46 lines · plain
1// RUN: mlir-opt %s -split-input-file -async-parallel-for=num-workers=-1 \2// RUN: | FileCheck %s --dump-input=always3 4// CHECK-LABEL: @num_worker_threads(5// CHECK: %[[MEMREF:.*]]: memref<?xf32>6func.func @num_worker_threads(%arg0: memref<?xf32>) {7 8 // CHECK-DAG: %[[scalingCstInit:.*]] = arith.constant 8.000000e+00 : f329 // CHECK-DAG: %[[bracketLowerBound4:.*]] = arith.constant 4 : index10 // CHECK-DAG: %[[scalingCst4:.*]] = arith.constant 4.000000e+00 : f3211 // CHECK-DAG: %[[bracketLowerBound8:.*]] = arith.constant 8 : index12 // CHECK-DAG: %[[scalingCst8:.*]] = arith.constant 2.000000e+00 : f3213 // CHECK-DAG: %[[bracketLowerBound16:.*]] = arith.constant 16 : index14 // CHECK-DAG: %[[scalingCst16:.*]] = arith.constant 1.000000e+00 : f3215 // CHECK-DAG: %[[bracketLowerBound32:.*]] = arith.constant 32 : index16 // CHECK-DAG: %[[scalingCst32:.*]] = arith.constant 8.000000e-01 : f3217 // CHECK-DAG: %[[bracketLowerBound64:.*]] = arith.constant 64 : index18 // CHECK-DAG: %[[scalingCst64:.*]] = arith.constant 6.000000e-01 : f3219 // CHECK: %[[workersIndex:.*]] = async.runtime.num_worker_threads : index20 // CHECK: %[[inBracket4:.*]] = arith.cmpi sgt, %[[workersIndex]], %[[bracketLowerBound4]] : index21 // CHECK: %[[scalingFactor4:.*]] = arith.select %[[inBracket4]], %[[scalingCst4]], %[[scalingCstInit]] : f3222 // CHECK: %[[inBracket8:.*]] = arith.cmpi sgt, %[[workersIndex]], %[[bracketLowerBound8]] : index23 // CHECK: %[[scalingFactor8:.*]] = arith.select %[[inBracket8]], %[[scalingCst8]], %[[scalingFactor4]] : f3224 // CHECK: %[[inBracket16:.*]] = arith.cmpi sgt, %[[workersIndex]], %[[bracketLowerBound16]] : index25 // CHECK: %[[scalingFactor16:.*]] = arith.select %[[inBracket16]], %[[scalingCst16]], %[[scalingFactor8]] : f3226 // CHECK: %[[inBracket32:.*]] = arith.cmpi sgt, %[[workersIndex]], %[[bracketLowerBound32]] : index27 // CHECK: %[[scalingFactor32:.*]] = arith.select %[[inBracket32]], %[[scalingCst32]], %[[scalingFactor16]] : f3228 // CHECK: %[[inBracket64:.*]] = arith.cmpi sgt, %[[workersIndex]], %[[bracketLowerBound64]] : index29 // CHECK: %[[scalingFactor64:.*]] = arith.select %[[inBracket64]], %[[scalingCst64]], %[[scalingFactor32]] : f3230 // CHECK: %[[workersInt:.*]] = arith.index_cast %[[workersIndex]] : index to i3231 // CHECK: %[[workersFloat:.*]] = arith.sitofp %[[workersInt]] : i32 to f3232 // CHECK: %[[scaledFloat:.*]] = arith.mulf %[[scalingFactor64]], %[[workersFloat]] : f3233 // CHECK: %[[scaledInt:.*]] = arith.fptosi %[[scaledFloat]] : f32 to i3234 // CHECK: %[[scaledIndex:.*]] = arith.index_cast %[[scaledInt]] : i32 to index35 36 %lb = arith.constant 0 : index37 %ub = arith.constant 100 : index38 %st = arith.constant 1 : index39 scf.parallel (%i) = (%lb) to (%ub) step (%st) {40 %one = arith.constant 1.0 : f3241 memref.store %one, %arg0[%i] : memref<?xf32>42 }43 44 return45}46