[X86] Add patterns to select (scalar_to_vector (loadf32)) as (V)MOVSSrm instead of COPY_TO_REGCLASS + (V)MOVSSrm_alt.

Similar for (V)MOVSD. Ultimately, I'd like to see about folding
scalar_to_vector+load to vzload. Which would select as (V)MOVSSrm
so this is closer to that.

llvm-svn: 364948

2 files changed, 24 insertions, 9 deletions

diff --git a/llvm/lib/Target/X86/X86InstrAVX512.td b/llvm/lib/Target/X86/X86InstrAVX512.td
index 2cdcb1e1f0e..b2dfc512932 100644
--- a/llvm/lib/Target/X86/X86InstrAVX512.td
+++ b/llvm/lib/Target/X86/X86InstrAVX512.td
@@ -4312,6 +4312,10 @@ let Predicates = [HasAVX512, OptForSpeed] in {
 }
 
 let Predicates = [HasAVX512] in {
+  def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
+            (VMOVSSZrm addr:$src)>;
+  def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
+            (VMOVSDZrm addr:$src)>;
 
   // Represent the same patterns above but in the form they appear for
   // 256-bit types
diff --git a/llvm/lib/Target/X86/X86InstrSSE.td b/llvm/lib/Target/X86/X86InstrSSE.td
index c79cf7ade88..56974c44b4d 100644
--- a/llvm/lib/Target/X86/X86InstrSSE.td
+++ b/llvm/lib/Target/X86/X86InstrSSE.td
@@ -263,6 +263,11 @@ let canFoldAsLoad = 1, isReMaterializable = 1 in {
 
 // Patterns
 let Predicates = [UseAVX] in {
+  def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
+            (VMOVSSrm addr:$src)>;
+  def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
+            (VMOVSDrm addr:$src)>;
+
   // Represent the same patterns above but in the form they appear for
   // 256-bit types
   def : Pat<(v8f32 (X86vzload addr:$src)),
@@ -290,17 +295,23 @@ let Predicates = [UseAVX, OptForSize] in {
               (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm)))), sub_xmm)>;
 }
 
-let Predicates = [UseSSE1] in {
-  let Predicates = [UseSSE1, NoSSE41_Or_OptForSize] in {
-  // Move scalar to XMM zero-extended, zeroing a VR128 then do a
-  // MOVSS to the lower bits.
-  def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
-            (MOVSSrr (v4f32 (V_SET0)), VR128:$src)>;
-  def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
-            (MOVSSrr (v4i32 (V_SET0)), VR128:$src)>;
-  }
+let Predicates = [UseSSE1, NoSSE41_Or_OptForSize] in {
+// Move scalar to XMM zero-extended, zeroing a VR128 then do a
+// MOVSS to the lower bits.
+def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
+          (MOVSSrr (v4f32 (V_SET0)), VR128:$src)>;
+def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
+          (MOVSSrr (v4i32 (V_SET0)), VR128:$src)>;
 }
 
+let Predicates = [UseSSE2] in
+def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
+          (MOVSDrm addr:$src)>;
+
+let Predicates = [UseSSE1] in
+def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
+          (MOVSSrm addr:$src)>;
+
 //===----------------------------------------------------------------------===//
 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
 //===----------------------------------------------------------------------===//