1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,SSE,SSE2
3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.2 | FileCheck %s --check-prefixes=CHECK,SSE,SSE42
4 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX,AVX2
5 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefixes=CHECK,AVX,AVX512F
6 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefixes=CHECK,AVX,AVX512VL
9 define <4 x i32> @combine_v4i32_abs_constant() {
10 ; SSE-LABEL: combine_v4i32_abs_constant:
12 ; SSE-NEXT: movaps {{.*#+}} xmm0 = [0,1,3,2147483648]
15 ; AVX-LABEL: combine_v4i32_abs_constant:
17 ; AVX-NEXT: vmovaps {{.*#+}} xmm0 = [0,1,3,2147483648]
19 %1 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> <i32 0, i32 -1, i32 3, i32 -2147483648>, i1 false)
23 define <16 x i16> @combine_v16i16_abs_constant() {
24 ; SSE-LABEL: combine_v16i16_abs_constant:
26 ; SSE-NEXT: movaps {{.*#+}} xmm0 = [0,1,1,3,3,7,7,255]
27 ; SSE-NEXT: movaps {{.*#+}} xmm1 = [255,4096,4096,32767,32767,32768,32768,0]
30 ; AVX-LABEL: combine_v16i16_abs_constant:
32 ; AVX-NEXT: vmovaps {{.*#+}} ymm0 = [0,1,1,3,3,7,7,255,255,4096,4096,32767,32767,32768,32768,0]
34 %1 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> <i16 0, i16 1, i16 -1, i16 3, i16 -3, i16 7, i16 -7, i16 255, i16 -255, i16 4096, i16 -4096, i16 32767, i16 -32767, i16 -32768, i16 32768, i16 65536>, i1 false)
38 ; fold (abs (abs x)) -> (abs x)
39 define i32 @combine_i32_abs_abs(i32 %a) {
40 ; CHECK-LABEL: combine_i32_abs_abs:
42 ; CHECK-NEXT: movl %edi, %eax
43 ; CHECK-NEXT: negl %eax
44 ; CHECK-NEXT: cmovsl %edi, %eax
46 %n1 = sub i32 zeroinitializer, %a
47 %b1 = icmp slt i32 %a, zeroinitializer
48 %a1 = select i1 %b1, i32 %n1, i32 %a
49 %n2 = sub i32 zeroinitializer, %a1
50 %b2 = icmp sgt i32 %a1, zeroinitializer
51 %a2 = select i1 %b2, i32 %a1, i32 %n2
55 define <8 x i16> @combine_v8i16_abs_abs(<8 x i16> %a) {
56 ; SSE2-LABEL: combine_v8i16_abs_abs:
58 ; SSE2-NEXT: pxor %xmm1, %xmm1
59 ; SSE2-NEXT: psubw %xmm0, %xmm1
60 ; SSE2-NEXT: pmaxsw %xmm1, %xmm0
63 ; SSE42-LABEL: combine_v8i16_abs_abs:
65 ; SSE42-NEXT: pabsw %xmm0, %xmm0
68 ; AVX-LABEL: combine_v8i16_abs_abs:
70 ; AVX-NEXT: vpabsw %xmm0, %xmm0
72 %a1 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a, i1 false)
73 %s2 = ashr <8 x i16> %a1, <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
74 %a2 = add <8 x i16> %a1, %s2
75 %x2 = xor <8 x i16> %a2, %s2
79 define <32 x i8> @combine_v32i8_abs_abs(<32 x i8> %a) {
80 ; SSE2-LABEL: combine_v32i8_abs_abs:
82 ; SSE2-NEXT: pxor %xmm2, %xmm2
83 ; SSE2-NEXT: pxor %xmm3, %xmm3
84 ; SSE2-NEXT: psubb %xmm0, %xmm3
85 ; SSE2-NEXT: pminub %xmm3, %xmm0
86 ; SSE2-NEXT: psubb %xmm1, %xmm2
87 ; SSE2-NEXT: pminub %xmm2, %xmm1
90 ; SSE42-LABEL: combine_v32i8_abs_abs:
92 ; SSE42-NEXT: pabsb %xmm0, %xmm0
93 ; SSE42-NEXT: pabsb %xmm1, %xmm1
96 ; AVX-LABEL: combine_v32i8_abs_abs:
98 ; AVX-NEXT: vpabsb %ymm0, %ymm0
100 %n1 = sub <32 x i8> zeroinitializer, %a
101 %b1 = icmp slt <32 x i8> %a, zeroinitializer
102 %a1 = select <32 x i1> %b1, <32 x i8> %n1, <32 x i8> %a
103 %a2 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a1, i1 false)
107 define <4 x i64> @combine_v4i64_abs_abs(<4 x i64> %a) {
108 ; SSE2-LABEL: combine_v4i64_abs_abs:
110 ; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
111 ; SSE2-NEXT: psrad $31, %xmm2
112 ; SSE2-NEXT: pxor %xmm2, %xmm0
113 ; SSE2-NEXT: psubq %xmm2, %xmm0
114 ; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
115 ; SSE2-NEXT: psrad $31, %xmm2
116 ; SSE2-NEXT: pxor %xmm2, %xmm1
117 ; SSE2-NEXT: psubq %xmm2, %xmm1
120 ; SSE42-LABEL: combine_v4i64_abs_abs:
122 ; SSE42-NEXT: movdqa %xmm0, %xmm2
123 ; SSE42-NEXT: pxor %xmm3, %xmm3
124 ; SSE42-NEXT: pxor %xmm4, %xmm4
125 ; SSE42-NEXT: psubq %xmm0, %xmm4
126 ; SSE42-NEXT: blendvpd %xmm0, %xmm4, %xmm2
127 ; SSE42-NEXT: psubq %xmm1, %xmm3
128 ; SSE42-NEXT: movdqa %xmm1, %xmm0
129 ; SSE42-NEXT: blendvpd %xmm0, %xmm3, %xmm1
130 ; SSE42-NEXT: movapd %xmm2, %xmm0
133 ; AVX2-LABEL: combine_v4i64_abs_abs:
135 ; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
136 ; AVX2-NEXT: vpsubq %ymm0, %ymm1, %ymm1
137 ; AVX2-NEXT: vblendvpd %ymm0, %ymm1, %ymm0, %ymm0
140 ; AVX512F-LABEL: combine_v4i64_abs_abs:
142 ; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
143 ; AVX512F-NEXT: vpabsq %zmm0, %zmm0
144 ; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 killed $zmm0
147 ; AVX512VL-LABEL: combine_v4i64_abs_abs:
149 ; AVX512VL-NEXT: vpabsq %ymm0, %ymm0
150 ; AVX512VL-NEXT: retq
151 %n1 = sub <4 x i64> zeroinitializer, %a
152 %b1 = icmp slt <4 x i64> %a, zeroinitializer
153 %a1 = select <4 x i1> %b1, <4 x i64> %n1, <4 x i64> %a
154 %n2 = sub <4 x i64> zeroinitializer, %a1
155 %b2 = icmp sgt <4 x i64> %a1, zeroinitializer
156 %a2 = select <4 x i1> %b2, <4 x i64> %a1, <4 x i64> %n2
160 ; fold (abs x) -> x iff not-negative
161 define <16 x i8> @combine_v16i8_abs_constant(<16 x i8> %a) {
162 ; SSE-LABEL: combine_v16i8_abs_constant:
164 ; SSE-NEXT: andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
167 ; AVX2-LABEL: combine_v16i8_abs_constant:
169 ; AVX2-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
172 ; AVX512F-LABEL: combine_v16i8_abs_constant:
174 ; AVX512F-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
177 ; AVX512VL-LABEL: combine_v16i8_abs_constant:
179 ; AVX512VL-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm0, %xmm0
180 ; AVX512VL-NEXT: retq
181 %1 = insertelement <16 x i8> undef, i8 15, i32 0
182 %2 = shufflevector <16 x i8> %1, <16 x i8> undef, <16 x i32> zeroinitializer
183 %3 = and <16 x i8> %a, %2
184 %4 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %3, i1 false)
188 define <8 x i32> @combine_v8i32_abs_pos(<8 x i32> %a) {
189 ; SSE-LABEL: combine_v8i32_abs_pos:
191 ; SSE-NEXT: psrld $1, %xmm0
192 ; SSE-NEXT: psrld $1, %xmm1
195 ; AVX-LABEL: combine_v8i32_abs_pos:
197 ; AVX-NEXT: vpsrld $1, %ymm0, %ymm0
199 %1 = lshr <8 x i32> %a, <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
200 %2 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %1, i1 false)
204 ; (abs x) upper bits are known zero if x has extra sign bits
205 define i32 @combine_i32_abs_zerosign(i32 %a) {
206 ; CHECK-LABEL: combine_i32_abs_zerosign:
208 ; CHECK-NEXT: xorl %eax, %eax
211 %2 = call i32 @llvm.abs.i32(i32 %1, i1 false)
212 %3 = and i32 %2, -524288 ; 0xFFF80000
216 define <8 x i16> @combine_v8i16_abs_zerosign(<8 x i16> %a) {
217 ; SSE-LABEL: combine_v8i16_abs_zerosign:
219 ; SSE-NEXT: xorps %xmm0, %xmm0
222 ; AVX-LABEL: combine_v8i16_abs_zerosign:
224 ; AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
226 %1 = ashr <8 x i16> %a, <i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14>
227 %2 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %1, i1 false)
228 %3 = and <8 x i16> %2, <i16 32768, i16 32768, i16 32768, i16 32768, i16 32768, i16 32768, i16 32768, i16 32768>
232 ; negative test - mask extends beyond known zero bits
233 define i32 @combine_i32_abs_zerosign_negative(i32 %a) {
234 ; CHECK-LABEL: combine_i32_abs_zerosign_negative:
236 ; CHECK-NEXT: sarl $3, %edi
237 ; CHECK-NEXT: movl %edi, %eax
238 ; CHECK-NEXT: negl %eax
239 ; CHECK-NEXT: cmovsl %edi, %eax
240 ; CHECK-NEXT: andl $536346624, %eax # imm = 0x1FF80000
243 %2 = call i32 @llvm.abs.i32(i32 %1, i1 false)
244 %3 = and i32 %2, -524288 ; 0xFFF80000
248 declare <16 x i8> @llvm.abs.v16i8(<16 x i8>, i1) nounwind readnone
249 declare <4 x i32> @llvm.abs.v4i32(<4 x i32>, i1) nounwind readnone
250 declare <8 x i16> @llvm.abs.v8i16(<8 x i16>, i1) nounwind readnone
252 declare <32 x i8> @llvm.abs.v32i8(<32 x i8>, i1) nounwind readnone
253 declare <8 x i32> @llvm.abs.v8i32(<8 x i32>, i1) nounwind readnone
254 declare <16 x i16> @llvm.abs.v16i16(<16 x i16>, i1) nounwind readnone