[X86] Add patterns for using zmm registers for v8i32/v8f32 vselect with the false...

We can use zmm move with zero masking for this. We already had patterns for using a masked move, but we didn't check for the zero masking case separately.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@321612 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td
index 0b31f876b386e30267ea56b68091f210674a156a..1f58b28d153f65e465e95fecd9c64171a629ca6c 100644 (file)
--- a/lib/Target/X86/X86InstrAVX512.td
+++ b/lib/Target/X86/X86InstrAVX512.td
@@ -3351,28 +3351,33 @@ def : Pat<(v16i32 (vselect (xor VK16:$mask, (v16i1 immAllOnesV)),
                            (v16i32 VR512:$src))),
                   (VMOVDQA32Zrrkz VK16WM:$mask, VR512:$src)>;
 
+multiclass mask_move_lowering<string InstrStr, X86VectorVTInfo Narrow,
+                              X86VectorVTInfo Wide> {
+ def : Pat<(Narrow.VT (vselect (Narrow.KVT Narrow.KRCWM:$mask),
+                               Narrow.RC:$src1, Narrow.RC:$src0)),
+           (EXTRACT_SUBREG
+            (Wide.VT
+             (!cast<Instruction>(InstrStr#"rrk")
+              (Wide.VT (INSERT_SUBREG (IMPLICIT_DEF), Narrow.RC:$src0, Narrow.SubRegIdx)),
+              (COPY_TO_REGCLASS Narrow.KRCWM:$mask, Wide.KRCWM),
+              (Wide.VT (INSERT_SUBREG (IMPLICIT_DEF), Narrow.RC:$src1, Narrow.SubRegIdx)))),
+            Narrow.SubRegIdx)>;
+
+ def : Pat<(Narrow.VT (vselect (Narrow.KVT Narrow.KRCWM:$mask),
+                               Narrow.RC:$src1, Narrow.ImmAllZerosV)),
+           (EXTRACT_SUBREG
+            (Wide.VT
+             (!cast<Instruction>(InstrStr#"rrkz")
+              (COPY_TO_REGCLASS Narrow.KRCWM:$mask, Wide.KRCWM),
+              (Wide.VT (INSERT_SUBREG (IMPLICIT_DEF), Narrow.RC:$src1, Narrow.SubRegIdx)))),
+            Narrow.SubRegIdx)>;
+}
+
 // Patterns for handling v8i1 selects of 256-bit vectors when VLX isn't
 // available. Use a 512-bit operation and extract.
 let Predicates = [HasAVX512, NoVLX] in {
-def : Pat<(v8f32 (vselect (v8i1 VK8WM:$mask), (v8f32 VR256X:$src1),
-                          (v8f32 VR256X:$src0))),
-          (EXTRACT_SUBREG
-           (v16f32
-            (VMOVAPSZrrk
-             (v16f32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src0, sub_ymm)),
-             (COPY_TO_REGCLASS VK8WM:$mask, VK16WM),
-             (v16f32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src1, sub_ymm)))),
-           sub_ymm)>;
-
-def : Pat<(v8i32 (vselect (v8i1 VK8WM:$mask), (v8i32 VR256X:$src1),
-                          (v8i32 VR256X:$src0))),
-          (EXTRACT_SUBREG
-           (v16i32
-            (VMOVDQA32Zrrk
-             (v16i32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src0, sub_ymm)),
-             (COPY_TO_REGCLASS VK8WM:$mask, VK16WM),
-             (v16i32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src1, sub_ymm)))),
-           sub_ymm)>;
+  defm : mask_move_lowering<"VMOVAPSZ", v8f32x_info, v16f32_info>;
+  defm : mask_move_lowering<"VMOVDQA32Z", v8i32x_info, v16i32_info>;
 }
 
 let Predicates = [HasAVX512] in {

diff --git a/test/CodeGen/X86/avx512-ext.ll b/test/CodeGen/X86/avx512-ext.ll
index 8c79415912179222b8ff7c50eda25d6b58c59012..a966235df2160c30a4bd14ead9e0c2b439fc194f 100644 (file)
--- a/test/CodeGen/X86/avx512-ext.ll
+++ b/test/CodeGen/X86/avx512-ext.ll
@@ -345,9 +345,8 @@ define <8 x i32> @zext_8x8mem_to_8x32(<8 x i8> *%i , <8 x i1> %mask) nounwind re
 ; KNL-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; KNL-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmq %zmm0, %zmm0, %k1
-; KNL-NEXT:    vpmovzxbd {{.*#+}} ymm1 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero,mem[4],zero,zero,zero,mem[5],zero,zero,zero,mem[6],zero,zero,zero,mem[7],zero,zero,zero
-; KNL-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; KNL-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
+; KNL-NEXT:    vpmovzxbd {{.*#+}} ymm0 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero,mem[4],zero,zero,zero,mem[5],zero,zero,zero,mem[6],zero,zero,zero,mem[7],zero,zero,zero
+; KNL-NEXT:    vmovdqa32 %zmm0, %zmm0 {%k1} {z}
 ; KNL-NEXT:    # kill: def %ymm0 killed %ymm0 killed %zmm0
 ; KNL-NEXT:    retq
 ;
@@ -369,9 +368,8 @@ define <8 x i32> @sext_8x8mem_to_8x32(<8 x i8> *%i , <8 x i1> %mask) nounwind re
 ; KNL-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; KNL-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmq %zmm0, %zmm0, %k1
-; KNL-NEXT:    vpmovsxbd (%rdi), %ymm1
-; KNL-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; KNL-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
+; KNL-NEXT:    vpmovsxbd (%rdi), %ymm0
+; KNL-NEXT:    vmovdqa32 %zmm0, %zmm0 {%k1} {z}
 ; KNL-NEXT:    # kill: def %ymm0 killed %ymm0 killed %zmm0
 ; KNL-NEXT:    retq
 ;
@@ -702,9 +700,8 @@ define <8 x i32> @zext_8x16mem_to_8x32(<8 x i16> *%i , <8 x i1> %mask) nounwind
 ; KNL-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; KNL-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmq %zmm0, %zmm0, %k1
-; KNL-NEXT:    vpmovzxwd {{.*#+}} ymm1 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero
-; KNL-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; KNL-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
+; KNL-NEXT:    vpmovzxwd {{.*#+}} ymm0 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero
+; KNL-NEXT:    vmovdqa32 %zmm0, %zmm0 {%k1} {z}
 ; KNL-NEXT:    # kill: def %ymm0 killed %ymm0 killed %zmm0
 ; KNL-NEXT:    retq
 ;
@@ -726,9 +723,8 @@ define <8 x i32> @sext_8x16mem_to_8x32mask(<8 x i16> *%i , <8 x i1> %mask) nounw
 ; KNL-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; KNL-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmq %zmm0, %zmm0, %k1
-; KNL-NEXT:    vpmovsxwd (%rdi), %ymm1
-; KNL-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; KNL-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
+; KNL-NEXT:    vpmovsxwd (%rdi), %ymm0
+; KNL-NEXT:    vmovdqa32 %zmm0, %zmm0 {%k1} {z}
 ; KNL-NEXT:    # kill: def %ymm0 killed %ymm0 killed %zmm0
 ; KNL-NEXT:    retq
 ;
@@ -760,9 +756,8 @@ define <8 x i32> @zext_8x16_to_8x32mask(<8 x i16> %a , <8 x i1> %mask) nounwind
 ; KNL-NEXT:    vpmovsxwq %xmm1, %zmm1
 ; KNL-NEXT:    vpsllq $63, %zmm1, %zmm1
 ; KNL-NEXT:    vptestmq %zmm1, %zmm1, %k1
-; KNL-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; KNL-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; KNL-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
+; KNL-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; KNL-NEXT:    vmovdqa32 %zmm0, %zmm0 {%k1} {z}
 ; KNL-NEXT:    # kill: def %ymm0 killed %ymm0 killed %zmm0
 ; KNL-NEXT:    retq
 ;