maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[InstCombine] fold umax/umin intrinsics based on demanded bits 

This is a direct translation of the select folds added with
D53033 / D53036 and another step towards canonicalization
using the intrinsics (see D98152).
This commit is contained in:
Sanjay Patel 2021-08-12 12:35:39 -04:00
parent 4b2fd1149d
commit 790c29ab86
2 changed files with 27 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
View File

@ -829,6 +829,29 @@ Value *InstCombinerImpl::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
KnownBitsComputed = true;
break;
}
case Intrinsic::umax: {
// UMax(A, C) == A if ...
// The lowest non-zero bit of DemandMask is higher than the highest
// non-zero bit of C.
const APInt *C;
unsigned CTZ = DemandedMask.countTrailingZeros();
if (match(II->getArgOperand(1), m_APInt(C)) &&
CTZ >= C->getActiveBits())
return II->getArgOperand(0);
break;
}
case Intrinsic::umin: {
// UMin(A, C) == A if ...
// The lowest non-zero bit of DemandMask is higher than the highest
// non-one bit of C.
// This comes from using DeMorgans on the above umax example.
const APInt *C;
unsigned CTZ = DemandedMask.countTrailingZeros();
if (match(II->getArgOperand(1), m_APInt(C)) &&
CTZ >= C->getBitWidth() - C->countLeadingOnes())
return II->getArgOperand(0);
break;
}
default: {
// Handle target specific intrinsics
Optional<Value *> V = targetSimplifyDemandedUseBitsIntrinsic(

12
llvm/test/Transforms/InstCombine/minmax-intrinsics.ll
View File

@ -976,8 +976,7 @@ define i8 @smax_uses(i8 %x, i8 %y, i8 %z) {
define i8 @umax_demand_lshr(i8 %x) {
; CHECK-LABEL: @umax_demand_lshr(
; CHECK-NEXT: [[M:%.*]] = call i8 @llvm.umax.i8(i8 [[X:%.*]], i8 15)
; CHECK-NEXT: [[R:%.*]] = lshr i8 [[M]], 4
; CHECK-NEXT: [[R:%.*]] = lshr i8 [[X:%.*]], 4
; CHECK-NEXT: ret i8 [[R]]
;
%m = call i8 @llvm.umax.i8(i8 %x, i8 15)
@ -987,8 +986,7 @@ define i8 @umax_demand_lshr(i8 %x) {
define i8 @umax_demand_and(i8 %x) {
; CHECK-LABEL: @umax_demand_and(
; CHECK-NEXT: [[M:%.*]] = call i8 @llvm.umax.i8(i8 [[X:%.*]], i8 1)
; CHECK-NEXT: [[R:%.*]] = and i8 [[M]], 10
; CHECK-NEXT: [[R:%.*]] = and i8 [[X:%.*]], 10
; CHECK-NEXT: ret i8 [[R]]
;
%m = call i8 @llvm.umax.i8(i8 1, i8 %x)
@ -998,8 +996,7 @@ define i8 @umax_demand_and(i8 %x) {
define i8 @umin_demand_or_31_30(i8 %x) {
; CHECK-LABEL: @umin_demand_or_31_30(
; CHECK-NEXT: [[M:%.*]] = call i8 @llvm.umin.i8(i8 [[X:%.*]], i8 -30)
; CHECK-NEXT: [[R:%.*]] = or i8 [[M]], 31
; CHECK-NEXT: [[R:%.*]] = or i8 [[X:%.*]], 31
; CHECK-NEXT: ret i8 [[R]]
;
%m = call i8 @llvm.umin.i8(i8 -30, i8 %x)
@ -1009,8 +1006,7 @@ define i8 @umin_demand_or_31_30(i8 %x) {
define i8 @umin_demand_and_7_8(i8 %x) {
; CHECK-LABEL: @umin_demand_and_7_8(
; CHECK-NEXT: [[M:%.*]] = call i8 @llvm.umin.i8(i8 [[X:%.*]], i8 -7)
; CHECK-NEXT: [[R:%.*]] = and i8 [[M]], -8
; CHECK-NEXT: [[R:%.*]] = and i8 [[X:%.*]], -8
; CHECK-NEXT: ret i8 [[R]]
;
%m = call i8 @llvm.umin.i8(i8 %x, i8 -7)