6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2024-07-08 18:30:04 +00:00
Code Issues Projects Releases Wiki Activity

add a note.

Browse Source 
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@43444 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Chris Lattner 2007-10-29 06:19:48 +00:00
parent cee291c505
commit dc4756bfda
1 changed files with 33 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

33
lib/Target/X86/README-SSE.txt
View File

@ -708,3 +708,36 @@ The instruction at (1) can be moved out of the main body of the loop.
//===---------------------------------------------------------------------===//
These functions:
#include <xmmintrin.h>
__m128i a;
void x(unsigned short n) {
a = _mm_slli_epi32 (a, n);
}
void y(unsigned n) {
a = _mm_slli_epi32 (a, n);
}
compile to ( -O3 -static -fomit-frame-pointer):
_x:
movzwl 4(%esp), %eax
movd %eax, %xmm0
movaps _a, %xmm1
pslld %xmm0, %xmm1
movaps %xmm1, _a
ret
_y:
movd 4(%esp), %xmm0
movaps _a, %xmm1
pslld %xmm0, %xmm1
movaps %xmm1, _a
ret
"y" looks good, but "x" does silly movzwl stuff around into a GPR. It seems
like movd would be sufficient in both cases as the value is already zero
extended in the 32-bit stack slot IIRC. For signed short, it should also be
save, as a really-signed value would be undefined for pslld.
//===---------------------------------------------------------------------===//