vpextrb $1, %xmm0, %ecx
|
|
; AVX512-NEXT: movsbl %al, %edx
|
|
; AVX512-NEXT: movsbl %cl, %eax
|
|
; AVX512-NEXT: addl %edx, %eax
|
|
; AVX512-NEXT: vzeroupper
|
|
; AVX512-NEXT: retq
|
|
entry:
|
|
br i1 poison, label %for.cond.cleanup, label %vector.body
|
|
|
|
vector.body: ; preds = %vector.body, %entry
|
|
%vec.phi = phi <16 x i16> [ zeroinitializer, %vector.body ], [ <i16 -8197, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, %entry ]
|
|
%0 = icmp eq i32 poison, poison
|
|
br i1 %0, label %middle.block, label %vector.body
|
|
|
|
middle.block: ; preds = %vector.body
|
|
%bin.rdx22 = xor <16 x i16> %vec.phi, <i16 256, i16 256, i16 256, i16 256, i16 256, i16 256, i16 256, i16 256, i16 256, i16 256, i16 256, i16 256, i16 256, i16 256, i16 256, i16 256>
|
|
%1 = tail call i16 @llvm.vector.reduce.xor.v16i16(<16 x i16> %bin.rdx22)
|
|
%2 = bitcast i16 %1 to <2 x i8>
|
|
br label %for.cond.cleanup
|
|
|
|
for.cond.cleanup: ; preds = %middle.block, %entry
|
|
%id15142.0.lcssa = phi <2 x i8> [ <i8 -5, i8 -33>, %entry ], [ %2, %middle.block ]
|
|
%vecext = extractelement <2 x i8> %id15142.0.lcssa, i64 0
|
|
%vecext8 = extractelement <2 x i8> %id15142.0.lcssa, i64 1
|
|
%conv7 = sext i8 %vecext to i32
|
|
%conv9 = sext i8 %vecext8 to i32
|
|
%res = add i32 %conv7, %conv9
|
|
ret i32 %res
|
|
}
|
|
declare i16 @llvm.vector.reduce.xor.v16i16(<16 x i16>)
|