| // RUN: mlir-opt %s -pass-pipeline='builtin.module(func.func(cse))' | FileCheck %s |
| |
| // Verify that acc.set (which writes CurrentDeviceIdResource, a root disjoint |
| // from DefaultResource) does not block CSE of identical memref.load operations |
| // (which read DefaultResource). The two resources are disjoint, so the write |
| // cannot conflict with the loads. |
| |
| // CHECK-LABEL: @cse_across_acc_set |
| func.func @cse_across_acc_set(%a: memref<10xf32>, %i: index) -> (f32, f32) { |
| %v1 = memref.load %a[%i] : memref<10xf32> |
| %c42 = arith.constant 42 : i32 |
| acc.set device_num(%c42 : i32) |
| %v2 = memref.load %a[%i] : memref<10xf32> |
| // CHECK: %[[V:.*]] = memref.load |
| // CHECK-NOT: memref.load |
| // CHECK: return %[[V]], %[[V]] : f32, f32 |
| return %v1, %v2 : f32, f32 |
| } |
| |
| // ----- |
| |
| // Two identical acc.gpu_shared_memory ops must not be CSE'd: each reserves a |
| // distinct workgroup-memory slot. |
| // CHECK-LABEL: @cse_gpu_shared_memory_not_merged |
| func.func @cse_gpu_shared_memory_not_merged() -> (memref<8xf32, #gpu.address_space<workgroup>>, memref<8xf32, #gpu.address_space<workgroup>>) { |
| %0 = acc.gpu_shared_memory {num_copies = 1 : i64, static_upper_bound_bytes = 256 : i64} |
| : () -> memref<8xf32, #gpu.address_space<workgroup>> |
| %1 = acc.gpu_shared_memory {num_copies = 1 : i64, static_upper_bound_bytes = 256 : i64} |
| : () -> memref<8xf32, #gpu.address_space<workgroup>> |
| // CHECK: acc.gpu_shared_memory |
| // CHECK: acc.gpu_shared_memory |
| return %0, %1 : memref<8xf32, #gpu.address_space<workgroup>>, memref<8xf32, #gpu.address_space<workgroup>> |
| } |
