summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--llvm/lib/Target/X86/X86InstrAVX512.td210
-rw-r--r--llvm/test/CodeGen/X86/avx512-logic.ll96
-rw-r--r--llvm/test/CodeGen/X86/avx512vl-logic.ll192
3 files changed, 306 insertions, 192 deletions
diff --git a/llvm/lib/Target/X86/X86InstrAVX512.td b/llvm/lib/Target/X86/X86InstrAVX512.td
index 21736b6c55b..28c943a2bc7 100644
--- a/llvm/lib/Target/X86/X86InstrAVX512.td
+++ b/llvm/lib/Target/X86/X86InstrAVX512.td
@@ -4235,6 +4235,216 @@ defm VOR : avx512_fp_binop_p<0x56, "vor", X86for, HasDQI,
defm VXOR : avx512_fp_binop_p<0x57, "vxor", X86fxor, HasDQI,
SSE_ALU_ITINS_P, 1>;
+// Patterns catch floating point selects with bitcasted integer logic ops.
+let Predicates = [HasVLX] in {
+ def : Pat<(v4f32 (vselect VK4WM:$mask,
+ (bitconvert (v2i64 (and VR128X:$src1, VR128X:$src2))),
+ VR128X:$src0)),
+ (VPANDDZ128rrk VR128X:$src0, VK4WM:$mask, VR128X:$src1, VR128X:$src2)>;
+ def : Pat<(v4f32 (vselect VK4WM:$mask,
+ (bitconvert (v2i64 (or VR128X:$src1, VR128X:$src2))),
+ VR128X:$src0)),
+ (VPORDZ128rrk VR128X:$src0, VK4WM:$mask, VR128X:$src1, VR128X:$src2)>;
+ def : Pat<(v4f32 (vselect VK4WM:$mask,
+ (bitconvert (v2i64 (xor VR128X:$src1, VR128X:$src2))),
+ VR128X:$src0)),
+ (VPXORDZ128rrk VR128X:$src0, VK4WM:$mask, VR128X:$src1, VR128X:$src2)>;
+ def : Pat<(v4f32 (vselect VK4WM:$mask,
+ (bitconvert (v2i64 (X86andnp VR128X:$src1, VR128X:$src2))),
+ VR128X:$src0)),
+ (VPANDNDZ128rrk VR128X:$src0, VK4WM:$mask, VR128X:$src1,
+ VR128X:$src2)>;
+
+ def : Pat<(v4f32 (vselect VK4WM:$mask,
+ (bitconvert (v2i64 (and VR128X:$src1, VR128X:$src2))),
+ (bitconvert (v4i32 immAllZerosV)))),
+ (VPANDDZ128rrkz VK4WM:$mask, VR128X:$src1, VR128X:$src2)>;
+ def : Pat<(v4f32 (vselect VK4WM:$mask,
+ (bitconvert (v2i64 (or VR128X:$src1, VR128X:$src2))),
+ (bitconvert (v4i32 immAllZerosV)))),
+ (VPORDZ128rrkz VK4WM:$mask, VR128X:$src1, VR128X:$src2)>;
+ def : Pat<(v4f32 (vselect VK4WM:$mask,
+ (bitconvert (v2i64 (xor VR128X:$src1, VR128X:$src2))),
+ (bitconvert (v4i32 immAllZerosV)))),
+ (VPXORDZ128rrkz VK4WM:$mask, VR128X:$src1, VR128X:$src2)>;
+ def : Pat<(v4f32 (vselect VK4WM:$mask,
+ (bitconvert (v2i64 (X86andnp VR128X:$src1, VR128X:$src2))),
+ (bitconvert (v4i32 immAllZerosV)))),
+ (VPANDNDZ128rrkz VK4WM:$mask, VR128X:$src1, VR128X:$src2)>;
+
+ def : Pat<(v2f64 (vselect VK2WM:$mask,
+ (bitconvert (v2i64 (and VR128X:$src1, VR128X:$src2))),
+ VR128X:$src0)),
+ (VPANDQZ128rrk VR128X:$src0, VK2WM:$mask, VR128X:$src1, VR128X:$src2)>;
+ def : Pat<(v2f64 (vselect VK2WM:$mask,
+ (bitconvert (v2i64 (or VR128X:$src1, VR128X:$src2))),
+ VR128X:$src0)),
+ (VPORQZ128rrk VR128X:$src0, VK2WM:$mask, VR128X:$src1, VR128X:$src2)>;
+ def : Pat<(v2f64 (vselect VK2WM:$mask,
+ (bitconvert (v2i64 (xor VR128X:$src1, VR128X:$src2))),
+ VR128X:$src0)),
+ (VPXORQZ128rrk VR128X:$src0, VK2WM:$mask, VR128X:$src1, VR128X:$src2)>;
+ def : Pat<(v2f64 (vselect VK2WM:$mask,
+ (bitconvert (v2i64 (X86andnp VR128X:$src1, VR128X:$src2))),
+ VR128X:$src0)),
+ (VPANDNQZ128rrk VR128X:$src0, VK2WM:$mask, VR128X:$src1, VR128X:$src2)>;
+
+ def : Pat<(v2f64 (vselect VK2WM:$mask,
+ (bitconvert (v2i64 (and VR128X:$src1, VR128X:$src2))),
+ (bitconvert (v4i32 immAllZerosV)))),
+ (VPANDQZ128rrkz VK2WM:$mask, VR128X:$src1, VR128X:$src2)>;
+ def : Pat<(v2f64 (vselect VK2WM:$mask,
+ (bitconvert (v2i64 (or VR128X:$src1, VR128X:$src2))),
+ (bitconvert (v4i32 immAllZerosV)))),
+ (VPORQZ128rrkz VK2WM:$mask, VR128X:$src1, VR128X:$src2)>;
+ def : Pat<(v2f64 (vselect VK2WM:$mask,
+ (bitconvert (v2i64 (xor VR128X:$src1, VR128X:$src2))),
+ (bitconvert (v4i32 immAllZerosV)))),
+ (VPXORQZ128rrkz VK2WM:$mask, VR128X:$src1, VR128X:$src2)>;
+ def : Pat<(v2f64 (vselect VK2WM:$mask,
+ (bitconvert (v2i64 (X86andnp VR128X:$src1, VR128X:$src2))),
+ (bitconvert (v4i32 immAllZerosV)))),
+ (VPANDNQZ128rrkz VK2WM:$mask, VR128X:$src1, VR128X:$src2)>;
+
+ def : Pat<(v8f32 (vselect VK8WM:$mask,
+ (bitconvert (v4i64 (and VR256X:$src1, VR256X:$src2))),
+ VR256X:$src0)),
+ (VPANDDZ256rrk VR256X:$src0, VK8WM:$mask, VR256X:$src1, VR256X:$src2)>;
+ def : Pat<(v8f32 (vselect VK8WM:$mask,
+ (bitconvert (v4i64 (or VR256X:$src1, VR256X:$src2))),
+ VR256X:$src0)),
+ (VPORDZ256rrk VR256X:$src0, VK8WM:$mask, VR256X:$src1, VR256X:$src2)>;
+ def : Pat<(v8f32 (vselect VK8WM:$mask,
+ (bitconvert (v4i64 (xor VR256X:$src1, VR256X:$src2))),
+ VR256X:$src0)),
+ (VPXORDZ256rrk VR256X:$src0, VK8WM:$mask, VR256X:$src1, VR256X:$src2)>;
+ def : Pat<(v8f32 (vselect VK8WM:$mask,
+ (bitconvert (v4i64 (X86andnp VR256X:$src1, VR256X:$src2))),
+ VR256X:$src0)),
+ (VPANDNDZ256rrk VR256X:$src0, VK8WM:$mask, VR256X:$src1, VR256X:$src2)>;
+
+ def : Pat<(v8f32 (vselect VK8WM:$mask,
+ (bitconvert (v4i64 (and VR256X:$src1, VR256X:$src2))),
+ (bitconvert (v8i32 immAllZerosV)))),
+ (VPANDDZ256rrkz VK8WM:$mask, VR256X:$src1, VR256X:$src2)>;
+ def : Pat<(v8f32 (vselect VK8WM:$mask,
+ (bitconvert (v4i64 (or VR256X:$src1, VR256X:$src2))),
+ (bitconvert (v8i32 immAllZerosV)))),
+ (VPORDZ256rrkz VK8WM:$mask, VR256X:$src1, VR256X:$src2)>;
+ def : Pat<(v8f32 (vselect VK8WM:$mask,
+ (bitconvert (v4i64 (xor VR256X:$src1, VR256X:$src2))),
+ (bitconvert (v8i32 immAllZerosV)))),
+ (VPXORDZ256rrkz VK8WM:$mask, VR256X:$src1, VR256X:$src2)>;
+ def : Pat<(v8f32 (vselect VK8WM:$mask,
+ (bitconvert (v4i64 (X86andnp VR256X:$src1, VR256X:$src2))),
+ (bitconvert (v8i32 immAllZerosV)))),
+ (VPANDNDZ256rrkz VK8WM:$mask, VR256X:$src1, VR256X:$src2)>;
+
+ def : Pat<(v4f64 (vselect VK4WM:$mask,
+ (bitconvert (v4i64 (and VR256X:$src1, VR256X:$src2))),
+ VR256X:$src0)),
+ (VPANDQZ256rrk VR256X:$src0, VK4WM:$mask, VR256X:$src1, VR256X:$src2)>;
+ def : Pat<(v4f64 (vselect VK4WM:$mask,
+ (bitconvert (v4i64 (or VR256X:$src1, VR256X:$src2))),
+ VR256X:$src0)),
+ (VPORQZ256rrk VR256X:$src0, VK4WM:$mask, VR256X:$src1, VR256X:$src2)>;
+ def : Pat<(v4f64 (vselect VK4WM:$mask,
+ (bitconvert (v4i64 (xor VR256X:$src1, VR256X:$src2))),
+ VR256X:$src0)),
+ (VPXORQZ256rrk VR256X:$src0, VK4WM:$mask, VR256X:$src1, VR256X:$src2)>;
+ def : Pat<(v4f64 (vselect VK4WM:$mask,
+ (bitconvert (v4i64 (X86andnp VR256X:$src1, VR256X:$src2))),
+ VR256X:$src0)),
+ (VPANDNQZ256rrk VR256X:$src0, VK4WM:$mask, VR256X:$src1, VR256X:$src2)>;
+
+ def : Pat<(v4f64 (vselect VK4WM:$mask,
+ (bitconvert (v4i64 (and VR256X:$src1, VR256X:$src2))),
+ (bitconvert (v8i32 immAllZerosV)))),
+ (VPANDQZ256rrkz VK4WM:$mask, VR256X:$src1, VR256X:$src2)>;
+ def : Pat<(v4f64 (vselect VK4WM:$mask,
+ (bitconvert (v4i64 (or VR256X:$src1, VR256X:$src2))),
+ (bitconvert (v8i32 immAllZerosV)))),
+ (VPORQZ256rrkz VK4WM:$mask, VR256X:$src1, VR256X:$src2)>;
+ def : Pat<(v4f64 (vselect VK4WM:$mask,
+ (bitconvert (v4i64 (xor VR256X:$src1, VR256X:$src2))),
+ (bitconvert (v8i32 immAllZerosV)))),
+ (VPXORQZ256rrkz VK4WM:$mask, VR256X:$src1, VR256X:$src2)>;
+ def : Pat<(v4f64 (vselect VK4WM:$mask,
+ (bitconvert (v4i64 (X86andnp VR256X:$src1, VR256X:$src2))),
+ (bitconvert (v8i32 immAllZerosV)))),
+ (VPANDNQZ256rrkz VK4WM:$mask, VR256X:$src1, VR256X:$src2)>;
+}
+
+let Predicates = [HasAVX512] in {
+ def : Pat<(v16f32 (vselect VK16WM:$mask,
+ (bitconvert (v8i64 (and VR512:$src1, VR512:$src2))),
+ VR512:$src0)),
+ (VPANDDZrrk VR512:$src0, VK16WM:$mask, VR512:$src1, VR512:$src2)>;
+ def : Pat<(v16f32 (vselect VK16WM:$mask,
+ (bitconvert (v8i64 (or VR512:$src1, VR512:$src2))),
+ VR512:$src0)),
+ (VPORDZrrk VR512:$src0, VK16WM:$mask, VR512:$src1, VR512:$src2)>;
+ def : Pat<(v16f32 (vselect VK16WM:$mask,
+ (bitconvert (v8i64 (xor VR512:$src1, VR512:$src2))),
+ VR512:$src0)),
+ (VPXORDZrrk VR512:$src0, VK16WM:$mask, VR512:$src1, VR512:$src2)>;
+ def : Pat<(v16f32 (vselect VK16WM:$mask,
+ (bitconvert (v8i64 (X86andnp VR512:$src1, VR512:$src2))),
+ VR512:$src0)),
+ (VPANDNDZrrk VR512:$src0, VK16WM:$mask, VR512:$src1, VR512:$src2)>;
+
+ def : Pat<(v16f32 (vselect VK16WM:$mask,
+ (bitconvert (v8i64 (and VR512:$src1, VR512:$src2))),
+ (bitconvert (v16i32 immAllZerosV)))),
+ (VPANDDZrrkz VK16WM:$mask, VR512:$src1, VR512:$src2)>;
+ def : Pat<(v16f32 (vselect VK16WM:$mask,
+ (bitconvert (v8i64 (or VR512:$src1, VR512:$src2))),
+ (bitconvert (v16i32 immAllZerosV)))),
+ (VPORDZrrkz VK16WM:$mask, VR512:$src1, VR512:$src2)>;
+ def : Pat<(v16f32 (vselect VK16WM:$mask,
+ (bitconvert (v8i64 (xor VR512:$src1, VR512:$src2))),
+ (bitconvert (v16i32 immAllZerosV)))),
+ (VPXORDZrrkz VK16WM:$mask, VR512:$src1, VR512:$src2)>;
+ def : Pat<(v16f32 (vselect VK16WM:$mask,
+ (bitconvert (v8i64 (X86andnp VR512:$src1, VR512:$src2))),
+ (bitconvert (v16i32 immAllZerosV)))),
+ (VPANDNDZrrkz VK16WM:$mask, VR512:$src1, VR512:$src2)>;
+
+ def : Pat<(v8f64 (vselect VK8WM:$mask,
+ (bitconvert (v8i64 (and VR512:$src1, VR512:$src2))),
+ VR512:$src0)),
+ (VPANDQZrrk VR512:$src0, VK8WM:$mask, VR512:$src1, VR512:$src2)>;
+ def : Pat<(v8f64 (vselect VK8WM:$mask,
+ (bitconvert (v8i64 (or VR512:$src1, VR512:$src2))),
+ VR512:$src0)),
+ (VPORQZrrk VR512:$src0, VK8WM:$mask, VR512:$src1, VR512:$src2)>;
+ def : Pat<(v8f64 (vselect VK8WM:$mask,
+ (bitconvert (v8i64 (xor VR512:$src1, VR512:$src2))),
+ VR512:$src0)),
+ (VPXORQZrrk VR512:$src0, VK8WM:$mask, VR512:$src1, VR512:$src2)>;
+ def : Pat<(v8f64 (vselect VK8WM:$mask,
+ (bitconvert (v8i64 (X86andnp VR512:$src1, VR512:$src2))),
+ VR512:$src0)),
+ (VPANDNQZrrk VR512:$src0, VK8WM:$mask, VR512:$src1, VR512:$src2)>;
+
+ def : Pat<(v8f64 (vselect VK8WM:$mask,
+ (bitconvert (v8i64 (and VR512:$src1, VR512:$src2))),
+ (bitconvert (v16i32 immAllZerosV)))),
+ (VPANDQZrrkz VK8WM:$mask, VR512:$src1, VR512:$src2)>;
+ def : Pat<(v8f64 (vselect VK8WM:$mask,
+ (bitconvert (v8i64 (or VR512:$src1, VR512:$src2))),
+ (bitconvert (v16i32 immAllZerosV)))),
+ (VPORQZrrkz VK8WM:$mask, VR512:$src1, VR512:$src2)>;
+ def : Pat<(v8f64 (vselect VK8WM:$mask,
+ (bitconvert (v8i64 (xor VR512:$src1, VR512:$src2))),
+ (bitconvert (v16i32 immAllZerosV)))),
+ (VPXORQZrrkz VK8WM:$mask, VR512:$src1, VR512:$src2)>;
+ def : Pat<(v8f64 (vselect VK8WM:$mask,
+ (bitconvert (v8i64 (X86andnp VR512:$src1, VR512:$src2))),
+ (bitconvert (v16i32 immAllZerosV)))),
+ (VPANDNQZrrkz VK8WM:$mask, VR512:$src1, VR512:$src2)>;
+}
+
multiclass avx512_fp_scalef_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
X86VectorVTInfo _> {
defm rr: AVX512_maskable<opc, MRMSrcReg, _, (outs _.RC:$dst),
diff --git a/llvm/test/CodeGen/X86/avx512-logic.ll b/llvm/test/CodeGen/X86/avx512-logic.ll
index b54c8a94058..ca603b11fbb 100644
--- a/llvm/test/CodeGen/X86/avx512-logic.ll
+++ b/llvm/test/CodeGen/X86/avx512-logic.ll
@@ -498,16 +498,14 @@ entry:
define <8 x double> @test_mm512_mask_xor_pd(<8 x double> %__W, i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B) {
; KNL-LABEL: test_mm512_mask_xor_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpxorq %zmm2, %zmm1, %zmm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmpd %zmm1, %zmm0, %zmm0 {%k1}
+; KNL-NEXT: vpxorq %zmm2, %zmm1, %zmm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm512_mask_xor_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vxorpd %zmm2, %zmm1, %zmm1
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vblendmpd %zmm1, %zmm0, %zmm0 {%k1}
+; SKX-NEXT: vxorpd %zmm2, %zmm1, %zmm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <8 x double> %__A to <8 x i64>
@@ -522,16 +520,14 @@ entry:
define <8 x double> @test_mm512_maskz_xor_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B) {
; KNL-LABEL: test_mm512_maskz_xor_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpxorq %zmm1, %zmm0, %zmm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovapd %zmm0, %zmm0 {%k1} {z}
+; KNL-NEXT: vpxorq %zmm1, %zmm0, %zmm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm512_maskz_xor_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vxorpd %zmm1, %zmm0, %zmm0
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vmovapd %zmm0, %zmm0 {%k1} {z}
+; SKX-NEXT: vxorpd %zmm1, %zmm0, %zmm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <8 x double> %__A to <8 x i64>
@@ -546,16 +542,14 @@ entry:
define <16 x float> @test_mm512_mask_xor_ps(<16 x float> %__W, i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B) {
; KNL-LABEL: test_mm512_mask_xor_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpxorq %zmm2, %zmm1, %zmm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmps %zmm1, %zmm0, %zmm0 {%k1}
+; KNL-NEXT: vpxord %zmm2, %zmm1, %zmm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm512_mask_xor_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vxorps %zmm2, %zmm1, %zmm1
; SKX-NEXT: kmovw %edi, %k1
-; SKX-NEXT: vblendmps %zmm1, %zmm0, %zmm0 {%k1}
+; SKX-NEXT: vxorps %zmm2, %zmm1, %zmm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <16 x float> %__A to <16 x i32>
@@ -570,16 +564,14 @@ entry:
define <16 x float> @test_mm512_maskz_xor_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B) {
; KNL-LABEL: test_mm512_maskz_xor_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpxorq %zmm1, %zmm0, %zmm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovaps %zmm0, %zmm0 {%k1} {z}
+; KNL-NEXT: vpxord %zmm1, %zmm0, %zmm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm512_maskz_xor_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vxorps %zmm1, %zmm0, %zmm0
; SKX-NEXT: kmovw %edi, %k1
-; SKX-NEXT: vmovaps %zmm0, %zmm0 {%k1} {z}
+; SKX-NEXT: vxorps %zmm1, %zmm0, %zmm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <16 x float> %__A to <16 x i32>
@@ -594,16 +586,14 @@ entry:
define <8 x double> @test_mm512_mask_or_pd(<8 x double> %__W, i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B) {
; KNL-LABEL: test_mm512_mask_or_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vporq %zmm1, %zmm2, %zmm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmpd %zmm1, %zmm0, %zmm0 {%k1}
+; KNL-NEXT: vporq %zmm1, %zmm2, %zmm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm512_mask_or_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vorpd %zmm1, %zmm2, %zmm1
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vblendmpd %zmm1, %zmm0, %zmm0 {%k1}
+; SKX-NEXT: vorpd %zmm1, %zmm2, %zmm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <8 x double> %__A to <8 x i64>
@@ -618,16 +608,14 @@ entry:
define <8 x double> @test_mm512_maskz_or_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B) {
; KNL-LABEL: test_mm512_maskz_or_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vporq %zmm0, %zmm1, %zmm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovapd %zmm0, %zmm0 {%k1} {z}
+; KNL-NEXT: vporq %zmm0, %zmm1, %zmm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm512_maskz_or_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vorpd %zmm0, %zmm1, %zmm0
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vmovapd %zmm0, %zmm0 {%k1} {z}
+; SKX-NEXT: vorpd %zmm0, %zmm1, %zmm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <8 x double> %__A to <8 x i64>
@@ -642,16 +630,14 @@ entry:
define <16 x float> @test_mm512_mask_or_ps(<16 x float> %__W, i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B) {
; KNL-LABEL: test_mm512_mask_or_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vporq %zmm1, %zmm2, %zmm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmps %zmm1, %zmm0, %zmm0 {%k1}
+; KNL-NEXT: vpord %zmm1, %zmm2, %zmm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm512_mask_or_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vorps %zmm1, %zmm2, %zmm1
; SKX-NEXT: kmovw %edi, %k1
-; SKX-NEXT: vblendmps %zmm1, %zmm0, %zmm0 {%k1}
+; SKX-NEXT: vorps %zmm1, %zmm2, %zmm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <16 x float> %__A to <16 x i32>
@@ -666,16 +652,14 @@ entry:
define <16 x float> @test_mm512_maskz_or_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B) {
; KNL-LABEL: test_mm512_maskz_or_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vporq %zmm0, %zmm1, %zmm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovaps %zmm0, %zmm0 {%k1} {z}
+; KNL-NEXT: vpord %zmm0, %zmm1, %zmm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm512_maskz_or_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vorps %zmm0, %zmm1, %zmm0
; SKX-NEXT: kmovw %edi, %k1
-; SKX-NEXT: vmovaps %zmm0, %zmm0 {%k1} {z}
+; SKX-NEXT: vorps %zmm0, %zmm1, %zmm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <16 x float> %__A to <16 x i32>
@@ -690,16 +674,14 @@ entry:
define <8 x double> @test_mm512_mask_and_pd(<8 x double> %__W, i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B) {
; KNL-LABEL: test_mm512_mask_and_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandq %zmm1, %zmm2, %zmm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmpd %zmm1, %zmm0, %zmm0 {%k1}
+; KNL-NEXT: vpandq %zmm1, %zmm2, %zmm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm512_mask_and_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandpd %zmm1, %zmm2, %zmm1
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vblendmpd %zmm1, %zmm0, %zmm0 {%k1}
+; SKX-NEXT: vandpd %zmm1, %zmm2, %zmm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <8 x double> %__A to <8 x i64>
@@ -714,16 +696,14 @@ entry:
define <8 x double> @test_mm512_maskz_and_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B) {
; KNL-LABEL: test_mm512_maskz_and_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandq %zmm0, %zmm1, %zmm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovapd %zmm0, %zmm0 {%k1} {z}
+; KNL-NEXT: vpandq %zmm0, %zmm1, %zmm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm512_maskz_and_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandpd %zmm0, %zmm1, %zmm0
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vmovapd %zmm0, %zmm0 {%k1} {z}
+; SKX-NEXT: vandpd %zmm0, %zmm1, %zmm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <8 x double> %__A to <8 x i64>
@@ -738,16 +718,14 @@ entry:
define <16 x float> @test_mm512_mask_and_ps(<16 x float> %__W, i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B) {
; KNL-LABEL: test_mm512_mask_and_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandq %zmm1, %zmm2, %zmm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmps %zmm1, %zmm0, %zmm0 {%k1}
+; KNL-NEXT: vpandd %zmm1, %zmm2, %zmm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm512_mask_and_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandps %zmm1, %zmm2, %zmm1
; SKX-NEXT: kmovw %edi, %k1
-; SKX-NEXT: vblendmps %zmm1, %zmm0, %zmm0 {%k1}
+; SKX-NEXT: vandps %zmm1, %zmm2, %zmm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <16 x float> %__A to <16 x i32>
@@ -762,16 +740,14 @@ entry:
define <16 x float> @test_mm512_maskz_and_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B) {
; KNL-LABEL: test_mm512_maskz_and_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandq %zmm0, %zmm1, %zmm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovaps %zmm0, %zmm0 {%k1} {z}
+; KNL-NEXT: vpandd %zmm0, %zmm1, %zmm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm512_maskz_and_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandps %zmm0, %zmm1, %zmm0
; SKX-NEXT: kmovw %edi, %k1
-; SKX-NEXT: vmovaps %zmm0, %zmm0 {%k1} {z}
+; SKX-NEXT: vandps %zmm0, %zmm1, %zmm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <16 x float> %__A to <16 x i32>
@@ -786,16 +762,14 @@ entry:
define <8 x double> @test_mm512_mask_andnot_pd(<8 x double> %__W, i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B) {
; KNL-LABEL: test_mm512_mask_andnot_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandnq %zmm2, %zmm1, %zmm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmpd %zmm1, %zmm0, %zmm0 {%k1}
+; KNL-NEXT: vpandnq %zmm2, %zmm1, %zmm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm512_mask_andnot_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandnpd %zmm2, %zmm1, %zmm1
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vblendmpd %zmm1, %zmm0, %zmm0 {%k1}
+; SKX-NEXT: vandnpd %zmm2, %zmm1, %zmm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <8 x double> %__A to <8 x i64>
@@ -811,16 +785,14 @@ entry:
define <8 x double> @test_mm512_maskz_andnot_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B) {
; KNL-LABEL: test_mm512_maskz_andnot_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovapd %zmm0, %zmm0 {%k1} {z}
+; KNL-NEXT: vpandnq %zmm1, %zmm0, %zmm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm512_maskz_andnot_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandnpd %zmm1, %zmm0, %zmm0
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vmovapd %zmm0, %zmm0 {%k1} {z}
+; SKX-NEXT: vandnpd %zmm1, %zmm0, %zmm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <8 x double> %__A to <8 x i64>
@@ -836,16 +808,14 @@ entry:
define <16 x float> @test_mm512_mask_andnot_ps(<16 x float> %__W, i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B) {
; KNL-LABEL: test_mm512_mask_andnot_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandnq %zmm2, %zmm1, %zmm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmps %zmm1, %zmm0, %zmm0 {%k1}
+; KNL-NEXT: vpandnd %zmm2, %zmm1, %zmm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm512_mask_andnot_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandnps %zmm2, %zmm1, %zmm1
; SKX-NEXT: kmovw %edi, %k1
-; SKX-NEXT: vblendmps %zmm1, %zmm0, %zmm0 {%k1}
+; SKX-NEXT: vandnps %zmm2, %zmm1, %zmm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <16 x float> %__A to <16 x i32>
@@ -861,16 +831,14 @@ entry:
define <16 x float> @test_mm512_maskz_andnot_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B) {
; KNL-LABEL: test_mm512_maskz_andnot_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovaps %zmm0, %zmm0 {%k1} {z}
+; KNL-NEXT: vpandnd %zmm1, %zmm0, %zmm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm512_maskz_andnot_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandnps %zmm1, %zmm0, %zmm0
; SKX-NEXT: kmovw %edi, %k1
-; SKX-NEXT: vmovaps %zmm0, %zmm0 {%k1} {z}
+; SKX-NEXT: vandnps %zmm1, %zmm0, %zmm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <16 x float> %__A to <16 x i32>
diff --git a/llvm/test/CodeGen/X86/avx512vl-logic.ll b/llvm/test/CodeGen/X86/avx512vl-logic.ll
index 011497eba8a..00fb5407ed9 100644
--- a/llvm/test/CodeGen/X86/avx512vl-logic.ll
+++ b/llvm/test/CodeGen/X86/avx512vl-logic.ll
@@ -222,16 +222,14 @@ entry:
define <4 x double> @test_mm256_mask_andnot_pd(<4 x double> %__W, i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B) {
; KNL-LABEL: test_mm256_mask_andnot_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandnq %ymm2, %ymm1, %ymm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmpd %ymm1, %ymm0, %ymm0 {%k1}
+; KNL-NEXT: vpandnq %ymm2, %ymm1, %ymm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm256_mask_andnot_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandnpd %ymm2, %ymm1, %ymm1
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vblendmpd %ymm1, %ymm0, %ymm0 {%k1}
+; SKX-NEXT: vandnpd %ymm2, %ymm1, %ymm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <4 x double> %__A to <4 x i64>
@@ -248,16 +246,14 @@ entry:
define <4 x double> @test_mm256_maskz_andnot_pd(i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B) {
; KNL-LABEL: test_mm256_maskz_andnot_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandnq %ymm1, %ymm0, %ymm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovapd %ymm0, %ymm0 {%k1} {z}
+; KNL-NEXT: vpandnq %ymm1, %ymm0, %ymm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm256_maskz_andnot_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandnpd %ymm1, %ymm0, %ymm0
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vmovapd %ymm0, %ymm0 {%k1} {z}
+; SKX-NEXT: vandnpd %ymm1, %ymm0, %ymm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <4 x double> %__A to <4 x i64>
@@ -274,16 +270,14 @@ entry:
define <2 x double> @test_mm_mask_andnot_pd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
; KNL-LABEL: test_mm_mask_andnot_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandnq %xmm2, %xmm1, %xmm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmpd %xmm1, %xmm0, %xmm0 {%k1}
+; KNL-NEXT: vpandnq %xmm2, %xmm1, %xmm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm_mask_andnot_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandnpd %xmm2, %xmm1, %xmm1
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vblendmpd %xmm1, %xmm0, %xmm0 {%k1}
+; SKX-NEXT: vandnpd %xmm2, %xmm1, %xmm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <2 x double> %__A to <2 x i64>
@@ -300,16 +294,14 @@ entry:
define <2 x double> @test_mm_maskz_andnot_pd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
; KNL-LABEL: test_mm_maskz_andnot_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandnq %xmm1, %xmm0, %xmm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovapd %xmm0, %xmm0 {%k1} {z}
+; KNL-NEXT: vpandnq %xmm1, %xmm0, %xmm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm_maskz_andnot_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandnpd %xmm1, %xmm0, %xmm0
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vmovapd %xmm0, %xmm0 {%k1} {z}
+; SKX-NEXT: vandnpd %xmm1, %xmm0, %xmm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <2 x double> %__A to <2 x i64>
@@ -326,16 +318,14 @@ entry:
define <8 x float> @test_mm256_mask_andnot_ps(<8 x float> %__W, i8 zeroext %__U, <8 x float> %__A, <8 x float> %__B) {
; KNL-LABEL: test_mm256_mask_andnot_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandnq %ymm2, %ymm1, %ymm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmps %ymm1, %ymm0, %ymm0 {%k1}
+; KNL-NEXT: vpandnd %ymm2, %ymm1, %ymm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm256_mask_andnot_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandnps %ymm2, %ymm1, %ymm1
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vblendmps %ymm1, %ymm0, %ymm0 {%k1}
+; SKX-NEXT: vandnps %ymm2, %ymm1, %ymm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <8 x float> %__A to <8 x i32>
@@ -351,16 +341,14 @@ entry:
define <8 x float> @test_mm256_maskz_andnot_ps(i8 zeroext %__U, <8 x float> %__A, <8 x float> %__B) {
; KNL-LABEL: test_mm256_maskz_andnot_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandnq %ymm1, %ymm0, %ymm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovaps %ymm0, %ymm0 {%k1} {z}
+; KNL-NEXT: vpandnd %ymm1, %ymm0, %ymm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm256_maskz_andnot_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandnps %ymm1, %ymm0, %ymm0
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vmovaps %ymm0, %ymm0 {%k1} {z}
+; SKX-NEXT: vandnps %ymm1, %ymm0, %ymm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <8 x float> %__A to <8 x i32>
@@ -376,16 +364,14 @@ entry:
define <4 x float> @test_mm_mask_andnot_ps(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
; KNL-LABEL: test_mm_mask_andnot_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandnq %xmm2, %xmm1, %xmm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmps %xmm1, %xmm0, %xmm0 {%k1}
+; KNL-NEXT: vpandnd %xmm2, %xmm1, %xmm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm_mask_andnot_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandnps %xmm2, %xmm1, %xmm1
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vblendmps %xmm1, %xmm0, %xmm0 {%k1}
+; SKX-NEXT: vandnps %xmm2, %xmm1, %xmm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <4 x float> %__A to <4 x i32>
@@ -402,16 +388,14 @@ entry:
define <4 x float> @test_mm_maskz_andnot_ps(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
; KNL-LABEL: test_mm_maskz_andnot_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandnq %xmm1, %xmm0, %xmm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovaps %xmm0, %xmm0 {%k1} {z}
+; KNL-NEXT: vpandnd %xmm1, %xmm0, %xmm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm_maskz_andnot_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandnps %xmm1, %xmm0, %xmm0
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vmovaps %xmm0, %xmm0 {%k1} {z}
+; SKX-NEXT: vandnps %xmm1, %xmm0, %xmm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <4 x float> %__A to <4 x i32>
@@ -428,16 +412,14 @@ entry:
define <4 x double> @test_mm256_mask_and_pd(<4 x double> %__W, i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B) {
; KNL-LABEL: test_mm256_mask_and_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandq %ymm1, %ymm2, %ymm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmpd %ymm1, %ymm0, %ymm0 {%k1}
+; KNL-NEXT: vpandq %ymm1, %ymm2, %ymm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm256_mask_and_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandpd %ymm1, %ymm2, %ymm1
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vblendmpd %ymm1, %ymm0, %ymm0 {%k1}
+; SKX-NEXT: vandpd %ymm1, %ymm2, %ymm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <4 x double> %__A to <4 x i64>
@@ -453,16 +435,14 @@ entry:
define <4 x double> @test_mm256_maskz_and_pd(i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B) {
; KNL-LABEL: test_mm256_maskz_and_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandq %ymm0, %ymm1, %ymm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovapd %ymm0, %ymm0 {%k1} {z}
+; KNL-NEXT: vpandq %ymm0, %ymm1, %ymm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm256_maskz_and_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandpd %ymm0, %ymm1, %ymm0
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vmovapd %ymm0, %ymm0 {%k1} {z}
+; SKX-NEXT: vandpd %ymm0, %ymm1, %ymm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <4 x double> %__A to <4 x i64>
@@ -478,16 +458,14 @@ entry:
define <2 x double> @test_mm_mask_and_pd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
; KNL-LABEL: test_mm_mask_and_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandq %xmm1, %xmm2, %xmm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmpd %xmm1, %xmm0, %xmm0 {%k1}
+; KNL-NEXT: vpandq %xmm1, %xmm2, %xmm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm_mask_and_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandpd %xmm1, %xmm2, %xmm1
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vblendmpd %xmm1, %xmm0, %xmm0 {%k1}
+; SKX-NEXT: vandpd %xmm1, %xmm2, %xmm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <2 x double> %__A to <2 x i64>
@@ -503,16 +481,14 @@ entry:
define <2 x double> @test_mm_maskz_and_pd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
; KNL-LABEL: test_mm_maskz_and_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandq %xmm0, %xmm1, %xmm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovapd %xmm0, %xmm0 {%k1} {z}
+; KNL-NEXT: vpandq %xmm0, %xmm1, %xmm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm_maskz_and_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandpd %xmm0, %xmm1, %xmm0
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vmovapd %xmm0, %xmm0 {%k1} {z}
+; SKX-NEXT: vandpd %xmm0, %xmm1, %xmm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <2 x double> %__A to <2 x i64>
@@ -528,16 +504,14 @@ entry:
define <8 x float> @test_mm256_mask_and_ps(<8 x float> %__W, i8 zeroext %__U, <8 x float> %__A, <8 x float> %__B) {
; KNL-LABEL: test_mm256_mask_and_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandq %ymm1, %ymm2, %ymm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmps %ymm1, %ymm0, %ymm0 {%k1}
+; KNL-NEXT: vpandd %ymm1, %ymm2, %ymm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm256_mask_and_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandps %ymm1, %ymm2, %ymm1
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vblendmps %ymm1, %ymm0, %ymm0 {%k1}
+; SKX-NEXT: vandps %ymm1, %ymm2, %ymm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <8 x float> %__A to <8 x i32>
@@ -552,16 +526,14 @@ entry:
define <8 x float> @test_mm256_maskz_and_ps(i8 zeroext %__U, <8 x float> %__A, <8 x float> %__B) {
; KNL-LABEL: test_mm256_maskz_and_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandq %ymm0, %ymm1, %ymm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovaps %ymm0, %ymm0 {%k1} {z}
+; KNL-NEXT: vpandd %ymm0, %ymm1, %ymm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm256_maskz_and_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandps %ymm0, %ymm1, %ymm0
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vmovaps %ymm0, %ymm0 {%k1} {z}
+; SKX-NEXT: vandps %ymm0, %ymm1, %ymm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <8 x float> %__A to <8 x i32>
@@ -576,16 +548,14 @@ entry:
define <4 x float> @test_mm_mask_and_ps(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
; KNL-LABEL: test_mm_mask_and_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandq %xmm1, %xmm2, %xmm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmps %xmm1, %xmm0, %xmm0 {%k1}
+; KNL-NEXT: vpandd %xmm1, %xmm2, %xmm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm_mask_and_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandps %xmm1, %xmm2, %xmm1
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vblendmps %xmm1, %xmm0, %xmm0 {%k1}
+; SKX-NEXT: vandps %xmm1, %xmm2, %xmm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <4 x float> %__A to <4 x i32>
@@ -601,16 +571,14 @@ entry:
define <4 x float> @test_mm_maskz_and_ps(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
; KNL-LABEL: test_mm_maskz_and_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpandq %xmm0, %xmm1, %xmm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovaps %xmm0, %xmm0 {%k1} {z}
+; KNL-NEXT: vpandd %xmm0, %xmm1, %xmm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm_maskz_and_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vandps %xmm0, %xmm1, %xmm0
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vmovaps %xmm0, %xmm0 {%k1} {z}
+; SKX-NEXT: vandps %xmm0, %xmm1, %xmm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <4 x float> %__A to <4 x i32>
@@ -626,16 +594,14 @@ entry:
define <4 x double> @test_mm256_mask_xor_pd(<4 x double> %__W, i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B) {
; KNL-LABEL: test_mm256_mask_xor_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpxorq %ymm2, %ymm1, %ymm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmpd %ymm1, %ymm0, %ymm0 {%k1}
+; KNL-NEXT: vpxorq %ymm2, %ymm1, %ymm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm256_mask_xor_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vxorpd %ymm2, %ymm1, %ymm1
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vblendmpd %ymm1, %ymm0, %ymm0 {%k1}
+; SKX-NEXT: vxorpd %ymm2, %ymm1, %ymm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <4 x double> %__A to <4 x i64>
@@ -651,16 +617,14 @@ entry:
define <4 x double> @test_mm256_maskz_xor_pd(i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B) {
; KNL-LABEL: test_mm256_maskz_xor_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpxorq %ymm1, %ymm0, %ymm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovapd %ymm0, %ymm0 {%k1} {z}
+; KNL-NEXT: vpxorq %ymm1, %ymm0, %ymm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm256_maskz_xor_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vxorpd %ymm1, %ymm0, %ymm0
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vmovapd %ymm0, %ymm0 {%k1} {z}
+; SKX-NEXT: vxorpd %ymm1, %ymm0, %ymm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <4 x double> %__A to <4 x i64>
@@ -676,16 +640,14 @@ entry:
define <2 x double> @test_mm_mask_xor_pd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
; KNL-LABEL: test_mm_mask_xor_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpxorq %xmm2, %xmm1, %xmm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmpd %xmm1, %xmm0, %xmm0 {%k1}
+; KNL-NEXT: vpxorq %xmm2, %xmm1, %xmm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm_mask_xor_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vxorpd %xmm2, %xmm1, %xmm1
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vblendmpd %xmm1, %xmm0, %xmm0 {%k1}
+; SKX-NEXT: vxorpd %xmm2, %xmm1, %xmm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <2 x double> %__A to <2 x i64>
@@ -701,16 +663,14 @@ entry:
define <2 x double> @test_mm_maskz_xor_pd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
; KNL-LABEL: test_mm_maskz_xor_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpxorq %xmm1, %xmm0, %xmm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovapd %xmm0, %xmm0 {%k1} {z}
+; KNL-NEXT: vpxorq %xmm1, %xmm0, %xmm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm_maskz_xor_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vxorpd %xmm1, %xmm0, %xmm0
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vmovapd %xmm0, %xmm0 {%k1} {z}
+; SKX-NEXT: vxorpd %xmm1, %xmm0, %xmm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <2 x double> %__A to <2 x i64>
@@ -726,16 +686,14 @@ entry:
define <8 x float> @test_mm256_mask_xor_ps(<8 x float> %__W, i8 zeroext %__U, <8 x float> %__A, <8 x float> %__B) {
; KNL-LABEL: test_mm256_mask_xor_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpxorq %ymm2, %ymm1, %ymm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmps %ymm1, %ymm0, %ymm0 {%k1}
+; KNL-NEXT: vpxord %ymm2, %ymm1, %ymm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm256_mask_xor_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vxorps %ymm2, %ymm1, %ymm1
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vblendmps %ymm1, %ymm0, %ymm0 {%k1}
+; SKX-NEXT: vxorps %ymm2, %ymm1, %ymm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <8 x float> %__A to <8 x i32>
@@ -750,16 +708,14 @@ entry:
define <8 x float> @test_mm256_maskz_xor_ps(i8 zeroext %__U, <8 x float> %__A, <8 x float> %__B) {
; KNL-LABEL: test_mm256_maskz_xor_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpxorq %ymm1, %ymm0, %ymm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovaps %ymm0, %ymm0 {%k1} {z}
+; KNL-NEXT: vpxord %ymm1, %ymm0, %ymm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm256_maskz_xor_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vxorps %ymm1, %ymm0, %ymm0
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vmovaps %ymm0, %ymm0 {%k1} {z}
+; SKX-NEXT: vxorps %ymm1, %ymm0, %ymm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <8 x float> %__A to <8 x i32>
@@ -774,16 +730,14 @@ entry:
define <4 x float> @test_mm_mask_xor_ps(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
; KNL-LABEL: test_mm_mask_xor_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpxorq %xmm2, %xmm1, %xmm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmps %xmm1, %xmm0, %xmm0 {%k1}
+; KNL-NEXT: vpxord %xmm2, %xmm1, %xmm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm_mask_xor_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vxorps %xmm2, %xmm1, %xmm1
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vblendmps %xmm1, %xmm0, %xmm0 {%k1}
+; SKX-NEXT: vxorps %xmm2, %xmm1, %xmm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <4 x float> %__A to <4 x i32>
@@ -799,16 +753,14 @@ entry:
define <4 x float> @test_mm_maskz_xor_ps(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
; KNL-LABEL: test_mm_maskz_xor_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vpxorq %xmm1, %xmm0, %xmm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovaps %xmm0, %xmm0 {%k1} {z}
+; KNL-NEXT: vpxord %xmm1, %xmm0, %xmm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm_maskz_xor_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vxorps %xmm1, %xmm0, %xmm0
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vmovaps %xmm0, %xmm0 {%k1} {z}
+; SKX-NEXT: vxorps %xmm1, %xmm0, %xmm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <4 x float> %__A to <4 x i32>
@@ -824,16 +776,14 @@ entry:
define <4 x double> @test_mm256_mask_or_pd(<4 x double> %__W, i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B) {
; KNL-LABEL: test_mm256_mask_or_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vporq %ymm1, %ymm2, %ymm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmpd %ymm1, %ymm0, %ymm0 {%k1}
+; KNL-NEXT: vporq %ymm1, %ymm2, %ymm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm256_mask_or_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vorpd %ymm1, %ymm2, %ymm1
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vblendmpd %ymm1, %ymm0, %ymm0 {%k1}
+; SKX-NEXT: vorpd %ymm1, %ymm2, %ymm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <4 x double> %__A to <4 x i64>
@@ -849,16 +799,14 @@ entry:
define <4 x double> @test_mm256_maskz_or_pd(i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B) {
; KNL-LABEL: test_mm256_maskz_or_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vporq %ymm0, %ymm1, %ymm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovapd %ymm0, %ymm0 {%k1} {z}
+; KNL-NEXT: vporq %ymm0, %ymm1, %ymm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm256_maskz_or_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vorpd %ymm0, %ymm1, %ymm0
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vmovapd %ymm0, %ymm0 {%k1} {z}
+; SKX-NEXT: vorpd %ymm0, %ymm1, %ymm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <4 x double> %__A to <4 x i64>
@@ -874,16 +822,14 @@ entry:
define <2 x double> @test_mm_mask_or_pd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
; KNL-LABEL: test_mm_mask_or_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vporq %xmm1, %xmm2, %xmm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmpd %xmm1, %xmm0, %xmm0 {%k1}
+; KNL-NEXT: vporq %xmm1, %xmm2, %xmm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm_mask_or_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vorpd %xmm1, %xmm2, %xmm1
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vblendmpd %xmm1, %xmm0, %xmm0 {%k1}
+; SKX-NEXT: vorpd %xmm1, %xmm2, %xmm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <2 x double> %__A to <2 x i64>
@@ -899,16 +845,14 @@ entry:
define <2 x double> @test_mm_maskz_or_pd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
; KNL-LABEL: test_mm_maskz_or_pd:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vporq %xmm0, %xmm1, %xmm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovapd %xmm0, %xmm0 {%k1} {z}
+; KNL-NEXT: vporq %xmm0, %xmm1, %xmm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm_maskz_or_pd:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vorpd %xmm0, %xmm1, %xmm0
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vmovapd %xmm0, %xmm0 {%k1} {z}
+; SKX-NEXT: vorpd %xmm0, %xmm1, %xmm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <2 x double> %__A to <2 x i64>
@@ -924,16 +868,14 @@ entry:
define <8 x float> @test_mm256_mask_or_ps(<8 x float> %__W, i8 zeroext %__U, <8 x float> %__A, <8 x float> %__B) {
; KNL-LABEL: test_mm256_mask_or_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vporq %ymm1, %ymm2, %ymm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmps %ymm1, %ymm0, %ymm0 {%k1}
+; KNL-NEXT: vpord %ymm1, %ymm2, %ymm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm256_mask_or_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vorps %ymm1, %ymm2, %ymm1
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vblendmps %ymm1, %ymm0, %ymm0 {%k1}
+; SKX-NEXT: vorps %ymm1, %ymm2, %ymm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <8 x float> %__A to <8 x i32>
@@ -948,16 +890,14 @@ entry:
define <8 x float> @test_mm256_maskz_or_ps(i8 zeroext %__U, <8 x float> %__A, <8 x float> %__B) {
; KNL-LABEL: test_mm256_maskz_or_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vporq %ymm0, %ymm1, %ymm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovaps %ymm0, %ymm0 {%k1} {z}
+; KNL-NEXT: vpord %ymm0, %ymm1, %ymm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm256_maskz_or_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vorps %ymm0, %ymm1, %ymm0
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vmovaps %ymm0, %ymm0 {%k1} {z}
+; SKX-NEXT: vorps %ymm0, %ymm1, %ymm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <8 x float> %__A to <8 x i32>
@@ -972,16 +912,14 @@ entry:
define <4 x float> @test_mm_mask_or_ps(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
; KNL-LABEL: test_mm_mask_or_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vporq %xmm1, %xmm2, %xmm1
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vblendmps %xmm1, %xmm0, %xmm0 {%k1}
+; KNL-NEXT: vpord %xmm1, %xmm2, %xmm0 {%k1}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm_mask_or_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vorps %xmm1, %xmm2, %xmm1
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vblendmps %xmm1, %xmm0, %xmm0 {%k1}
+; SKX-NEXT: vorps %xmm1, %xmm2, %xmm0 {%k1}
; SKX-NEXT: retq
entry:
%0 = bitcast <4 x float> %__A to <4 x i32>
@@ -997,16 +935,14 @@ entry:
define <4 x float> @test_mm_maskz_or_ps(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
; KNL-LABEL: test_mm_maskz_or_ps:
; KNL: ## BB#0: ## %entry
-; KNL-NEXT: vporq %xmm0, %xmm1, %xmm0
; KNL-NEXT: kmovw %edi, %k1
-; KNL-NEXT: vmovaps %xmm0, %xmm0 {%k1} {z}
+; KNL-NEXT: vpord %xmm0, %xmm1, %xmm0 {%k1} {z}
; KNL-NEXT: retq
;
; SKX-LABEL: test_mm_maskz_or_ps:
; SKX: ## BB#0: ## %entry
-; SKX-NEXT: vorps %xmm0, %xmm1, %xmm0
; SKX-NEXT: kmovb %edi, %k1
-; SKX-NEXT: vmovaps %xmm0, %xmm0 {%k1} {z}
+; SKX-NEXT: vorps %xmm0, %xmm1, %xmm0 {%k1} {z}
; SKX-NEXT: retq
entry:
%0 = bitcast <4 x float> %__A to <4 x i32>
OpenPOWER on IntegriCloud