Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fixing copypasta in codegen/flow tests that included ABI ops. #16140

Merged
merged 1 commit into from
Jan 17, 2024
Merged

Fixing copypasta in codegen/flow tests that included ABI ops. #16140

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
203 changes: 64 additions & 139 deletions compiler/src/iree/compiler/Codegen/Common/CPU/test/llvmcpu_materialize_encoding.mlir
View file
Open in desktop

Large diffs are not rendered by default.

27 changes: 0 additions & 27 deletions compiler/src/iree/compiler/Codegen/LLVMGPU/test/set_transform_strategy_batch_matmul.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,33 +51,6 @@ module attributes {hal.device.targets = [#device_target_cuda]} {
}
}
}
func.func @batch_matmul(%arg0: !hal.buffer_view, %arg1: !hal.buffer_view, %arg2: !hal.buffer_view) -> !hal.buffer_view attributes {iree.abi.stub} {
%c1310720 = arith.constant 1310720 : index
%c5242880 = arith.constant 5242880 : index
%c13107200 = arith.constant 13107200 : index
%c0 = arith.constant 0 : index
%c320 = arith.constant 320 : index
%c553648160_i32 = arith.constant 553648160 : i32
%c1_i32 = arith.constant 1 : i32
%c128 = arith.constant 128 : index
%c80 = arith.constant 80 : index
%c32 = arith.constant 32 : index
hal.buffer_view.assert<%arg0 : !hal.buffer_view> message("input 0") shape([%c128, %c80, %c32]) type(%c553648160_i32) encoding(%c1_i32)
%0 = stream.tensor.import %arg0 : !hal.buffer_view -> tensor<128x80x32xf32> in !stream.resource<external>{%c1310720}
hal.buffer_view.assert<%arg1 : !hal.buffer_view> message("input 1") shape([%c128, %c32, %c320]) type(%c553648160_i32) encoding(%c1_i32)
%1 = stream.tensor.import %arg1 : !hal.buffer_view -> tensor<128x32x320xf32> in !stream.resource<external>{%c5242880}
%2 = stream.resource.alloc uninitialized : !stream.resource<external>{%c13107200}
%3 = stream.cmd.execute with(%0 as %arg3: !stream.resource<external>{%c1310720}, %1 as %arg4: !stream.resource<external>{%c5242880}, %2 as %arg5: !stream.resource<external>{%c13107200}) {
stream.cmd.dispatch @batch_matmul_dispatch_0::@cuda_nvptx_fb::@batch_matmul_dispatch_0_generic_128x80x320x32_f32 {
ro %arg3[%c0 for %c1310720] : !stream.resource<external>{%c1310720},
ro %arg4[%c0 for %c5242880] : !stream.resource<external>{%c5242880},
wo %arg5[%c0 for %c13107200] : !stream.resource<external>{%c13107200}
} attributes {hal.interface.bindings = [#hal.interface.binding<0, 0>, #hal.interface.binding<0, 1>, #hal.interface.binding<0, 2>]}
} => !stream.timepoint
%4 = stream.timepoint.await %3 => %2 : !stream.resource<external>{%c13107200}
%5 = stream.tensor.export %4 : tensor<128x80x320xf32> in !stream.resource<external>{%c13107200} -> !hal.buffer_view
return %5 : !hal.buffer_view
}
}


Expand Down
20 changes: 8 additions & 12 deletions ...src/iree/compiler/Dialect/Flow/Transforms/test/clone_producers_into_dispatch_regions.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -96,17 +96,14 @@ func.func @complex_create(%real : f32, %imag : f32, %input: tensor<4x2xcomplex<f

// -----

#map = affine_map<() -> ()>
func.func @use_in_dispatch_count(%arg0: !hal.buffer_view, %arg1: !hal.buffer_view) -> !hal.buffer_view {
func.func @use_in_dispatch_count(%arg0: tensor<1xi32>, %arg1: tensor<1xi32>) -> tensor<i32> {
%c1 = arith.constant 1 : index
%c2_i32 = arith.constant 2 : i32
%c0 = arith.constant 0 : index
%0 = hal.tensor.import %arg0 "input 0" : !hal.buffer_view -> tensor<1xi32>
%1 = hal.tensor.import %arg1 "input 1" : !hal.buffer_view -> tensor<1xi32>
%2 = tensor.empty() : tensor<i32>
%extracted = tensor.extract %0[%c1] : tensor<1xi32>
%extracted = tensor.extract %arg0[%c1] : tensor<1xi32>
%4 = flow.dispatch.region -> (tensor<i32>) {
%6 = linalg.generic {indexing_maps = [#map], iterator_types = []} outs(%2 : tensor<i32>) {
%6 = linalg.generic {indexing_maps = [affine_map<() -> ()>], iterator_types = []} outs(%2 : tensor<i32>) {
^bb0(%out: i32):
%7 = arith.addi %extracted, %c2_i32 : i32
linalg.yield %7 : i32
Expand All @@ -115,8 +112,7 @@ func.func @use_in_dispatch_count(%arg0: !hal.buffer_view, %arg1: !hal.buffer_vie
} count() -> (index, index, index) {
flow.return %c1, %c1, %c1 : index, index, index
}
%5 = hal.tensor.export %4 "output 0" : tensor<i32> -> !hal.buffer_view
return %5 : !hal.buffer_view
return %4 : tensor<i32>
}


Expand Down Expand Up @@ -194,16 +190,16 @@ module {
func.func @clone_dequantization_like(%arg0: tensor<32x1x16x1x8xi16>, %arg1: tensor<32x344x16x32x8xi4>) -> tensor<32x1x344x1x32xi32> {
%c0_i32 = arith.constant 0 : i32
%0 = tensor.empty() : tensor<32x1x16x1x8xi32>
%1 = linalg.generic {indexing_maps = [#map, #map],
iterator_types = ["parallel", "parallel", "parallel", "parallel", "parallel"]}
%1 = linalg.generic {indexing_maps = [#map, #map],
iterator_types = ["parallel", "parallel", "parallel", "parallel", "parallel"]}
ins(%arg0 : tensor<32x1x16x1x8xi16>) outs(%0 : tensor<32x1x16x1x8xi32>) {
^bb0(%in: i16, %out: i32):
%7 = arith.extsi %in : i16 to i32
linalg.yield %7 : i32
} -> tensor<32x1x16x1x8xi32>
%2 = tensor.empty() : tensor<32x344x16x32x8xi32>
%3 = linalg.generic {indexing_maps = [#map, #map],
iterator_types = ["parallel", "parallel", "parallel", "parallel", "parallel"]}
%3 = linalg.generic {indexing_maps = [#map, #map],
iterator_types = ["parallel", "parallel", "parallel", "parallel", "parallel"]}
ins(%arg1 : tensor<32x344x16x32x8xi4>) outs(%2 : tensor<32x344x16x32x8xi32>) {
^bb0(%in: i4, %out: i32):
%7 = arith.extui %in : i4 to i32
Expand Down
12 changes: 5 additions & 7 deletions ...er/src/iree/compiler/Dialect/Flow/Transforms/test/collapse_linalg_generic_on_tensors.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -459,24 +459,22 @@ func.func @collapse12() -> (!type,!type,!type,!type) {

// -----

func.func @multi_reduce_dim(%arg0: !hal.buffer_view) -> !hal.buffer_view attributes {iree.abi.stub} {
func.func @multi_reduce_dim(%arg0: tensor<2x32x10x4096xf32>) -> tensor<2x32x1x1xf32> {
%cst = arith.constant -0.000000e+00 : f32
%0 = hal.tensor.import %arg0 : !hal.buffer_view -> tensor<2x32x10x4096xf32>
%1 = tensor.empty() : tensor<2x32xf32>
%2 = linalg.fill ins(%cst : f32) outs(%1 : tensor<2x32xf32>) -> tensor<2x32xf32>
%3 = linalg.generic {indexing_maps = [affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>, affine_map<(d0, d1, d2, d3) -> (d0, d1)>], iterator_types = ["parallel", "parallel", "reduction", "reduction"]} ins(%0 : tensor<2x32x10x4096xf32>) outs(%2 : tensor<2x32xf32>) {
%3 = linalg.generic {indexing_maps = [affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>, affine_map<(d0, d1, d2, d3) -> (d0, d1)>], iterator_types = ["parallel", "parallel", "reduction", "reduction"]} ins(%arg0 : tensor<2x32x10x4096xf32>) outs(%2 : tensor<2x32xf32>) {
^bb0(%arg1: f32, %arg2: f32):
%6 = arith.addf %arg1, %arg2 : f32
linalg.yield %6 : f32
} -> tensor<2x32xf32>
%4 = tensor.expand_shape %3 [[0], [1, 2, 3]] : tensor<2x32xf32> into tensor<2x32x1x1xf32>
%5 = hal.tensor.export %4 : tensor<2x32x1x1xf32> -> !hal.buffer_view
return %5 : !hal.buffer_view
return %4 : tensor<2x32x1x1xf32>
}

// Check that we collapse dimensions.
// CHECK-LABEL: @multi_reduce_dim(
// CHECK-DAG: %[[ARG0:.+]] = hal.tensor.import
// CHECK-LABEL: @multi_reduce_dim
// CHECK-SAME: (%[[ARG0:.+]]: tensor<2x32x10x4096xf32>)
// CHECK: %[[COLLAPSE:.+]] = tensor.collapse_shape %[[ARG0]] {{\[}}[0, 1], [2, 3]{{\]}}
// CHECK: %[[DISPATCH:.+]] = flow.dispatch.region
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<64xf32>
Expand Down
16 changes: 6 additions & 10 deletions compiler/src/iree/compiler/Dialect/Flow/Transforms/test/collapse_reduction.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,18 +1,16 @@
// RUN: iree-opt --split-input-file -iree-flow-collapse-dims %s | FileCheck %s

func.func @multi_reduce_dim(%arg0: !hal.buffer_view) -> !hal.buffer_view attributes {iree.abi.stub} {
func.func @multi_reduce_dim(%arg0: tensor<2x32x10x4096xf32>) -> tensor<2x32x1x1xf32> {
%cst = arith.constant -0.000000e+00 : f32
%0 = hal.tensor.import %arg0 : !hal.buffer_view -> tensor<2x32x10x4096xf32>
%1 = tensor.empty() : tensor<2x32xf32>
%2 = linalg.fill ins(%cst : f32) outs(%1 : tensor<2x32xf32>) -> tensor<2x32xf32>
%3 = linalg.generic {indexing_maps = [affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>, affine_map<(d0, d1, d2, d3) -> (d0, d1)>], iterator_types = ["parallel", "parallel", "reduction", "reduction"]} ins(%0 : tensor<2x32x10x4096xf32>) outs(%2 : tensor<2x32xf32>) {
%3 = linalg.generic {indexing_maps = [affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>, affine_map<(d0, d1, d2, d3) -> (d0, d1)>], iterator_types = ["parallel", "parallel", "reduction", "reduction"]} ins(%arg0 : tensor<2x32x10x4096xf32>) outs(%2 : tensor<2x32xf32>) {
^bb0(%arg1: f32, %arg2: f32):
%6 = arith.addf %arg1, %arg2 : f32
linalg.yield %6 : f32
} -> tensor<2x32xf32>
%4 = tensor.expand_shape %3 [[0], [1, 2, 3]] : tensor<2x32xf32> into tensor<2x32x1x1xf32>
%5 = hal.tensor.export %4 : tensor<2x32x1x1xf32> -> !hal.buffer_view
return %5 : !hal.buffer_view
return %4 : tensor<2x32x1x1xf32>
}

// Check that we collapse dimensions.
Expand Down Expand Up @@ -42,22 +40,20 @@ func.func @input_broadcast(%arg0: tensor<4x8xf32>, %arg1: tensor<4xf32>) -> tens

// Collapsing should not happen to ops in flow.dispatch.region or flow.dispatch.workgroups

func.func @multi_reduce_dim_dispatch(%arg0: !hal.buffer_view) -> !hal.buffer_view attributes {iree.abi.stub} {
func.func @multi_reduce_dim_dispatch(%arg0: tensor<2x32x10x4096xf32>) -> tensor<2x32x1x1xf32> {
%cst = arith.constant -0.000000e+00 : f32
%0 = hal.tensor.import %arg0 : !hal.buffer_view -> tensor<2x32x10x4096xf32>
%1 = tensor.empty() : tensor<2x32xf32>
%2 = linalg.fill ins(%cst : f32) outs(%1 : tensor<2x32xf32>) -> tensor<2x32xf32>
%3 = flow.dispatch.region -> (tensor<2x32xf32>) {
%6 = linalg.generic {indexing_maps = [affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>, affine_map<(d0, d1, d2, d3) -> (d0, d1)>], iterator_types = ["parallel", "parallel", "reduction", "reduction"]} ins(%0 : tensor<2x32x10x4096xf32>) outs(%2 : tensor<2x32xf32>) {
%6 = linalg.generic {indexing_maps = [affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>, affine_map<(d0, d1, d2, d3) -> (d0, d1)>], iterator_types = ["parallel", "parallel", "reduction", "reduction"]} ins(%arg0 : tensor<2x32x10x4096xf32>) outs(%2 : tensor<2x32xf32>) {
^bb0(%arg1: f32, %arg2: f32):
%7 = arith.addf %arg1, %arg2 : f32
linalg.yield %7 : f32
} -> tensor<2x32xf32>
flow.return %6 : tensor<2x32xf32>
}
%4 = tensor.expand_shape %3 [[0], [1, 2, 3]] : tensor<2x32xf32> into tensor<2x32x1x1xf32>
%5 = hal.tensor.export %4 : tensor<2x32x1x1xf32> -> !hal.buffer_view
return %5 : !hal.buffer_view
return %4 : tensor<2x32x1x1xf32>
}

// CHECK: @multi_reduce_dim_dispatch
Expand Down
7 changes: 2 additions & 5 deletions compiler/src/iree/compiler/Dialect/Flow/Transforms/test/dispatch_linalg_on_tensors.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -790,7 +790,7 @@ func.func @dynamic_slice(%arg0: tensor<?x?xi32>, %arg1: tensor<i32>, %arg2: tens

// -----

func.func @dynamic_dot() -> !hal.buffer_view attributes {iree.abi.stub} {
func.func @dynamic_dot() -> tensor<?x?xf32> {
%c0 = arith.constant 0 : index
%c1 = arith.constant 1 : index
%cst = arith.constant 0.000000e+00 : f32
Expand All @@ -801,10 +801,7 @@ func.func @dynamic_dot() -> !hal.buffer_view attributes {iree.abi.stub} {
%4 = tensor.empty(%2, %3) : tensor<?x?xf32>
%5 = linalg.fill ins(%cst : f32) outs(%4 : tensor<?x?xf32>) -> tensor<?x?xf32>
%6 = linalg.matmul ins(%0, %1 : tensor<?x?xf32>, tensor<?x?xf32>) outs(%5 : tensor<?x?xf32>) -> tensor<?x?xf32>
%7 = tensor.dim %6, %c0 : tensor<?x?xf32>
%8 = tensor.dim %6, %c1 : tensor<?x?xf32>
%9 = hal.tensor.export %6 : tensor<?x?xf32>{%7, %8} -> !hal.buffer_view
return %9 : !hal.buffer_view
return %6 : tensor<?x?xf32>
}
// CHECK-LABEL: func.func @dynamic_dot()
// CHECK-NOT: linalg.fill
Expand Down
Loading