69 lines · plain
1// RUN: mlir-opt %s \2// RUN: | mlir-opt -gpu-lower-to-nvvm-pipeline="cubin-format=%gpu_compilation_format allow-pattern-rollback=0" \3// RUN: | mlir-runner \4// RUN: --shared-libs=%mlir_cuda_runtime \5// RUN: --shared-libs=%mlir_runner_utils \6// RUN: --entry-point-result=void \7// RUN: | FileCheck %s8 9func.func @main() {10 %data = memref.alloc() : memref<2x6xi32>11 %sum = memref.alloc() : memref<2xi32>12 %cst0 = arith.constant 0 : i3213 %cst1 = arith.constant 1 : i3214 %cst2 = arith.constant 2 : i3215 %cst4 = arith.constant 4 : i3216 %cst8 = arith.constant 8 : i3217 %cst16 = arith.constant 16 : i3218 19 %cst3 = arith.constant 3 : i3220 %cst6 = arith.constant 6 : i3221 %cst7 = arith.constant 7 : i3222 %cst10 = arith.constant 10 : i3223 %cst11 = arith.constant 11 : i3224 25 %c0 = arith.constant 0 : index26 %c1 = arith.constant 1 : index27 %c2 = arith.constant 2 : index28 %c3 = arith.constant 3 : index29 %c4 = arith.constant 4 : index30 %c5 = arith.constant 5 : index31 %c6 = arith.constant 6 : index32 33 %cast_data = memref.cast %data : memref<2x6xi32> to memref<*xi32>34 gpu.host_register %cast_data : memref<*xi32>35 %cast_sum = memref.cast %sum : memref<2xi32> to memref<*xi32>36 gpu.host_register %cast_sum : memref<*xi32>37 38 memref.store %cst0, %data[%c0, %c0] : memref<2x6xi32>39 memref.store %cst1, %data[%c0, %c1] : memref<2x6xi32>40 memref.store %cst2, %data[%c0, %c2] : memref<2x6xi32>41 memref.store %cst4, %data[%c0, %c3] : memref<2x6xi32>42 memref.store %cst8, %data[%c0, %c4] : memref<2x6xi32>43 memref.store %cst16, %data[%c0, %c5] : memref<2x6xi32>44 45 memref.store %cst2, %data[%c1, %c0] : memref<2x6xi32>46 memref.store %cst3, %data[%c1, %c1] : memref<2x6xi32>47 memref.store %cst6, %data[%c1, %c2] : memref<2x6xi32>48 memref.store %cst7, %data[%c1, %c3] : memref<2x6xi32>49 memref.store %cst10, %data[%c1, %c4] : memref<2x6xi32>50 memref.store %cst11, %data[%c1, %c5] : memref<2x6xi32>51 52 // OR53 gpu.launch blocks(%bx, %by, %bz) in (%grid_x = %c2, %grid_y = %c1, %grid_z = %c1)54 threads(%tx, %ty, %tz) in (%block_x = %c6, %block_y = %c1, %block_z = %c1) {55 %val = memref.load %data[%bx, %tx] : memref<2x6xi32>56 %reduced = gpu.all_reduce or %val uniform {} : (i32) -> (i32)57 memref.store %reduced, %sum[%bx] : memref<2xi32>58 gpu.terminator59 }60 61 call @printMemrefI32(%cast_sum) : (memref<*xi32>) -> ()62 // CHECK: [31, 15]63 64 return65}66 67func.func private @printMemrefI32(memref<*xi32>)68 69