mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-26 12:43:36 +01:00
b21f9b16ed
As discussed in: https://llvm.org/PR49055 We invert instcombine's add->or transform here because it makes it easier to identify factorization transforms like the mul in the motivating test. This extends the logic added with: https://reviews.llvm.org/rG70472f3 https://reviews.llvm.org/rG93f3d7f (I intentionally kept the formatting fix in this patch to provide more context about the calling logic.)
65 lines
2.2 KiB
LLVM
65 lines
2.2 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
|
|
; RUN: opt < %s -reassociate -S | FileCheck %s
|
|
|
|
define i32 @shl_add(i8 %x) {
|
|
; CHECK-LABEL: @shl_add(
|
|
; CHECK-NEXT: [[CONV:%.*]] = zext i8 [[X:%.*]] to i32
|
|
; CHECK-NEXT: [[SHL:%.*]] = shl nuw nsw i32 [[CONV]], 8
|
|
; CHECK-NEXT: [[SHL2:%.*]] = shl nuw nsw i32 [[CONV]], 16
|
|
; CHECK-NEXT: [[ADD:%.*]] = add nuw nsw i32 [[SHL]], [[SHL2]]
|
|
; CHECK-NEXT: ret i32 [[ADD]]
|
|
;
|
|
%conv = zext i8 %x to i32
|
|
%shl = shl nuw nsw i32 %conv, 8
|
|
%shl2 = shl nuw nsw i32 %conv, 16
|
|
%add = or i32 %shl, %shl2
|
|
ret i32 %add
|
|
}
|
|
|
|
; If we don't know that operands have no common bits set,
|
|
; we can't convert the `or` into an `add`.
|
|
define i32 @test1(i32 %a, i32 %b) {
|
|
; CHECK-LABEL: @test1(
|
|
; CHECK-NEXT: [[C:%.*]] = or i32 [[B:%.*]], [[A:%.*]]
|
|
; CHECK-NEXT: [[C_PLUS_ONE:%.*]] = add i32 [[C]], 1
|
|
; CHECK-NEXT: ret i32 [[C_PLUS_ONE]]
|
|
;
|
|
%c = or i32 %a, %b
|
|
%c.plus.one = add i32 %c, 1
|
|
ret i32 %c.plus.one
|
|
}
|
|
|
|
; But if we *do* know that operands have no common bits set,
|
|
; we *can* convert the `or` into an `add`.
|
|
define i32 @test2(i32 %x, i32 %y) {
|
|
; CHECK-LABEL: @test2(
|
|
; CHECK-NEXT: [[X_NUMLZ:%.*]] = tail call i32 @llvm.ctlz.i32(i32 [[X:%.*]], i1 true), [[RNG0:!range !.*]]
|
|
; CHECK-NEXT: [[RES:%.*]] = add nuw nsw i32 [[X_NUMLZ]], -32
|
|
; CHECK-NEXT: [[RES_PLUS_ONE:%.*]] = add i32 [[RES]], [[Y:%.*]]
|
|
; CHECK-NEXT: ret i32 [[RES_PLUS_ONE]]
|
|
;
|
|
%x.numlz = tail call i32 @llvm.ctlz.i32(i32 %x, i1 true), !range !0
|
|
%res = or i32 %x.numlz, -32
|
|
%res.plus.one = add i32 %res, %y
|
|
ret i32 %res.plus.one
|
|
}
|
|
|
|
; And that allows reassociation in general.
|
|
define i32 @test3(i32 %x, i32 %bit) {
|
|
; CHECK-LABEL: @test3(
|
|
; CHECK-NEXT: [[X_NUMLZ:%.*]] = tail call i32 @llvm.ctlz.i32(i32 [[X:%.*]], i1 true), [[RNG0]]
|
|
; CHECK-NEXT: [[BIT_PLUS_ONE:%.*]] = add i32 [[BIT:%.*]], -31
|
|
; CHECK-NEXT: [[RES:%.*]] = add i32 [[BIT_PLUS_ONE]], [[X_NUMLZ]]
|
|
; CHECK-NEXT: ret i32 [[RES]]
|
|
;
|
|
%x.numlz = tail call i32 @llvm.ctlz.i32(i32 %x, i1 true), !range !0
|
|
%zero.minus.x.numactivebits = or i32 %x.numlz, -32
|
|
%bit.plus.one = add i32 %bit, 1
|
|
%res = add i32 %bit.plus.one, %zero.minus.x.numactivebits
|
|
ret i32 %res
|
|
}
|
|
|
|
declare i32 @llvm.ctlz.i32(i32, i1 immarg) #2
|
|
|
|
!0 = !{i32 0, i32 33}
|