Skip to content

Commit

Permalink
[BugFix][MetaSchedule] MultiLevelTilingTensorCore generates inconsist…
Browse files Browse the repository at this point in the history 
…ent thread-binding sketch for batched matmul (#17012)

* [BugFix][MetaSchedule] MultiLevelTilingTensorCore generates inconsistent thread-binding sketch for batched matmul

* Update testcase test_meta_schedule_schedule_rule_mlt_tc.py::test_conv_1x1

---------

Co-authored-by: tsu-bin <[email protected]>
  • Loading branch information
@tsu-bin
tsu-bin and tsu-bin committed Jun 28, 2024
1 parent a84adaf commit 4a5e22e
Show file tree
Hide file tree
Showing 4 changed files with 70 additions and 50 deletions.
23 changes: 22 additions & 1 deletion src/meta_schedule/schedule_rule/multi_level_tiling.cc
View file
Original file line number Diff line number Diff line change
Expand Up @@ -190,7 +190,8 @@ std::pair<Array<tir::ExprRV>, Array<tir::LoopRV>> MultiLevelTilingNode::SplitLoo
return {factors, splits};
}

std::vector<State> MultiLevelTilingNode::TileLoopNest(State state) const {
std::vector<State> MultiLevelTilingNode::TileLoopNest(State state,
int tile_inner_most_space_loop_num) const {
Schedule& sch = state->sch;
const BlockRV& block_rv = state->block_rv;
// Step 1. Assuming trivial binding, pair the loops and their iter-var-types
Expand All @@ -199,6 +200,16 @@ std::vector<State> MultiLevelTilingNode::TileLoopNest(State state) const {
ICHECK_EQ(loops.size(), iter_types.size());
// Step 2. For each loop axis, tile it
int64_t spatial_loop_product = 1;

int total_spatial_loop_num = 0;
std::for_each(iter_types.begin(), iter_types.end(), [&](const auto& iter_type) {
if (iter_type == IterVarType::kDataPar) total_spatial_loop_num++;
});
CHECK_GE(total_spatial_loop_num, tile_inner_most_space_loop_num);
if (tile_inner_most_space_loop_num < 0) tile_inner_most_space_loop_num = total_spatial_loop_num;
int outer_most_spatial_loop_skipped_num = total_spatial_loop_num - tile_inner_most_space_loop_num;

Array<LoopRV> skipped_outer_spatial_loops;
std::vector<Array<LoopRV>> tiles(s_indices_.size() + r_indices_.size());
state->tile_factors.resize(tiles.size());
std::vector<Array<tir::ExprRV>> tile_factors;
Expand All @@ -208,6 +219,11 @@ std::vector<State> MultiLevelTilingNode::TileLoopNest(State state) const {
const std::vector<int>* idx = nullptr;

if (iter_types[i] == IterVarType::kDataPar) {
if (outer_most_spatial_loop_skipped_num > 0) {
skipped_outer_spatial_loops.push_back(loop);
outer_most_spatial_loop_skipped_num--;
continue;
}
idx = &s_indices_;
if (spatial_loop_product != -1) {
if (const int64_t* extent = tir::GetLoopIntExtent(sch->Get(loop).get())) {
Expand Down Expand Up @@ -241,6 +257,11 @@ std::vector<State> MultiLevelTilingNode::TileLoopNest(State state) const {
sch->Reorder(support::ConcatArrayList<LoopRV>(tiles.begin(), tiles.end()));
// Step 4. Bind the tiles to threads
int n_binds = std::min(tile_binds.size(), tiles.size());
if (skipped_outer_spatial_loops.size() && n_binds) {
auto& the_first_tile = tiles[0];
the_first_tile.insert(the_first_tile.begin(), skipped_outer_spatial_loops.begin(),
skipped_outer_spatial_loops.end());
}
for (int i = 0; i < n_binds; ++i) {
LoopRV fused = sch->Fuse(tiles[i]);
sch->Bind(fused, tile_binds[i]);
Expand Down
2 changes: 1 addition & 1 deletion src/meta_schedule/schedule_rule/multi_level_tiling.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -162,7 +162,7 @@ class MultiLevelTilingNode : public ScheduleRuleNode {
// SubRule 1. add write cache
std::vector<State> AddWriteReuse(State state) const;
// SubRule 2. tile the loop nest
std::vector<State> TileLoopNest(State state) const;
std::vector<State> TileLoopNest(State state, int tile_inner_most_space_loop_num = -1) const;
// SubRule 3. add read cache
std::vector<State> AddReadReuse(State state) const;
// SubRule 4. add async pipeline
Expand Down
2 changes: 1 addition & 1 deletion src/meta_schedule/schedule_rule/multi_level_tiling_tensor_core.cc
View file
Original file line number Diff line number Diff line change
Expand Up @@ -251,7 +251,7 @@ std::vector<State> MultiLevelTilingTensorCoreNode::ApplySubRules(std::vector<Sta
});
states = SubRule(std::move(states), [&](State state) {
TensorCoreState tc_state = Downcast<TensorCoreState>(state);
return tc_state->is_mma ? MMATileLoopNest(tc_state) : TileLoopNest(state);
return tc_state->is_mma ? MMATileLoopNest(tc_state) : TileLoopNest(state, 2);
});
states = SubRule(std::move(states), [&](State state) {
return TransformIntermediateOutputLayout(Downcast<TensorCoreState>(state));
Expand Down
93 changes: 46 additions & 47 deletions tests/python/meta_schedule/test_meta_schedule_schedule_rule_mlt_tc.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -903,39 +903,39 @@ def test_conv_1x1():
def conv2d_1x1_0(inputs: T.Buffer((1, 16, 16, 64), "float16"), weight: T.Buffer((1, 1, 64, 64), "float16"), conv2d_nhwc: T.Buffer((1, 16, 16, 64), "float32")):
T.func_attr({"global_symbol": "main", "tir.noalias": T.bool(True)})
# with T.block("root"):
conv2d_nhwc_reindex_shared = T.alloc_buffer((2, 2, 8, 2, 16, 16), scope="shared")
conv2d_nhwc_reindex_shared_wmma_accumulator = T.alloc_buffer((2, 2, 8, 2, 16, 16), scope="wmma.accumulator")
conv2d_nhwc_reindex_shared = T.alloc_buffer((2, 1, 8, 4, 16, 16), scope="shared")
conv2d_nhwc_reindex_shared_wmma_accumulator = T.alloc_buffer((2, 1, 8, 4, 16, 16), scope="wmma.accumulator")
PadInput_reindex_shared = T.alloc_buffer((256, 64), "float16", scope="shared")
weight_reindex_shared = T.alloc_buffer((1, 1, 64, 64), "float16", scope="shared")
PadInput_reindex_shared_wmma_matrix_a = T.alloc_buffer((256, 64), "float16", scope="wmma.matrix_a")
weight_reindex_shared_wmma_matrix_b = T.alloc_buffer((1, 1, 64, 64), "float16", scope="wmma.matrix_b")
for ax0_0_ax1_0_ax2_0_0_ax3_0_0_fused in T.thread_binding(4, thread="blockIdx.y"):
for ax0_1_ax1_1_ax2_0_1_ax3_0_1_fused in T.thread_binding(1, thread="blockIdx.x"):
for ax0_2_ax1_2_ax2_0_2_ax3_0_2_fused in T.thread_binding(1, thread="threadIdx.y"):
for ax4_0_0 in range(1):
for ax0_ax1_ax2_0_0_ax3_0_0_fused in T.thread_binding(1, thread="blockIdx.y"):
for ax2_0_1_ax3_0_1_fused in T.thread_binding(1, thread="blockIdx.x"):
for ax2_0_2_ax3_0_2_fused in T.thread_binding(2, thread="threadIdx.y"):
for ax4_0_0 in range(2):
for ax0_ax1_fused in range(8192):
with T.block("PadInput_reindex_shared"):
v0 = T.axis.spatial(256, ax0_0_ax1_0_ax2_0_0_ax3_0_0_fused // 2 * 128 + ax0_ax1_fused // 64)
v1 = T.axis.spatial(64, ax0_ax1_fused % 64)
v0 = T.axis.spatial(256, ax0_ax1_fused // 32)
v1 = T.axis.spatial(64, ax4_0_0 * 32 + ax0_ax1_fused % 32)
T.reads(inputs[0, v0 // 16, v0 % 16, v1])
T.writes(PadInput_reindex_shared[v0, v1])
T.block_attr({"buffer_dim_align": [[0, 0, 32, 8]], "meta_schedule.cooperative_fetch": 2})
T.block_attr({"buffer_dim_align": [[0, 0, 32, 8]], "meta_schedule.cooperative_fetch": 8})
PadInput_reindex_shared[v0, v1] = inputs[0, v0 // 16, v0 % 16, v1]
for ax0_ax1_ax2_ax3_fused in range(2048):
with T.block("weight_reindex_shared"):
v0 = T.axis.spatial(1, 0)
v1 = T.axis.spatial(1, 0)
v2 = T.axis.spatial(64, ax0_ax1_ax2_ax3_fused // 32)
v3 = T.axis.spatial(64, ax0_0_ax1_0_ax2_0_0_ax3_0_0_fused % 2 * 32 + ax0_ax1_ax2_ax3_fused % 32)
v2 = T.axis.spatial(64, ax4_0_0 * 32 + ax0_ax1_ax2_ax3_fused // 64)
v3 = T.axis.spatial(64, ax0_ax1_ax2_ax3_fused % 64)
T.reads(weight[v0, v1, v2, v3])
T.writes(weight_reindex_shared[v0, v1, v2, v3])
T.block_attr({"buffer_dim_align": [[0, 2, 32, 8]], "meta_schedule.cooperative_fetch": 8})
T.block_attr({"buffer_dim_align": [[0, 2, 32, 8]], "meta_schedule.cooperative_fetch": 4})
weight_reindex_shared[v0, v1, v2, v3] = weight[v0, v1, v2, v3]
for ax4_0_1 in range(1):
for ax0_0, ax1_0 in T.grid(8, 4):
for ax0_0, ax1_0 in T.grid(8, 2):
with T.block("PadInput_reindex_shared_wmma.matrix_a_o"):
v0_o = T.axis.spatial(16, ax0_0_ax1_0_ax2_0_0_ax3_0_0_fused // 2 * 8 + ax0_0)
v1_o = T.axis.spatial(4, ax1_0)
v0_o = T.axis.spatial(16, ax2_0_2_ax3_0_2_fused * 8 + ax0_0)
v1_o = T.axis.spatial(4, ax4_0_0 * 2 + ax1_0)
T.reads(PadInput_reindex_shared[v0_o * 16:v0_o * 16 + 16, v1_o * 16:v1_o * 16 + 16])
T.writes(PadInput_reindex_shared_wmma_matrix_a[v0_o * 16:v0_o * 16 + 16, v1_o * 16:v1_o * 16 + 16])
T.block_attr({"meta_schedule.auto_tensorize": "wmma_load_16x16x16_f16_a_shared"})
Expand All @@ -945,10 +945,11 @@ def conv2d_1x1_0(inputs: T.Buffer((1, 16, 16, 64), "float16"), weight: T.Buffer(
T.reads(PadInput_reindex_shared[v0_o * 16 + v0_i, v1_o * 16 + v1_i])
T.writes(PadInput_reindex_shared_wmma_matrix_a[v0_o * 16 + v0_i, v1_o * 16 + v1_i])
PadInput_reindex_shared_wmma_matrix_a[v0_o * 16 + v0_i, v1_o * 16 + v1_i] = PadInput_reindex_shared[v0_o * 16 + v0_i, v1_o * 16 + v1_i]
for ax0, ax1, ax2_0, ax3_0 in T.grid(1, 1, 4, 2):
for ax0, ax1, ax2_0, ax3_0 in T.grid(1, 1, 2, 4):
with T.block("weight_reindex_shared_wmma.matrix_b_o"):
v0_o, v1_o, v2_o = T.axis.remap("SSS", [ax0, ax1, ax2_0])
v3_o = T.axis.spatial(4, ax0_0_ax1_0_ax2_0_0_ax3_0_0_fused % 2 * 2 + ax3_0)
v0_o, v1_o = T.axis.remap("SS", [ax0, ax1])
v2_o = T.axis.spatial(4, ax4_0_0 * 2 + ax2_0)
v3_o = T.axis.spatial(4, ax3_0)
T.reads(weight_reindex_shared[v0_o, v1_o, v2_o * 16:v2_o * 16 + 16, v3_o * 16:v3_o * 16 + 16])
T.writes(weight_reindex_shared_wmma_matrix_b[v0_o, v1_o, v2_o * 16:v2_o * 16 + 16, v3_o * 16:v3_o * 16 + 16])
T.block_attr({"meta_schedule.auto_tensorize": "wmma_load_16x16x16_f16_b_shared"})
Expand All @@ -958,38 +959,38 @@ def conv2d_1x1_0(inputs: T.Buffer((1, 16, 16, 64), "float16"), weight: T.Buffer(
T.reads(weight_reindex_shared[v0_o, v1_o, v2_o * 16 + v2_i, v3_o * 16 + v3_i])
T.writes(weight_reindex_shared_wmma_matrix_b[v0_o, v1_o, v2_o * 16 + v2_i, v3_o * 16 + v3_i])
weight_reindex_shared_wmma_matrix_b[v0_o, v1_o, v2_o * 16 + v2_i, v3_o * 16 + v3_i] = weight_reindex_shared[v0_o, v1_o, v2_o * 16 + v2_i, v3_o * 16 + v3_i]
for ax0_3, ax1_3, ax2_0_3, ax3_0_3, ax4_0_2, ax0_4, ax1_4, ax2_0_4, ax3_0_4 in T.grid(1, 1, 8, 2, 4, 1, 1, 1, 1):
for ax2_0_3, ax3_0_3, ax4_0_2, ax2_0_4, ax3_0_4 in T.grid(8, 1, 2, 1, 4):
with T.block("conv2d_nhwc_o"):
v0_o = T.axis.spatial(1, ax0_3 + ax0_4)
v1_o = T.axis.spatial(1, ax1_3 + ax1_4)
v2_o = T.axis.spatial(16, ax0_0_ax1_0_ax2_0_0_ax3_0_0_fused // 2 * 8 + ax2_0_3 + ax2_0_4)
v3_o = T.axis.spatial(4, ax0_0_ax1_0_ax2_0_0_ax3_0_0_fused % 2 * 2 + ax3_0_3 + ax3_0_4)
v4_o = T.axis.reduce(4, ax4_0_0 * 4 + ax4_0_1 * 4 + ax4_0_2)
v0_o = T.axis.spatial(1, 0)
v1_o = T.axis.spatial(1, 0)
v2_o = T.axis.spatial(16, ax2_0_2_ax3_0_2_fused * 8 + ax2_0_3 + ax2_0_4)
v3_o = T.axis.spatial(4, ax3_0_3 * 4 + ax3_0_4)
v4_o = T.axis.reduce(4, ax4_0_0 * 2 + ax4_0_1 * 2 + ax4_0_2)
T.reads(PadInput_reindex_shared_wmma_matrix_a[v2_o * 16:v2_o * 16 + 16, v4_o * 16:v4_o * 16 + 16], weight_reindex_shared_wmma_matrix_b[v0_o, v1_o, v4_o * 16:v4_o * 16 + 16, v3_o * 16:v3_o * 16 + 16])
T.writes(conv2d_nhwc_reindex_shared_wmma_accumulator[v2_o // 8, v3_o // 2, v2_o % 8, v3_o % 2, 0:16, 0:16])
T.writes(conv2d_nhwc_reindex_shared_wmma_accumulator[v2_o // 8, 0, v2_o % 8, v3_o, 0:16, 0:16])
T.block_attr({"meta_schedule.auto_tensorize": "wmma_sync_16x16x16_f16f16f32", "meta_schedule.auto_tensorize_init": "wmma_fill_16x16x16_f32", "warp_execution": 1})
with T.init():
for ax2_1, ax3_1 in T.grid(16, 16):
with T.block("conv2d_nhwc_init"):
v2_i_init, v3_i_init = T.axis.remap("SS", [ax2_1, ax3_1])
T.reads()
T.writes(conv2d_nhwc_reindex_shared_wmma_accumulator[v2_o // 8, v3_o // 2, v2_o % 8, v3_o % 2, v2_i_init, v3_i_init])
conv2d_nhwc_reindex_shared_wmma_accumulator[v2_o // 8, v3_o // 2, v2_o % 8, v3_o % 2, v2_i_init, v3_i_init] = T.float32(0)
T.writes(conv2d_nhwc_reindex_shared_wmma_accumulator[v2_o // 8, 0, v2_o % 8, v3_o, v2_i_init, v3_i_init])
conv2d_nhwc_reindex_shared_wmma_accumulator[v2_o // 8, 0, v2_o % 8, v3_o, v2_i_init, v3_i_init] = T.float32(0)
for ax2_1, ax3_1, ax4_1 in T.grid(16, 16, 16):
with T.block("conv2d_nhwc"):
v2_i, v3_i, v4_i = T.axis.remap("SSR", [ax2_1, ax3_1, ax4_1])
T.reads(conv2d_nhwc_reindex_shared_wmma_accumulator[v2_o // 8, v3_o // 2, v2_o % 8, v3_o % 2, v2_i, v3_i], PadInput_reindex_shared_wmma_matrix_a[v2_o * 16 + v2_i, v4_o * 16 + v4_i], weight_reindex_shared_wmma_matrix_b[v0_o, v1_o, v4_o * 16 + v4_i, v3_o * 16 + v3_i])
T.writes(conv2d_nhwc_reindex_shared_wmma_accumulator[v2_o // 8, v3_o // 2, v2_o % 8, v3_o % 2, v2_i, v3_i])
T.reads(conv2d_nhwc_reindex_shared_wmma_accumulator[v2_o // 8, 0, v2_o % 8, v3_o, v2_i, v3_i], PadInput_reindex_shared_wmma_matrix_a[v2_o * 16 + v2_i, v4_o * 16 + v4_i], weight_reindex_shared_wmma_matrix_b[v0_o, v1_o, v4_o * 16 + v4_i, v3_o * 16 + v3_i])
T.writes(conv2d_nhwc_reindex_shared_wmma_accumulator[v2_o // 8, 0, v2_o % 8, v3_o, v2_i, v3_i])
T.block_attr({"meta_schedule.tiling_structure": "SSSRRSRS"})
conv2d_nhwc_reindex_shared_wmma_accumulator[v2_o // 8, v3_o // 2, v2_o % 8, v3_o % 2, v2_i, v3_i] = conv2d_nhwc_reindex_shared_wmma_accumulator[v2_o // 8, v3_o // 2, v2_o % 8, v3_o % 2, v2_i, v3_i] + T.Cast("float32", PadInput_reindex_shared_wmma_matrix_a[v2_o * 16 + v2_i, v4_o * 16 + v4_i]) * T.Cast("float32", weight_reindex_shared_wmma_matrix_b[v0_o, v1_o, v4_o * 16 + v4_i, v3_o * 16 + v3_i])
conv2d_nhwc_reindex_shared_wmma_accumulator[v2_o // 8, 0, v2_o % 8, v3_o, v2_i, v3_i] = conv2d_nhwc_reindex_shared_wmma_accumulator[v2_o // 8, 0, v2_o % 8, v3_o, v2_i, v3_i] + T.Cast("float32", PadInput_reindex_shared_wmma_matrix_a[v2_o * 16 + v2_i, v4_o * 16 + v4_i]) * T.Cast("float32", weight_reindex_shared_wmma_matrix_b[v0_o, v1_o, v4_o * 16 + v4_i, v3_o * 16 + v3_i])
for ax2 in range(8):
for ax0_ax1_fused in T.thread_binding(1, thread="threadIdx.y"):
for ax2_1, ax3 in T.grid(1, 2):
for ax0_ax1_fused in T.thread_binding(2, thread="threadIdx.y"):
for ax2_1, ax3 in T.grid(1, 4):
with T.block("conv2d_nhwc_reindex_shared_wmma.accumulator_o"):
v0_o = T.axis.spatial(2, ax0_0_ax1_0_ax2_0_0_ax3_0_0_fused // 2)
v1_o = T.axis.spatial(2, ax0_0_ax1_0_ax2_0_0_ax3_0_0_fused % 2)
v0_o = T.axis.spatial(2, ax0_ax1_fused)
v1_o = T.axis.spatial(1, 0)
v2_o = T.axis.spatial(8, ax2 + ax2_1)
v3_o = T.axis.spatial(2, ax3)
v3_o = T.axis.spatial(4, ax3)
v4_o = T.axis.spatial(1, 0)
v5_o = T.axis.spatial(1, 0)
T.reads(conv2d_nhwc_reindex_shared_wmma_accumulator[v0_o, v1_o, v2_o, v3_o, 0:16, 0:16])
Expand All @@ -1001,29 +1002,27 @@ def conv2d_1x1_0(inputs: T.Buffer((1, 16, 16, 64), "float16"), weight: T.Buffer(
T.reads(conv2d_nhwc_reindex_shared_wmma_accumulator[v0_o, v1_o, v2_o, v3_o, v4_i, v5_i])
T.writes(conv2d_nhwc_reindex_shared[v0_o, v1_o, v2_o, v3_o, v4_i, v5_i])
conv2d_nhwc_reindex_shared[v0_o, v1_o, v2_o, v3_o, v4_i, v5_i] = conv2d_nhwc_reindex_shared_wmma_accumulator[v0_o, v1_o, v2_o, v3_o, v4_i, v5_i]
for ax0_ax1_ax3_ax4_ax5_fused in range(512):
for ax0_ax1_ax3_ax4_ax5_fused in range(2048):
with T.block("conv2d_nhwc_reindex_shared"):
v0 = T.axis.spatial(2, ax0_0_ax1_0_ax2_0_0_ax3_0_0_fused // 2)
v1 = T.axis.spatial(2, ax0_0_ax1_0_ax2_0_0_ax3_0_0_fused % 2)
v0 = T.axis.spatial(2, ax0_ax1_ax3_ax4_ax5_fused // 1024)
v1 = T.axis.spatial(1, 0)
v2 = T.axis.spatial(8, ax2)
v3 = T.axis.spatial(2, ax0_ax1_ax3_ax4_ax5_fused // 256)
v3 = T.axis.spatial(4, ax0_ax1_ax3_ax4_ax5_fused % 1024 // 256)
v4 = T.axis.spatial(16, ax0_ax1_ax3_ax4_ax5_fused % 256 // 16)
v5 = T.axis.spatial(16, ax0_ax1_ax3_ax4_ax5_fused % 16)
T.reads(conv2d_nhwc_reindex_shared[v0, v1, v2, v3, v4, v5])
T.writes(conv2d_nhwc[0, (v4 + v2 * 16 + v0 * 128) // 16, (v4 + v2 * 16 + v0 * 128) % 16, v5 + v3 * 16 + v1 * 32])
T.writes(conv2d_nhwc[0, (v4 + v2 * 16 + v0 * 128) // 16, (v4 + v2 * 16 + v0 * 128) % 16, v5 + v3 * 16])
T.block_attr({"meta_schedule.cooperative_fetch": 1})
conv2d_nhwc[0, (v4 + v2 * 16 + v0 * 128) // 16, (v4 + v2 * 16 + v0 * 128) % 16, v5 + v3 * 16 + v1 * 32] = conv2d_nhwc_reindex_shared[v0, v1, v2, v3, v4, v5]
conv2d_nhwc[0, (v4 + v2 * 16 + v0 * 128) // 16, (v4 + v2 * 16 + v0 * 128) % 16, v5 + v3 * 16] = conv2d_nhwc_reindex_shared[v0, v1, v2, v3, v4, v5]
# fmt: on

decision_0 = [
("SamplePerfectTile", [1, 1, 1, 1, 1]),
("SamplePerfectTile", [1, 1, 1, 1, 1]),
("SamplePerfectTile", [2, 1, 1, 8, 1]),
("SamplePerfectTile", [2, 1, 1, 2, 1]),
("SamplePerfectTile", [1, 1, 4]),
("SamplePerfectTile", [1, 1, 2, 8, 1]),
("SamplePerfectTile", [1, 1, 1, 1, 4]),
("SamplePerfectTile", [2, 1, 2]),
("SampleCategorical", 0),
("SampleCategorical", 1),
("SampleCategorical", 3),
("SampleCategorical", 2),
]

mod = te.create_prim_func(
Expand Down

0 comments on commit 4a5e22e

Please sign in to comment.