Searched refs:kWarpSize (Results 1 – 4 of 4) sorted by relevance
| /external/tensorflow/tensorflow/core/kernels/ |
| D | bias_op_gpu.cu.cc | 207 static constexpr int32 kWarpSize = 32; in compute() local
226 if (config.thread_per_block < kWarpSize) { in compute()
227 config.thread_per_block = kWarpSize; in compute()
|
| /external/tensorflow/tensorflow/compiler/xla/service/gpu/ |
| D | ir_emission_utils.h | 59 constexpr int64 kWarpSize = 32; variable
|
| D | ir_emission_utils.cc | 218 {all_warps_mask, value, offset, builder->getInt32(kWarpSize - 1)}, {}, in EmitFullWarpShuffleDown()
236 builder->getInt32(kWarpSize - 1)}, in EmitFullWarpShuffleDown()
|
| D | ir_emitter_unnested.cc | 3121 mapping_scheme->GetNumberOfThreadsForDimensionX() == kWarpSize ? x in EmitKernel()
3241 reduced_output_dims, /*tile_size_y=*/kWarpSize, in EmitHlo021Tile()
3242 /*tile_size_x=*/kWarpSize, /*req_block_sizes=*/{1, 1, 1}, in EmitHlo021Tile()
3244 /*num_threads_x=*/kWarpSize, &b_); in EmitHlo021Tile()
3672 tile_size_x = kWarpSize * 16; in ComputeMappingSchemeAndReductionKind()
3673 num_threads_x = kWarpSize; in ComputeMappingSchemeAndReductionKind()
3705 num_threads_x = kWarpSize; in ComputeMappingSchemeAndReductionKind()
3706 if (width % (kWarpSize * 64) == 0) { in ComputeMappingSchemeAndReductionKind()
3707 tile_size_x = kWarpSize * 64; in ComputeMappingSchemeAndReductionKind()
3709 tile_size_x = kWarpSize * 8; in ComputeMappingSchemeAndReductionKind()
|