Lines Matching defs:forOp

76 static Operation::operand_range getLowerBoundOperands(AffineForOp forOp) {
77   return forOp.getLowerBoundOperands();
81 static Operation::operand_range getUpperBoundOperands(AffineForOp forOp) {
82   return forOp.getUpperBoundOperands();
87 static Value getOrCreateStep(AffineForOp forOp, OpBuilder &builder) {
88   return builder.create<arith::ConstantIndexOp>(forOp.getLoc(),
89                                                 forOp.getStepAsInt());
94 static Value getOrEmitLowerBound(AffineForOp forOp, OpBuilder &builder) {
95   return lowerAffineLowerBound(forOp, builder);
100 static Value getOrEmitUpperBound(AffineForOp forOp, OpBuilder &builder) {
101   return lowerAffineUpperBound(forOp, builder);
110 static LogicalResult checkAffineLoopNestMappableImpl(AffineForOp forOp,
112   Region &limit = forOp.getRegion();
114     Operation *nested = &forOp.getBody()->front();
115     if (!areValuesDefinedAbove(getLowerBoundOperands(forOp), limit) ||
116         !areValuesDefinedAbove(getUpperBoundOperands(forOp), limit))
117       return forOp.emitError(
126     auto begin = forOp.getBody()->begin(), end = forOp.getBody()->end();
127     if (forOp.getBody()->empty() || std::next(begin, 2) != end)
128       return forOp.emitError("expected perfectly nested loops in the body");
130     if (!(forOp = dyn_cast<AffineForOp>(nested)))
136 static LogicalResult checkAffineLoopNestMappable(AffineForOp forOp,
145     return forOp.emitError("cannot map to more than 3 block dimensions");
148     return forOp.emitError("cannot map to more than 3 thread dimensions");
150   return checkAffineLoopNestMappableImpl(forOp, numBlockDims + numThreadDims);
157   std::optional<AffineForOp> collectBounds(AffineForOp forOp,
175 // mapping a loop nest of depth "numLoops" rooted at "forOp" to a GPU kernel.
180 AffineLoopToGpuConverter::collectBounds(AffineForOp forOp, unsigned numLoops) {
181   OpBuilder builder(forOp.getOperation());
186   AffineForOp currentLoop = forOp;
281 static LogicalResult convertAffineLoopNestToGPULaunch(AffineForOp forOp,
284   if (failed(checkAffineLoopNestMappable(forOp, numBlockDims, numThreadDims)))
289       converter.collectBounds(forOp, numBlockDims + numThreadDims);
292   converter.createLaunch(forOp, *maybeInnerLoop, numBlockDims, numThreadDims);
297 LogicalResult mlir::convertAffineLoopNestToGPULaunch(AffineForOp forOp,
300   return ::convertAffineLoopNestToGPULaunch(forOp, numBlockDims, numThreadDims);