63 lines · plain
1// RUN: mlir-opt --test-gpu-rewrite -split-input-file %s | FileCheck %s2 3module {4 // CHECK-LABEL: func.func @shuffleF645 // CHECK-SAME: (%[[SZ:.*]]: index, %[[VALUE:.*]]: f64, %[[OFF:.*]]: i32, %[[WIDTH:.*]]: i32, %[[MEM:.*]]: memref<f64, 1>) {6 func.func @shuffleF64(%sz : index, %value: f64, %offset: i32, %width: i32, %mem: memref<f64, 1>) {7 gpu.launch blocks(%bx, %by, %bz) in (%grid_x = %sz, %grid_y = %sz, %grid_z = %sz)8 threads(%tx, %ty, %tz) in (%block_x = %sz, %block_y = %sz, %block_z = %sz) {9 // CHECK: %[[INTVAL:.*]] = arith.bitcast %[[VALUE]] : f64 to i6410 // CHECK-NEXT: %[[LO:.*]] = arith.trunci %[[INTVAL]] : i64 to i3211 // CHECK-NEXT: %[[HI64:.*]] = arith.shrui %[[INTVAL]], %[[C32:.*]] : i6412 // CHECK-NEXT: %[[HI:.*]] = arith.trunci %[[HI64]] : i64 to i3213 // CHECK-NEXT: %[[SH1:.*]], %[[V1:.*]] = gpu.shuffle xor %[[LO]], %[[OFF]], %[[WIDTH]] : i3214 // CHECK-NEXT: %[[SH2:.*]], %[[V2:.*]] = gpu.shuffle xor %[[HI]], %[[OFF]], %[[WIDTH]] : i3215 // CHECK-NEXT: %[[LOSH:.*]] = arith.extui %[[SH1]] : i32 to i6416 // CHECK-NEXT: %[[HISHTMP:.*]] = arith.extui %[[SH2]] : i32 to i6417 // CHECK-NEXT: %[[HISH:.*]] = arith.shli %[[HISHTMP]], %[[C32]] : i6418 // CHECK-NEXT: %[[SHFLINT:.*]] = arith.ori %[[HISH]], %[[LOSH]] : i6419 // CHECK-NEXT: = arith.bitcast %[[SHFLINT]] : i64 to f6420 %shfl, %pred = gpu.shuffle xor %value, %offset, %width : f6421 memref.store %shfl, %mem[] : memref<f64, 1>22 gpu.terminator23 }24 return25 }26}27 28// -----29 30module {31 // CHECK-LABEL: func.func @shuffleI6432 // CHECK-SAME: (%[[SZ:.*]]: index, %[[VALUE:.*]]: i64, %[[OFF:.*]]: i32, %[[WIDTH:.*]]: i32, %[[MEM:.*]]: memref<i64, 1>) {33 func.func @shuffleI64(%sz : index, %value: i64, %offset: i32, %width: i32, %mem: memref<i64, 1>) {34 gpu.launch blocks(%bx, %by, %bz) in (%grid_x = %sz, %grid_y = %sz, %grid_z = %sz)35 threads(%tx, %ty, %tz) in (%block_x = %sz, %block_y = %sz, %block_z = %sz) {36 // CHECK: %[[LO:.*]] = arith.trunci %[[VALUE]] : i64 to i3237 // CHECK-NEXT: %[[HI64:.*]] = arith.shrui %[[VALUE]], %[[C32:.*]] : i6438 // CHECK-NEXT: %[[HI:.*]] = arith.trunci %[[HI64]] : i64 to i3239 // CHECK-NEXT: %[[SH1:.*]], %[[V1:.*]] = gpu.shuffle xor %[[LO]], %[[OFF]], %[[WIDTH]] : i3240 // CHECK-NEXT: %[[SH2:.*]], %[[V2:.*]] = gpu.shuffle xor %[[HI]], %[[OFF]], %[[WIDTH]] : i3241 // CHECK-NEXT: %[[LOSH:.*]] = arith.extui %[[SH1]] : i32 to i6442 // CHECK-NEXT: %[[HISHTMP:.*]] = arith.extui %[[SH2]] : i32 to i6443 // CHECK-NEXT: %[[HISH:.*]] = arith.shli %[[HISHTMP]], %[[C32]] : i6444 // CHECK-NEXT: %[[SHFLINT:.*]] = arith.ori %[[HISH]], %[[LOSH]] : i6445 %shfl, %pred = gpu.shuffle xor %value, %offset, %width : i6446 memref.store %shfl, %mem[] : memref<i64, 1>47 gpu.terminator48 }49 return50 }51}52 53// -----54 55// CHECK-LABEL: @gpu_shuffle_unsupported56func.func @gpu_shuffle_unsupported(%arg0 : vector<4xf16>) -> vector<4xf16> {57 %offset = arith.constant 4 : i3258 %width = arith.constant 64 : i3259 // CHECK: gpu.shuffle xor %{{.*}}, %{{.*}}, %{{.*}} : vector<4xf16>60 %shfl, %pred = gpu.shuffle xor %arg0, %offset, %width : vector<4xf16>61 return %shfl : vector<4xf16>62}63