https://bugs.llvm.org/show_bug.cgi?id=42986
Bug ID: 42986
Summary: Load/store combine fails to merge complex address math
Product: libraries
Version: trunk
Hardware: PC
OS: Windows NT
Status: NEW
Severity: enhancement
Priority: P
Component: Common Code Generator Code
Assignee: unassignedb...@nondot.org
Reporter: llvm-...@redking.me.uk
CC: craig.top...@gmail.com, llvm-bugs@lists.llvm.org,
spatel+l...@rotateright.com
Current Codegen: https://godbolt.org/z/LnSMCV
void cpy(const float *src, float *dst,
unsigned src_offset, unsigned dst_offset, unsigned dst_stride) {
float x0 = src[src_offset + 0];
float x1 = src[src_offset + 1];
dst[dst_offset + dst_stride + 0] = x0;
dst[dst_offset + dst_stride + 1] = x1;
}
We fail to recognise the sequential load/stores, and merge to a single i64/f64
load/store pair.
clang -g0 -O3 -march=btver2
cpy:
movl %edx, %eax
incl %edx
movl (%rdi,%rax,4), %eax
movl (%rdi,%rdx,4), %edx
leal (%r8,%rcx), %edi
movl %eax, (%rsi,%rdi,4)
leal 1(%r8,%rcx), %eax
movl %edx, (%rsi,%rax,4)
retq
Th -Oz case is even worse:
clang -g0 -Oz -march=btver2
cpy:
movl %edx, %eax
movl (%rdi,%rax,4), %eax
pushq $1
popq %r9
addl %r9d, %edx
addl %r8d, %ecx
movl (%rdi,%rdx,4), %edx
movl %eax, (%rsi,%rcx,4)
addl %r9d, %ecx
movl %edx, (%rsi,%rcx,4)
retq
--
You are receiving this mail because:
You are on the CC list for the bug._______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
