Reapply a more appropriate solution than in r137114. AVX supports

v4f64 = sitofp v4i32. This fix PR10559.
Also add support for v4i32 = fptosi v4f64.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@137128 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 79d455a326c4c30cc39cf0a72211e696612cefd0..2b8180aac04c79c7b148dde83e7876fc6bddb91b 100644 (file)
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -3305,6 +3305,11 @@ def CVTPD2DQrm  : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
 def CVTPD2DQrr  : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                        "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
 
+def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
+          (VCVTPD2DQYrr VR256:$src)>;
+def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
+          (VCVTPD2DQYrm addr:$src)>;
+
 // Convert Packed DW Integers to Packed Double FP
 let Predicates = [HasAVX] in {
 def VCVTDQ2PDrm  : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
@@ -3333,6 +3338,11 @@ def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
           (VCVTPD2DQYrm addr:$src)>;
 
+def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
+          (VCVTDQ2PDYrr VR128:$src)>;
+def : Pat<(v4f64 (sint_to_fp (memopv4i32 addr:$src))),
+          (VCVTDQ2PDYrm addr:$src)>;
+
 //===---------------------------------------------------------------------===//
 // SSE3 - Move Instructions
 //===---------------------------------------------------------------------===//

diff --git a/test/CodeGen/X86/avx-cvt.ll b/test/CodeGen/X86/avx-cvt.ll
index e45010c903edab99e887ea7a12342b038fabe6ca..6c0bd58074d49ba8fd7017be32ea596f9656a073 100644 (file)
--- a/test/CodeGen/X86/avx-cvt.ll
+++ b/test/CodeGen/X86/avx-cvt.ll
@@ -12,6 +12,18 @@ define <8 x i32> @fptosi00(<8 x float> %a) nounwind {
   ret <8 x i32> %b
 }
 
+; CHECK: vcvtdq2pd %xmm
+define <4 x double> @sitofp01(<4 x i32> %a) {
+  %b = sitofp <4 x i32> %a to <4 x double>
+  ret <4 x double> %b
+}
+
+; CHECK: vcvtpd2dqy %ymm
+define <4 x i32> @fptosi01(<4 x double> %a) {
+  %b = fptosi <4 x double> %a to <4 x i32>
+  ret <4 x i32> %b
+}
+
 ; CHECK: vcvtpd2psy %ymm
 ; CHECK-NEXT: vcvtpd2psy %ymm
 ; CHECK-NEXT: vinsertf128 $1