; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=CHECK,SSE
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=CHECK,AVX,AVX1
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX,AVX2
declare <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16>, <8 x i16>) nounwind readnone
declare <8 x i16> @llvm.x86.ssse3.pmadd.ub.sw.128(<16 x i8>, <16 x i8>) nounwind readnone
define <4 x i32> @combine_pmaddwd_zero(<8 x i16> %a0, <8 x i16> %a1) {
; SSE-LABEL: combine_pmaddwd_zero:
; SSE: # %bb.0:
; SSE-NEXT: xorps %xmm0, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: combine_pmaddwd_zero:
; AVX: # %bb.0:
; AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
; AVX-NEXT: retq
%1 = call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> %a0, <8 x i16> zeroinitializer)
ret <4 x i32> %1
}
define <4 x i32> @combine_pmaddwd_zero_commute(<8 x i16> %a0, <8 x i16> %a1) {
; SSE-LABEL: combine_pmaddwd_zero_commute:
; SSE: # %bb.0:
; SSE-NEXT: xorps %xmm0, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: combine_pmaddwd_zero_commute:
; AVX: # %bb.0:
; AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
; AVX-NEXT: retq
%1 = call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> zeroinitializer, <8 x i16> %a0)
ret <4 x i32> %1
}
define <8 x i32> @combine_pmaddwd_concat(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> %a2, <8 x i16> %a3) {
; SSE-LABEL: combine_pmaddwd_concat:
; SSE: # %bb.0:
; SSE-NEXT: pmaddwd %xmm1, %xmm0
; SSE-NEXT: pmaddwd %xmm3, %xmm2
; SSE-NEXT: movdqa %xmm2, %xmm1
; SSE-NEXT: retq
;
; AVX1-LABEL: combine_pmaddwd_concat:
; AVX1: # %bb.0:
; AVX1-NEXT: vpmaddwd %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vpmaddwd %xmm3, %xmm2, %xmm1
; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: combine_pmaddwd_concat:
; AVX2: # %bb.0:
; AVX2-NEXT: # kill: def $xmm1 killed $xmm1 def $ymm1
; AVX2-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm0
; AVX2-NEXT: vinserti128 $1, %xmm3, %ymm1, %ymm1
; AVX2-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm0
; AVX2-NEXT: vpmaddwd %ymm1, %ymm0, %ymm0
; AVX2-NEXT: retq
%1 = call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> %a0, <8 x i16> %a1)
%2 = call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> %a2, <8 x i16> %a3)
%3 = shufflevector <4 x i32> %1, <4 x i32> %2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
ret <8 x i32> %3
}
define <8 x i32> @combine_pmaddwd_concat_freeze(<8 x i16> %a0, <8 x i16> %a1) {
; SSE-LABEL: combine_pmaddwd_concat_freeze:
; SSE: # %bb.0:
; SSE-NEXT: pmovsxbw {{.*#+}} xmm2 = [1,1,1,1,1,1,1,1]
; SSE-NEXT: pmaddwd %xmm2, %xmm0
; SSE-NEXT: pmaddwd %xmm2, %xmm1
; SSE-NEXT: retq
;
; AVX1-LABEL: combine_pmaddwd_concat_freeze:
; AVX1: # %bb.0:
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm2 = [1,1,1,1,1,1,1,1]
; AVX1-NEXT: vpmaddwd %xmm2, %xmm0, %xmm0
; AVX1-NEXT: vpmaddwd %xmm2, %xmm1, %xmm1
; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: combine_pmaddwd_concat_freeze:
; AVX2: # %bb.0:
; AVX2-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm0
; AVX2-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm0
; AVX2-NEXT: vpmaddwd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0 # [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX2-NEXT: retq
%lo = call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> %a0, <8 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>)
%hi = call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> %a1, <8 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>)
%flo = freeze <4 x i32> %lo
%fhi = freeze <4 x i32> %hi
%res = shufflevector <4 x i32> %flo, <4 x i32> %fhi, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
ret <8 x i32> %res
}
define <4 x i32> @combine_pmaddwd_demandedelts(<8 x i16> %a0, <8 x i16> %a1) {
; SSE-LABEL: combine_pmaddwd_demandedelts:
; SSE: # %bb.0:
; SSE-NEXT: pmaddwd %xmm1, %xmm0
; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE-NEXT: retq
;
; AVX1-LABEL: combine_pmaddwd_demandedelts:
; AVX1: # %bb.0:
; AVX1-NEXT: vpmaddwd %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; AVX1-NEXT: retq
;
; AVX2-LABEL: combine_pmaddwd_demandedelts:
; AVX2: # %bb.0:
; AVX2-NEXT: vpmaddwd %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpbroadcastd %xmm0, %xmm0
; AVX2-NEXT: retq
%1 = shufflevector <8 x i16> %a0, <8 x i16> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 4, i32 4, i32 4>
%2 = shufflevector <8 x i16> %a1, <8 x i16> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 7, i32 7, i32 7, i32 7>
%3 = call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> %1, <8 x i16> %2)
%4 = shufflevector <4 x i32> %3, <4 x i32> poison, <4 x i32> zeroinitializer
ret <4 x i32> %4
}
; [2]: (-5*13)+(6*-15) = -155 = 4294967141
define <4 x i32> @combine_pmaddwd_constant() {
; SSE-LABEL: combine_pmaddwd_constant:
; SSE: # %bb.0:
; SSE-NEXT: movaps {{.*#+}} xmm0 = [19,17,4294967141,271]
; SSE-NEXT: retq
;
; AVX-LABEL: combine_pmaddwd_constant:
; AVX: # %bb.0:
; AVX-NEXT: vmovaps {{.*#+}} xmm0 = [19,17,4294967141,271]
; AVX-NEXT: retq
%1 = call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> <i16 -1, i16 2, i16 3, i16 -4, i16 -5, i16 6, i16 7, i16 -8>, <8 x i16> <i16 -5, i16 7, i16 -9, i16 -11, i16 13, i16 -15, i16 17, i16 -19>)
ret <4 x i32> %1
}
; ensure we don't assume pmaddwd performs add nsw
; [0]: (-32768*-32768)+(-32768*-32768) = 0x80000000 = 2147483648
define <4 x i32> @combine_pmaddwd_constant_nsw() {
; SSE-LABEL: combine_pmaddwd_constant_nsw:
; SSE: # %bb.0:
; SSE-NEXT: movaps {{.*#+}} xmm0 = [2147483648,2147483648,2147483648,2147483648]
; SSE-NEXT: retq
;
; AVX-LABEL: combine_pmaddwd_constant_nsw:
; AVX: # %bb.0:
; AVX-NEXT: vbroadcastss {{.*#+}} xmm0 = [2147483648,2147483648,2147483648,2147483648]
; AVX-NEXT: retq
%1 = insertelement <8 x i16> undef, i16 32768, i32 0
%2 = shufflevector <8 x i16> %1, <8 x i16> undef, <8 x i32> zeroinitializer
%3 = call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> %2, <8 x i16> %2)
ret <4 x i32> %3
}
define <8 x i16> @combine_pmaddubsw_zero(<16 x i8> %a0, <16 x i8> %a1) {
; SSE-LABEL: combine_pmaddubsw_zero:
; SSE: # %bb.0:
; SSE-NEXT: xorps %xmm0, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: combine_pmaddubsw_zero:
; AVX: # %bb.0:
; AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
; AVX-NEXT: retq
%1 = call <8 x i16> @llvm.x86.ssse3.pmadd.ub.sw.128(<16 x i8> %a0, <16 x i8> zeroinitializer)
ret <8 x i16> %1
}
define <8 x i16> @combine_pmaddubsw_zero_commute(<16 x i8> %a0, <16 x i8> %a1) {
; SSE-LABEL: combine_pmaddubsw_zero_commute:
; SSE: # %bb.0:
; SSE-NEXT: xorps %xmm0, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: combine_pmaddubsw_zero_commute:
; AVX: # %bb.0:
; AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
; AVX-NEXT: retq
%1 = call <8 x i16> @llvm.x86.ssse3.pmadd.ub.sw.128(<16 x i8> zeroinitializer, <16 x i8> %a0)
ret <8 x i16> %1
}
define <16 x i16> @combine_pmaddubsw_concat(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> %a2, <16 x i8> %a3) {
; SSE-LABEL: combine_pmaddubsw_concat:
; SSE: # %bb.0:
; SSE-NEXT: pmaddubsw %xmm1, %xmm0
; SSE-NEXT: pmaddubsw %xmm3, %xmm2
; SSE-NEXT: movdqa %xmm2, %xmm1
; SSE-NEXT: retq
;
; AVX1-LABEL: combine_pmaddubsw_concat:
; AVX1: # %bb.0:
; AVX1-NEXT: vpmaddubsw %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vpmaddubsw %xmm3, %xmm2, %xmm1
; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: combine_pmaddubsw_concat:
; AVX2: # %bb.0:
; AVX2-NEXT: # kill: def $xmm1 killed $xmm1 def $ymm1
; AVX2-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm0
; AVX2-NEXT: vinserti128 $1, %xmm3, %ymm1, %ymm1
; AVX2-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm0
; AVX2-NEXT: vpmaddubsw %ymm1, %ymm0, %ymm0
; AVX2-NEXT: retq
%1 = call <8 x i16> @llvm.x86.ssse3.pmadd.ub.sw.128(<16 x i8> %a0, <16 x i8> %a1)
%2 = call <8 x i16> @llvm.x86.ssse3.pmadd.ub.sw.128(<16 x i8> %a2, <16 x i8> %a3)
%3 = shufflevector <8 x i16> %1, <8 x i16> %2, <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>
ret <16 x i16> %3
}
define <16 x i16> @combine_pmaddubsw_concat_freeze(<16 x i8> %a0, <16 x i8> %a1) {
; SSE-LABEL: combine_pmaddubsw_concat_freeze:
; SSE: # %bb.0:
; SSE-NEXT: movdqa {{.*#+}} xmm2 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; SSE-NEXT: pmaddubsw %xmm2, %xmm0
; SSE-NEXT: pmaddubsw %xmm2, %xmm1
; SSE-NEXT: retq
;
; AVX1-LABEL: combine_pmaddubsw_concat_freeze:
; AVX1: # %bb.0:
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm2 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX1-NEXT: vpmaddubsw %xmm2, %xmm0, %xmm0
; AVX1-NEXT: vpmaddubsw %xmm2, %xmm1, %xmm1
; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: combine_pmaddubsw_concat_freeze:
; AVX2: # %bb.0:
; AVX2-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm0
; AVX2-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm0
; AVX2-NEXT: vpmaddubsw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0 # [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX2-NEXT: retq
%lo = call <8 x i16> @llvm.x86.ssse3.pmadd.ub.sw.128(<16 x i8> %a0, <16 x i8> <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>)
%hi = call <8 x i16> @llvm.x86.ssse3.pmadd.ub.sw.128(<16 x i8> %a1, <16 x i8> <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>)
%flo = freeze <8 x i16> %lo
%fhi = freeze <8 x i16> %hi
%res = shufflevector <8 x i16> %flo, <8 x i16> %fhi, <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>
ret <16 x i16> %res
}
define <8 x i16> @combine_pmaddubsw_demandedelts(<16 x i8> %a0, <16 x i8> %a1) {
; SSE-LABEL: combine_pmaddubsw_demandedelts:
; SSE: # %bb.0:
; SSE-NEXT: pmaddubsw %xmm1, %xmm0
; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; SSE-NEXT: retq
;
; AVX1-LABEL: combine_pmaddubsw_demandedelts:
; AVX1: # %bb.0:
; AVX1-NEXT: vpmaddubsw %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
; AVX1-NEXT: retq
;
; AVX2-LABEL: combine_pmaddubsw_demandedelts:
; AVX2: # %bb.0:
; AVX2-NEXT: vpmaddubsw %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpbroadcastd %xmm0, %xmm0
; AVX2-NEXT: retq
%1 = shufflevector <16 x i8> %a0, <16 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8>
%2 = shufflevector <16 x i8> %a1, <16 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
%3 = call <8 x i16> @llvm.x86.ssse3.pmadd.ub.sw.128(<16 x i8> %1, <16 x i8> %2)
%4 = shufflevector <8 x i16> %3, <8 x i16> poison, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
ret <8 x i16> %4
}
; [3]: ((uint16_t)-6*7)+(7*-8) = (250*7)+(7*-8) = 1694
define i32 @combine_pmaddubsw_constant() {
; CHECK-LABEL: combine_pmaddubsw_constant:
; CHECK: # %bb.0:
; CHECK-NEXT: movl $1694, %eax # imm = 0x69E
; CHECK-NEXT: retq
%1 = call <8 x i16> @llvm.x86.ssse3.pmadd.ub.sw.128(<16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 -6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, <16 x i8> <i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 -8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 16>)
%2 = extractelement <8 x i16> %1, i32 3
%3 = sext i16 %2 to i32
ret i32 %3
}
; [0]: add_sat_i16(((uint16_t)-1*-128),((uint16_t)-1*-128)_ = add_sat_i16(255*-128),(255*-128)) = sat_i16(-65280) = -32768
define i32 @combine_pmaddubsw_constant_sat() {
; CHECK-LABEL: combine_pmaddubsw_constant_sat:
; CHECK: # %bb.0:
; CHECK-NEXT: movl $-32768, %eax # imm = 0x8000
; CHECK-NEXT: retq
%1 = call <8 x i16> @llvm.x86.ssse3.pmadd.ub.sw.128(<16 x i8> <i8 -1, i8 -1, i8 2, i8 3, i8 4, i8 5, i8 -6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, <16 x i8> <i8 -128, i8 -128, i8 3, i8 4, i8 5, i8 6, i8 7, i8 -8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 16>)
%2 = extractelement <8 x i16> %1, i32 0
%3 = sext i16 %2 to i32
ret i32 %3
}
; Constant folding PMADDWD was causing an infinite loop in the PCMPGT commuting between 2 constant values.
define i1 @pmaddwd_pcmpgt_infinite_loop() {
; CHECK-LABEL: pmaddwd_pcmpgt_infinite_loop:
; CHECK: # %bb.0:
; CHECK-NEXT: movb $1, %al
; CHECK-NEXT: retq
%1 = tail call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> <i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768>, <8 x i16> <i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768>)
%2 = icmp eq <4 x i32> %1, <i32 -2147483648, i32 -2147483648, i32 -2147483648, i32 -2147483648>
%3 = select <4 x i1> %2, <4 x i32> <i32 2147483647, i32 2147483647, i32 2147483647, i32 2147483647>, <4 x i32> zeroinitializer
%4 = add <4 x i32> %3, <i32 -8, i32 -9, i32 -10, i32 -11>
%.not = trunc <4 x i32> %3 to <4 x i1>
%5 = icmp sgt <4 x i32> %4, <i32 2147483640, i32 2147483639, i32 2147483638, i32 2147483637>
%6 = select <4 x i1> %.not, <4 x i1> %5, <4 x i1> zeroinitializer
%7 = bitcast <4 x i1> %6 to i4
%8 = icmp eq i4 %7, 0
ret i1 %8
}
Codebase Browser
llvm