mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-22 18:54:02 +01:00
8998944672
This reverses the existing transform that would uniformly canonicalize any 'xor' after any shift. In the case of logical shifts, that turns a 'not' into an arbitrary 'xor' with constant, and that's probably not as good for analysis, SCEV, or codegen. The SCEV motivating case is discussed in: http://bugs.llvm.org/PR47136 There's an analysis motivating case at: http://bugs.llvm.org/PR38781 I did draft a patch that would do the same for 'ashr' but that's questionable because it's just swapping the position of a 'not' and uncovers at least 2 missing folds that we would probably need to deal with as preliminary steps. Alive proofs: https://rise4fun.com/Alive/BBV Name: shift right of 'not' Pre: C2 == (-1 u>> C1) %a = lshr i8 %x, C1 %r = xor i8 %a, C2 => %n = xor i8 %x, -1 %r = lshr i8 %n, C1 Name: shift left of 'not' Pre: C2 == (-1 << C1) %a = shl i8 %x, C1 %r = xor i8 %a, C2 => %n = xor i8 %x, -1 %r = shl i8 %n, C1 Name: ashr of 'not' %a = ashr i8 %x, C1 %r = xor i8 %a, -1 => %n = xor i8 %x, -1 %r = ashr i8 %n, C1 Differential Revision: https://reviews.llvm.org/D86243
151 lines
4.4 KiB
LLVM
151 lines
4.4 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
|
|
; RUN: opt -instcombine -S < %s | FileCheck %s
|
|
|
|
; PR5438
|
|
|
|
define i32 @test1(i32 %a, i32 %b) nounwind readnone {
|
|
; CHECK-LABEL: @test1(
|
|
; CHECK-NEXT: [[TMP1:%.*]] = xor i32 [[B:%.*]], [[A:%.*]]
|
|
; CHECK-NEXT: [[TMP2:%.*]] = xor i32 [[TMP1]], -1
|
|
; CHECK-NEXT: [[DOTLOBIT_NOT:%.*]] = lshr i32 [[TMP2]], 31
|
|
; CHECK-NEXT: ret i32 [[DOTLOBIT_NOT]]
|
|
;
|
|
%t0 = icmp sgt i32 %a, -1
|
|
%t1 = icmp slt i32 %b, 0
|
|
%t2 = xor i1 %t1, %t0
|
|
%t3 = zext i1 %t2 to i32
|
|
ret i32 %t3
|
|
}
|
|
|
|
; TODO: This optimizes partially but not all the way.
|
|
define i32 @test2(i32 %a, i32 %b) nounwind readnone {
|
|
; CHECK-LABEL: @test2(
|
|
; CHECK-NEXT: [[TMP1:%.*]] = xor i32 [[A:%.*]], [[B:%.*]]
|
|
; CHECK-NEXT: [[TMP2:%.*]] = lshr i32 [[TMP1]], 3
|
|
; CHECK-NEXT: [[DOTLOBIT:%.*]] = and i32 [[TMP2]], 1
|
|
; CHECK-NEXT: [[TMP3:%.*]] = xor i32 [[DOTLOBIT]], 1
|
|
; CHECK-NEXT: ret i32 [[TMP3]]
|
|
;
|
|
%t0 = and i32 %a, 8
|
|
%t1 = and i32 %b, 8
|
|
%t2 = icmp eq i32 %t0, %t1
|
|
%t3 = zext i1 %t2 to i32
|
|
ret i32 %t3
|
|
}
|
|
|
|
define i32 @test3(i32 %a, i32 %b) nounwind readnone {
|
|
; CHECK-LABEL: @test3(
|
|
; CHECK-NEXT: [[T2_UNSHIFTED:%.*]] = xor i32 [[A:%.*]], [[B:%.*]]
|
|
; CHECK-NEXT: [[TMP1:%.*]] = xor i32 [[T2_UNSHIFTED]], -1
|
|
; CHECK-NEXT: [[T2_UNSHIFTED_LOBIT_NOT:%.*]] = lshr i32 [[TMP1]], 31
|
|
; CHECK-NEXT: ret i32 [[T2_UNSHIFTED_LOBIT_NOT]]
|
|
;
|
|
%t0 = lshr i32 %a, 31
|
|
%t1 = lshr i32 %b, 31
|
|
%t2 = icmp eq i32 %t0, %t1
|
|
%t3 = zext i1 %t2 to i32
|
|
ret i32 %t3
|
|
}
|
|
|
|
; TODO this should optimize but doesn't due to missing vector support in InstCombiner::foldICmpEquality.
|
|
define <2 x i32> @test3vec(<2 x i32> %a, <2 x i32> %b) nounwind readnone {
|
|
; CHECK-LABEL: @test3vec(
|
|
; CHECK-NEXT: [[T0:%.*]] = lshr <2 x i32> [[A:%.*]], <i32 31, i32 31>
|
|
; CHECK-NEXT: [[T1:%.*]] = lshr <2 x i32> [[B:%.*]], <i32 31, i32 31>
|
|
; CHECK-NEXT: [[T2:%.*]] = icmp eq <2 x i32> [[T0]], [[T1]]
|
|
; CHECK-NEXT: [[T3:%.*]] = zext <2 x i1> [[T2]] to <2 x i32>
|
|
; CHECK-NEXT: ret <2 x i32> [[T3]]
|
|
;
|
|
%t0 = lshr <2 x i32> %a, <i32 31, i32 31>
|
|
%t1 = lshr <2 x i32> %b, <i32 31, i32 31>
|
|
%t2 = icmp eq <2 x i32> %t0, %t1
|
|
%t3 = zext <2 x i1> %t2 to <2 x i32>
|
|
ret <2 x i32> %t3
|
|
}
|
|
|
|
; Variation on @test3: checking the 2nd bit in a situation where the 5th bit
|
|
; is one, not zero.
|
|
define i32 @test3i(i32 %a, i32 %b) nounwind readnone {
|
|
; CHECK-LABEL: @test3i(
|
|
; CHECK-NEXT: [[T01:%.*]] = xor i32 [[A:%.*]], [[B:%.*]]
|
|
; CHECK-NEXT: [[TMP1:%.*]] = xor i32 [[T01]], -1
|
|
; CHECK-NEXT: [[T4:%.*]] = lshr i32 [[TMP1]], 31
|
|
; CHECK-NEXT: ret i32 [[T4]]
|
|
;
|
|
%t0 = lshr i32 %a, 29
|
|
%t1 = lshr i32 %b, 29
|
|
%t2 = or i32 %t0, 35
|
|
%t3 = or i32 %t1, 35
|
|
%t4 = icmp eq i32 %t2, %t3
|
|
%t5 = zext i1 %t4 to i32
|
|
ret i32 %t5
|
|
}
|
|
|
|
define i1 @test4a(i32 %a) {
|
|
; CHECK-LABEL: @test4a(
|
|
; CHECK-NEXT: [[C:%.*]] = icmp slt i32 [[A:%.*]], 1
|
|
; CHECK-NEXT: ret i1 [[C]]
|
|
;
|
|
%l = ashr i32 %a, 31
|
|
%na = sub i32 0, %a
|
|
%r = lshr i32 %na, 31
|
|
%signum = or i32 %l, %r
|
|
%c = icmp slt i32 %signum, 1
|
|
ret i1 %c
|
|
}
|
|
|
|
define <2 x i1> @test4a_vec(<2 x i32> %a) {
|
|
; CHECK-LABEL: @test4a_vec(
|
|
; CHECK-NEXT: [[C:%.*]] = icmp slt <2 x i32> [[A:%.*]], <i32 1, i32 1>
|
|
; CHECK-NEXT: ret <2 x i1> [[C]]
|
|
;
|
|
%l = ashr <2 x i32> %a, <i32 31, i32 31>
|
|
%na = sub <2 x i32> zeroinitializer, %a
|
|
%r = lshr <2 x i32> %na, <i32 31, i32 31>
|
|
%signum = or <2 x i32> %l, %r
|
|
%c = icmp slt <2 x i32> %signum, <i32 1, i32 1>
|
|
ret <2 x i1> %c
|
|
}
|
|
|
|
define i1 @test4b(i64 %a) {
|
|
; CHECK-LABEL: @test4b(
|
|
; CHECK-NEXT: [[C:%.*]] = icmp slt i64 [[A:%.*]], 1
|
|
; CHECK-NEXT: ret i1 [[C]]
|
|
;
|
|
%l = ashr i64 %a, 63
|
|
%na = sub i64 0, %a
|
|
%r = lshr i64 %na, 63
|
|
%signum = or i64 %l, %r
|
|
%c = icmp slt i64 %signum, 1
|
|
ret i1 %c
|
|
}
|
|
|
|
define i1 @test4c(i64 %a) {
|
|
; CHECK-LABEL: @test4c(
|
|
; CHECK-NEXT: [[C:%.*]] = icmp slt i64 [[A:%.*]], 1
|
|
; CHECK-NEXT: ret i1 [[C]]
|
|
;
|
|
%l = ashr i64 %a, 63
|
|
%na = sub i64 0, %a
|
|
%r = lshr i64 %na, 63
|
|
%signum = or i64 %l, %r
|
|
%signum.trunc = trunc i64 %signum to i32
|
|
%c = icmp slt i32 %signum.trunc, 1
|
|
ret i1 %c
|
|
}
|
|
|
|
define <2 x i1> @test4c_vec(<2 x i64> %a) {
|
|
; CHECK-LABEL: @test4c_vec(
|
|
; CHECK-NEXT: [[C:%.*]] = icmp slt <2 x i64> [[A:%.*]], <i64 1, i64 1>
|
|
; CHECK-NEXT: ret <2 x i1> [[C]]
|
|
;
|
|
%l = ashr <2 x i64> %a, <i64 63, i64 63>
|
|
%na = sub <2 x i64> zeroinitializer, %a
|
|
%r = lshr <2 x i64> %na, <i64 63, i64 63>
|
|
%signum = or <2 x i64> %l, %r
|
|
%signum.trunc = trunc <2 x i64> %signum to <2 x i32>
|
|
%c = icmp slt <2 x i32> %signum.trunc, <i32 1, i32 1>
|
|
ret <2 x i1> %c
|
|
}
|
|
|