[X86] Add patterns to select MOVLPDrm from MOVSD+load and MOVHPD from UNPCKL+load.

These narrow the load so we can only do it if the load isn't
volatile.

There also tests in vector-shuffle-128-v4.ll that this should
support, but we don't seem to fold bitcast+load on pre-sse4.2
targets due to the slow unaligned mem 16 flag.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@365266 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index dc2db1e8e614c1e2edb027314c15b56f63fc0388..76530adc1527fa70c569d145ec892b9d0c6d9eda 100644 (file)
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -748,6 +748,13 @@ let Predicates = [UseSSE2] in {
             (MOVLPDrm VR128:$src1, addr:$src2)>;
 }
 
+let Predicates = [UseSSE2, NoSSE41_Or_OptForSize] in {
+  // Use MOVLPD to load into the low bits from a full vector unless we can use
+  // BLENDPD.
+  def : Pat<(X86Movsd VR128:$src1, (v2f64 (nonvolatile_load addr:$src2))),
+            (MOVLPDrm VR128:$src1, addr:$src2)>;
+}
+
 //===----------------------------------------------------------------------===//
 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
 //===----------------------------------------------------------------------===//
@@ -2075,6 +2082,13 @@ let Predicates = [HasAVX1Only] in {
             (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
 }
 
+let Predicates = [UseSSE2] in {
+  // Use MOVHPD if the load isn't aligned enough for UNPCKLPD.
+  def : Pat<(v2f64 (X86Unpckl VR128:$src1,
+                              (v2f64 (nonvolatile_load addr:$src2)))),
+            (MOVHPDrm VR128:$src1, addr:$src2)>;
+}
+
 //===----------------------------------------------------------------------===//
 // SSE 1 & 2 - Extract Floating-Point Sign mask
 //===----------------------------------------------------------------------===//

diff --git a/test/CodeGen/X86/vector-shuffle-128-v2.ll b/test/CodeGen/X86/vector-shuffle-128-v2.ll
index 761855e5c628fa036fa5cac4038ac206d0650e8a..2b42d33000afa0b88ae3a987cc0d09a548a7723e 100644 (file)
--- a/test/CodeGen/X86/vector-shuffle-128-v2.ll
+++ b/test/CodeGen/X86/vector-shuffle-128-v2.ll
@@ -1309,8 +1309,7 @@ define <2 x double> @shuffle_mem_v2f64_31(<2 x double> %a, <2 x double>* %b) {
 define <2 x double> @shuffle_mem_v2f64_02(<2 x double> %a, <2 x double>* %pb) {
 ; SSE-LABEL: shuffle_mem_v2f64_02:
 ; SSE:       # %bb.0:
-; SSE-NEXT:    movups (%rdi), %xmm1
-; SSE-NEXT:    movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSE-NEXT:    movhpd {{.*#+}} xmm0 = xmm0[0],mem[0]
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: shuffle_mem_v2f64_02:
@@ -1325,20 +1324,17 @@ define <2 x double> @shuffle_mem_v2f64_02(<2 x double> %a, <2 x double>* %pb) {
 define <2 x double> @shuffle_mem_v2f64_21(<2 x double> %a, <2 x double>* %pb) {
 ; SSE2-LABEL: shuffle_mem_v2f64_21:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movupd (%rdi), %xmm1
-; SSE2-NEXT:    movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
+; SSE2-NEXT:    movlpd {{.*#+}} xmm0 = mem[0],xmm0[1]
 ; SSE2-NEXT:    retq
 ;
 ; SSE3-LABEL: shuffle_mem_v2f64_21:
 ; SSE3:       # %bb.0:
-; SSE3-NEXT:    movupd (%rdi), %xmm1
-; SSE3-NEXT:    movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
+; SSE3-NEXT:    movlpd {{.*#+}} xmm0 = mem[0],xmm0[1]
 ; SSE3-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_mem_v2f64_21:
 ; SSSE3:       # %bb.0:
-; SSSE3-NEXT:    movupd (%rdi), %xmm1
-; SSSE3-NEXT:    movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
+; SSSE3-NEXT:    movlpd {{.*#+}} xmm0 = mem[0],xmm0[1]
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: shuffle_mem_v2f64_21: