[X86][SSE] Vector double -> float conversion memory folding (cvtpd2ps)

Added a missing memory folding relationship for the (V)CVTPD2PS instruction - we can safely fold these for stack reloads.

Differential Revision: http://reviews.llvm.org/D6663

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@224383 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrInfo.cpp b/lib/Target/X86/X86InstrInfo.cpp
index b31d086f867b91793e2e2c5fb5c57a62ed31e83c..d745ba68bb25e73191942c35ff7b7101d9975ad8 100644 (file)
--- a/lib/Target/X86/X86InstrInfo.cpp
+++ b/lib/Target/X86/X86InstrInfo.cpp
@@ -450,6 +450,7 @@ X86InstrInfo::X86InstrInfo(X86Subtarget &STI)
     { X86::CVTSS2SIrr,      X86::CVTSS2SIrm,          0 },
     { X86::CVTDQ2PSrr,      X86::CVTDQ2PSrm,          TB_ALIGN_16 },
     { X86::CVTPD2DQrr,      X86::CVTPD2DQrm,          TB_ALIGN_16 },
+    { X86::CVTPD2PSrr,      X86::CVTPD2PSrm,          TB_ALIGN_16 },
     { X86::CVTPS2DQrr,      X86::CVTPS2DQrm,          TB_ALIGN_16 },
     { X86::CVTTPD2DQrr,     X86::CVTTPD2DQrm,         TB_ALIGN_16 },
     { X86::CVTTPS2DQrr,     X86::CVTTPS2DQrm,         TB_ALIGN_16 },
@@ -531,6 +532,7 @@ X86InstrInfo::X86InstrInfo(X86Subtarget &STI)
     { X86::VCVTSS2SIrr,     X86::VCVTSS2SIrm,         0 },
     { X86::VCVTDQ2PSrr,     X86::VCVTDQ2PSrm,         0 },
     { X86::VCVTPD2DQrr,     X86::VCVTPD2DQXrm,        0 },
+    { X86::VCVTPD2PSrr,     X86::VCVTPD2PSXrm,        0 },
     { X86::VCVTPS2DQrr,     X86::VCVTPS2DQrm,         0 },
     { X86::VCVTTPD2DQrr,    X86::VCVTTPD2DQXrm,       0 },
     { X86::VCVTTPS2DQrr,    X86::VCVTTPS2DQrm,        0 },
@@ -569,6 +571,7 @@ X86InstrInfo::X86InstrInfo(X86Subtarget &STI)
     // AVX 256-bit foldable instructions
     { X86::VCVTDQ2PSYrr,    X86::VCVTDQ2PSYrm,        0 },
     { X86::VCVTPD2DQYrr,    X86::VCVTPD2DQYrm,        0 },
+    { X86::VCVTPD2PSYrr,    X86::VCVTPD2PSYrm,        0 },
     { X86::VCVTPS2DQYrr,    X86::VCVTPS2DQYrm,        0 },
     { X86::VCVTTPD2DQYrr,   X86::VCVTTPD2DQYrm,       0 },
     { X86::VCVTTPS2DQYrr,   X86::VCVTTPS2DQYrm,       0 },

diff --git a/test/CodeGen/X86/avx1-stack-reload-folding.ll b/test/CodeGen/X86/avx1-stack-reload-folding.ll
index 2e669b0fe125e62a3d810fa308d776223d46926a..480e3dd63060cd52679f7d5ad2ba29fd8ea8ad24 100644 (file)
--- a/test/CodeGen/X86/avx1-stack-reload-folding.ll
+++ b/test/CodeGen/X86/avx1-stack-reload-folding.ll
@@ -37,6 +37,21 @@ define void @stack_fold_cvtdq2ps(<128 x i32>* %a, <128 x i32>* %b, <128 x float>
   ret void\r
 }\r
 \r
+define void @stack_fold_cvtpd2ps(<128 x double>* %a, <128 x double>* %b, <128 x float>* %c) {\r
+  ;CHECK-LABEL: stack_fold_cvtpd2ps\r
+  ;CHECK:   vcvtpd2psy {{[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 32-byte Folded Reload\r
+\r
+  %1 = load <128 x double>* %a\r
+  %2 = load <128 x double>* %b\r
+  %3 = fadd <128 x double> %1, %2\r
+  %4 = fsub <128 x double> %1, %2\r
+  %5 = fptrunc <128 x double> %3 to <128 x float>\r
+  %6 = fptrunc <128 x double> %4 to <128 x float>\r
+  %7 = fadd <128 x float> %5, %6\r
+  store <128 x float> %7, <128 x float>* %c\r
+  ret void\r
+}\r
+\r
 define void @stack_fold_cvttpd2dq(<64 x double>* %a, <64 x double>* %b, <64 x i32>* %c) #0 {\r
   ;CHECK-LABEL: stack_fold_cvttpd2dq\r
   ;CHECK:  vcvttpd2dqy {{[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 32-byte Folded Reload\r