[PATCH] D141974: [X86] Removing 'TuningSlow3OpsLEA' from ICL config
Noah Goldstein via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Tue Jan 17 19:14:18 PST 2023
goldstein.w.n added a comment.
In D141974#4060943 <https://reviews.llvm.org/D141974#4060943>, @pengfei wrote:
> I'm not sure of that. There's no description in Intel SOM. I also googled about it when writting #60043 and just found LEA was improved in Golden Cove, e.g., https://www.hardwaretimes.com/intel-golden-cove-core-architecture-deep-dive-vs-zen-3-and-sunny-cove/
Fairly certain its been changed. uops.info is generally reliable also tested just now on ICX:
.global _start
.p2align 6
.text
_start:
movl $10000000, %eax
xorl %edx, %edx
loop:
leaq 1(%rdx, %rax, 8), %rdx
decl %eax
jnz loop
movl $60, %eax
xorl %edi, %edi
syscall
Results in:
10,002,046 cycles
4,678,897 p0
4,678,955 p1
5,321,396 p5
5,321,559 p6
Should be 30,000,000 cycles if was 3c latency.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D141974/new/
https://reviews.llvm.org/D141974
More information about the llvm-commits
mailing list