[llvm] [LoopUnroll] Consider simplified operands while retrieving TTI instruction cost (PR #70929)
Sergey Kachkov via llvm-commits
llvm-commits at lists.llvm.org
Thu Nov 16 03:02:22 PST 2023
================
@@ -50,7 +50,7 @@ for.end:
}
; CHECK-LABEL: @test_func_addrspacecast_cost_nonfree(
-; CHECK: br i1 %exitcond
+; CHECK-NOT: br i1 %exitcond
----------------
skachkov-sc wrote:
Unfortunately, with this change it's now impossbile to find threshold that keeps old behavour for all tests in this file. The savings in test_func_addrspacecast_cost_nonfree are higher (I assume that for AMDGPU target GEP with constant offset is cheaper in addrspace(3)), so there is no threshold where first 2 tests are unrolled and the last is not.
https://github.com/llvm/llvm-project/pull/70929
More information about the llvm-commits
mailing list