RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-26 04:32:44 +01:00
Code Issues Projects Releases Wiki Activity

[SCCP] Handle bitcast of vector constants.

Browse Source 
Vectors where all elements have the same known constant range are treated as a
single constant range in the lattice. When bitcasting such vectors, there is a
mis-match between the width of the lattice value (single constant range) and
the original operands (vector). Go to overdefined in that case.

Fixes PR47991.
This commit is contained in:
Florian Hahn 2020-11-03 10:32:38 +00:00
parent be5f4f896c
commit 8985e25f47
2 changed files with 64 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
lib/Transforms/Scalar/SCCP.cpp
View File

@ -843,6 +843,16 @@ void SCCPSolver::visitCastInst(CastInst &I) {
auto &LV = getValueState(&I);
ConstantRange OpRange = OpSt.getConstantRange();
Type *DestTy = I.getDestTy();
// Vectors where all elements have the same known constant range are treated
// as a single constant range in the lattice. When bitcasting such vectors,
// there is a mis-match between the width of the lattice value (single
// constant range) and the original operands (vector). Go to overdefined in
// that case.
if (I.getOpcode() == Instruction::BitCast &&
I.getOperand(0)->getType()->isVectorTy() &&
OpRange.getBitWidth() < DL.getTypeSizeInBits(DestTy))
return (void)markOverdefined(&I);
ConstantRange Res =
OpRange.castOp(I.getOpcode(), DL.getTypeSizeInBits(DestTy));
mergeInValue(LV, &I, ValueLatticeElement::getRange(Res));

55
test/Transforms/SCCP/vector-bitcast.ll
View File

@ -1,11 +1,23 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -sccp -S < %s | FileCheck %s
target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128-n8:16:32-S128"
; CHECK: store volatile <2 x i64> zeroinitializer, <2 x i64>* %p
; rdar://11324230
declare void @use(i1)
define void @foo(<2 x i64>* %p) nounwind {
; CHECK-LABEL: @foo(
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[WHILE_BODY_I:%.*]]
; CHECK: while.body.i:
; CHECK-NEXT: [[VWORKEXPONENT_I_033:%.*]] = phi <4 x i32> [ [[SUB_I_I:%.*]], [[WHILE_BODY_I]] ], [ <i32 939524096, i32 939524096, i32 939524096, i32 939524096>, [[ENTRY:%.*]] ]
; CHECK-NEXT: [[SUB_I_I]] = add <4 x i32> [[VWORKEXPONENT_I_033]], <i32 -8388608, i32 -8388608, i32 -8388608, i32 -8388608>
; CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x i32> [[SUB_I_I]] to <2 x i64>
; CHECK-NEXT: store volatile <2 x i64> zeroinitializer, <2 x i64>* [[P:%.*]], align 16
; CHECK-NEXT: br label [[WHILE_BODY_I]]
;
entry:
br label %while.body.i
@ -18,3 +30,44 @@ while.body.i: ; preds = %while.body.i, %entr
br label %while.body.i
}
%union.V512 = type { <16 x float> }
@i8_mix = dso_local global %union.V512 zeroinitializer
declare <64 x i8> @llvm.abs.v64i8(<64 x i8>, i1 immarg)
; Test for PR47991.
define void @vec_cast_abs() {
; CHECK-LABEL: @vec_cast_abs(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP1:%.*]] = load <64 x i8>, <64 x i8>* bitcast (%union.V512* @i8_mix to <64 x i8>*), align 64
; CHECK-NEXT: [[TMP2:%.*]] = tail call <64 x i8> @llvm.abs.v64i8(<64 x i8> [[TMP1]], i1 false)
; CHECK-NEXT: [[TMP3:%.*]] = bitcast <64 x i8> [[TMP2]] to i512
; CHECK-NEXT: [[CMP_1:%.*]] = icmp eq i512 [[TMP3]], 12
; CHECK-NEXT: call void @use(i1 [[CMP_1]])
; CHECK-NEXT: [[CMP_2:%.*]] = icmp ult i512 [[TMP3]], 500
; CHECK-NEXT: call void @use(i1 [[CMP_2]])
; CHECK-NEXT: [[TMP4:%.*]] = trunc i512 [[TMP3]] to i32
; CHECK-NEXT: [[CMP_3:%.*]] = icmp eq i32 [[TMP4]], 12
; CHECK-NEXT: call void @use(i1 [[CMP_3]])
; CHECK-NEXT: [[CMP_4:%.*]] = icmp ult i32 [[TMP4]], 500
; CHECK-NEXT: call void @use(i1 [[CMP_3]])
; CHECK-NEXT: ret void
;
entry:
%tmp1 = load <64 x i8>, <64 x i8>* bitcast (%union.V512* @i8_mix to <64 x i8>*)
%tmp2 = tail call <64 x i8> @llvm.abs.v64i8(<64 x i8> %tmp1, i1 false)
%tmp3 = bitcast <64 x i8> %tmp2 to i512
%cmp.1 = icmp eq i512 %tmp3, 12
call void @use(i1 %cmp.1)
%cmp.2 = icmp ult i512 %tmp3, 500
call void @use(i1 %cmp.2)
%tmp4 = trunc i512 %tmp3 to i32
%cmp.3 = icmp eq i32 %tmp4, 12
call void @use(i1 %cmp.3)
%cmp.4 = icmp ult i32 %tmp4, 500
call void @use(i1 %cmp.3)
ret void
}