https://bugs.llvm.org/show_bug.cgi?id=42870
Bug ID: 42870
Summary: Regression: LLVM9 trunk misoptimization of
llvm.x86.sse.movmsk.ps for i586-unknown-linux-gnu
Product: libraries
Version: trunk
Hardware: PC
OS: All
Status: NEW
Severity: release blocker
Priority: P
Component: Backend: X86
Assignee: [email protected]
Reporter: [email protected]
CC: [email protected], [email protected],
[email protected], [email protected]
LLVM8 opt and LLVM9 (trunk) opt, do optimize llvm.x86.sse.movmsk.ps
differently. LLVM8 leaves it as is, while LLVM9 transforms it to (see godbolt:
https://godbolt.org/z/35BloZ):
define i32 @_ZN7example15_mm_movemask_ps17h9d7ca884d8f840c4E(<4 x float>*
noalias nocapture readonly dereferenceable(16) %a) unnamed_addr #0 {
start:
%0 = bitcast <4 x float>* %a to <4 x i32>*
%1 = load <4 x i32>, <4 x i32>* %0, align 16
%2 = icmp slt <4 x i32> %1, zeroinitializer
%3 = bitcast <4 x i1> %2 to i4
%4 = zext i4 %3 to i32
ret i32 %4
}
As a consequence, LLVM8 produces the following machine code for mm_movemask_ps
on i586-unknown-linux-gnu:
example::_mm_movemask_ps: # @example::_mm_movemask_ps
mov eax, dword ptr [esp + 4]
movaps xmm0, xmmword ptr [eax]
movmskps eax, xmm0
ret
while LLVM9 produces:
example::_mm_movemask_ps: # @example::_mm_movemask_ps
mov eax, dword ptr [esp + 4]
cmp dword ptr [eax], 0
sets cl
cmp dword ptr [eax + 4], 0
sets dl
add dl, dl
or dl, cl
cmp dword ptr [eax + 8], 0
sets cl
cmp dword ptr [eax + 12], 0
sets al
add al, al
or al, cl
shl al, 2
or al, dl
movzx eax, al
ret
--
You are receiving this mail because:
You are on the CC list for the bug._______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs