Make DAGCombiner work on vector bitshifts with constant splat vectors.

llvm-svn: 204071

2 files changed, 155 insertions, 24 deletions

diff --git a/llvm/test/CodeGen/X86/avx2-vector-shifts.ll b/llvm/test/CodeGen/X86/avx2-vector-shifts.ll
index 4868e4b4797..4ae2905ef22 100644
--- a/llvm/test/CodeGen/X86/avx2-vector-shifts.ll
+++ b/llvm/test/CodeGen/X86/avx2-vector-shifts.ll
@@ -9,7 +9,7 @@ entry:
 }
 
 ; CHECK-LABEL: test_sllw_1:
-; CHECK: vpsllw  $0, %ymm0, %ymm0
+; CHECK-NOT: vpsllw  $0, %ymm0, %ymm0
 ; CHECK: ret
 
 define <16 x i16> @test_sllw_2(<16 x i16> %InVec) {
@@ -39,7 +39,7 @@ entry:
 }
 
 ; CHECK-LABEL: test_slld_1:
-; CHECK: vpslld  $0, %ymm0, %ymm0
+; CHECK-NOT: vpslld  $0, %ymm0, %ymm0
 ; CHECK: ret
 
 define <8 x i32> @test_slld_2(<8 x i32> %InVec) {
@@ -69,7 +69,7 @@ entry:
 }
 
 ; CHECK-LABEL: test_sllq_1:
-; CHECK: vpsllq  $0, %ymm0, %ymm0
+; CHECK-NOT: vpsllq  $0, %ymm0, %ymm0
 ; CHECK: ret
 
 define <4 x i64> @test_sllq_2(<4 x i64> %InVec) {
@@ -101,7 +101,7 @@ entry:
 }
 
 ; CHECK-LABEL: test_sraw_1:
-; CHECK: vpsraw  $0, %ymm0, %ymm0
+; CHECK-NOT: vpsraw  $0, %ymm0, %ymm0
 ; CHECK: ret
 
 define <16 x i16> @test_sraw_2(<16 x i16> %InVec) {
@@ -131,7 +131,7 @@ entry:
 }
 
 ; CHECK-LABEL: test_srad_1:
-; CHECK: vpsrad  $0, %ymm0, %ymm0
+; CHECK-NOT: vpsrad  $0, %ymm0, %ymm0
 ; CHECK: ret
 
 define <8 x i32> @test_srad_2(<8 x i32> %InVec) {
@@ -163,7 +163,7 @@ entry:
 }
 
 ; CHECK-LABEL: test_srlw_1:
-; CHECK: vpsrlw  $0, %ymm0, %ymm0
+; CHECK-NOT: vpsrlw  $0, %ymm0, %ymm0
 ; CHECK: ret
 
 define <16 x i16> @test_srlw_2(<16 x i16> %InVec) {
@@ -193,7 +193,7 @@ entry:
 }
 
 ; CHECK-LABEL: test_srld_1:
-; CHECK: vpsrld  $0, %ymm0, %ymm0
+; CHECK-NOT: vpsrld  $0, %ymm0, %ymm0
 ; CHECK: ret
 
 define <8 x i32> @test_srld_2(<8 x i32> %InVec) {
@@ -223,7 +223,7 @@ entry:
 }
 
 ; CHECK-LABEL: test_srlq_1:
-; CHECK: vpsrlq  $0, %ymm0, %ymm0
+; CHECK-NOT: vpsrlq  $0, %ymm0, %ymm0
 ; CHECK: ret
 
 define <4 x i64> @test_srlq_2(<4 x i64> %InVec) {
@@ -245,3 +245,14 @@ entry:
 ; CHECK-LABEL: test_srlq_3:
 ; CHECK: vpsrlq $63, %ymm0, %ymm0
 ; CHECK: ret
+
+; CHECK-LABEL: @srl_trunc_and_v4i64
+; CHECK: vpand
+; CHECK-NEXT: vpsrlvd
+; CHECK: ret
+define <4 x i32> @srl_trunc_and_v4i64(<4 x i32> %x, <4 x i64> %y) nounwind {
+  %and = and <4 x i64> %y, <i64 8, i64 8, i64 8, i64 8>
+  %trunc = trunc <4 x i64> %and to <4 x i32>
+  %sra = lshr <4 x i32> %x, %trunc
+  ret <4 x i32> %sra
+}
diff --git a/llvm/test/CodeGen/X86/sse2-vector-shifts.ll b/llvm/test/CodeGen/X86/sse2-vector-shifts.ll
index 47a01ff2583..7c8d5e57889 100644
--- a/llvm/test/CodeGen/X86/sse2-vector-shifts.ll
+++ b/llvm/test/CodeGen/X86/sse2-vector-shifts.ll
@@ -9,8 +9,8 @@ entry:
 }
 
 ; CHECK-LABEL: test_sllw_1:
-; CHECK: psllw   $0, %xmm0
-; CHECK-NEXT: ret
+; CHECK-NOT: psllw   $0, %xmm0
+; CHECK: ret
 
 define <8 x i16> @test_sllw_2(<8 x i16> %InVec) {
 entry:
@@ -39,8 +39,8 @@ entry:
 }
 
 ; CHECK-LABEL: test_slld_1:
-; CHECK: pslld   $0, %xmm0
-; CHECK-NEXT: ret
+; CHECK-NOT: pslld   $0, %xmm0
+; CHECK: ret
 
 define <4 x i32> @test_slld_2(<4 x i32> %InVec) {
 entry:
@@ -69,8 +69,8 @@ entry:
 }
 
 ; CHECK-LABEL: test_sllq_1:
-; CHECK: psllq   $0, %xmm0
-; CHECK-NEXT: ret
+; CHECK-NOT: psllq   $0, %xmm0
+; CHECK: ret
 
 define <2 x i64> @test_sllq_2(<2 x i64> %InVec) {
 entry:
@@ -101,8 +101,8 @@ entry:
 }
 
 ; CHECK-LABEL: test_sraw_1:
-; CHECK: psraw   $0, %xmm0
-; CHECK-NEXT: ret
+; CHECK-NOT: psraw   $0, %xmm0
+; CHECK: ret
 
 define <8 x i16> @test_sraw_2(<8 x i16> %InVec) {
 entry:
@@ -131,8 +131,8 @@ entry:
 }
 
 ; CHECK-LABEL: test_srad_1:
-; CHECK: psrad   $0, %xmm0
-; CHECK-NEXT: ret
+; CHECK-NOT: psrad   $0, %xmm0
+; CHECK: ret
 
 define <4 x i32> @test_srad_2(<4 x i32> %InVec) {
 entry:
@@ -163,8 +163,8 @@ entry:
 }
 
 ; CHECK-LABEL: test_srlw_1:
-; CHECK: psrlw   $0, %xmm0
-; CHECK-NEXT: ret
+; CHECK-NOT: psrlw   $0, %xmm0
+; CHECK: ret
 
 define <8 x i16> @test_srlw_2(<8 x i16> %InVec) {
 entry:
@@ -193,8 +193,8 @@ entry:
 }
 
 ; CHECK-LABEL: test_srld_1:
-; CHECK: psrld   $0, %xmm0
-; CHECK-NEXT: ret
+; CHECK-NOT: psrld   $0, %xmm0
+; CHECK: ret
 
 define <4 x i32> @test_srld_2(<4 x i32> %InVec) {
 entry:
@@ -223,8 +223,8 @@ entry:
 }
 
 ; CHECK-LABEL: test_srlq_1:
-; CHECK: psrlq   $0, %xmm0
-; CHECK-NEXT: ret
+; CHECK-NOT: psrlq   $0, %xmm0
+; CHECK: ret
 
 define <2 x i64> @test_srlq_2(<2 x i64> %InVec) {
 entry:
@@ -245,3 +245,123 @@ entry:
 ; CHECK-LABEL: test_srlq_3:
 ; CHECK: psrlq $63, %xmm0
 ; CHECK-NEXT: ret
+
+
+; CHECK-LABEL: sra_sra_v4i32:
+; CHECK: psrad $6, %xmm0
+; CHECK-NEXT: retq
+define <4 x i32> @sra_sra_v4i32(<4 x i32> %x) nounwind {
+  %sra0 = ashr <4 x i32> %x, <i32 2, i32 2, i32 2, i32 2>
+  %sra1 = ashr <4 x i32> %sra0, <i32 4, i32 4, i32 4, i32 4>
+  ret <4 x i32> %sra1
+}
+
+; CHECK-LABEL: @srl_srl_v4i32
+; CHECK: psrld $6, %xmm0
+; CHECK-NEXT: ret
+define <4 x i32> @srl_srl_v4i32(<4 x i32> %x) nounwind {
+  %srl0 = lshr <4 x i32> %x, <i32 2, i32 2, i32 2, i32 2>
+  %srl1 = lshr <4 x i32> %srl0, <i32 4, i32 4, i32 4, i32 4>
+  ret <4 x i32> %srl1
+}
+
+; CHECK-LABEL: @srl_shl_v4i32
+; CHECK: andps
+; CHECK-NEXT: retq
+define <4 x i32> @srl_shl_v4i32(<4 x i32> %x) nounwind {
+  %srl0 = shl <4 x i32> %x, <i32 4, i32 4, i32 4, i32 4>
+  %srl1 = lshr <4 x i32> %srl0, <i32 4, i32 4, i32 4, i32 4>
+  ret <4 x i32> %srl1
+}
+
+; CHECK-LABEL: @srl_sra_31_v4i32
+; CHECK: psrld $31, %xmm0
+; CHECK-NEXT: ret
+define <4 x i32> @srl_sra_31_v4i32(<4 x i32> %x, <4 x i32> %y) nounwind {
+  %sra = ashr <4 x i32> %x, %y
+  %srl1 = lshr <4 x i32> %sra, <i32 31, i32 31, i32 31, i32 31>
+  ret <4 x i32> %srl1
+}
+
+; CHECK-LABEL: @shl_shl_v4i32
+; CHECK: pslld $6, %xmm0
+; CHECK-NEXT: ret
+define <4 x i32> @shl_shl_v4i32(<4 x i32> %x) nounwind {
+  %shl0 = shl <4 x i32> %x, <i32 2, i32 2, i32 2, i32 2>
+  %shl1 = shl <4 x i32> %shl0, <i32 4, i32 4, i32 4, i32 4>
+  ret <4 x i32> %shl1
+}
+
+; CHECK-LABEL: @shl_sra_v4i32
+; CHECK: andps
+; CHECK-NEXT: ret
+define <4 x i32> @shl_sra_v4i32(<4 x i32> %x) nounwind {
+  %shl0 = ashr <4 x i32> %x, <i32 4, i32 4, i32 4, i32 4>
+  %shl1 = shl <4 x i32> %shl0, <i32 4, i32 4, i32 4, i32 4>
+  ret <4 x i32> %shl1
+}
+
+; CHECK-LABEL: @shl_srl_v4i32
+; CHECK: pslld $3, %xmm0
+; CHECK-NEXT: pand
+; CHECK-NEXT: ret
+define <4 x i32> @shl_srl_v4i32(<4 x i32> %x) nounwind {
+  %shl0 = lshr <4 x i32> %x, <i32 2, i32 2, i32 2, i32 2>
+  %shl1 = shl <4 x i32> %shl0, <i32 5, i32 5, i32 5, i32 5>
+  ret <4 x i32> %shl1
+}
+
+; CHECK-LABEL: @shl_zext_srl_v4i32
+; CHECK: andps
+; CHECK-NEXT: ret
+define <4 x i32> @shl_zext_srl_v4i32(<4 x i16> %x) nounwind {
+  %srl = lshr <4 x i16> %x, <i16 2, i16 2, i16 2, i16 2>
+  %zext = zext <4 x i16> %srl to <4 x i32>
+  %shl = shl <4 x i32> %zext, <i32 2, i32 2, i32 2, i32 2>
+  ret <4 x i32> %shl
+}
+
+; CHECK: @sra_trunc_srl_v4i32
+; CHECK: psrad $19, %xmm0
+; CHECK-NEXT: retq
+define <4 x i16> @sra_trunc_srl_v4i32(<4 x i32> %x) nounwind {
+  %srl = lshr <4 x i32> %x, <i32 16, i32 16, i32 16, i32 16>
+  %trunc = trunc <4 x i32> %srl to <4 x i16>
+  %sra = ashr <4 x i16> %trunc, <i16 3, i16 3, i16 3, i16 3>
+  ret <4 x i16> %sra
+}
+
+; CHECK-LABEL: @shl_zext_shl_v4i32
+; CHECK: pand
+; CHECK-NEXT: pslld $19, %xmm0
+; CHECK-NEXT: ret
+define <4 x i32> @shl_zext_shl_v4i32(<4 x i16> %x) nounwind {
+  %shl0 = shl <4 x i16> %x, <i16 2, i16 2, i16 2, i16 2>
+  %ext = zext <4 x i16> %shl0 to <4 x i32>
+  %shl1 = shl <4 x i32> %ext, <i32 17, i32 17, i32 17, i32 17>
+  ret <4 x i32> %shl1
+}
+
+; CHECK-LABEL: @sra_v4i32
+; CHECK: psrad $3, %xmm0
+; CHECK-NEXT: ret
+define <4 x i32> @sra_v4i32(<4 x i32> %x) nounwind {
+  %sra = ashr <4 x i32> %x, <i32 3, i32 3, i32 3, i32 3>
+  ret <4 x i32> %sra
+}
+
+; CHECK-LABEL: @srl_v4i32
+; CHECK: psrld $3, %xmm0
+; CHECK-NEXT: ret
+define <4 x i32> @srl_v4i32(<4 x i32> %x) nounwind {
+  %sra = lshr <4 x i32> %x, <i32 3, i32 3, i32 3, i32 3>
+  ret <4 x i32> %sra
+}
+
+; CHECK-LABEL: @shl_v4i32
+; CHECK: pslld $3, %xmm0
+; CHECK-NEXT: ret
+define <4 x i32> @shl_v4i32(<4 x i32> %x) nounwind {
+  %sra = shl <4 x i32> %x, <i32 3, i32 3, i32 3, i32 3>
+  ret <4 x i32> %sra
+}