Skip to content

[ValueTracking] Add support for vector_reduce_{s,u}{min,max} in isKnownNonZero/computeKnownBits. #88169

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 3 commits into from
Closed

[ValueTracking] Add support for vector_reduce_{s,u}{min,max} in isKnownNonZero/computeKnownBits. #88169

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
2ee1477
[InstCombine] Add tests for non-zero/knownbits of `vector_reduce_{s,u…
goldsteinn Apr 9, 2024
6a359c1
[ValueTracking] Add support for `vector_reduce_{s,u}{min,max}` in `is…
goldsteinn Apr 9, 2024
ed78299
[ValueTracking] Add support for `vector_reduce_{s,u}{min,max}` in `co…
goldsteinn Apr 9, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 14 additions & 0 deletions llvm/lib/Analysis/ValueTracking.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1621,6 +1621,14 @@ static void computeKnownBitsFromOperator(const Operator *I,
computeKnownBits(I->getOperand(1), Known2, Depth + 1, Q);
Known = KnownBits::ssub_sat(Known, Known2);
break;
// for min/max reduce, any bit common to each element in the input vec
// is set in the output.
case Intrinsic::vector_reduce_umax:
case Intrinsic::vector_reduce_umin:
case Intrinsic::vector_reduce_smax:
case Intrinsic::vector_reduce_smin:
computeKnownBits(I->getOperand(0), Known, Depth + 1, Q);
break;
case Intrinsic::umin:
computeKnownBits(I->getOperand(0), Known, Depth + 1, Q);
computeKnownBits(I->getOperand(1), Known2, Depth + 1, Q);
Expand Down Expand Up @@ -2824,6 +2832,12 @@ static bool isKnownNonZeroFromOperator(const Operator *I,
return isNonZeroAdd(DemandedElts, Depth, Q, BitWidth,
II->getArgOperand(0), II->getArgOperand(1),
/*NSW=*/true, /* NUW=*/false);
// umin/smin/smax/smin of all non-zero elements is always non-zero.
case Intrinsic::vector_reduce_umax:
case Intrinsic::vector_reduce_umin:
case Intrinsic::vector_reduce_smax:
case Intrinsic::vector_reduce_smin:
return isKnownNonZero(II->getArgOperand(0), Depth, Q);
case Intrinsic::umax:
case Intrinsic::uadd_sat:
return isKnownNonZero(II->getArgOperand(1), DemandedElts, Depth, Q) ||
Expand Down
274 changes: 274 additions & 0 deletions llvm/test/Transforms/InstCombine/vector-reduce-min-max-known.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,274 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -passes=instcombine -S | FileCheck %s

define i1 @vec_reduce_umax_non_zero(<4 x i8> %xx) {
; CHECK-LABEL: @vec_reduce_umax_non_zero(
; CHECK-NEXT: [[X:%.*]] = add nuw <4 x i8> [[XX:%.*]], <i8 0, i8 1, i8 0, i8 0>
; CHECK-NEXT: [[V:%.*]] = call i8 @llvm.vector.reduce.umax.v4i8(<4 x i8> [[X]])
; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[V]], 0
; CHECK-NEXT: ret i1 [[R]]
;
%x = add nuw <4 x i8> %xx, <i8 0, i8 1, i8 0, i8 0>
%v = call i8 @llvm.vector.reduce.umax(<4 x i8> %x)
%r = icmp eq i8 %v, 0
ret i1 %r
}

define i1 @vec_reduce_umax_non_zero_fail(<4 x i8> %xx) {
; CHECK-LABEL: @vec_reduce_umax_non_zero_fail(
; CHECK-NEXT: [[X:%.*]] = add nsw <4 x i8> [[XX:%.*]], <i8 1, i8 1, i8 1, i8 1>
; CHECK-NEXT: [[V:%.*]] = call i8 @llvm.vector.reduce.umax.v4i8(<4 x i8> [[X]])
; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[V]], 0
; CHECK-NEXT: ret i1 [[R]]
;
%x = add nsw <4 x i8> %xx, <i8 1, i8 1, i8 1, i8 1>
%v = call i8 @llvm.vector.reduce.umax(<4 x i8> %x)
%r = icmp eq i8 %v, 0
ret i1 %r
}

define i1 @vec_reduce_umin_non_zero(<4 x i8> %xx) {
; CHECK-LABEL: @vec_reduce_umin_non_zero(
; CHECK-NEXT: ret i1 false
;
%x = add nuw <4 x i8> %xx, <i8 1, i8 1, i8 1, i8 1>
%v = call i8 @llvm.vector.reduce.umin(<4 x i8> %x)
%r = icmp eq i8 %v, 0
ret i1 %r
}

define i1 @vec_reduce_umin_non_zero_fail(<4 x i8> %xx) {
; CHECK-LABEL: @vec_reduce_umin_non_zero_fail(
; CHECK-NEXT: [[X:%.*]] = add nuw <4 x i8> [[XX:%.*]], <i8 0, i8 1, i8 1, i8 1>
; CHECK-NEXT: [[V:%.*]] = call i8 @llvm.vector.reduce.umin.v4i8(<4 x i8> [[X]])
; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[V]], 0
; CHECK-NEXT: ret i1 [[R]]
;
%x = add nuw <4 x i8> %xx, <i8 0, i8 1, i8 1, i8 1>
%v = call i8 @llvm.vector.reduce.umin(<4 x i8> %x)
%r = icmp eq i8 %v, 0
ret i1 %r
}

define i1 @vec_reduce_smax_non_zero0(<4 x i8> %xx) {
; CHECK-LABEL: @vec_reduce_smax_non_zero0(
; CHECK-NEXT: ret i1 false
;
%x = add nuw <4 x i8> %xx, <i8 1, i8 1, i8 1, i8 1>
%v = call i8 @llvm.vector.reduce.smax(<4 x i8> %x)
%r = icmp eq i8 %v, 0
ret i1 %r
}

define i1 @vec_reduce_smax_non_zero1(<4 x i8> %xx) {
; CHECK-LABEL: @vec_reduce_smax_non_zero1(
; CHECK-NEXT: [[X0:%.*]] = and <4 x i8> [[XX:%.*]], <i8 127, i8 -1, i8 -1, i8 -1>
; CHECK-NEXT: [[X:%.*]] = or <4 x i8> [[X0]], <i8 1, i8 0, i8 0, i8 0>
; CHECK-NEXT: [[V:%.*]] = call i8 @llvm.vector.reduce.smax.v4i8(<4 x i8> [[X]])
; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[V]], 0
; CHECK-NEXT: ret i1 [[R]]
;
%x0 = and <4 x i8> %xx, <i8 127, i8 255, i8 255, i8 255>
%x = or <4 x i8> %x0, <i8 1, i8 0, i8 0, i8 0>
%v = call i8 @llvm.vector.reduce.smax(<4 x i8> %x)
%r = icmp eq i8 %v, 0
ret i1 %r
}

define i1 @vec_reduce_smax_non_zero_fail(<4 x i8> %xx) {
; CHECK-LABEL: @vec_reduce_smax_non_zero_fail(
; CHECK-NEXT: [[X0:%.*]] = and <4 x i8> [[XX:%.*]], <i8 127, i8 -1, i8 -1, i8 -1>
; CHECK-NEXT: [[X:%.*]] = add nuw <4 x i8> [[X0]], <i8 1, i8 0, i8 0, i8 0>
; CHECK-NEXT: [[V:%.*]] = call i8 @llvm.vector.reduce.smax.v4i8(<4 x i8> [[X]])
; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[V]], 0
; CHECK-NEXT: ret i1 [[R]]
;
%x0 = and <4 x i8> %xx, <i8 127, i8 255, i8 255, i8 255>
%x = add nuw <4 x i8> %x0, <i8 1, i8 0, i8 0, i8 0>
%v = call i8 @llvm.vector.reduce.smax(<4 x i8> %x)
%r = icmp eq i8 %v, 0
ret i1 %r
}

define i1 @vec_reduce_smin_non_zero0(<4 x i8> %xx) {
; CHECK-LABEL: @vec_reduce_smin_non_zero0(
; CHECK-NEXT: ret i1 false
;
%x = add nuw <4 x i8> %xx, <i8 1, i8 1, i8 1, i8 1>
%v = call i8 @llvm.vector.reduce.smin(<4 x i8> %x)
%r = icmp eq i8 %v, 0
ret i1 %r
}

define i1 @vec_reduce_smin_non_zero1(<4 x i8> %xx) {
; CHECK-LABEL: @vec_reduce_smin_non_zero1(
; CHECK-NEXT: [[X:%.*]] = or <4 x i8> [[XX:%.*]], <i8 0, i8 0, i8 0, i8 -128>
; CHECK-NEXT: [[V:%.*]] = call i8 @llvm.vector.reduce.smin.v4i8(<4 x i8> [[X]])
; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[V]], 0
; CHECK-NEXT: ret i1 [[R]]
;
%x = or <4 x i8> %xx, <i8 0, i8 0, i8 0, i8 128>
%v = call i8 @llvm.vector.reduce.smin(<4 x i8> %x)
%r = icmp eq i8 %v, 0
ret i1 %r
}

define i1 @vec_reduce_smin_non_zero_fail(<4 x i8> %xx) {
; CHECK-LABEL: @vec_reduce_smin_non_zero_fail(
; CHECK-NEXT: [[X0:%.*]] = or <4 x i8> [[XX:%.*]], <i8 0, i8 0, i8 0, i8 -128>
; CHECK-NEXT: [[X:%.*]] = add <4 x i8> [[X0]], <i8 0, i8 0, i8 0, i8 1>
; CHECK-NEXT: [[V:%.*]] = call i8 @llvm.vector.reduce.smin.v4i8(<4 x i8> [[X]])
; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[V]], 0
; CHECK-NEXT: ret i1 [[R]]
;
%x0 = or <4 x i8> %xx, <i8 0, i8 0, i8 0, i8 128>
%x = add <4 x i8> %x0, <i8 0, i8 0, i8 0, i8 1>
%v = call i8 @llvm.vector.reduce.smin(<4 x i8> %x)
%r = icmp eq i8 %v, 0
ret i1 %r
}

define i8 @vec_reduce_umax_known0(<4 x i8> %xx) {
; CHECK-LABEL: @vec_reduce_umax_known0(
; CHECK-NEXT: ret i8 1
;
%x = or <4 x i8> %xx, <i8 1, i8 1, i8 1, i8 1>
%v = call i8 @llvm.vector.reduce.umax(<4 x i8> %x)
%r = and i8 %v, 1
ret i8 %r
}

define i8 @vec_reduce_umax_known1(<4 x i8> %xx) {
; CHECK-LABEL: @vec_reduce_umax_known1(
; CHECK-NEXT: [[X:%.*]] = or <4 x i8> [[XX:%.*]], <i8 1, i8 1, i8 1, i8 -128>
; CHECK-NEXT: [[V:%.*]] = call i8 @llvm.vector.reduce.umax.v4i8(<4 x i8> [[X]])
; CHECK-NEXT: [[R:%.*]] = and i8 [[V]], -128
; CHECK-NEXT: ret i8 [[R]]
;
%x = or <4 x i8> %xx, <i8 1, i8 1, i8 1, i8 128>
%v = call i8 @llvm.vector.reduce.umax(<4 x i8> %x)
%r = and i8 %v, 128
ret i8 %r
}

define i8 @vec_reduce_umax_known_fail0(<4 x i8> %xx) {
; CHECK-LABEL: @vec_reduce_umax_known_fail0(
; CHECK-NEXT: [[X:%.*]] = or <4 x i8> [[XX:%.*]], <i8 1, i8 1, i8 1, i8 -128>
; CHECK-NEXT: [[V:%.*]] = call i8 @llvm.vector.reduce.umax.v4i8(<4 x i8> [[X]])
; CHECK-NEXT: [[R:%.*]] = and i8 [[V]], 1
; CHECK-NEXT: ret i8 [[R]]
;
%x = or <4 x i8> %xx, <i8 1, i8 1, i8 1, i8 128>
%v = call i8 @llvm.vector.reduce.umax(<4 x i8> %x)
%r = and i8 %v, 1
ret i8 %r
}

define i8 @vec_reduce_umax_known_fail1(<4 x i8> %xx) {
; CHECK-LABEL: @vec_reduce_umax_known_fail1(
; CHECK-NEXT: [[X:%.*]] = or <4 x i8> [[XX:%.*]], <i8 1, i8 2, i8 4, i8 8>
; CHECK-NEXT: [[V:%.*]] = call i8 @llvm.vector.reduce.umax.v4i8(<4 x i8> [[X]])
; CHECK-NEXT: [[R:%.*]] = and i8 [[V]], 1
; CHECK-NEXT: ret i8 [[R]]
;
%x = or <4 x i8> %xx, <i8 1, i8 2, i8 4, i8 8>
%v = call i8 @llvm.vector.reduce.umax(<4 x i8> %x)
%r = and i8 %v, 1
ret i8 %r
}

define i8 @vec_reduce_umin_known0(<4 x i8> %xx) {
; CHECK-LABEL: @vec_reduce_umin_known0(
; CHECK-NEXT: ret i8 1
;
%x = or <4 x i8> %xx, <i8 1, i8 1, i8 1, i8 1>
%v = call i8 @llvm.vector.reduce.umin(<4 x i8> %x)
%r = and i8 %v, 1
ret i8 %r
}

define i8 @vec_reduce_umin_known1(<4 x i8> %xx) {
; CHECK-LABEL: @vec_reduce_umin_known1(
; CHECK-NEXT: [[X:%.*]] = and <4 x i8> [[XX:%.*]], <i8 127, i8 -1, i8 -1, i8 -1>
; CHECK-NEXT: [[V:%.*]] = call i8 @llvm.vector.reduce.umin.v4i8(<4 x i8> [[X]])
; CHECK-NEXT: [[R:%.*]] = and i8 [[V]], -128
; CHECK-NEXT: ret i8 [[R]]
;
%x = and <4 x i8> %xx, <i8 127, i8 255, i8 255, i8 255>
%v = call i8 @llvm.vector.reduce.umin(<4 x i8> %x)
%r = and i8 %v, 128
ret i8 %r
}

define i8 @vec_reduce_umin_known_fail0(<4 x i8> %xx) {
; CHECK-LABEL: @vec_reduce_umin_known_fail0(
; CHECK-NEXT: [[X:%.*]] = or <4 x i8> [[XX:%.*]], <i8 1, i8 0, i8 0, i8 0>
; CHECK-NEXT: [[V:%.*]] = call i8 @llvm.vector.reduce.umin.v4i8(<4 x i8> [[X]])
; CHECK-NEXT: [[R:%.*]] = and i8 [[V]], 1
; CHECK-NEXT: ret i8 [[R]]
;
%x0 = and <4 x i8> %xx, <i8 127, i8 255, i8 255, i8 255>
%x = or <4 x i8> %xx, <i8 1, i8 0, i8 0, i8 0>
%v = call i8 @llvm.vector.reduce.umin(<4 x i8> %x)
%r = and i8 %v, 1
ret i8 %r
}

define i8 @vec_reduce_umin_known_fail1(<4 x i8> %xx) {
; CHECK-LABEL: @vec_reduce_umin_known_fail1(
; CHECK-NEXT: [[X:%.*]] = or <4 x i8> [[XX:%.*]], <i8 1, i8 2, i8 4, i8 8>
; CHECK-NEXT: [[V:%.*]] = call i8 @llvm.vector.reduce.umin.v4i8(<4 x i8> [[X]])
; CHECK-NEXT: [[R:%.*]] = and i8 [[V]], 1
; CHECK-NEXT: ret i8 [[R]]
;
%x = or <4 x i8> %xx, <i8 1, i8 2, i8 4, i8 8>
%v = call i8 @llvm.vector.reduce.umin(<4 x i8> %x)
%r = and i8 %v, 1
ret i8 %r
}

define i8 @vec_reduce_smax_known(<4 x i8> %xx) {
; CHECK-LABEL: @vec_reduce_smax_known(
; CHECK-NEXT: ret i8 4
;
%x = or <4 x i8> %xx, <i8 4, i8 4, i8 4, i8 5>
%v = call i8 @llvm.vector.reduce.smax(<4 x i8> %x)
%r = and i8 %v, 4
ret i8 %r
}

define i8 @vec_reduce_smax_known_fail(<4 x i8> %xx) {
; CHECK-LABEL: @vec_reduce_smax_known_fail(
; CHECK-NEXT: [[X:%.*]] = or <4 x i8> [[XX:%.*]], <i8 4, i8 4, i8 8, i8 5>
; CHECK-NEXT: [[V:%.*]] = call i8 @llvm.vector.reduce.umax.v4i8(<4 x i8> [[X]])
; CHECK-NEXT: [[R:%.*]] = and i8 [[V]], 4
; CHECK-NEXT: ret i8 [[R]]
;
%x = or <4 x i8> %xx, <i8 4, i8 4, i8 8, i8 5>
%v = call i8 @llvm.vector.reduce.umax(<4 x i8> %x)
%r = and i8 %v, 4
ret i8 %r
}

define i8 @vec_reduce_smin_known(<4 x i8> %xx) {
; CHECK-LABEL: @vec_reduce_smin_known(
; CHECK-NEXT: ret i8 8
;
%x = or <4 x i8> %xx, <i8 8, i8 24, i8 56, i8 9>
%v = call i8 @llvm.vector.reduce.smin(<4 x i8> %x)
%r = and i8 %v, 8
ret i8 %r
}

define i8 @vec_reduce_smin_known_fail(<4 x i8> %xx) {
; CHECK-LABEL: @vec_reduce_smin_known_fail(
; CHECK-NEXT: [[X:%.*]] = or <4 x i8> [[XX:%.*]], <i8 8, i8 23, i8 56, i8 9>
; CHECK-NEXT: [[V:%.*]] = call i8 @llvm.vector.reduce.smin.v4i8(<4 x i8> [[X]])
; CHECK-NEXT: [[R:%.*]] = and i8 [[V]], 8
; CHECK-NEXT: ret i8 [[R]]
;
%x = or <4 x i8> %xx, <i8 8, i8 23, i8 56, i8 9>
%v = call i8 @llvm.vector.reduce.smin(<4 x i8> %x)
%r = and i8 %v, 8
ret i8 %r
}