; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+ssse3 | FileCheck %s --check-prefixes=SSE3,SSE3-SLOW
; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+ssse3,fast-hops | FileCheck %s --check-prefixes=SSE3,SSE3-FAST
; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX-SLOW,AVX1-SLOW
; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx,fast-hops | FileCheck %s --check-prefixes=AVX,AVX-FAST,AVX1-FAST
; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX-SLOW,AVX2-SLOW
; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx2,fast-hops | FileCheck %s --check-prefixes=AVX,AVX-FAST,AVX2-FAST
; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx512vl | FileCheck %s --check-prefixes=AVX,AVX-SLOW,AVX512-SLOW
; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx512vl,fast-hops | FileCheck %s --check-prefixes=AVX,AVX-FAST,AVX512-FAST
; 128-bit vectors, 16/32-bit, add/sub
define i32 @extract_extract01_v4i32_add_i32(<4 x i32> %x) {
; SSE3-SLOW-LABEL: extract_extract01_v4i32_add_i32:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: movd %xmm0, %ecx
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract01_v4i32_add_i32:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddd %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract01_v4i32_add_i32:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vmovd %xmm0, %ecx
; AVX-SLOW-NEXT: vpextrd $1, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract01_v4i32_add_i32:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: retq
%x0 = extractelement <4 x i32> %x, i32 0
%x1 = extractelement <4 x i32> %x, i32 1
%x01 = add i32 %x0, %x1
ret i32 %x01
}
define i32 @extract_extract23_v4i32_add_i32(<4 x i32> %x) {
; SSE3-SLOW-LABEL: extract_extract23_v4i32_add_i32:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-SLOW-NEXT: movd %xmm1, %ecx
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm0[3,3,3,3]
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract23_v4i32_add_i32:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddd %xmm0, %xmm0
; SSE3-FAST-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract23_v4i32_add_i32:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vextractps $2, %xmm0, %ecx
; AVX-SLOW-NEXT: vextractps $3, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract23_v4i32_add_i32:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vpextrd $1, %xmm0, %eax
; AVX-FAST-NEXT: retq
%x0 = extractelement <4 x i32> %x, i32 2
%x1 = extractelement <4 x i32> %x, i32 3
%x01 = add i32 %x0, %x1
ret i32 %x01
}
define i32 @extract_extract01_v4i32_add_i32_commute(<4 x i32> %x) {
; SSE3-SLOW-LABEL: extract_extract01_v4i32_add_i32_commute:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: movd %xmm0, %ecx
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract01_v4i32_add_i32_commute:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddd %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract01_v4i32_add_i32_commute:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vmovd %xmm0, %ecx
; AVX-SLOW-NEXT: vpextrd $1, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract01_v4i32_add_i32_commute:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: retq
%x0 = extractelement <4 x i32> %x, i32 0
%x1 = extractelement <4 x i32> %x, i32 1
%x01 = add i32 %x1, %x0
ret i32 %x01
}
define i32 @extract_extract23_v4i32_add_i32_commute(<4 x i32> %x) {
; SSE3-SLOW-LABEL: extract_extract23_v4i32_add_i32_commute:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-SLOW-NEXT: movd %xmm1, %ecx
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm0[3,3,3,3]
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract23_v4i32_add_i32_commute:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddd %xmm0, %xmm0
; SSE3-FAST-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract23_v4i32_add_i32_commute:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vextractps $2, %xmm0, %ecx
; AVX-SLOW-NEXT: vextractps $3, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract23_v4i32_add_i32_commute:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vpextrd $1, %xmm0, %eax
; AVX-FAST-NEXT: retq
%x0 = extractelement <4 x i32> %x, i32 2
%x1 = extractelement <4 x i32> %x, i32 3
%x01 = add i32 %x1, %x0
ret i32 %x01
}
define i16 @extract_extract01_v8i16_add_i16(<8 x i16> %x) {
; SSE3-SLOW-LABEL: extract_extract01_v8i16_add_i16:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: movd %xmm0, %ecx
; SSE3-SLOW-NEXT: pextrw $1, %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract01_v8i16_add_i16:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddw %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract01_v8i16_add_i16:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vmovd %xmm0, %ecx
; AVX-SLOW-NEXT: vpextrw $1, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract01_v8i16_add_i16:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddw %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-FAST-NEXT: retq
%x0 = extractelement <8 x i16> %x, i32 0
%x1 = extractelement <8 x i16> %x, i32 1
%x01 = add i16 %x0, %x1
ret i16 %x01
}
define i16 @extract_extract45_v8i16_add_i16(<8 x i16> %x) {
; SSE3-SLOW-LABEL: extract_extract45_v8i16_add_i16:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: pextrw $4, %xmm0, %ecx
; SSE3-SLOW-NEXT: pextrw $5, %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract45_v8i16_add_i16:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddw %xmm0, %xmm0
; SSE3-FAST-NEXT: pextrw $2, %xmm0, %eax
; SSE3-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract45_v8i16_add_i16:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vpextrw $4, %xmm0, %ecx
; AVX-SLOW-NEXT: vpextrw $5, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract45_v8i16_add_i16:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddw %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vpextrw $2, %xmm0, %eax
; AVX-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-FAST-NEXT: retq
%x0 = extractelement <8 x i16> %x, i32 4
%x1 = extractelement <8 x i16> %x, i32 5
%x01 = add i16 %x0, %x1
ret i16 %x01
}
define i16 @extract_extract01_v8i16_add_i16_commute(<8 x i16> %x) {
; SSE3-SLOW-LABEL: extract_extract01_v8i16_add_i16_commute:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: movd %xmm0, %ecx
; SSE3-SLOW-NEXT: pextrw $1, %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract01_v8i16_add_i16_commute:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddw %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract01_v8i16_add_i16_commute:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vmovd %xmm0, %ecx
; AVX-SLOW-NEXT: vpextrw $1, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract01_v8i16_add_i16_commute:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddw %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-FAST-NEXT: retq
%x0 = extractelement <8 x i16> %x, i32 0
%x1 = extractelement <8 x i16> %x, i32 1
%x01 = add i16 %x1, %x0
ret i16 %x01
}
define i16 @extract_extract45_v8i16_add_i16_commute(<8 x i16> %x) {
; SSE3-SLOW-LABEL: extract_extract45_v8i16_add_i16_commute:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: pextrw $4, %xmm0, %ecx
; SSE3-SLOW-NEXT: pextrw $5, %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract45_v8i16_add_i16_commute:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddw %xmm0, %xmm0
; SSE3-FAST-NEXT: pextrw $2, %xmm0, %eax
; SSE3-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract45_v8i16_add_i16_commute:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vpextrw $4, %xmm0, %ecx
; AVX-SLOW-NEXT: vpextrw $5, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract45_v8i16_add_i16_commute:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddw %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vpextrw $2, %xmm0, %eax
; AVX-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-FAST-NEXT: retq
%x0 = extractelement <8 x i16> %x, i32 4
%x1 = extractelement <8 x i16> %x, i32 5
%x01 = add i16 %x1, %x0
ret i16 %x01
}
define i32 @extract_extract01_v4i32_sub_i32(<4 x i32> %x) {
; SSE3-SLOW-LABEL: extract_extract01_v4i32_sub_i32:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE3-SLOW-NEXT: movd %xmm0, %ecx
; SSE3-SLOW-NEXT: subl %ecx, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract01_v4i32_sub_i32:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phsubd %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract01_v4i32_sub_i32:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vmovd %xmm0, %eax
; AVX-SLOW-NEXT: vpextrd $1, %xmm0, %ecx
; AVX-SLOW-NEXT: subl %ecx, %eax
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract01_v4i32_sub_i32:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphsubd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: retq
%x0 = extractelement <4 x i32> %x, i32 0
%x1 = extractelement <4 x i32> %x, i32 1
%x01 = sub i32 %x0, %x1
ret i32 %x01
}
define i32 @extract_extract23_v4i32_sub_i32(<4 x i32> %x) {
; SSE3-SLOW-LABEL: extract_extract23_v4i32_sub_i32:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-SLOW-NEXT: movd %xmm1, %eax
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm0[3,3,3,3]
; SSE3-SLOW-NEXT: movd %xmm0, %ecx
; SSE3-SLOW-NEXT: subl %ecx, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract23_v4i32_sub_i32:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phsubd %xmm0, %xmm0
; SSE3-FAST-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract23_v4i32_sub_i32:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vextractps $2, %xmm0, %eax
; AVX-SLOW-NEXT: vextractps $3, %xmm0, %ecx
; AVX-SLOW-NEXT: subl %ecx, %eax
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract23_v4i32_sub_i32:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphsubd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vpextrd $1, %xmm0, %eax
; AVX-FAST-NEXT: retq
%x0 = extractelement <4 x i32> %x, i32 2
%x1 = extractelement <4 x i32> %x, i32 3
%x01 = sub i32 %x0, %x1
ret i32 %x01
}
define i32 @extract_extract01_v4i32_sub_i32_commute(<4 x i32> %x) {
; SSE3-LABEL: extract_extract01_v4i32_sub_i32_commute:
; SSE3: # %bb.0:
; SSE3-NEXT: movd %xmm0, %ecx
; SSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE3-NEXT: movd %xmm0, %eax
; SSE3-NEXT: subl %ecx, %eax
; SSE3-NEXT: retq
;
; AVX-LABEL: extract_extract01_v4i32_sub_i32_commute:
; AVX: # %bb.0:
; AVX-NEXT: vmovd %xmm0, %ecx
; AVX-NEXT: vpextrd $1, %xmm0, %eax
; AVX-NEXT: subl %ecx, %eax
; AVX-NEXT: retq
%x0 = extractelement <4 x i32> %x, i32 0
%x1 = extractelement <4 x i32> %x, i32 1
%x01 = sub i32 %x1, %x0
ret i32 %x01
}
define i32 @extract_extract23_v4i32_sub_i32_commute(<4 x i32> %x) {
; SSE3-LABEL: extract_extract23_v4i32_sub_i32_commute:
; SSE3: # %bb.0:
; SSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-NEXT: movd %xmm1, %ecx
; SSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[3,3,3,3]
; SSE3-NEXT: movd %xmm0, %eax
; SSE3-NEXT: subl %ecx, %eax
; SSE3-NEXT: retq
;
; AVX-LABEL: extract_extract23_v4i32_sub_i32_commute:
; AVX: # %bb.0:
; AVX-NEXT: vextractps $2, %xmm0, %ecx
; AVX-NEXT: vextractps $3, %xmm0, %eax
; AVX-NEXT: subl %ecx, %eax
; AVX-NEXT: retq
%x0 = extractelement <4 x i32> %x, i32 2
%x1 = extractelement <4 x i32> %x, i32 3
%x01 = sub i32 %x1, %x0
ret i32 %x01
}
define i16 @extract_extract01_v8i16_sub_i16(<8 x i16> %x) {
; SSE3-SLOW-LABEL: extract_extract01_v8i16_sub_i16:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: pextrw $1, %xmm0, %ecx
; SSE3-SLOW-NEXT: subl %ecx, %eax
; SSE3-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract01_v8i16_sub_i16:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phsubw %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract01_v8i16_sub_i16:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vmovd %xmm0, %eax
; AVX-SLOW-NEXT: vpextrw $1, %xmm0, %ecx
; AVX-SLOW-NEXT: subl %ecx, %eax
; AVX-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract01_v8i16_sub_i16:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphsubw %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-FAST-NEXT: retq
%x0 = extractelement <8 x i16> %x, i32 0
%x1 = extractelement <8 x i16> %x, i32 1
%x01 = sub i16 %x0, %x1
ret i16 %x01
}
define i16 @extract_extract23_v8i16_sub_i16(<8 x i16> %x) {
; SSE3-SLOW-LABEL: extract_extract23_v8i16_sub_i16:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: pextrw $2, %xmm0, %eax
; SSE3-SLOW-NEXT: pextrw $3, %xmm0, %ecx
; SSE3-SLOW-NEXT: subl %ecx, %eax
; SSE3-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract23_v8i16_sub_i16:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phsubw %xmm0, %xmm0
; SSE3-FAST-NEXT: pextrw $1, %xmm0, %eax
; SSE3-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract23_v8i16_sub_i16:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vpextrw $2, %xmm0, %eax
; AVX-SLOW-NEXT: vpextrw $3, %xmm0, %ecx
; AVX-SLOW-NEXT: subl %ecx, %eax
; AVX-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract23_v8i16_sub_i16:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphsubw %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vpextrw $1, %xmm0, %eax
; AVX-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-FAST-NEXT: retq
%x0 = extractelement <8 x i16> %x, i32 2
%x1 = extractelement <8 x i16> %x, i32 3
%x01 = sub i16 %x0, %x1
ret i16 %x01
}
define i16 @extract_extract01_v8i16_sub_i16_commute(<8 x i16> %x) {
; SSE3-LABEL: extract_extract01_v8i16_sub_i16_commute:
; SSE3: # %bb.0:
; SSE3-NEXT: movd %xmm0, %ecx
; SSE3-NEXT: pextrw $1, %xmm0, %eax
; SSE3-NEXT: subl %ecx, %eax
; SSE3-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-NEXT: retq
;
; AVX-LABEL: extract_extract01_v8i16_sub_i16_commute:
; AVX: # %bb.0:
; AVX-NEXT: vmovd %xmm0, %ecx
; AVX-NEXT: vpextrw $1, %xmm0, %eax
; AVX-NEXT: subl %ecx, %eax
; AVX-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-NEXT: retq
%x0 = extractelement <8 x i16> %x, i32 0
%x1 = extractelement <8 x i16> %x, i32 1
%x01 = sub i16 %x1, %x0
ret i16 %x01
}
define i16 @extract_extract23_v8i16_sub_i16_commute(<8 x i16> %x) {
; SSE3-LABEL: extract_extract23_v8i16_sub_i16_commute:
; SSE3: # %bb.0:
; SSE3-NEXT: pextrw $2, %xmm0, %ecx
; SSE3-NEXT: pextrw $3, %xmm0, %eax
; SSE3-NEXT: subl %ecx, %eax
; SSE3-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-NEXT: retq
;
; AVX-LABEL: extract_extract23_v8i16_sub_i16_commute:
; AVX: # %bb.0:
; AVX-NEXT: vpextrw $2, %xmm0, %ecx
; AVX-NEXT: vpextrw $3, %xmm0, %eax
; AVX-NEXT: subl %ecx, %eax
; AVX-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-NEXT: retq
%x0 = extractelement <8 x i16> %x, i32 2
%x1 = extractelement <8 x i16> %x, i32 3
%x01 = sub i16 %x1, %x0
ret i16 %x01
}
; 256-bit vectors, i32/i16, add/sub
define i32 @extract_extract01_v8i32_add_i32(<8 x i32> %x) {
; SSE3-SLOW-LABEL: extract_extract01_v8i32_add_i32:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: movd %xmm0, %ecx
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract01_v8i32_add_i32:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddd %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract01_v8i32_add_i32:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vmovd %xmm0, %ecx
; AVX-SLOW-NEXT: vpextrd $1, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract01_v8i32_add_i32:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: vzeroupper
; AVX-FAST-NEXT: retq
%x0 = extractelement <8 x i32> %x, i32 0
%x1 = extractelement <8 x i32> %x, i32 1
%x01 = add i32 %x0, %x1
ret i32 %x01
}
define i32 @extract_extract23_v8i32_add_i32(<8 x i32> %x) {
; SSE3-SLOW-LABEL: extract_extract23_v8i32_add_i32:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-SLOW-NEXT: movd %xmm1, %ecx
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm0[3,3,3,3]
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract23_v8i32_add_i32:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddd %xmm0, %xmm0
; SSE3-FAST-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract23_v8i32_add_i32:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vextractps $2, %xmm0, %ecx
; AVX-SLOW-NEXT: vextractps $3, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract23_v8i32_add_i32:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vpextrd $1, %xmm0, %eax
; AVX-FAST-NEXT: vzeroupper
; AVX-FAST-NEXT: retq
%x0 = extractelement <8 x i32> %x, i32 2
%x1 = extractelement <8 x i32> %x, i32 3
%x01 = add i32 %x0, %x1
ret i32 %x01
}
define i32 @extract_extract67_v8i32_add_i32(<8 x i32> %x) {
; SSE3-SLOW-LABEL: extract_extract67_v8i32_add_i32:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm1[2,3,2,3]
; SSE3-SLOW-NEXT: movd %xmm0, %ecx
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm1[3,3,3,3]
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract67_v8i32_add_i32:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddd %xmm1, %xmm1
; SSE3-FAST-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract67_v8i32_add_i32:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vextractf128 $1, %ymm0, %xmm0
; AVX-SLOW-NEXT: vextractps $2, %xmm0, %ecx
; AVX-SLOW-NEXT: vextractps $3, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX1-FAST-LABEL: extract_extract67_v8i32_add_i32:
; AVX1-FAST: # %bb.0:
; AVX1-FAST-NEXT: vextractf128 $1, %ymm0, %xmm0
; AVX1-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX1-FAST-NEXT: vpextrd $1, %xmm0, %eax
; AVX1-FAST-NEXT: vzeroupper
; AVX1-FAST-NEXT: retq
;
; AVX2-FAST-LABEL: extract_extract67_v8i32_add_i32:
; AVX2-FAST: # %bb.0:
; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm0
; AVX2-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX2-FAST-NEXT: vpextrd $1, %xmm0, %eax
; AVX2-FAST-NEXT: vzeroupper
; AVX2-FAST-NEXT: retq
;
; AVX512-FAST-LABEL: extract_extract67_v8i32_add_i32:
; AVX512-FAST: # %bb.0:
; AVX512-FAST-NEXT: vextracti128 $1, %ymm0, %xmm0
; AVX512-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX512-FAST-NEXT: vpextrd $1, %xmm0, %eax
; AVX512-FAST-NEXT: vzeroupper
; AVX512-FAST-NEXT: retq
%x0 = extractelement <8 x i32> %x, i32 6
%x1 = extractelement <8 x i32> %x, i32 7
%x01 = add i32 %x0, %x1
ret i32 %x01
}
define i32 @extract_extract01_v8i32_add_i32_commute(<8 x i32> %x) {
; SSE3-SLOW-LABEL: extract_extract01_v8i32_add_i32_commute:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: movd %xmm0, %ecx
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract01_v8i32_add_i32_commute:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddd %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract01_v8i32_add_i32_commute:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vmovd %xmm0, %ecx
; AVX-SLOW-NEXT: vpextrd $1, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract01_v8i32_add_i32_commute:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: vzeroupper
; AVX-FAST-NEXT: retq
%x0 = extractelement <8 x i32> %x, i32 0
%x1 = extractelement <8 x i32> %x, i32 1
%x01 = add i32 %x1, %x0
ret i32 %x01
}
define i32 @extract_extract23_v8i32_add_i32_commute(<8 x i32> %x) {
; SSE3-SLOW-LABEL: extract_extract23_v8i32_add_i32_commute:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-SLOW-NEXT: movd %xmm1, %ecx
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm0[3,3,3,3]
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract23_v8i32_add_i32_commute:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddd %xmm0, %xmm0
; SSE3-FAST-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract23_v8i32_add_i32_commute:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vextractps $2, %xmm0, %ecx
; AVX-SLOW-NEXT: vextractps $3, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract23_v8i32_add_i32_commute:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vpextrd $1, %xmm0, %eax
; AVX-FAST-NEXT: vzeroupper
; AVX-FAST-NEXT: retq
%x0 = extractelement <8 x i32> %x, i32 2
%x1 = extractelement <8 x i32> %x, i32 3
%x01 = add i32 %x1, %x0
ret i32 %x01
}
define i32 @extract_extract67_v8i32_add_i32_commute(<8 x i32> %x) {
; SSE3-SLOW-LABEL: extract_extract67_v8i32_add_i32_commute:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm1[2,3,2,3]
; SSE3-SLOW-NEXT: movd %xmm0, %ecx
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm1[3,3,3,3]
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract67_v8i32_add_i32_commute:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddd %xmm1, %xmm1
; SSE3-FAST-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract67_v8i32_add_i32_commute:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vextractf128 $1, %ymm0, %xmm0
; AVX-SLOW-NEXT: vextractps $2, %xmm0, %ecx
; AVX-SLOW-NEXT: vextractps $3, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX1-FAST-LABEL: extract_extract67_v8i32_add_i32_commute:
; AVX1-FAST: # %bb.0:
; AVX1-FAST-NEXT: vextractf128 $1, %ymm0, %xmm0
; AVX1-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX1-FAST-NEXT: vpextrd $1, %xmm0, %eax
; AVX1-FAST-NEXT: vzeroupper
; AVX1-FAST-NEXT: retq
;
; AVX2-FAST-LABEL: extract_extract67_v8i32_add_i32_commute:
; AVX2-FAST: # %bb.0:
; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm0
; AVX2-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX2-FAST-NEXT: vpextrd $1, %xmm0, %eax
; AVX2-FAST-NEXT: vzeroupper
; AVX2-FAST-NEXT: retq
;
; AVX512-FAST-LABEL: extract_extract67_v8i32_add_i32_commute:
; AVX512-FAST: # %bb.0:
; AVX512-FAST-NEXT: vextracti128 $1, %ymm0, %xmm0
; AVX512-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX512-FAST-NEXT: vpextrd $1, %xmm0, %eax
; AVX512-FAST-NEXT: vzeroupper
; AVX512-FAST-NEXT: retq
%x0 = extractelement <8 x i32> %x, i32 6
%x1 = extractelement <8 x i32> %x, i32 7
%x01 = add i32 %x1, %x0
ret i32 %x01
}
define i16 @extract_extract01_v16i16_add_i16(<16 x i16> %x) {
; SSE3-SLOW-LABEL: extract_extract01_v16i16_add_i16:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: movd %xmm0, %ecx
; SSE3-SLOW-NEXT: pextrw $1, %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract01_v16i16_add_i16:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddw %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract01_v16i16_add_i16:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vmovd %xmm0, %ecx
; AVX-SLOW-NEXT: vpextrw $1, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract01_v16i16_add_i16:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddw %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-FAST-NEXT: vzeroupper
; AVX-FAST-NEXT: retq
%x0 = extractelement <16 x i16> %x, i32 0
%x1 = extractelement <16 x i16> %x, i32 1
%x01 = add i16 %x0, %x1
ret i16 %x01
}
define i16 @extract_extract23_v16i16_add_i16(<16 x i16> %x) {
; SSE3-SLOW-LABEL: extract_extract23_v16i16_add_i16:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: pextrw $2, %xmm0, %ecx
; SSE3-SLOW-NEXT: pextrw $3, %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract23_v16i16_add_i16:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddw %xmm0, %xmm0
; SSE3-FAST-NEXT: pextrw $1, %xmm0, %eax
; SSE3-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract23_v16i16_add_i16:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vpextrw $2, %xmm0, %ecx
; AVX-SLOW-NEXT: vpextrw $3, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract23_v16i16_add_i16:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddw %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vpextrw $1, %xmm0, %eax
; AVX-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-FAST-NEXT: vzeroupper
; AVX-FAST-NEXT: retq
%x0 = extractelement <16 x i16> %x, i32 2
%x1 = extractelement <16 x i16> %x, i32 3
%x01 = add i16 %x0, %x1
ret i16 %x01
}
define i16 @extract_extract89_v16i16_add_i16(<16 x i16> %x) {
; SSE3-SLOW-LABEL: extract_extract89_v16i16_add_i16:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: movd %xmm1, %ecx
; SSE3-SLOW-NEXT: pextrw $1, %xmm1, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract89_v16i16_add_i16:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddw %xmm1, %xmm1
; SSE3-FAST-NEXT: movd %xmm1, %eax
; SSE3-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-FAST-NEXT: retq
;
; AVX1-SLOW-LABEL: extract_extract89_v16i16_add_i16:
; AVX1-SLOW: # %bb.0:
; AVX1-SLOW-NEXT: vextractf128 $1, %ymm0, %xmm0
; AVX1-SLOW-NEXT: vmovd %xmm0, %ecx
; AVX1-SLOW-NEXT: vpextrw $1, %xmm0, %eax
; AVX1-SLOW-NEXT: addl %ecx, %eax
; AVX1-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; AVX1-SLOW-NEXT: vzeroupper
; AVX1-SLOW-NEXT: retq
;
; AVX1-FAST-LABEL: extract_extract89_v16i16_add_i16:
; AVX1-FAST: # %bb.0:
; AVX1-FAST-NEXT: vextractf128 $1, %ymm0, %xmm0
; AVX1-FAST-NEXT: vphaddw %xmm0, %xmm0, %xmm0
; AVX1-FAST-NEXT: vmovd %xmm0, %eax
; AVX1-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; AVX1-FAST-NEXT: vzeroupper
; AVX1-FAST-NEXT: retq
;
; AVX2-SLOW-LABEL: extract_extract89_v16i16_add_i16:
; AVX2-SLOW: # %bb.0:
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm0
; AVX2-SLOW-NEXT: vmovd %xmm0, %ecx
; AVX2-SLOW-NEXT: vpextrw $1, %xmm0, %eax
; AVX2-SLOW-NEXT: addl %ecx, %eax
; AVX2-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; AVX2-SLOW-NEXT: vzeroupper
; AVX2-SLOW-NEXT: retq
;
; AVX2-FAST-LABEL: extract_extract89_v16i16_add_i16:
; AVX2-FAST: # %bb.0:
; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm0
; AVX2-FAST-NEXT: vphaddw %xmm0, %xmm0, %xmm0
; AVX2-FAST-NEXT: vmovd %xmm0, %eax
; AVX2-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; AVX2-FAST-NEXT: vzeroupper
; AVX2-FAST-NEXT: retq
;
; AVX512-SLOW-LABEL: extract_extract89_v16i16_add_i16:
; AVX512-SLOW: # %bb.0:
; AVX512-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm0
; AVX512-SLOW-NEXT: vmovd %xmm0, %ecx
; AVX512-SLOW-NEXT: vpextrw $1, %xmm0, %eax
; AVX512-SLOW-NEXT: addl %ecx, %eax
; AVX512-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; AVX512-SLOW-NEXT: vzeroupper
; AVX512-SLOW-NEXT: retq
;
; AVX512-FAST-LABEL: extract_extract89_v16i16_add_i16:
; AVX512-FAST: # %bb.0:
; AVX512-FAST-NEXT: vextracti128 $1, %ymm0, %xmm0
; AVX512-FAST-NEXT: vphaddw %xmm0, %xmm0, %xmm0
; AVX512-FAST-NEXT: vmovd %xmm0, %eax
; AVX512-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; AVX512-FAST-NEXT: vzeroupper
; AVX512-FAST-NEXT: retq
%x0 = extractelement <16 x i16> %x, i32 8
%x1 = extractelement <16 x i16> %x, i32 9
%x01 = add i16 %x0, %x1
ret i16 %x01
}
define i16 @extract_extract01_v16i16_add_i16_commute(<16 x i16> %x) {
; SSE3-SLOW-LABEL: extract_extract01_v16i16_add_i16_commute:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: movd %xmm0, %ecx
; SSE3-SLOW-NEXT: pextrw $1, %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract01_v16i16_add_i16_commute:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddw %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract01_v16i16_add_i16_commute:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vmovd %xmm0, %ecx
; AVX-SLOW-NEXT: vpextrw $1, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract01_v16i16_add_i16_commute:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddw %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-FAST-NEXT: vzeroupper
; AVX-FAST-NEXT: retq
%x0 = extractelement <16 x i16> %x, i32 0
%x1 = extractelement <16 x i16> %x, i32 1
%x01 = add i16 %x1, %x0
ret i16 %x01
}
define i16 @extract_extract45_v16i16_add_i16_commute(<16 x i16> %x) {
; SSE3-SLOW-LABEL: extract_extract45_v16i16_add_i16_commute:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: pextrw $4, %xmm0, %ecx
; SSE3-SLOW-NEXT: pextrw $5, %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract45_v16i16_add_i16_commute:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddw %xmm0, %xmm0
; SSE3-FAST-NEXT: pextrw $2, %xmm0, %eax
; SSE3-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract45_v16i16_add_i16_commute:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vpextrw $4, %xmm0, %ecx
; AVX-SLOW-NEXT: vpextrw $5, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract45_v16i16_add_i16_commute:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddw %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vpextrw $2, %xmm0, %eax
; AVX-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-FAST-NEXT: vzeroupper
; AVX-FAST-NEXT: retq
%x0 = extractelement <16 x i16> %x, i32 4
%x1 = extractelement <16 x i16> %x, i32 5
%x01 = add i16 %x1, %x0
ret i16 %x01
}
define i16 @extract_extract89_v16i16_add_i16_commute(<16 x i16> %x) {
; SSE3-SLOW-LABEL: extract_extract89_v16i16_add_i16_commute:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: movd %xmm1, %ecx
; SSE3-SLOW-NEXT: pextrw $1, %xmm1, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract89_v16i16_add_i16_commute:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddw %xmm1, %xmm1
; SSE3-FAST-NEXT: movd %xmm1, %eax
; SSE3-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-FAST-NEXT: retq
;
; AVX1-SLOW-LABEL: extract_extract89_v16i16_add_i16_commute:
; AVX1-SLOW: # %bb.0:
; AVX1-SLOW-NEXT: vextractf128 $1, %ymm0, %xmm0
; AVX1-SLOW-NEXT: vmovd %xmm0, %ecx
; AVX1-SLOW-NEXT: vpextrw $1, %xmm0, %eax
; AVX1-SLOW-NEXT: addl %ecx, %eax
; AVX1-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; AVX1-SLOW-NEXT: vzeroupper
; AVX1-SLOW-NEXT: retq
;
; AVX1-FAST-LABEL: extract_extract89_v16i16_add_i16_commute:
; AVX1-FAST: # %bb.0:
; AVX1-FAST-NEXT: vextractf128 $1, %ymm0, %xmm0
; AVX1-FAST-NEXT: vphaddw %xmm0, %xmm0, %xmm0
; AVX1-FAST-NEXT: vmovd %xmm0, %eax
; AVX1-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; AVX1-FAST-NEXT: vzeroupper
; AVX1-FAST-NEXT: retq
;
; AVX2-SLOW-LABEL: extract_extract89_v16i16_add_i16_commute:
; AVX2-SLOW: # %bb.0:
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm0
; AVX2-SLOW-NEXT: vmovd %xmm0, %ecx
; AVX2-SLOW-NEXT: vpextrw $1, %xmm0, %eax
; AVX2-SLOW-NEXT: addl %ecx, %eax
; AVX2-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; AVX2-SLOW-NEXT: vzeroupper
; AVX2-SLOW-NEXT: retq
;
; AVX2-FAST-LABEL: extract_extract89_v16i16_add_i16_commute:
; AVX2-FAST: # %bb.0:
; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm0
; AVX2-FAST-NEXT: vphaddw %xmm0, %xmm0, %xmm0
; AVX2-FAST-NEXT: vmovd %xmm0, %eax
; AVX2-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; AVX2-FAST-NEXT: vzeroupper
; AVX2-FAST-NEXT: retq
;
; AVX512-SLOW-LABEL: extract_extract89_v16i16_add_i16_commute:
; AVX512-SLOW: # %bb.0:
; AVX512-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm0
; AVX512-SLOW-NEXT: vmovd %xmm0, %ecx
; AVX512-SLOW-NEXT: vpextrw $1, %xmm0, %eax
; AVX512-SLOW-NEXT: addl %ecx, %eax
; AVX512-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; AVX512-SLOW-NEXT: vzeroupper
; AVX512-SLOW-NEXT: retq
;
; AVX512-FAST-LABEL: extract_extract89_v16i16_add_i16_commute:
; AVX512-FAST: # %bb.0:
; AVX512-FAST-NEXT: vextracti128 $1, %ymm0, %xmm0
; AVX512-FAST-NEXT: vphaddw %xmm0, %xmm0, %xmm0
; AVX512-FAST-NEXT: vmovd %xmm0, %eax
; AVX512-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; AVX512-FAST-NEXT: vzeroupper
; AVX512-FAST-NEXT: retq
%x0 = extractelement <16 x i16> %x, i32 8
%x1 = extractelement <16 x i16> %x, i32 9
%x01 = add i16 %x1, %x0
ret i16 %x01
}
define i32 @extract_extract01_v8i32_sub_i32(<8 x i32> %x) {
; SSE3-SLOW-LABEL: extract_extract01_v8i32_sub_i32:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE3-SLOW-NEXT: movd %xmm0, %ecx
; SSE3-SLOW-NEXT: subl %ecx, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract01_v8i32_sub_i32:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phsubd %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract01_v8i32_sub_i32:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vmovd %xmm0, %eax
; AVX-SLOW-NEXT: vpextrd $1, %xmm0, %ecx
; AVX-SLOW-NEXT: subl %ecx, %eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract01_v8i32_sub_i32:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphsubd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: vzeroupper
; AVX-FAST-NEXT: retq
%x0 = extractelement <8 x i32> %x, i32 0
%x1 = extractelement <8 x i32> %x, i32 1
%x01 = sub i32 %x0, %x1
ret i32 %x01
}
define i32 @extract_extract23_v8i32_sub_i32(<8 x i32> %x) {
; SSE3-SLOW-LABEL: extract_extract23_v8i32_sub_i32:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-SLOW-NEXT: movd %xmm1, %eax
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm0[3,3,3,3]
; SSE3-SLOW-NEXT: movd %xmm0, %ecx
; SSE3-SLOW-NEXT: subl %ecx, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract23_v8i32_sub_i32:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phsubd %xmm0, %xmm0
; SSE3-FAST-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract23_v8i32_sub_i32:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vextractps $2, %xmm0, %eax
; AVX-SLOW-NEXT: vextractps $3, %xmm0, %ecx
; AVX-SLOW-NEXT: subl %ecx, %eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract23_v8i32_sub_i32:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphsubd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vpextrd $1, %xmm0, %eax
; AVX-FAST-NEXT: vzeroupper
; AVX-FAST-NEXT: retq
%x0 = extractelement <8 x i32> %x, i32 2
%x1 = extractelement <8 x i32> %x, i32 3
%x01 = sub i32 %x0, %x1
ret i32 %x01
}
define i32 @extract_extract67_v8i32_sub_i32(<8 x i32> %x) {
; SSE3-SLOW-LABEL: extract_extract67_v8i32_sub_i32:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm1[2,3,2,3]
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm1[3,3,3,3]
; SSE3-SLOW-NEXT: movd %xmm0, %ecx
; SSE3-SLOW-NEXT: subl %ecx, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract67_v8i32_sub_i32:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phsubd %xmm1, %xmm1
; SSE3-FAST-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract67_v8i32_sub_i32:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vextractf128 $1, %ymm0, %xmm0
; AVX-SLOW-NEXT: vextractps $2, %xmm0, %eax
; AVX-SLOW-NEXT: vextractps $3, %xmm0, %ecx
; AVX-SLOW-NEXT: subl %ecx, %eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX1-FAST-LABEL: extract_extract67_v8i32_sub_i32:
; AVX1-FAST: # %bb.0:
; AVX1-FAST-NEXT: vextractf128 $1, %ymm0, %xmm0
; AVX1-FAST-NEXT: vphsubd %xmm0, %xmm0, %xmm0
; AVX1-FAST-NEXT: vpextrd $1, %xmm0, %eax
; AVX1-FAST-NEXT: vzeroupper
; AVX1-FAST-NEXT: retq
;
; AVX2-FAST-LABEL: extract_extract67_v8i32_sub_i32:
; AVX2-FAST: # %bb.0:
; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm0
; AVX2-FAST-NEXT: vphsubd %xmm0, %xmm0, %xmm0
; AVX2-FAST-NEXT: vpextrd $1, %xmm0, %eax
; AVX2-FAST-NEXT: vzeroupper
; AVX2-FAST-NEXT: retq
;
; AVX512-FAST-LABEL: extract_extract67_v8i32_sub_i32:
; AVX512-FAST: # %bb.0:
; AVX512-FAST-NEXT: vextracti128 $1, %ymm0, %xmm0
; AVX512-FAST-NEXT: vphsubd %xmm0, %xmm0, %xmm0
; AVX512-FAST-NEXT: vpextrd $1, %xmm0, %eax
; AVX512-FAST-NEXT: vzeroupper
; AVX512-FAST-NEXT: retq
%x0 = extractelement <8 x i32> %x, i32 6
%x1 = extractelement <8 x i32> %x, i32 7
%x01 = sub i32 %x0, %x1
ret i32 %x01
}
; Negative test...or get hoppy and negate?
define i32 @extract_extract01_v8i32_sub_i32_commute(<8 x i32> %x) {
; SSE3-LABEL: extract_extract01_v8i32_sub_i32_commute:
; SSE3: # %bb.0:
; SSE3-NEXT: movd %xmm0, %ecx
; SSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE3-NEXT: movd %xmm0, %eax
; SSE3-NEXT: subl %ecx, %eax
; SSE3-NEXT: retq
;
; AVX-LABEL: extract_extract01_v8i32_sub_i32_commute:
; AVX: # %bb.0:
; AVX-NEXT: vmovd %xmm0, %ecx
; AVX-NEXT: vpextrd $1, %xmm0, %eax
; AVX-NEXT: subl %ecx, %eax
; AVX-NEXT: vzeroupper
; AVX-NEXT: retq
%x0 = extractelement <8 x i32> %x, i32 0
%x1 = extractelement <8 x i32> %x, i32 1
%x01 = sub i32 %x1, %x0
ret i32 %x01
}
define i16 @extract_extract01_v16i16_sub_i16(<16 x i16> %x) {
; SSE3-SLOW-LABEL: extract_extract01_v16i16_sub_i16:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: pextrw $1, %xmm0, %ecx
; SSE3-SLOW-NEXT: subl %ecx, %eax
; SSE3-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract01_v16i16_sub_i16:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phsubw %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract01_v16i16_sub_i16:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vmovd %xmm0, %eax
; AVX-SLOW-NEXT: vpextrw $1, %xmm0, %ecx
; AVX-SLOW-NEXT: subl %ecx, %eax
; AVX-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract01_v16i16_sub_i16:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphsubw %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-FAST-NEXT: vzeroupper
; AVX-FAST-NEXT: retq
%x0 = extractelement <16 x i16> %x, i32 0
%x1 = extractelement <16 x i16> %x, i32 1
%x01 = sub i16 %x0, %x1
ret i16 %x01
}
; Negative test...or get hoppy and negate?
define i16 @extract_extract01_v16i16_sub_i16_commute(<16 x i16> %x) {
; SSE3-LABEL: extract_extract01_v16i16_sub_i16_commute:
; SSE3: # %bb.0:
; SSE3-NEXT: movd %xmm0, %ecx
; SSE3-NEXT: pextrw $1, %xmm0, %eax
; SSE3-NEXT: subl %ecx, %eax
; SSE3-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-NEXT: retq
;
; AVX-LABEL: extract_extract01_v16i16_sub_i16_commute:
; AVX: # %bb.0:
; AVX-NEXT: vmovd %xmm0, %ecx
; AVX-NEXT: vpextrw $1, %xmm0, %eax
; AVX-NEXT: subl %ecx, %eax
; AVX-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-NEXT: vzeroupper
; AVX-NEXT: retq
%x0 = extractelement <16 x i16> %x, i32 0
%x1 = extractelement <16 x i16> %x, i32 1
%x01 = sub i16 %x1, %x0
ret i16 %x01
}
; 512-bit vectors, i32/i16, add/sub
define i32 @extract_extract01_v16i32_add_i32(<16 x i32> %x) {
; SSE3-SLOW-LABEL: extract_extract01_v16i32_add_i32:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: movd %xmm0, %ecx
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract01_v16i32_add_i32:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddd %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract01_v16i32_add_i32:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vmovd %xmm0, %ecx
; AVX-SLOW-NEXT: vpextrd $1, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract01_v16i32_add_i32:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: vzeroupper
; AVX-FAST-NEXT: retq
%x0 = extractelement <16 x i32> %x, i32 0
%x1 = extractelement <16 x i32> %x, i32 1
%x01 = add i32 %x0, %x1
ret i32 %x01
}
define i32 @extract_extract01_v16i32_add_i32_commute(<16 x i32> %x) {
; SSE3-SLOW-LABEL: extract_extract01_v16i32_add_i32_commute:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: movd %xmm0, %ecx
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract01_v16i32_add_i32_commute:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddd %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract01_v16i32_add_i32_commute:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vmovd %xmm0, %ecx
; AVX-SLOW-NEXT: vpextrd $1, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract01_v16i32_add_i32_commute:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: vzeroupper
; AVX-FAST-NEXT: retq
%x0 = extractelement <16 x i32> %x, i32 0
%x1 = extractelement <16 x i32> %x, i32 1
%x01 = add i32 %x1, %x0
ret i32 %x01
}
define i16 @extract_extract01_v32i16_add_i16(<32 x i16> %x) {
; SSE3-SLOW-LABEL: extract_extract01_v32i16_add_i16:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: movd %xmm0, %ecx
; SSE3-SLOW-NEXT: pextrw $1, %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract01_v32i16_add_i16:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddw %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract01_v32i16_add_i16:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vmovd %xmm0, %ecx
; AVX-SLOW-NEXT: vpextrw $1, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract01_v32i16_add_i16:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddw %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-FAST-NEXT: vzeroupper
; AVX-FAST-NEXT: retq
%x0 = extractelement <32 x i16> %x, i32 0
%x1 = extractelement <32 x i16> %x, i32 1
%x01 = add i16 %x0, %x1
ret i16 %x01
}
define i16 @extract_extract01_v32i16_add_i16_commute(<32 x i16> %x) {
; SSE3-SLOW-LABEL: extract_extract01_v32i16_add_i16_commute:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: movd %xmm0, %ecx
; SSE3-SLOW-NEXT: pextrw $1, %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract01_v32i16_add_i16_commute:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddw %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract01_v32i16_add_i16_commute:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vmovd %xmm0, %ecx
; AVX-SLOW-NEXT: vpextrw $1, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract01_v32i16_add_i16_commute:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddw %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-FAST-NEXT: vzeroupper
; AVX-FAST-NEXT: retq
%x0 = extractelement <32 x i16> %x, i32 0
%x1 = extractelement <32 x i16> %x, i32 1
%x01 = add i16 %x1, %x0
ret i16 %x01
}
define i32 @extract_extract01_v16i32_sub_i32(<16 x i32> %x) {
; SSE3-SLOW-LABEL: extract_extract01_v16i32_sub_i32:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE3-SLOW-NEXT: movd %xmm0, %ecx
; SSE3-SLOW-NEXT: subl %ecx, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract01_v16i32_sub_i32:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phsubd %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract01_v16i32_sub_i32:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vmovd %xmm0, %eax
; AVX-SLOW-NEXT: vpextrd $1, %xmm0, %ecx
; AVX-SLOW-NEXT: subl %ecx, %eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract01_v16i32_sub_i32:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphsubd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: vzeroupper
; AVX-FAST-NEXT: retq
%x0 = extractelement <16 x i32> %x, i32 0
%x1 = extractelement <16 x i32> %x, i32 1
%x01 = sub i32 %x0, %x1
ret i32 %x01
}
define i32 @extract_extract01_v16i32_sub_i32_commute(<16 x i32> %x) {
; SSE3-LABEL: extract_extract01_v16i32_sub_i32_commute:
; SSE3: # %bb.0:
; SSE3-NEXT: movd %xmm0, %ecx
; SSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE3-NEXT: movd %xmm0, %eax
; SSE3-NEXT: subl %ecx, %eax
; SSE3-NEXT: retq
;
; AVX-LABEL: extract_extract01_v16i32_sub_i32_commute:
; AVX: # %bb.0:
; AVX-NEXT: vmovd %xmm0, %ecx
; AVX-NEXT: vpextrd $1, %xmm0, %eax
; AVX-NEXT: subl %ecx, %eax
; AVX-NEXT: vzeroupper
; AVX-NEXT: retq
%x0 = extractelement <16 x i32> %x, i32 0
%x1 = extractelement <16 x i32> %x, i32 1
%x01 = sub i32 %x1, %x0
ret i32 %x01
}
define i16 @extract_extract01_v32i16_sub_i16(<32 x i16> %x) {
; SSE3-SLOW-LABEL: extract_extract01_v32i16_sub_i16:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: pextrw $1, %xmm0, %ecx
; SSE3-SLOW-NEXT: subl %ecx, %eax
; SSE3-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract01_v32i16_sub_i16:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phsubw %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract01_v32i16_sub_i16:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vmovd %xmm0, %eax
; AVX-SLOW-NEXT: vpextrw $1, %xmm0, %ecx
; AVX-SLOW-NEXT: subl %ecx, %eax
; AVX-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract01_v32i16_sub_i16:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphsubw %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-FAST-NEXT: vzeroupper
; AVX-FAST-NEXT: retq
%x0 = extractelement <32 x i16> %x, i32 0
%x1 = extractelement <32 x i16> %x, i32 1
%x01 = sub i16 %x0, %x1
ret i16 %x01
}
define i16 @extract_extract01_v32i16_sub_i16_commute(<32 x i16> %x) {
; SSE3-LABEL: extract_extract01_v32i16_sub_i16_commute:
; SSE3: # %bb.0:
; SSE3-NEXT: movd %xmm0, %ecx
; SSE3-NEXT: pextrw $1, %xmm0, %eax
; SSE3-NEXT: subl %ecx, %eax
; SSE3-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-NEXT: retq
;
; AVX-LABEL: extract_extract01_v32i16_sub_i16_commute:
; AVX: # %bb.0:
; AVX-NEXT: vmovd %xmm0, %ecx
; AVX-NEXT: vpextrw $1, %xmm0, %eax
; AVX-NEXT: subl %ecx, %eax
; AVX-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-NEXT: vzeroupper
; AVX-NEXT: retq
%x0 = extractelement <32 x i16> %x, i32 0
%x1 = extractelement <32 x i16> %x, i32 1
%x01 = sub i16 %x1, %x0
ret i16 %x01
}
; Check output when 1 or both extracts have extra uses.
define i32 @extract_extract01_v4i32_add_i32_uses1(<4 x i32> %x, ptr %p) {
; SSE3-SLOW-LABEL: extract_extract01_v4i32_add_i32_uses1:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: movd %xmm0, %ecx
; SSE3-SLOW-NEXT: movd %xmm0, (%rdi)
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract01_v4i32_add_i32_uses1:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: movd %xmm0, (%rdi)
; SSE3-FAST-NEXT: phaddd %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract01_v4i32_add_i32_uses1:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vmovd %xmm0, %ecx
; AVX-SLOW-NEXT: vmovd %xmm0, (%rdi)
; AVX-SLOW-NEXT: vpextrd $1, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract01_v4i32_add_i32_uses1:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vmovd %xmm0, (%rdi)
; AVX-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: retq
%x0 = extractelement <4 x i32> %x, i32 0
store i32 %x0, ptr %p
%x1 = extractelement <4 x i32> %x, i32 1
%x01 = add i32 %x0, %x1
ret i32 %x01
}
define i32 @extract_extract01_v4i32_add_i32_uses2(<4 x i32> %x, ptr %p) {
; SSE3-SLOW-LABEL: extract_extract01_v4i32_add_i32_uses2:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: movd %xmm0, %ecx
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: addl %ecx, %eax
; SSE3-SLOW-NEXT: movd %xmm0, (%rdi)
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: extract_extract01_v4i32_add_i32_uses2:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
; SSE3-FAST-NEXT: movd %xmm1, (%rdi)
; SSE3-FAST-NEXT: phaddd %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: extract_extract01_v4i32_add_i32_uses2:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vmovd %xmm0, %ecx
; AVX-SLOW-NEXT: vpextrd $1, %xmm0, %eax
; AVX-SLOW-NEXT: addl %ecx, %eax
; AVX-SLOW-NEXT: vpextrd $1, %xmm0, (%rdi)
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: extract_extract01_v4i32_add_i32_uses2:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vpextrd $1, %xmm0, (%rdi)
; AVX-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: retq
%x0 = extractelement <4 x i32> %x, i32 0
%x1 = extractelement <4 x i32> %x, i32 1
store i32 %x1, ptr %p
%x01 = add i32 %x0, %x1
ret i32 %x01
}
define i32 @extract_extract01_v4i32_add_i32_uses3(<4 x i32> %x, ptr %p1, ptr %p2) {
; SSE3-LABEL: extract_extract01_v4i32_add_i32_uses3:
; SSE3: # %bb.0:
; SSE3-NEXT: movd %xmm0, %ecx
; SSE3-NEXT: movd %xmm0, (%rdi)
; SSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE3-NEXT: movd %xmm0, %eax
; SSE3-NEXT: addl %ecx, %eax
; SSE3-NEXT: movd %xmm0, (%rsi)
; SSE3-NEXT: retq
;
; AVX-LABEL: extract_extract01_v4i32_add_i32_uses3:
; AVX: # %bb.0:
; AVX-NEXT: vmovd %xmm0, %ecx
; AVX-NEXT: vmovd %xmm0, (%rdi)
; AVX-NEXT: vpextrd $1, %xmm0, %eax
; AVX-NEXT: addl %ecx, %eax
; AVX-NEXT: vpextrd $1, %xmm0, (%rsi)
; AVX-NEXT: retq
%x0 = extractelement <4 x i32> %x, i32 0
store i32 %x0, ptr %p1
%x1 = extractelement <4 x i32> %x, i32 1
store i32 %x1, ptr %p2
%x01 = add i32 %x0, %x1
ret i32 %x01
}
; PR33758: https://bugs.llvm.org/show_bug.cgi?id=33758
define i32 @partial_reduction_add_v8i32(<8 x i32> %x) {
; SSE3-SLOW-LABEL: partial_reduction_add_v8i32:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-SLOW-NEXT: paddd %xmm0, %xmm1
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
; SSE3-SLOW-NEXT: paddd %xmm1, %xmm0
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: partial_reduction_add_v8i32:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-FAST-NEXT: paddd %xmm0, %xmm1
; SSE3-FAST-NEXT: phaddd %xmm1, %xmm1
; SSE3-FAST-NEXT: movd %xmm1, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: partial_reduction_add_v8i32:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; AVX-SLOW-NEXT: vpaddd %xmm1, %xmm0, %xmm0
; AVX-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
; AVX-SLOW-NEXT: vpaddd %xmm1, %xmm0, %xmm0
; AVX-SLOW-NEXT: vmovd %xmm0, %eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: partial_reduction_add_v8i32:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: vzeroupper
; AVX-FAST-NEXT: retq
%x23 = shufflevector <8 x i32> %x, <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%x0213 = add <8 x i32> %x, %x23
%x13 = shufflevector <8 x i32> %x0213, <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%x0123 = add <8 x i32> %x0213, %x13
%r = extractelement <8 x i32> %x0123, i32 0
ret i32 %r
}
define i32 @partial_reduction_add_v16i32(<16 x i32> %x) {
; SSE3-SLOW-LABEL: partial_reduction_add_v16i32:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-SLOW-NEXT: paddd %xmm0, %xmm1
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
; SSE3-SLOW-NEXT: paddd %xmm1, %xmm0
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: partial_reduction_add_v16i32:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-FAST-NEXT: paddd %xmm0, %xmm1
; SSE3-FAST-NEXT: phaddd %xmm1, %xmm1
; SSE3-FAST-NEXT: movd %xmm1, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: partial_reduction_add_v16i32:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; AVX-SLOW-NEXT: vpaddd %xmm1, %xmm0, %xmm0
; AVX-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
; AVX-SLOW-NEXT: vpaddd %xmm1, %xmm0, %xmm0
; AVX-SLOW-NEXT: vmovd %xmm0, %eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: partial_reduction_add_v16i32:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: vzeroupper
; AVX-FAST-NEXT: retq
%x23 = shufflevector <16 x i32> %x, <16 x i32> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%x0213 = add <16 x i32> %x, %x23
%x13 = shufflevector <16 x i32> %x0213, <16 x i32> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%x0123 = add <16 x i32> %x0213, %x13
%r = extractelement <16 x i32> %x0123, i32 0
ret i32 %r
}
define i32 @partial_reduction_sub_v8i32(<8 x i32> %x) {
; SSE3-SLOW-LABEL: partial_reduction_sub_v8i32:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-SLOW-NEXT: psubd %xmm1, %xmm0
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
; SSE3-SLOW-NEXT: psubd %xmm1, %xmm0
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: partial_reduction_sub_v8i32:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-FAST-NEXT: psubd %xmm1, %xmm0
; SSE3-FAST-NEXT: phsubd %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: partial_reduction_sub_v8i32:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; AVX-SLOW-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; AVX-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
; AVX-SLOW-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; AVX-SLOW-NEXT: vmovd %xmm0, %eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: partial_reduction_sub_v8i32:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; AVX-FAST-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; AVX-FAST-NEXT: vphsubd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: vzeroupper
; AVX-FAST-NEXT: retq
%x23 = shufflevector <8 x i32> %x, <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%x0213 = sub <8 x i32> %x, %x23
%x13 = shufflevector <8 x i32> %x0213, <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%x0123 = sub <8 x i32> %x0213, %x13
%r = extractelement <8 x i32> %x0123, i32 0
ret i32 %r
}
define i32 @partial_reduction_sub_v16i32(<16 x i32> %x) {
; SSE3-SLOW-LABEL: partial_reduction_sub_v16i32:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-SLOW-NEXT: psubd %xmm1, %xmm0
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
; SSE3-SLOW-NEXT: psubd %xmm1, %xmm0
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: partial_reduction_sub_v16i32:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-FAST-NEXT: psubd %xmm1, %xmm0
; SSE3-FAST-NEXT: phsubd %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: partial_reduction_sub_v16i32:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; AVX-SLOW-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; AVX-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
; AVX-SLOW-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; AVX-SLOW-NEXT: vmovd %xmm0, %eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX1-FAST-LABEL: partial_reduction_sub_v16i32:
; AVX1-FAST: # %bb.0:
; AVX1-FAST-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; AVX1-FAST-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; AVX1-FAST-NEXT: vphsubd %xmm0, %xmm0, %xmm0
; AVX1-FAST-NEXT: vmovd %xmm0, %eax
; AVX1-FAST-NEXT: vzeroupper
; AVX1-FAST-NEXT: retq
;
; AVX2-FAST-LABEL: partial_reduction_sub_v16i32:
; AVX2-FAST: # %bb.0:
; AVX2-FAST-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; AVX2-FAST-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; AVX2-FAST-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
; AVX2-FAST-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; AVX2-FAST-NEXT: vmovd %xmm0, %eax
; AVX2-FAST-NEXT: vzeroupper
; AVX2-FAST-NEXT: retq
;
; AVX512-FAST-LABEL: partial_reduction_sub_v16i32:
; AVX512-FAST: # %bb.0:
; AVX512-FAST-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; AVX512-FAST-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; AVX512-FAST-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
; AVX512-FAST-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; AVX512-FAST-NEXT: vmovd %xmm0, %eax
; AVX512-FAST-NEXT: vzeroupper
; AVX512-FAST-NEXT: retq
%x23 = shufflevector <16 x i32> %x, <16 x i32> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%x0213 = sub <16 x i32> %x, %x23
%x13 = shufflevector <16 x i32> %x0213, <16 x i32> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%x0123 = sub <16 x i32> %x0213, %x13
%r = extractelement <16 x i32> %x0123, i32 0
ret i32 %r
}
; https://bugs.chromium.org/p/chromium/issues/detail?id=1195353
define <2 x i64> @negative_extract_v16i16_v8i16(<4 x i64> %a0) {
; SSE3-LABEL: negative_extract_v16i16_v8i16:
; SSE3: # %bb.0:
; SSE3-NEXT: paddw %xmm1, %xmm0
; SSE3-NEXT: retq
;
; AVX1-SLOW-LABEL: negative_extract_v16i16_v8i16:
; AVX1-SLOW: # %bb.0:
; AVX1-SLOW-NEXT: vextractf128 $1, %ymm0, %xmm1
; AVX1-SLOW-NEXT: vpaddw %xmm1, %xmm0, %xmm0
; AVX1-SLOW-NEXT: vzeroupper
; AVX1-SLOW-NEXT: retq
;
; AVX1-FAST-LABEL: negative_extract_v16i16_v8i16:
; AVX1-FAST: # %bb.0:
; AVX1-FAST-NEXT: vextractf128 $1, %ymm0, %xmm1
; AVX1-FAST-NEXT: vpaddw %xmm1, %xmm0, %xmm0
; AVX1-FAST-NEXT: vzeroupper
; AVX1-FAST-NEXT: retq
;
; AVX2-SLOW-LABEL: negative_extract_v16i16_v8i16:
; AVX2-SLOW: # %bb.0:
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX2-SLOW-NEXT: vpaddw %xmm1, %xmm0, %xmm0
; AVX2-SLOW-NEXT: vzeroupper
; AVX2-SLOW-NEXT: retq
;
; AVX2-FAST-LABEL: negative_extract_v16i16_v8i16:
; AVX2-FAST: # %bb.0:
; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX2-FAST-NEXT: vpaddw %xmm1, %xmm0, %xmm0
; AVX2-FAST-NEXT: vzeroupper
; AVX2-FAST-NEXT: retq
;
; AVX512-SLOW-LABEL: negative_extract_v16i16_v8i16:
; AVX512-SLOW: # %bb.0:
; AVX512-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX512-SLOW-NEXT: vpaddw %xmm1, %xmm0, %xmm0
; AVX512-SLOW-NEXT: vzeroupper
; AVX512-SLOW-NEXT: retq
;
; AVX512-FAST-LABEL: negative_extract_v16i16_v8i16:
; AVX512-FAST: # %bb.0:
; AVX512-FAST-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX512-FAST-NEXT: vpaddw %xmm1, %xmm0, %xmm0
; AVX512-FAST-NEXT: vzeroupper
; AVX512-FAST-NEXT: retq
%s = shufflevector <4 x i64> %a0, <4 x i64> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
%b = bitcast <4 x i64> %a0 to <16 x i16>
%c = bitcast <4 x i64> %s to <16 x i16>
%d = add <16 x i16> %b, %c
%e = bitcast <16 x i16> %d to <4 x i64>
%f = shufflevector <4 x i64> %e, <4 x i64> undef, <2 x i32> <i32 0, i32 1>
ret <2 x i64> %f
}
; PR42023 - https://bugs.llvm.org/show_bug.cgi?id=42023
define i16 @hadd16_8(<8 x i16> %x223) {
; SSE3-SLOW-LABEL: hadd16_8:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-SLOW-NEXT: paddw %xmm0, %xmm1
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
; SSE3-SLOW-NEXT: paddw %xmm1, %xmm0
; SSE3-SLOW-NEXT: movdqa %xmm0, %xmm1
; SSE3-SLOW-NEXT: psrld $16, %xmm1
; SSE3-SLOW-NEXT: paddw %xmm0, %xmm1
; SSE3-SLOW-NEXT: movd %xmm1, %eax
; SSE3-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: hadd16_8:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddw %xmm0, %xmm0
; SSE3-FAST-NEXT: phaddw %xmm0, %xmm0
; SSE3-FAST-NEXT: phaddw %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: hadd16_8:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; AVX-SLOW-NEXT: vpaddw %xmm1, %xmm0, %xmm0
; AVX-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
; AVX-SLOW-NEXT: vpaddw %xmm1, %xmm0, %xmm0
; AVX-SLOW-NEXT: vpsrld $16, %xmm0, %xmm1
; AVX-SLOW-NEXT: vpaddw %xmm1, %xmm0, %xmm0
; AVX-SLOW-NEXT: vmovd %xmm0, %eax
; AVX-SLOW-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: hadd16_8:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddw %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vphaddw %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vphaddw %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-FAST-NEXT: retq
%x224 = shufflevector <8 x i16> %x223, <8 x i16> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
%x225 = add <8 x i16> %x223, %x224
%x226 = shufflevector <8 x i16> %x225, <8 x i16> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%x227 = add <8 x i16> %x225, %x226
%x228 = shufflevector <8 x i16> %x227, <8 x i16> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%x229 = add <8 x i16> %x227, %x228
%x230 = extractelement <8 x i16> %x229, i32 0
ret i16 %x230
}
define i32 @hadd32_4(<4 x i32> %x225) {
; SSE3-SLOW-LABEL: hadd32_4:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-SLOW-NEXT: paddd %xmm0, %xmm1
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
; SSE3-SLOW-NEXT: paddd %xmm1, %xmm0
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: hadd32_4:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: phaddd %xmm0, %xmm0
; SSE3-FAST-NEXT: phaddd %xmm0, %xmm0
; SSE3-FAST-NEXT: movd %xmm0, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: hadd32_4:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; AVX-SLOW-NEXT: vpaddd %xmm1, %xmm0, %xmm0
; AVX-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
; AVX-SLOW-NEXT: vpaddd %xmm1, %xmm0, %xmm0
; AVX-SLOW-NEXT: vmovd %xmm0, %eax
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: hadd32_4:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: retq
%x226 = shufflevector <4 x i32> %x225, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
%x227 = add <4 x i32> %x225, %x226
%x228 = shufflevector <4 x i32> %x227, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
%x229 = add <4 x i32> %x227, %x228
%x230 = extractelement <4 x i32> %x229, i32 0
ret i32 %x230
}
define i32 @hadd32_8(<8 x i32> %x225) {
; SSE3-SLOW-LABEL: hadd32_8:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-SLOW-NEXT: paddd %xmm0, %xmm1
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
; SSE3-SLOW-NEXT: paddd %xmm1, %xmm0
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: hadd32_8:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-FAST-NEXT: paddd %xmm0, %xmm1
; SSE3-FAST-NEXT: phaddd %xmm1, %xmm1
; SSE3-FAST-NEXT: movd %xmm1, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: hadd32_8:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; AVX-SLOW-NEXT: vpaddd %xmm1, %xmm0, %xmm0
; AVX-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
; AVX-SLOW-NEXT: vpaddd %xmm1, %xmm0, %xmm0
; AVX-SLOW-NEXT: vmovd %xmm0, %eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: hadd32_8:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: vzeroupper
; AVX-FAST-NEXT: retq
%x226 = shufflevector <8 x i32> %x225, <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%x227 = add <8 x i32> %x225, %x226
%x228 = shufflevector <8 x i32> %x227, <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%x229 = add <8 x i32> %x227, %x228
%x230 = extractelement <8 x i32> %x229, i32 0
ret i32 %x230
}
define i32 @hadd32_16(<16 x i32> %x225) {
; SSE3-SLOW-LABEL: hadd32_16:
; SSE3-SLOW: # %bb.0:
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-SLOW-NEXT: paddd %xmm0, %xmm1
; SSE3-SLOW-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
; SSE3-SLOW-NEXT: paddd %xmm1, %xmm0
; SSE3-SLOW-NEXT: movd %xmm0, %eax
; SSE3-SLOW-NEXT: retq
;
; SSE3-FAST-LABEL: hadd32_16:
; SSE3-FAST: # %bb.0:
; SSE3-FAST-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-FAST-NEXT: paddd %xmm0, %xmm1
; SSE3-FAST-NEXT: phaddd %xmm1, %xmm1
; SSE3-FAST-NEXT: movd %xmm1, %eax
; SSE3-FAST-NEXT: retq
;
; AVX-SLOW-LABEL: hadd32_16:
; AVX-SLOW: # %bb.0:
; AVX-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; AVX-SLOW-NEXT: vpaddd %xmm1, %xmm0, %xmm0
; AVX-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
; AVX-SLOW-NEXT: vpaddd %xmm1, %xmm0, %xmm0
; AVX-SLOW-NEXT: vmovd %xmm0, %eax
; AVX-SLOW-NEXT: vzeroupper
; AVX-SLOW-NEXT: retq
;
; AVX-FAST-LABEL: hadd32_16:
; AVX-FAST: # %bb.0:
; AVX-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-FAST-NEXT: vmovd %xmm0, %eax
; AVX-FAST-NEXT: vzeroupper
; AVX-FAST-NEXT: retq
%x226 = shufflevector <16 x i32> %x225, <16 x i32> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%x227 = add <16 x i32> %x225, %x226
%x228 = shufflevector <16 x i32> %x227, <16 x i32> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%x229 = add <16 x i32> %x227, %x228
%x230 = extractelement <16 x i32> %x229, i32 0
ret i32 %x230
}
define i16 @hadd16_8_optsize(<8 x i16> %x223) optsize {
; SSE3-LABEL: hadd16_8_optsize:
; SSE3: # %bb.0:
; SSE3-NEXT: phaddw %xmm0, %xmm0
; SSE3-NEXT: phaddw %xmm0, %xmm0
; SSE3-NEXT: phaddw %xmm0, %xmm0
; SSE3-NEXT: movd %xmm0, %eax
; SSE3-NEXT: # kill: def $ax killed $ax killed $eax
; SSE3-NEXT: retq
;
; AVX-LABEL: hadd16_8_optsize:
; AVX: # %bb.0:
; AVX-NEXT: vphaddw %xmm0, %xmm0, %xmm0
; AVX-NEXT: vphaddw %xmm0, %xmm0, %xmm0
; AVX-NEXT: vphaddw %xmm0, %xmm0, %xmm0
; AVX-NEXT: vmovd %xmm0, %eax
; AVX-NEXT: # kill: def $ax killed $ax killed $eax
; AVX-NEXT: retq
%x224 = shufflevector <8 x i16> %x223, <8 x i16> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
%x225 = add <8 x i16> %x223, %x224
%x226 = shufflevector <8 x i16> %x225, <8 x i16> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%x227 = add <8 x i16> %x225, %x226
%x228 = shufflevector <8 x i16> %x227, <8 x i16> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%x229 = add <8 x i16> %x227, %x228
%x230 = extractelement <8 x i16> %x229, i32 0
ret i16 %x230
}
define i32 @hadd32_4_optsize(<4 x i32> %x225) optsize {
; SSE3-LABEL: hadd32_4_optsize:
; SSE3: # %bb.0:
; SSE3-NEXT: phaddd %xmm0, %xmm0
; SSE3-NEXT: phaddd %xmm0, %xmm0
; SSE3-NEXT: movd %xmm0, %eax
; SSE3-NEXT: retq
;
; AVX-LABEL: hadd32_4_optsize:
; AVX: # %bb.0:
; AVX-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-NEXT: vmovd %xmm0, %eax
; AVX-NEXT: retq
%x226 = shufflevector <4 x i32> %x225, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
%x227 = add <4 x i32> %x225, %x226
%x228 = shufflevector <4 x i32> %x227, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
%x229 = add <4 x i32> %x227, %x228
%x230 = extractelement <4 x i32> %x229, i32 0
ret i32 %x230
}
define i32 @hadd32_4_pgso(<4 x i32> %x225) !prof !14 {
; SSE3-LABEL: hadd32_4_pgso:
; SSE3: # %bb.0:
; SSE3-NEXT: phaddd %xmm0, %xmm0
; SSE3-NEXT: phaddd %xmm0, %xmm0
; SSE3-NEXT: movd %xmm0, %eax
; SSE3-NEXT: retq
;
; AVX-LABEL: hadd32_4_pgso:
; AVX: # %bb.0:
; AVX-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-NEXT: vmovd %xmm0, %eax
; AVX-NEXT: retq
%x226 = shufflevector <4 x i32> %x225, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
%x227 = add <4 x i32> %x225, %x226
%x228 = shufflevector <4 x i32> %x227, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
%x229 = add <4 x i32> %x227, %x228
%x230 = extractelement <4 x i32> %x229, i32 0
ret i32 %x230
}
define i32 @hadd32_8_optsize(<8 x i32> %x225) optsize {
; SSE3-LABEL: hadd32_8_optsize:
; SSE3: # %bb.0:
; SSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-NEXT: paddd %xmm0, %xmm1
; SSE3-NEXT: phaddd %xmm1, %xmm1
; SSE3-NEXT: movd %xmm1, %eax
; SSE3-NEXT: retq
;
; AVX-LABEL: hadd32_8_optsize:
; AVX: # %bb.0:
; AVX-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-NEXT: vmovd %xmm0, %eax
; AVX-NEXT: vzeroupper
; AVX-NEXT: retq
%x226 = shufflevector <8 x i32> %x225, <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%x227 = add <8 x i32> %x225, %x226
%x228 = shufflevector <8 x i32> %x227, <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%x229 = add <8 x i32> %x227, %x228
%x230 = extractelement <8 x i32> %x229, i32 0
ret i32 %x230
}
define i32 @hadd32_16_optsize(<16 x i32> %x225) optsize {
; SSE3-LABEL: hadd32_16_optsize:
; SSE3: # %bb.0:
; SSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; SSE3-NEXT: paddd %xmm0, %xmm1
; SSE3-NEXT: phaddd %xmm1, %xmm1
; SSE3-NEXT: movd %xmm1, %eax
; SSE3-NEXT: retq
;
; AVX-LABEL: hadd32_16_optsize:
; AVX: # %bb.0:
; AVX-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-NEXT: vphaddd %xmm0, %xmm0, %xmm0
; AVX-NEXT: vmovd %xmm0, %eax
; AVX-NEXT: vzeroupper
; AVX-NEXT: retq
%x226 = shufflevector <16 x i32> %x225, <16 x i32> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%x227 = add <16 x i32> %x225, %x226
%x228 = shufflevector <16 x i32> %x227, <16 x i32> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%x229 = add <16 x i32> %x227, %x228
%x230 = extractelement <16 x i32> %x229, i32 0
ret i32 %x230
}
!llvm.module.flags = !{!0}
!0 = !{i32 1, !"ProfileSummary", !1}
!1 = !{!2, !3, !4, !5, !6, !7, !8, !9}
!2 = !{!"ProfileFormat", !"InstrProf"}
!3 = !{!"TotalCount", i64 10000}
!4 = !{!"MaxCount", i64 10}
!5 = !{!"MaxInternalCount", i64 1}
!6 = !{!"MaxFunctionCount", i64 1000}
!7 = !{!"NumCounts", i64 3}
!8 = !{!"NumFunctions", i64 3}
!9 = !{!"DetailedSummary", !10}
!10 = !{!11, !12, !13}
!11 = !{i32 10000, i64 100, i32 1}
!12 = !{i32 999000, i64 100, i32 1}
!13 = !{i32 999999, i64 1, i32 2}
!14 = !{!"function_entry_count", i64 0}