4a54e3eed3
This inverts a fold recently added to IR with:
3491f2f4b0
We can put -bidirectional on the Alive2 examples to show that
the reverse transforms work:
https://alive2.llvm.org/ce/z/8iVQwB
The motivation for the IR change was to improve matching to
'fabs' in IR (see https://github.com/llvm/llvm-project/issues/38828 ),
but it regressed x86 codegen for 'not-quite-fabs' patterns like
(X > -X) ? X : -X.
Ie, when there is no fast-math (nsz), the cmp+select is not a proper
fabs operation, but it does map nicely to the unusual NAN semantics
of MINSS/MAXSS.
I drafted this as a target-independent fold, but it doesn't appear to
help any other targets and seems to cause regressions for SystemZ at
least.
Differential Revision: https://reviews.llvm.org/D122726
43 lines
1.2 KiB
LLVM
43 lines
1.2 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc < %s -mtriple=i686-- -mcpu=i386 | FileCheck %s --check-prefix=BASE
|
|
; RUN: llc < %s -mtriple=i686-- -mcpu=pentiumpro | FileCheck %s --check-prefix=CMOV
|
|
; PR6679
|
|
|
|
define float @foo(float* %col) {
|
|
; BASE-LABEL: foo:
|
|
; BASE: # %bb.0:
|
|
; BASE-NEXT: movl {{[0-9]+}}(%esp), %eax
|
|
; BASE-NEXT: flds (%eax)
|
|
; BASE-NEXT: fld %st(0)
|
|
; BASE-NEXT: fchs
|
|
; BASE-NEXT: fucom %st(1)
|
|
; BASE-NEXT: fnstsw %ax
|
|
; BASE-NEXT: # kill: def $ah killed $ah killed $ax
|
|
; BASE-NEXT: sahf
|
|
; BASE-NEXT: ja .LBB0_2
|
|
; BASE-NEXT: # %bb.1:
|
|
; BASE-NEXT: fstp %st(0)
|
|
; BASE-NEXT: fldz
|
|
; BASE-NEXT: fxch %st(1)
|
|
; BASE-NEXT: .LBB0_2:
|
|
; BASE-NEXT: fstp %st(1)
|
|
; BASE-NEXT: retl
|
|
;
|
|
; CMOV-LABEL: foo:
|
|
; CMOV: # %bb.0:
|
|
; CMOV-NEXT: movl {{[0-9]+}}(%esp), %eax
|
|
; CMOV-NEXT: flds (%eax)
|
|
; CMOV-NEXT: fld %st(0)
|
|
; CMOV-NEXT: fchs
|
|
; CMOV-NEXT: fucomi %st(1), %st
|
|
; CMOV-NEXT: fxch %st(1)
|
|
; CMOV-NEXT: fcmovnbe %st(1), %st
|
|
; CMOV-NEXT: fstp %st(1)
|
|
; CMOV-NEXT: retl
|
|
%t = load float, float* %col
|
|
%t16 = fcmp olt float %t, 0.0
|
|
%t20 = fsub float -0.000000e+00, %t
|
|
%ift = select i1 %t16, float %t20, float %t
|
|
ret float %ift
|
|
}
|