diff test/CodeGen/X86/avx2-vector-shifts.ll @ 77:54457678186b LLVM3.6

LLVM 3.6
author Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
date Mon, 08 Sep 2014 22:06:00 +0900
parents 95c75e76d11b
children afa8332a0e37
line wrap: on
line diff
--- a/test/CodeGen/X86/avx2-vector-shifts.ll	Thu Dec 12 15:22:36 2013 +0900
+++ b/test/CodeGen/X86/avx2-vector-shifts.ll	Mon Sep 08 22:06:00 2014 +0900
@@ -9,7 +9,7 @@
 }
 
 ; 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) {
@@ -24,12 +24,12 @@
 
 define <16 x i16> @test_sllw_3(<16 x i16> %InVec) {
 entry:
-  %shl = shl <16 x i16> %InVec, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+  %shl = shl <16 x i16> %InVec, <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
   ret <16 x i16> %shl
 }
 
 ; CHECK-LABEL: test_sllw_3:
-; CHECK: vxorps  %ymm0, %ymm0, %ymm0
+; CHECK: vpsllw $15, %ymm0, %ymm0
 ; CHECK: ret
 
 define <8 x i32> @test_slld_1(<8 x i32> %InVec) {
@@ -39,7 +39,7 @@
 }
 
 ; 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) {
@@ -52,14 +52,24 @@
 ; CHECK: vpaddd  %ymm0, %ymm0, %ymm0
 ; CHECK: ret
 
+define <8 x i32> @test_vpslld_var(i32 %shift) {
+  %amt = insertelement <8 x i32> undef, i32 %shift, i32 0
+  %tmp = shl <8 x i32> <i32 192, i32 193, i32 194, i32 195, i32 196, i32 197, i32 198, i32 199>, %amt
+  ret <8 x i32> %tmp
+}
+
+; CHECK-LABEL: test_vpslld_var:
+; CHECK: vpslld %xmm0, %ymm1, %ymm0
+; CHECK: ret
+
 define <8 x i32> @test_slld_3(<8 x i32> %InVec) {
 entry:
-  %shl = shl <8 x i32> %InVec, <i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32>
+  %shl = shl <8 x i32> %InVec, <i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31>
   ret <8 x i32> %shl
 }
 
 ; CHECK-LABEL: test_slld_3:
-; CHECK: vxorps  %ymm0, %ymm0, %ymm0
+; CHECK: vpslld $31, %ymm0, %ymm0
 ; CHECK: ret
 
 define <4 x i64> @test_sllq_1(<4 x i64> %InVec) {
@@ -69,7 +79,7 @@
 }
 
 ; 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) {
@@ -84,12 +94,12 @@
 
 define <4 x i64> @test_sllq_3(<4 x i64> %InVec) {
 entry:
-  %shl = shl <4 x i64> %InVec, <i64 64, i64 64, i64 64, i64 64>
+  %shl = shl <4 x i64> %InVec, <i64 63, i64 63, i64 63, i64 63>
   ret <4 x i64> %shl
 }
 
 ; CHECK-LABEL: test_sllq_3:
-; CHECK: vxorps  %ymm0, %ymm0, %ymm0
+; CHECK: vpsllq $63, %ymm0, %ymm0
 ; CHECK: ret
 
 ; AVX2 Arithmetic Shift
@@ -101,7 +111,7 @@
 }
 
 ; 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) {
@@ -116,7 +126,7 @@
 
 define <16 x i16> @test_sraw_3(<16 x i16> %InVec) {
 entry:
-  %shl = ashr <16 x i16> %InVec, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+  %shl = ashr <16 x i16> %InVec, <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
   ret <16 x i16> %shl
 }
 
@@ -131,7 +141,7 @@
 }
 
 ; 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) {
@@ -146,7 +156,7 @@
 
 define <8 x i32> @test_srad_3(<8 x i32> %InVec) {
 entry:
-  %shl = ashr <8 x i32> %InVec, <i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32>
+  %shl = ashr <8 x i32> %InVec, <i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31>
   ret <8 x i32> %shl
 }
 
@@ -163,7 +173,7 @@
 }
 
 ; 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) {
@@ -178,12 +188,12 @@
 
 define <16 x i16> @test_srlw_3(<16 x i16> %InVec) {
 entry:
-  %shl = lshr <16 x i16> %InVec, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+  %shl = lshr <16 x i16> %InVec, <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
   ret <16 x i16> %shl
 }
 
 ; CHECK-LABEL: test_srlw_3:
-; CHECK: vxorps  %ymm0, %ymm0, %ymm0
+; CHECK: vpsrlw $15, %ymm0, %ymm0
 ; CHECK: ret
 
 define <8 x i32> @test_srld_1(<8 x i32> %InVec) {
@@ -193,7 +203,7 @@
 }
 
 ; 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) {
@@ -208,12 +218,12 @@
 
 define <8 x i32> @test_srld_3(<8 x i32> %InVec) {
 entry:
-  %shl = lshr <8 x i32> %InVec, <i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32>
+  %shl = lshr <8 x i32> %InVec, <i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31>
   ret <8 x i32> %shl
 }
 
 ; CHECK-LABEL: test_srld_3:
-; CHECK: vxorps  %ymm0, %ymm0, %ymm0
+; CHECK: vpsrld $31, %ymm0, %ymm0
 ; CHECK: ret
 
 define <4 x i64> @test_srlq_1(<4 x i64> %InVec) {
@@ -223,7 +233,7 @@
 }
 
 ; 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) {
@@ -238,10 +248,21 @@
 
 define <4 x i64> @test_srlq_3(<4 x i64> %InVec) {
 entry:
-  %shl = lshr <4 x i64> %InVec, <i64 64, i64 64, i64 64, i64 64>
+  %shl = lshr <4 x i64> %InVec, <i64 63, i64 63, i64 63, i64 63>
   ret <4 x i64> %shl
 }
 
 ; CHECK-LABEL: test_srlq_3:
-; CHECK: vxorps  %ymm0, %ymm0, %ymm0
+; 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
+}