[X86][MMX] Remove completed _mm_cvtsi32_si64 todo

rL322525 - mmx zero constant support
rL322553 - mmx i32 zero extended value
rL326497 - mmx i64 general constant handling

Not all constants are folded, we generate some on the GPRs (similar to SSE build vector) where appropriate

llvm-svn: 326673

diff --git a/llvm/lib/Target/X86/README-MMX.txt b/llvm/lib/Target/X86/README-MMX.txt
index a6c8616..f3a625e 100644 (file)
--- a/llvm/lib/Target/X86/README-MMX.txt
+++ b/llvm/lib/Target/X86/README-MMX.txt
@@ -40,32 +40,3 @@ _qux:
         movl    %edx, 4(%eax)
         addl    $12, %esp
         ret     $4
-
-//===---------------------------------------------------------------------===//
-
-We generate crappy code for this:
-
-__m64 t() {
-  return _mm_cvtsi32_si64(1);
-}
-
-_t:
-       subl    $12, %esp
-       movl    $1, %eax
-       movd    %eax, %mm0
-       movq    %mm0, (%esp)
-       movl    (%esp), %eax
-       movl    4(%esp), %edx
-       addl    $12, %esp
-       ret
-
-The extra stack traffic is covered in the previous entry. But the other reason
-is we are not smart about materializing constants in MMX registers. With -m64
-
-       movl    $1, %eax
-       movd    %eax, %mm0
-       movd    %mm0, %rax
-       ret
-
-We should be using a constantpool load instead:
-       movq    LC0(%rip), %rax