[CIR] Fix Lowering/ptrstride.cir xfailed after rebasing #1302

AmrDeveloper · 2025-01-29T18:56:40Z

Fixing Lit test after rebasing

CIRToLLVMPtrStrideOpLowering will not emit casting in this case because the width is equal to *layoutWidth

clangir/clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp

Lines 967 to 999 in d329c96

    
           if (indexOp && layoutWidth && width != *layoutWidth) { 
        
             // If the index comes from a subtraction, make sure the extension happens 
        
             // before it. To achieve that, look at unary minus, which already got 
        
             // lowered to "sub 0, x". 
        
             auto sub = dyn_cast<mlir::LLVM::SubOp>(indexOp); 
        
             auto unary = dyn_cast_if_present<cir::UnaryOp>( 
        
                 ptrStrideOp.getStride().getDefiningOp()); 
        
             bool rewriteSub = 
        
                 unary && unary.getKind() == cir::UnaryOpKind::Minus && sub; 
        
             if (rewriteSub) 
        
               index = indexOp->getOperand(1); 
        
             // Handle the cast 
        
             auto llvmDstType = mlir::IntegerType::get(ctx, *layoutWidth); 
        
             index = getLLVMIntCast(rewriter, index, llvmDstType, 
        
                                    ptrStrideOp.getStride().getType().isUnsigned(), 
        
                                    width, *layoutWidth); 
        
             // Rewrite the sub in front of extensions/trunc 
        
             if (rewriteSub) { 
        
               index = rewriter.create<mlir::LLVM::SubOp>( 
        
                   index.getLoc(), index.getType(), 
        
                   rewriter.create<mlir::LLVM::ConstantOp>( 
        
                       index.getLoc(), index.getType(), 
        
                       mlir::IntegerAttr::get(index.getType(), 0)), 
        
                   index); 
        
               rewriter.eraseOp(sub); 
        
             } 
        
           } 
        
           rewriter.replaceOpWithNewOp<mlir::LLVM::GEPOp>( 
        
               ptrStrideOp, resultTy, elementTy, adaptor.getBase(), index); 
        
           return mlir::success();

Fixes: #1295

AmrDeveloper · 2025-01-29T18:59:36Z

Is there a reason to expect a cast instruction in this case? From my understanding should emit GEPOp?

@bcardosolopes, @lanza

[CIR] Fix Lowering/ptrstride.cir xfailed after rebasing

3624e4a

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[CIR] Fix Lowering/ptrstride.cir xfailed after rebasing #1302

[CIR] Fix Lowering/ptrstride.cir xfailed after rebasing #1302

AmrDeveloper commented Jan 29, 2025 •

edited

Loading

AmrDeveloper commented Jan 29, 2025

	if (indexOp && layoutWidth && width != *layoutWidth) {
	// If the index comes from a subtraction, make sure the extension happens
	// before it. To achieve that, look at unary minus, which already got
	// lowered to "sub 0, x".
	auto sub = dyn_cast<mlir::LLVM::SubOp>(indexOp);
	auto unary = dyn_cast_if_present<cir::UnaryOp>(
	ptrStrideOp.getStride().getDefiningOp());
	bool rewriteSub =
	unary && unary.getKind() == cir::UnaryOpKind::Minus && sub;
	if (rewriteSub)
	index = indexOp->getOperand(1);

	// Handle the cast
	auto llvmDstType = mlir::IntegerType::get(ctx, *layoutWidth);
	index = getLLVMIntCast(rewriter, index, llvmDstType,
	ptrStrideOp.getStride().getType().isUnsigned(),
	width, *layoutWidth);

	// Rewrite the sub in front of extensions/trunc
	if (rewriteSub) {
	index = rewriter.create<mlir::LLVM::SubOp>(
	index.getLoc(), index.getType(),
	rewriter.create<mlir::LLVM::ConstantOp>(
	index.getLoc(), index.getType(),
	mlir::IntegerAttr::get(index.getType(), 0)),
	index);
	rewriter.eraseOp(sub);
	}
	}

	rewriter.replaceOpWithNewOp<mlir::LLVM::GEPOp>(
	ptrStrideOp, resultTy, elementTy, adaptor.getBase(), index);
	return mlir::success();

[CIR] Fix Lowering/ptrstride.cir xfailed after rebasing #1302

Are you sure you want to change the base?

[CIR] Fix Lowering/ptrstride.cir xfailed after rebasing #1302

Conversation

AmrDeveloper commented Jan 29, 2025 • edited Loading

AmrDeveloper commented Jan 29, 2025

AmrDeveloper commented Jan 29, 2025 •

edited

Loading