[X86] Add more patterns to use moves to zero the upper portions of a vector register that I missed in r312450.

llvm-svn: 312459

llvm/lib/Target/X86/X86InstrAVX512.td

@@ -4044,6 +4044,51 @@ let Predicates = [HasAVX512, NoVLX] in {
   // will zero the upper bits.
   // TODO: Is there a safe way to detect whether the producing instruction
   // already zeroed the upper bits?
+
+  // 128->512 register form.
+  def : Pat<(v8f64 (insert_subvector (bitconvert (v16i32 immAllZerosV)),
+                                     (v2f64 VR128:$src), (iPTR 0))),
+            (SUBREG_TO_REG (i64 0), (VMOVAPDrr VR128:$src), sub_xmm)>;
+  def : Pat<(v16f32 (insert_subvector (bitconvert (v16i32 immAllZerosV)),
+                                      (v4f32 VR128:$src), (iPTR 0))),
+            (SUBREG_TO_REG (i64 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
+  def : Pat<(v8i64 (insert_subvector (bitconvert (v16i32 immAllZerosV)),
+                                      (v2i64 VR128:$src), (iPTR 0))),
+            (SUBREG_TO_REG (i64 0), (VMOVDQArr VR128:$src), sub_xmm)>;
+  def : Pat<(v16i32 (insert_subvector (bitconvert (v16i32 immAllZerosV)),
+                                      (v4i32 VR128:$src), (iPTR 0))),
+            (SUBREG_TO_REG (i64 0), (VMOVDQArr VR128:$src), sub_xmm)>;
+  def : Pat<(v32i16 (insert_subvector (bitconvert (v16i32 immAllZerosV)),
+                                      (v8i16 VR128:$src), (iPTR 0))),
+            (SUBREG_TO_REG (i64 0), (VMOVDQArr VR128:$src), sub_xmm)>;
+  def : Pat<(v64i8 (insert_subvector (bitconvert (v16i32 immAllZerosV)),
+                                     (v16i8 VR128:$src), (iPTR 0))),
+            (SUBREG_TO_REG (i64 0), (VMOVDQArr VR128:$src), sub_xmm)>;
+
+  // 128->512 memory form.
+  def : Pat<(v8f64 (insert_subvector (bitconvert (v16i32 immAllZerosV)),
+                                     (loadv2f64 addr:$src), (iPTR 0))),
+            (SUBREG_TO_REG (i64 0), (VMOVAPDrm addr:$src), sub_xmm)>;
+  def : Pat<(v16f32 (insert_subvector (bitconvert (v16i32 immAllZerosV)),
+                                      (loadv4f32 addr:$src), (iPTR 0))),
+            (SUBREG_TO_REG (i64 0), (VMOVAPSrm addr:$src), sub_xmm)>;
+  def : Pat<(v8i64 (insert_subvector (bitconvert (v16i32 immAllZerosV)),
+                                      (loadv2i64 addr:$src), (iPTR 0))),
+            (SUBREG_TO_REG (i64 0), (VMOVDQArm addr:$src), sub_xmm)>;
+  def : Pat<(v16i32 (insert_subvector (bitconvert (v16i32 immAllZerosV)),
+                                      (bc_v4i32 (loadv2i64 addr:$src)),
+                                      (iPTR 0))),
+            (SUBREG_TO_REG (i64 0), (VMOVDQArm addr:$src), sub_xmm)>;
+  def : Pat<(v32i16 (insert_subvector (bitconvert (v16i32 immAllZerosV)),
+                                      (bc_v8i16 (loadv2i64 addr:$src)),
+                                      (iPTR 0))),
+            (SUBREG_TO_REG (i64 0), (VMOVDQArm addr:$src), sub_xmm)>;
+  def : Pat<(v64i8 (insert_subvector (bitconvert (v16i32 immAllZerosV)),
+                                     (bc_v16i8 (loadv2i64 addr:$src)),
+                                     (iPTR 0))),
+            (SUBREG_TO_REG (i64 0), (VMOVDQArm addr:$src), sub_xmm)>;
+
+  // 256->512 register form.
   def : Pat<(v8f64 (insert_subvector (bitconvert (v16i32 immAllZerosV)),
                                      (v4f64 VR256:$src), (iPTR 0))),
             (SUBREG_TO_REG (i64 0), (VMOVAPDYrr VR256:$src), sub_ymm)>;
@@ -4063,6 +4108,7 @@ let Predicates = [HasAVX512, NoVLX] in {
                                      (v32i8 VR256:$src), (iPTR 0))),
             (SUBREG_TO_REG (i64 0), (VMOVDQAYrr VR256:$src), sub_ymm)>;
 
+  // 256->512 memory form.
   def : Pat<(v8f64 (insert_subvector (bitconvert (v16i32 immAllZerosV)),
                                      (loadv4f64 addr:$src), (iPTR 0))),
             (SUBREG_TO_REG (i64 0), (VMOVAPDYrm addr:$src), sub_ymm)>;

llvm/test/CodeGen/X86/compress_expand.ll

@@ -226,8 +226,7 @@ define void @test11(i64* %base, <2 x i64> %V, <2 x i1> %mask) {
 ; KNL-NEXT:    # kill: %XMM0<def> %XMM0<kill> %ZMM0<def>
 ; KNL-NEXT:    vpsllq $63, %xmm1, %xmm1
 ; KNL-NEXT:    vpsraq $63, %zmm1, %zmm1
-; KNL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; KNL-NEXT:    vmovdqa %xmm1, %xmm1
-; KNL-NEXT:    vinserti32x4 $0, %xmm1, %zmm2, %zmm1
 ; KNL-NEXT:    vpsllq $63, %zmm1, %zmm1
 ; KNL-NEXT:    vptestmq %zmm1, %zmm1, %k1
 ; KNL-NEXT:    vpcompressq %zmm0, (%rdi) {%k1}
@@ -249,8 +248,7 @@ define void @test12(float* %base, <4 x float> %V, <4 x i1> %mask) {
 ; KNL-NEXT:    # kill: %XMM0<def> %XMM0<kill> %ZMM0<def>
 ; KNL-NEXT:    vpslld $31, %xmm1, %xmm1
 ; KNL-NEXT:    vpsrad $31, %xmm1, %xmm1
-; KNL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; KNL-NEXT:    vmovdqa %xmm1, %xmm1
-; KNL-NEXT:    vinserti32x4 $0, %xmm1, %zmm2, %zmm1
 ; KNL-NEXT:    vpslld $31, %zmm1, %zmm1
 ; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k1
 ; KNL-NEXT:    vcompressps %zmm0, (%rdi) {%k1}
@@ -275,8 +273,7 @@ define <2 x float> @test13(float* %base, <2 x float> %src0, <2 x i32> %trigger)
 ; KNL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
 ; KNL-NEXT:    vpcmpeqq %xmm2, %xmm1, %xmm1
 ; KNL-NEXT:    vinsertps {{.*#+}} xmm1 = xmm1[0,2],zero,zero
-; KNL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; KNL-NEXT:    vmovaps %xmm1, %xmm1
-; KNL-NEXT:    vinserti32x4 $0, %xmm1, %zmm2, %zmm1
 ; KNL-NEXT:    vpslld $31, %zmm1, %zmm1
 ; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k1
 ; KNL-NEXT:    vexpandps (%rdi), %zmm0 {%k1}
@@ -303,8 +300,7 @@ define void @test14(float* %base, <2 x float> %V, <2 x i32> %trigger) {
 ; KNL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
 ; KNL-NEXT:    vpcmpeqq %xmm2, %xmm1, %xmm1
 ; KNL-NEXT:    vinsertps {{.*#+}} xmm1 = xmm1[0,2],zero,zero
-; KNL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; KNL-NEXT:    vmovaps %xmm1, %xmm1
-; KNL-NEXT:    vinserti32x4 $0, %xmm1, %zmm2, %zmm1
 ; KNL-NEXT:    vpslld $31, %zmm1, %zmm1
 ; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k1
 ; KNL-NEXT:    vcompressps %zmm0, (%rdi) {%k1}

llvm/test/CodeGen/X86/masked_gather_scatter.ll

@@ -883,8 +883,7 @@ define <2 x double> @test17(double* %base, <2 x i32> %ind, <2 x i1> %mask, <2 x
 ; KNL_64:       # BB#0:
 ; KNL_64-NEXT:    # kill: %XMM2<def> %XMM2<kill> %ZMM2<def>
 ; KNL_64-NEXT:    # kill: %XMM0<def> %XMM0<kill> %ZMM0<def>
-; KNL_64-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; KNL_64-NEXT:    vmovdqa %xmm1, %xmm1
-; KNL_64-NEXT:    vinserti32x4 $0, %xmm1, %zmm3, %zmm1
 ; KNL_64-NEXT:    vpsllq $63, %zmm1, %zmm1
 ; KNL_64-NEXT:    vptestmq %zmm1, %zmm1, %k1
 ; KNL_64-NEXT:    vgatherqpd (%rdi,%zmm0,8), %zmm2 {%k1}
@@ -896,8 +895,7 @@ define <2 x double> @test17(double* %base, <2 x i32> %ind, <2 x i1> %mask, <2 x
 ; KNL_32:       # BB#0:
 ; KNL_32-NEXT:    # kill: %XMM2<def> %XMM2<kill> %ZMM2<def>
 ; KNL_32-NEXT:    # kill: %XMM0<def> %XMM0<kill> %ZMM0<def>
-; KNL_32-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; KNL_32-NEXT:    vmovdqa %xmm1, %xmm1
-; KNL_32-NEXT:    vinserti32x4 $0, %xmm1, %zmm3, %zmm1
 ; KNL_32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; KNL_32-NEXT:    vpsllq $63, %zmm1, %zmm1
 ; KNL_32-NEXT:    vptestmq %zmm1, %zmm1, %k1
@@ -1084,8 +1082,7 @@ define void @test21(<2 x i32>%a1, <2 x i32*> %ptr, <2 x i1>%mask) {
 ; KNL_64-LABEL: test21:
 ; KNL_64:       # BB#0:
 ; KNL_64-NEXT:    # kill: %XMM1<def> %XMM1<kill> %ZMM1<def>
-; KNL_64-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; KNL_64-NEXT:    vmovdqa %xmm2, %xmm2
-; KNL_64-NEXT:    vinserti32x4 $0, %xmm2, %zmm3, %zmm2
 ; KNL_64-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
 ; KNL_64-NEXT:    vpsllq $63, %zmm2, %zmm2
 ; KNL_64-NEXT:    vptestmq %zmm2, %zmm2, %k1
@@ -1096,8 +1093,7 @@ define void @test21(<2 x i32>%a1, <2 x i32*> %ptr, <2 x i1>%mask) {
 ; KNL_32-LABEL: test21:
 ; KNL_32:       # BB#0:
 ; KNL_32-NEXT:    # kill: %XMM1<def> %XMM1<kill> %ZMM1<def>
-; KNL_32-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; KNL_32-NEXT:    vmovdqa %xmm2, %xmm2
-; KNL_32-NEXT:    vinserti32x4 $0, %xmm2, %zmm3, %zmm2
 ; KNL_32-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
 ; KNL_32-NEXT:    vpsllq $63, %zmm2, %zmm2
 ; KNL_32-NEXT:    vptestmq %zmm2, %zmm2, %k1
@@ -1250,8 +1246,7 @@ define <2 x i32> @test23(i32* %base, <2 x i32> %ind, <2 x i1> %mask, <2 x i32> %
 ; KNL_64:       # BB#0:
 ; KNL_64-NEXT:    # kill: %XMM2<def> %XMM2<kill> %ZMM2<def>
 ; KNL_64-NEXT:    # kill: %XMM0<def> %XMM0<kill> %ZMM0<def>
-; KNL_64-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; KNL_64-NEXT:    vmovdqa %xmm1, %xmm1
-; KNL_64-NEXT:    vinserti32x4 $0, %xmm1, %zmm3, %zmm1
 ; KNL_64-NEXT:    vpsllq $63, %zmm1, %zmm1
 ; KNL_64-NEXT:    vptestmq %zmm1, %zmm1, %k1
 ; KNL_64-NEXT:    vpgatherqq (%rdi,%zmm0,8), %zmm2 {%k1}
@@ -1263,8 +1258,7 @@ define <2 x i32> @test23(i32* %base, <2 x i32> %ind, <2 x i1> %mask, <2 x i32> %
 ; KNL_32:       # BB#0:
 ; KNL_32-NEXT:    # kill: %XMM2<def> %XMM2<kill> %ZMM2<def>
 ; KNL_32-NEXT:    # kill: %XMM0<def> %XMM0<kill> %ZMM0<def>
-; KNL_32-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; KNL_32-NEXT:    vmovdqa %xmm1, %xmm1
-; KNL_32-NEXT:    vinserti32x4 $0, %xmm1, %zmm3, %zmm1
 ; KNL_32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; KNL_32-NEXT:    vpsllq $63, %zmm1, %zmm1
 ; KNL_32-NEXT:    vptestmq %zmm1, %zmm1, %k1
@@ -1312,8 +1306,7 @@ define <2 x i32> @test24(i32* %base, <2 x i32> %ind) {
 ; KNL_32:       # BB#0:
 ; KNL_32-NEXT:    # kill: %XMM0<def> %XMM0<kill> %ZMM0<def>
 ; KNL_32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; KNL_32-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; KNL_32-NEXT:    vmovdqa {{.*#+}} xmm1 = [1,0,1,0]
-; KNL_32-NEXT:    vinserti32x4 $0, {{\.LCPI.*}}, %zmm1, %zmm1
 ; KNL_32-NEXT:    vpsllq $63, %zmm1, %zmm1
 ; KNL_32-NEXT:    vptestmq %zmm1, %zmm1, %k1
 ; KNL_32-NEXT:    vpgatherqq (%eax,%zmm0,8), %zmm1 {%k1}
@@ -1346,8 +1339,7 @@ define <2 x i64> @test25(i64* %base, <2 x i32> %ind, <2 x i1> %mask, <2 x i64> %
 ; KNL_64:       # BB#0:
 ; KNL_64-NEXT:    # kill: %XMM2<def> %XMM2<kill> %ZMM2<def>
 ; KNL_64-NEXT:    # kill: %XMM0<def> %XMM0<kill> %ZMM0<def>
-; KNL_64-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; KNL_64-NEXT:    vmovdqa %xmm1, %xmm1
-; KNL_64-NEXT:    vinserti32x4 $0, %xmm1, %zmm3, %zmm1
 ; KNL_64-NEXT:    vpsllq $63, %zmm1, %zmm1
 ; KNL_64-NEXT:    vptestmq %zmm1, %zmm1, %k1
 ; KNL_64-NEXT:    vpgatherqq (%rdi,%zmm0,8), %zmm2 {%k1}
@@ -1359,8 +1351,7 @@ define <2 x i64> @test25(i64* %base, <2 x i32> %ind, <2 x i1> %mask, <2 x i64> %
 ; KNL_32:       # BB#0:
 ; KNL_32-NEXT:    # kill: %XMM2<def> %XMM2<kill> %ZMM2<def>
 ; KNL_32-NEXT:    # kill: %XMM0<def> %XMM0<kill> %ZMM0<def>
-; KNL_32-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; KNL_32-NEXT:    vmovdqa %xmm1, %xmm1
-; KNL_32-NEXT:    vinserti32x4 $0, %xmm1, %zmm3, %zmm1
 ; KNL_32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; KNL_32-NEXT:    vpsllq $63, %zmm1, %zmm1
 ; KNL_32-NEXT:    vptestmq %zmm1, %zmm1, %k1
@@ -1408,8 +1399,7 @@ define <2 x i64> @test26(i64* %base, <2 x i32> %ind, <2 x i64> %src0) {
 ; KNL_32-NEXT:    # kill: %XMM1<def> %XMM1<kill> %ZMM1<def>
 ; KNL_32-NEXT:    # kill: %XMM0<def> %XMM0<kill> %ZMM0<def>
 ; KNL_32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; KNL_32-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; KNL_32-NEXT:    vmovdqa {{.*#+}} xmm2 = [1,0,1,0]
-; KNL_32-NEXT:    vinserti32x4 $0, {{\.LCPI.*}}, %zmm2, %zmm2
 ; KNL_32-NEXT:    vpsllq $63, %zmm2, %zmm2
 ; KNL_32-NEXT:    vptestmq %zmm2, %zmm2, %k1
 ; KNL_32-NEXT:    vpgatherqq (%eax,%zmm0,8), %zmm1 {%k1}
@@ -1500,8 +1490,7 @@ define void @test28(<2 x i32>%a1, <2 x i32*> %ptr) {
 ; KNL_32:       # BB#0:
 ; KNL_32-NEXT:    # kill: %XMM1<def> %XMM1<kill> %ZMM1<def>
 ; KNL_32-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; KNL_32-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; KNL_32-NEXT:    vmovdqa {{.*#+}} xmm2 = [1,0,1,0]
-; KNL_32-NEXT:    vinserti32x4 $0, {{\.LCPI.*}}, %zmm2, %zmm2
 ; KNL_32-NEXT:    vpsllq $63, %zmm2, %zmm2
 ; KNL_32-NEXT:    vptestmq %zmm2, %zmm2, %k1
 ; KNL_32-NEXT:    vpscatterqd %ymm0, (,%zmm1) {%k1}