https://llvm.org/bugs/show_bug.cgi?id=30926
Bug ID: 30926
Summary: [X86] cvtsi2ss, cvtsi2sd, and cvtss2sd intrinsic
sequences from clang produce unnecessary vmovss/vmovsd
instructions
Product: libraries
Version: trunk
Hardware: PC
OS: All
Status: NEW
Severity: normal
Priority: P
Component: Backend: X86
Assignee: unassignedb...@nondot.org
Reporter: craig.top...@gmail.com
CC: llvm-bugs@lists.llvm.org
Classification: Unclassified
The code emitted by clang's intrinsics for cvtsi2ss, cvtsi2sd, and cvtss2sd is
lowered to a code sequence that involves and extra movss/movsd or a blend.
For example:
_test_mm_cvtsi32_sd:
vcvtsi2sdl %edi, %xmm1, %xmm1
vblendpd $1, %xmm1, %xmm0, %xmm0
retq
To fix this we need to add more patterns for detecting X86ISD::MOVSS/SD or
blend and emit the intrinsic version of the instruction instead. Similar to
what is being done for ADD/SUB/MUL/DIV.
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
