Fix for PR2687: Add patterns to match sint_to_fp and fp_to_sint for <2 x

i32>.  This is a little messy, but it works.

We should really get rid of the intrinsics, though, since they map
perfectly well to standard LLVM instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@55864 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 3c1d7c77c8aa7781101957e9e8160b211f7b96ce..06641446d6e0e646ad282d49dd4a99e56282a119 100644 (file)
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -3080,6 +3080,10 @@ def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
           (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>;
 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
           (Int_CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>;
+def : Pat<(v2f64 (sint_to_fp (v2i32 VR64:$src))),
+          (Int_CVTPI2PDrr VR64:$src)>, Requires<[HasSSE2]>;
+def : Pat<(v2i32 (fp_to_sint (v2f64 VR128:$src))),
+          (Int_CVTTPD2PIrr VR128:$src)>, Requires<[HasSSE2]>;
 
 // Use movaps / movups for SSE integer load / store (one byte shorter).
 def : Pat<(alignedloadv4i32 addr:$src),

diff --git a/test/CodeGen/X86/2008-09-05-sinttofp-2xi32.ll b/test/CodeGen/X86/2008-09-05-sinttofp-2xi32.ll
new file mode 100644 (file)
index 0000000..ffe10d4
--- /dev/null
+++ b/test/CodeGen/X86/2008-09-05-sinttofp-2xi32.ll
@@ -0,0 +1,15 @@
+; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse2 | grep cvttpd2pi | count 1
+; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse2 | grep cvtpi2pd | count 1
+; PR2687
+
+define <2 x double> @a(<2 x i32> %x) nounwind {
+entry:
+  %y = sitofp <2 x i32> %x to <2 x double>
+  ret <2 x double> %y
+}
+
+define <2 x i32> @b(<2 x double> %x) nounwind {
+entry:
+  %y = fptosi <2 x double> %x to <2 x i32>
+  ret <2 x i32> %y
+}