Allow a GR64 to be moved into an MMX register via the "movd" instruction.

Still need to have JIT generate this code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@37863 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrMMX.td b/lib/Target/X86/X86InstrMMX.td
index 897d8f2aab2e921d734d8b0f5d00d0e37b21e030..e3cfa0059f318004fabbbdd5dfdffa4742f4d48e 100644 (file)
--- a/lib/Target/X86/X86InstrMMX.td
+++ b/lib/Target/X86/X86InstrMMX.td
@@ -183,6 +183,9 @@ def MMX_MOVD64rm : MMXI<0x6E, MRMSrcMem, (ops VR64:$dst, i32mem:$src),
 def MMX_MOVD64mr : MMXI<0x7E, MRMDestMem, (ops i32mem:$dst, VR64:$src),
                         "movd {$src, $dst|$dst, $src}", []>;
 
+def MMX_MOVD64to64rr : MMXI<0x6E, MRMSrcReg, (ops VR64:$dst, GR64:$src),
+                            "movd {$src, $dst|$dst, $src}", []>;
+
 def MMX_MOVQ64rr : MMXI<0x6F, MRMSrcReg, (ops VR64:$dst, VR64:$src),
                         "movq {$src, $dst|$dst, $src}", []>;
 def MMX_MOVQ64rm : MMXI<0x6F, MRMSrcMem, (ops VR64:$dst, i64mem:$src),
@@ -552,6 +555,16 @@ def : Pat<(v1i64 (bitconvert (v2i32 VR64:$src))), (v1i64 VR64:$src)>;
 def : Pat<(v1i64 (bitconvert (v4i16 VR64:$src))), (v1i64 VR64:$src)>;
 def : Pat<(v1i64 (bitconvert (v8i8  VR64:$src))), (v1i64 VR64:$src)>;
 
+// 64-bit bit convert.
+def : Pat<(v1i64 (bitconvert (i64 GR64:$src))),
+          (MMX_MOVD64to64rr GR64:$src)>;
+def : Pat<(v2i32 (bitconvert (i64 GR64:$src))),
+          (MMX_MOVD64to64rr GR64:$src)>;
+def : Pat<(v4i16 (bitconvert (i64 GR64:$src))),
+          (MMX_MOVD64to64rr GR64:$src)>;
+def : Pat<(v8i8  (bitconvert (i64 GR64:$src))),
+          (MMX_MOVD64to64rr GR64:$src)>;
+
 def MMX_X86s2vec : SDNode<"X86ISD::S2VEC",  SDTypeProfile<1, 1, []>, []>;
 
 // Move scalar to XMM zero-extended

diff --git a/test/CodeGen/X86/2007-06-15-IntToMMX.ll b/test/CodeGen/X86/2007-06-15-IntToMMX.ll
index a72de1cd9870cfa8cc237176b655d0fa03e1be19..e608ac3ecb97e9dc201744cc1e82692e1e5889cb 100644 (file)
--- a/test/CodeGen/X86/2007-06-15-IntToMMX.ll
+++ b/test/CodeGen/X86/2007-06-15-IntToMMX.ll
@@ -1,5 +1,4 @@
 ; RUN: llvm-as < %s | llc -march=x86-64 -mattr=+mmx | grep paddusw
-; XFAIL: *
 @R = external global <1 x i64>          ; <<1 x i64>*> [#uses=1]
 
 define void @foo(<1 x i64> %A, <1 x i64> %B) {

diff --git a/test/CodeGen/X86/2007-07-03-GR64ToVR64.ll b/test/CodeGen/X86/2007-07-03-GR64ToVR64.ll
new file mode 100644 (file)
index 0000000..b2b1a94
--- /dev/null
+++ b/test/CodeGen/X86/2007-07-03-GR64ToVR64.ll
@@ -0,0 +1,20 @@
+; RUN: llvm-as < %s | llc -march=x86-64 -mattr=+mmx | grep {movd %rsi, %mm0} &&
+; RUN: llvm-as < %s | llc -march=x86-64 -mattr=+mmx | grep {movd %rdi, %mm1} &&
+; RUN: llvm-as < %s | llc -march=x86-64 -mattr=+mmx | grep {paddusw %mm0, %mm1}
+
+@R = external global <1 x i64>         ; <<1 x i64>*> [#uses=1]
+
+define void @foo(<1 x i64> %A, <1 x i64> %B) {
+entry:
+       %tmp4 = bitcast <1 x i64> %B to <4 x i16>               ; <<4 x i16>> [#uses=1]
+       %tmp6 = bitcast <1 x i64> %A to <4 x i16>               ; <<4 x i16>> [#uses=1]
+       %tmp7 = tail call <4 x i16> @llvm.x86.mmx.paddus.w( <4 x i16> %tmp6, <4 x i16> %tmp4 )          ; <<4 x i16>> [#uses=1]
+       %tmp8 = bitcast <4 x i16> %tmp7 to <1 x i64>            ; <<1 x i64>> [#uses=1]
+       store <1 x i64> %tmp8, <1 x i64>* @R
+       tail call void @llvm.x86.mmx.emms( )
+       ret void
+}
+
+declare <4 x i16> @llvm.x86.mmx.paddus.w(<4 x i16>, <4 x i16>)
+
+declare void @llvm.x86.mmx.emms()