Add v4f64 -> v2f32 fp_round support. Also add a testcase to exercise

the legalizer. This commit together with the two previous ones fixes
PR10495.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@136654 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index c6dede979dea0562454ce9b603bb5e173dfb7991..b00b00d047b171de0d149658aa646e6dd01fc3ff 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -966,6 +966,7 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
 
     setOperationAction(ISD::FP_TO_SINT,         MVT::v8i32, Legal);
     setOperationAction(ISD::SINT_TO_FP,         MVT::v8i32, Legal);
+    setOperationAction(ISD::FP_ROUND,           MVT::v4f32, Legal);
 
     setOperationAction(ISD::CONCAT_VECTORS,     MVT::v4f64,  Custom);
     setOperationAction(ISD::CONCAT_VECTORS,     MVT::v4i64,  Custom);

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 066f7c01a53afa3b30675e1d34d53ce6e0f35852..acd0c6b7a6556d40e9f52fa00d485a98076d6347 100644 (file)
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -1054,6 +1054,12 @@ def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
           (VCVTTPS2DQYrm addr:$src)>;
 
+// Match fround for 128/256-bit conversions
+def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
+          (VCVTPD2PSYrr VR256:$src)>;
+def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
+          (VCVTPD2PSYrm addr:$src)>;
+
 //===----------------------------------------------------------------------===//
 // SSE 1 & 2 - Compare Instructions
 //===----------------------------------------------------------------------===//

diff --git a/test/CodeGen/X86/avx-256-cvt.ll b/test/CodeGen/X86/avx-256-cvt.ll
index 8bf67568b6a349c814cf0bded881475fab1c80e2..d97327906cab1c3ec0cbf6044fcaedf7d2062b61 100644 (file)
--- a/test/CodeGen/X86/avx-256-cvt.ll
+++ b/test/CodeGen/X86/avx-256-cvt.ll
@@ -12,3 +12,10 @@ define <8 x i32> @funcB(<8 x float> %a) nounwind {
   ret <8 x i32> %b
 }
 
+; CHECK: vcvtpd2psy %ymm
+; CHECK-NEXT: vcvtpd2psy %ymm
+; CHECK-NEXT: vinsertf128 $1
+define <8 x float> @funcC(<8 x double> %b) nounwind {
+  %a = fptrunc <8 x double> %b to <8 x float>
+  ret <8 x float> %a
+}