; testz(~X,Y) -> testc(X,Y)
;
-define i32 @ptestz_128_invert(<2 x i64> %c, <2 x i64> %d, i32 %a, i32 %b) {
-; CHECK-LABEL: ptestz_128_invert:
+define i32 @ptestz_128_invert0(<2 x i64> %c, <2 x i64> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: ptestz_128_invert0:
; CHECK: # %bb.0:
; CHECK-NEXT: movl %edi, %eax
; CHECK-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
ret i32 %t4
}
-define i32 @ptestz_256_invert(<4 x i64> %c, <4 x i64> %d, i32 %a, i32 %b) {
-; CHECK-LABEL: ptestz_256_invert:
+define i32 @ptestz_256_invert0(<4 x i64> %c, <4 x i64> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: ptestz_256_invert0:
; CHECK: # %bb.0:
; CHECK-NEXT: movl %edi, %eax
; CHECK-NEXT: vxorps %xmm2, %xmm2, %xmm2
}
;
+; testz(X,~Y) -> testc(Y,X)
+;
+
+define i32 @ptestz_128_invert1(<2 x i64> %c, <2 x i64> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: ptestz_128_invert1:
+; CHECK: # %bb.0:
+; CHECK-NEXT: movl %edi, %eax
+; CHECK-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
+; CHECK-NEXT: vpxor %xmm2, %xmm1, %xmm1
+; CHECK-NEXT: vptest %xmm1, %xmm0
+; CHECK-NEXT: cmovnel %esi, %eax
+; CHECK-NEXT: retq
+ %t1 = xor <2 x i64> %d, <i64 -1, i64 -1>
+ %t2 = call i32 @llvm.x86.sse41.ptestz(<2 x i64> %c, <2 x i64> %t1)
+ %t3 = icmp ne i32 %t2, 0
+ %t4 = select i1 %t3, i32 %a, i32 %b
+ ret i32 %t4
+}
+
+define i32 @ptestz_256_invert1(<4 x i64> %c, <4 x i64> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: ptestz_256_invert1:
+; CHECK: # %bb.0:
+; CHECK-NEXT: movl %edi, %eax
+; CHECK-NEXT: vxorps %xmm2, %xmm2, %xmm2
+; CHECK-NEXT: vcmptrueps %ymm2, %ymm2, %ymm2
+; CHECK-NEXT: vxorps %ymm2, %ymm1, %ymm1
+; CHECK-NEXT: vptest %ymm1, %ymm0
+; CHECK-NEXT: cmovnel %esi, %eax
+; CHECK-NEXT: vzeroupper
+; CHECK-NEXT: retq
+ %t1 = xor <4 x i64> %d, <i64 -1, i64 -1, i64 -1, i64 -1>
+ %t2 = call i32 @llvm.x86.avx.ptestz.256(<4 x i64> %c, <4 x i64> %t1)
+ %t3 = icmp ne i32 %t2, 0
+ %t4 = select i1 %t3, i32 %a, i32 %b
+ ret i32 %t4
+}
+
+;
; testc(~X,Y) -> testz(X,Y)
;
-define i32 @ptestc_128_invert(<2 x i64> %c, <2 x i64> %d, i32 %a, i32 %b) {
-; CHECK-LABEL: ptestc_128_invert:
+define i32 @ptestc_128_invert0(<2 x i64> %c, <2 x i64> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: ptestc_128_invert0:
; CHECK: # %bb.0:
; CHECK-NEXT: movl %edi, %eax
; CHECK-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
ret i32 %t4
}
-define i32 @ptestc_256_invert(<4 x i64> %c, <4 x i64> %d, i32 %a, i32 %b) {
-; CHECK-LABEL: ptestc_256_invert:
+define i32 @ptestc_256_invert0(<4 x i64> %c, <4 x i64> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: ptestc_256_invert0:
; CHECK: # %bb.0:
; CHECK-NEXT: movl %edi, %eax
; CHECK-NEXT: vxorps %xmm2, %xmm2, %xmm2
; testnzc(~X,Y) -> testnzc(X,Y)
;
-define i32 @ptestnzc_128_invert(<2 x i64> %c, <2 x i64> %d, i32 %a, i32 %b) {
-; CHECK-LABEL: ptestnzc_128_invert:
+define i32 @ptestnzc_128_invert0(<2 x i64> %c, <2 x i64> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: ptestnzc_128_invert0:
; CHECK: # %bb.0:
; CHECK-NEXT: movl %edi, %eax
; CHECK-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
ret i32 %t4
}
-define i32 @ptestnzc_256_invert(<4 x i64> %c, <4 x i64> %d, i32 %a, i32 %b) {
-; CHECK-LABEL: ptestnzc_256_invert:
+define i32 @ptestnzc_256_invert0(<4 x i64> %c, <4 x i64> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: ptestnzc_256_invert0:
; CHECK: # %bb.0:
; CHECK-NEXT: movl %edi, %eax
; CHECK-NEXT: vxorps %xmm2, %xmm2, %xmm2
ret i32 %t4
}
+;
+; testz(-1,X) -> testz(X,X)
+;
+
+define i32 @ptestz_128_allones0(<2 x i64> %c, i32 %a, i32 %b) {
+; CHECK-LABEL: ptestz_128_allones0:
+; CHECK: # %bb.0:
+; CHECK-NEXT: movl %edi, %eax
+; CHECK-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
+; CHECK-NEXT: vptest %xmm0, %xmm1
+; CHECK-NEXT: cmovnel %esi, %eax
+; CHECK-NEXT: retq
+ %t1 = call i32 @llvm.x86.sse41.ptestz(<2 x i64> <i64 -1, i64 -1>, <2 x i64> %c)
+ %t2 = icmp ne i32 %t1, 0
+ %t3 = select i1 %t2, i32 %a, i32 %b
+ ret i32 %t3
+}
+
+define i32 @ptestz_256_allones0(<4 x i64> %c, i32 %a, i32 %b) {
+; CHECK-LABEL: ptestz_256_allones0:
+; CHECK: # %bb.0:
+; CHECK-NEXT: movl %edi, %eax
+; CHECK-NEXT: vxorps %xmm1, %xmm1, %xmm1
+; CHECK-NEXT: vcmptrueps %ymm1, %ymm1, %ymm1
+; CHECK-NEXT: vptest %ymm0, %ymm1
+; CHECK-NEXT: cmovnel %esi, %eax
+; CHECK-NEXT: vzeroupper
+; CHECK-NEXT: retq
+ %t1 = call i32 @llvm.x86.avx.ptestz.256(<4 x i64> <i64 -1, i64 -1, i64 -1, i64 -1>, <4 x i64> %c)
+ %t2 = icmp ne i32 %t1, 0
+ %t3 = select i1 %t2, i32 %a, i32 %b
+ ret i32 %t3
+}
+
+;
+; testz(X,-1) -> testz(X,X)
+;
+
+define i32 @ptestz_128_allones1(<2 x i64> %c, i32 %a, i32 %b) {
+; CHECK-LABEL: ptestz_128_allones1:
+; CHECK: # %bb.0:
+; CHECK-NEXT: movl %edi, %eax
+; CHECK-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
+; CHECK-NEXT: vptest %xmm1, %xmm0
+; CHECK-NEXT: cmovnel %esi, %eax
+; CHECK-NEXT: retq
+ %t1 = call i32 @llvm.x86.sse41.ptestz(<2 x i64> %c, <2 x i64> <i64 -1, i64 -1>)
+ %t2 = icmp ne i32 %t1, 0
+ %t3 = select i1 %t2, i32 %a, i32 %b
+ ret i32 %t3
+}
+
+define i32 @ptestz_256_allones1(<4 x i64> %c, i32 %a, i32 %b) {
+; CHECK-LABEL: ptestz_256_allones1:
+; CHECK: # %bb.0:
+; CHECK-NEXT: movl %edi, %eax
+; CHECK-NEXT: vxorps %xmm1, %xmm1, %xmm1
+; CHECK-NEXT: vcmptrueps %ymm1, %ymm1, %ymm1
+; CHECK-NEXT: vptest %ymm1, %ymm0
+; CHECK-NEXT: cmovnel %esi, %eax
+; CHECK-NEXT: vzeroupper
+; CHECK-NEXT: retq
+ %t1 = call i32 @llvm.x86.avx.ptestz.256(<4 x i64> %c, <4 x i64> <i64 -1, i64 -1, i64 -1, i64 -1>)
+ %t2 = icmp ne i32 %t1, 0
+ %t3 = select i1 %t2, i32 %a, i32 %b
+ ret i32 %t3
+}
+
define zeroext i1 @PR38522(<16 x i8>* %x, <16 x i8>* %y) {
; CHECK-LABEL: PR38522:
; CHECK: # %bb.0: # %start
; testz(~X,Y) -> testc(X,Y)
;
-define i32 @testpdz_128_invert(<2 x double> %c, <2 x double> %d, i32 %a, i32 %b) {
-; CHECK-LABEL: testpdz_128_invert:
+define i32 @testpdz_128_invert0(<2 x double> %c, <2 x double> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: testpdz_128_invert0:
; CHECK: # %bb.0:
; CHECK-NEXT: movl %edi, %eax
; CHECK-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
ret i32 %t5
}
-define i32 @testpdz_256_invert(<4 x double> %c, <4 x double> %d, i32 %a, i32 %b) {
-; CHECK-LABEL: testpdz_256_invert:
+define i32 @testpdz_256_invert0(<4 x double> %c, <4 x double> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: testpdz_256_invert0:
; CHECK: # %bb.0:
; CHECK-NEXT: movl %edi, %eax
; CHECK-NEXT: vxorps %xmm2, %xmm2, %xmm2
}
;
+; testz(X,~Y) -> testc(Y,X)
+;
+
+define i32 @testpdz_128_invert1(<2 x double> %c, <2 x double> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: testpdz_128_invert1:
+; CHECK: # %bb.0:
+; CHECK-NEXT: movl %edi, %eax
+; CHECK-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
+; CHECK-NEXT: vpxor %xmm2, %xmm1, %xmm1
+; CHECK-NEXT: vtestpd %xmm1, %xmm0
+; CHECK-NEXT: cmovnel %esi, %eax
+; CHECK-NEXT: retq
+ %t0 = bitcast <2 x double> %d to <2 x i64>
+ %t1 = xor <2 x i64> %t0, <i64 -1, i64 -1>
+ %t2 = bitcast <2 x i64> %t1 to <2 x double>
+ %t3 = call i32 @llvm.x86.avx.vtestz.pd(<2 x double> %c, <2 x double> %t2)
+ %t4 = icmp ne i32 %t3, 0
+ %t5 = select i1 %t4, i32 %a, i32 %b
+ ret i32 %t5
+}
+
+define i32 @testpdz_256_invert1(<4 x double> %c, <4 x double> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: testpdz_256_invert1:
+; CHECK: # %bb.0:
+; CHECK-NEXT: movl %edi, %eax
+; CHECK-NEXT: vxorps %xmm2, %xmm2, %xmm2
+; CHECK-NEXT: vcmptrueps %ymm2, %ymm2, %ymm2
+; CHECK-NEXT: vxorps %ymm2, %ymm1, %ymm1
+; CHECK-NEXT: vtestpd %ymm1, %ymm0
+; CHECK-NEXT: cmovnel %esi, %eax
+; CHECK-NEXT: vzeroupper
+; CHECK-NEXT: retq
+ %t0 = bitcast <4 x double> %d to <4 x i64>
+ %t1 = xor <4 x i64> %t0, <i64 -1, i64 -1, i64 -1, i64 -1>
+ %t2 = bitcast <4 x i64> %t1 to <4 x double>
+ %t3 = call i32 @llvm.x86.avx.vtestz.pd.256(<4 x double> %c, <4 x double> %t2)
+ %t4 = icmp ne i32 %t3, 0
+ %t5 = select i1 %t4, i32 %a, i32 %b
+ ret i32 %t5
+}
+
+;
; testc(~X,Y) -> testz(X,Y)
;
-define i32 @testpdc_128_invert(<2 x double> %c, <2 x double> %d, i32 %a, i32 %b) {
-; CHECK-LABEL: testpdc_128_invert:
+define i32 @testpdc_128_invert0(<2 x double> %c, <2 x double> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: testpdc_128_invert0:
; CHECK: # %bb.0:
; CHECK-NEXT: movl %edi, %eax
; CHECK-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
ret i32 %t5
}
-define i32 @testpdc_256_invert(<4 x double> %c, <4 x double> %d, i32 %a, i32 %b) {
-; CHECK-LABEL: testpdc_256_invert:
+define i32 @testpdc_256_invert0(<4 x double> %c, <4 x double> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: testpdc_256_invert0:
; CHECK: # %bb.0:
; CHECK-NEXT: movl %edi, %eax
; CHECK-NEXT: vxorps %xmm2, %xmm2, %xmm2
; testnzc(~X,Y) -> testnzc(X,Y)
;
-define i32 @testpdnzc_128_invert(<2 x double> %c, <2 x double> %d, i32 %a, i32 %b) {
-; CHECK-LABEL: testpdnzc_128_invert:
+define i32 @testpdnzc_128_invert0(<2 x double> %c, <2 x double> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: testpdnzc_128_invert0:
; CHECK: # %bb.0:
; CHECK-NEXT: movl %edi, %eax
; CHECK-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
ret i32 %t5
}
-define i32 @testpdnzc_256_invert(<4 x double> %c, <4 x double> %d, i32 %a, i32 %b) {
-; CHECK-LABEL: testpdnzc_256_invert:
+define i32 @testpdnzc_256_invert0(<4 x double> %c, <4 x double> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: testpdnzc_256_invert0:
; CHECK: # %bb.0:
; CHECK-NEXT: movl %edi, %eax
; CHECK-NEXT: vxorps %xmm2, %xmm2, %xmm2
; testz(~X,Y) -> testc(X,Y)
;
-define i32 @testpsz_128_invert(<4 x float> %c, <4 x float> %d, i32 %a, i32 %b) {
-; CHECK-LABEL: testpsz_128_invert:
+define i32 @testpsz_128_invert0(<4 x float> %c, <4 x float> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: testpsz_128_invert0:
; CHECK: # %bb.0:
; CHECK-NEXT: movl %edi, %eax
; CHECK-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
ret i32 %t5
}
-define i32 @testpsz_256_invert(<8 x float> %c, <8 x float> %d, i32 %a, i32 %b) {
-; CHECK-LABEL: testpsz_256_invert:
+define i32 @testpsz_256_invert0(<8 x float> %c, <8 x float> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: testpsz_256_invert0:
; CHECK: # %bb.0:
; CHECK-NEXT: movl %edi, %eax
; CHECK-NEXT: vxorps %xmm2, %xmm2, %xmm2
}
;
+; testz(X,~Y) -> testc(Y,X)
+;
+
+define i32 @testpsz_128_invert1(<4 x float> %c, <4 x float> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: testpsz_128_invert1:
+; CHECK: # %bb.0:
+; CHECK-NEXT: movl %edi, %eax
+; CHECK-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
+; CHECK-NEXT: vpxor %xmm2, %xmm1, %xmm1
+; CHECK-NEXT: vtestps %xmm1, %xmm0
+; CHECK-NEXT: cmovnel %esi, %eax
+; CHECK-NEXT: retq
+ %t0 = bitcast <4 x float> %d to <2 x i64>
+ %t1 = xor <2 x i64> %t0, <i64 -1, i64 -1>
+ %t2 = bitcast <2 x i64> %t1 to <4 x float>
+ %t3 = call i32 @llvm.x86.avx.vtestz.ps(<4 x float> %c, <4 x float> %t2)
+ %t4 = icmp ne i32 %t3, 0
+ %t5 = select i1 %t4, i32 %a, i32 %b
+ ret i32 %t5
+}
+
+define i32 @testpsz_256_invert1(<8 x float> %c, <8 x float> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: testpsz_256_invert1:
+; CHECK: # %bb.0:
+; CHECK-NEXT: movl %edi, %eax
+; CHECK-NEXT: vxorps %xmm2, %xmm2, %xmm2
+; CHECK-NEXT: vcmptrueps %ymm2, %ymm2, %ymm2
+; CHECK-NEXT: vxorps %ymm2, %ymm1, %ymm1
+; CHECK-NEXT: vtestps %ymm1, %ymm0
+; CHECK-NEXT: cmovnel %esi, %eax
+; CHECK-NEXT: vzeroupper
+; CHECK-NEXT: retq
+ %t0 = bitcast <8 x float> %d to <4 x i64>
+ %t1 = xor <4 x i64> %t0, <i64 -1, i64 -1, i64 -1, i64 -1>
+ %t2 = bitcast <4 x i64> %t1 to <8 x float>
+ %t3 = call i32 @llvm.x86.avx.vtestz.ps.256(<8 x float> %c, <8 x float> %t2)
+ %t4 = icmp ne i32 %t3, 0
+ %t5 = select i1 %t4, i32 %a, i32 %b
+ ret i32 %t5
+}
+
+;
; testc(~X,Y) -> testz(X,Y)
;
-define i32 @testpsc_128_invert(<4 x float> %c, <4 x float> %d, i32 %a, i32 %b) {
-; CHECK-LABEL: testpsc_128_invert:
+define i32 @testpsc_128_invert0(<4 x float> %c, <4 x float> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: testpsc_128_invert0:
; CHECK: # %bb.0:
; CHECK-NEXT: movl %edi, %eax
; CHECK-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
ret i32 %t5
}
-define i32 @testpsc_256_invert(<8 x float> %c, <8 x float> %d, i32 %a, i32 %b) {
-; CHECK-LABEL: testpsc_256_invert:
+define i32 @testpsc_256_invert0(<8 x float> %c, <8 x float> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: testpsc_256_invert0:
; CHECK: # %bb.0:
; CHECK-NEXT: movl %edi, %eax
; CHECK-NEXT: vxorps %xmm2, %xmm2, %xmm2
; testnzc(~X,Y) -> testnzc(X,Y)
;
-define i32 @testpsnzc_128_invert(<4 x float> %c, <4 x float> %d, i32 %a, i32 %b) {
-; CHECK-LABEL: testpsnzc_128_invert:
+define i32 @testpsnzc_128_invert0(<4 x float> %c, <4 x float> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: testpsnzc_128_invert0:
; CHECK: # %bb.0:
; CHECK-NEXT: movl %edi, %eax
; CHECK-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
ret i32 %t5
}
-define i32 @testpsnzc_256_invert(<8 x float> %c, <8 x float> %d, i32 %a, i32 %b) {
-; CHECK-LABEL: testpsnzc_256_invert:
+define i32 @testpsnzc_256_invert0(<8 x float> %c, <8 x float> %d, i32 %a, i32 %b) {
+; CHECK-LABEL: testpsnzc_256_invert0:
; CHECK: # %bb.0:
; CHECK-NEXT: movl %edi, %eax
; CHECK-NEXT: vxorps %xmm2, %xmm2, %xmm2