-
Notifications
You must be signed in to change notification settings - Fork 6
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
Showing
1 changed file
with
69 additions
and
7 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -1,13 +1,75 @@ | ||
| // RUN: enzymexlamlir-opt %s --enzyme-wrap="infn=main outfn= retTys=enzyme_dup argTys=enzyme_dup,enzyme_dup mode=ForwardMode" | FileCheck %s --check-prefix=FORWARD | ||
| // RUN: enzymexlamlir-opt %s --enzyme-wrap="infn=main outfn= retTys=enzyme_active argTys=enzyme_active,enzyme_active mode=ReverseModeCombined" --canonicalize --remove-unnecessary-enzyme-ops | FileCheck %s --check-prefix=REVERSE | ||
| // RUN: enzymexlamlir-opt %s --enzyme-wrap="infn=dot_general_nobatch outfn= retTys=enzyme_dup argTys=enzyme_dup,enzyme_dup mode=ForwardMode" | FileCheck %s --check-prefix=FORWARD | ||
| // RUN: enzymexlamlir-opt %s --enzyme-wrap="infn=dot_general_nobatch outfn= retTys=enzyme_active argTys=enzyme_active,enzyme_active mode=ReverseModeCombined" --canonicalize --remove-unnecessary-enzyme-ops | FileCheck %s --check-prefix=REVERSE | ||
|
|
||
| func.func @main(%a : tensor<2x3xf32>, %b : tensor<4x2xf32>) -> tensor<3x4xf32> { | ||
| func.func @dot_general_nobatch(%a : tensor<2x3xf32>, %b : tensor<4x2xf32>) -> tensor<3x4xf32> { | ||
| %c = "stablehlo.dot_general"(%a, %b) { | ||
| lhs_batching_dimensions = [1], | ||
| rhs_batching_dimensions = [0], | ||
| lhs_contracting_dimensions = [0], | ||
| rhs_contracting_dimensions = [1], | ||
| dot_dimension_numbers = #stablehlo.dot< | ||
| lhs_batching_dimensions = [], | ||
| rhs_batching_dimensions = [], | ||
| lhs_contracting_dimensions = [0], | ||
| rhs_contracting_dimensions = [1], | ||
| >, | ||
| precision_config = [#stablehlo<precision DEFAULT>, #stablehlo<precision DEFAULT>] | ||
| } : (tensor<2x3xf32>, tensor<4x2xf32>) -> tensor<3x4xf32> | ||
| func.return %c : tensor<3x4xf32> | ||
| } | ||
|
|
||
| // FORWARD: func.func @dot_general_nobatch(%arg0: tensor<2x3xf32>, %arg1: tensor<2x3xf32>, %arg2: tensor<4x2xf32>, %arg3: tensor<4x2xf32>) -> (tensor<3x4xf32>, tensor<3x4xf32>) { | ||
| // FORWARD-NEXT: %0 = stablehlo.dot_general %arg1, %arg2, contracting_dims = [0] x [1], precision = [DEFAULT, DEFAULT] : (tensor<2x3xf32>, tensor<4x2xf32>) -> tensor<3x4xf32> | ||
| // FORWARD-NEXT: %1 = stablehlo.dot_general %arg0, %arg3, contracting_dims = [0] x [1], precision = [DEFAULT, DEFAULT] : (tensor<2x3xf32>, tensor<4x2xf32>) -> tensor<3x4xf32> | ||
| // FORWARD-NEXT: %2 = stablehlo.add %0, %1 : tensor<3x4xf32> | ||
| // FORWARD-NEXT: %3 = stablehlo.dot_general %arg0, %arg2, contracting_dims = [0] x [1], precision = [DEFAULT, DEFAULT] : (tensor<2x3xf32>, tensor<4x2xf32>) -> tensor<3x4xf32> | ||
| // FORWARD-NEXT: return %3, %2 : tensor<3x4xf32>, tensor<3x4xf32> | ||
| // FORWARD-NEXT: } | ||
|
|
||
| // REVERSE: func.func @dot_general_nobatch(%arg0: tensor<2x3xf32>, %arg1: tensor<4x2xf32>, %arg2: tensor<3x4xf32>) -> (tensor<2x3xf32>, tensor<4x2xf32>) { | ||
| // REVERSE-NEXT: %cst = arith.constant dense<0.000000e+00> : tensor<3x4xf32> | ||
| // REVERSE-NEXT: %cst_0 = arith.constant dense<0.000000e+00> : tensor<2x3xf32> | ||
| // REVERSE-NEXT: %cst_1 = arith.constant dense<0.000000e+00> : tensor<4x2xf32> | ||
| // REVERSE-NEXT: %0 = arith.addf %arg2, %cst : tensor<3x4xf32> | ||
| // REVERSE-NEXT: %1 = stablehlo.dot_general %0, %arg1, contracting_dims = [1] x [0], precision = [DEFAULT, DEFAULT] : (tensor<3x4xf32>, tensor<4x2xf32>) -> tensor<3x2xf32> | ||
| // REVERSE-NEXT: %2 = stablehlo.transpose %1, dims = [1, 0] : (tensor<3x2xf32>) -> tensor<2x3xf32> | ||
| // REVERSE-NEXT: %3 = arith.addf %2, %cst_0 : tensor<2x3xf32> | ||
| // REVERSE-NEXT: %4 = stablehlo.dot_general %arg0, %0, contracting_dims = [1] x [0], precision = [DEFAULT, DEFAULT] : (tensor<2x3xf32>, tensor<3x4xf32>) -> tensor<2x4xf32> | ||
| // REVERSE-NEXT: %5 = stablehlo.transpose %4, dims = [1, 0] : (tensor<2x4xf32>) -> tensor<4x2xf32> | ||
| // REVERSE-NEXT: %6 = arith.addf %5, %cst_1 : tensor<4x2xf32> | ||
| // REVERSE-NEXT: return %3, %6 : tensor<2x3xf32>, tensor<4x2xf32> | ||
| // REVERSE-NEXT: } | ||
|
|
||
| // RUN: enzymexlamlir-opt %s --enzyme-wrap="infn=dot_general_batch outfn= retTys=enzyme_dup argTys=enzyme_dup,enzyme_dup mode=ForwardMode" | FileCheck %s --check-prefix=FORWARD-BATCH | ||
| // RUN: enzymexlamlir-opt %s --enzyme-wrap="infn=dot_general_batch outfn= retTys=enzyme_active argTys=enzyme_active,enzyme_active mode=ReverseModeCombined" --canonicalize --remove-unnecessary-enzyme-ops | FileCheck %s --check-prefix=REVERSE-BATCH | ||
|
|
||
| func.func @dot_general_batch(%a : tensor<2x3x8xf32>, %b : tensor<4x2x8xf32>) -> tensor<8x3x4xf32> { | ||
| %c = "stablehlo.dot_general"(%a, %b) { | ||
| dot_dimension_numbers = #stablehlo.dot< | ||
| lhs_batching_dimensions = [2], | ||
| rhs_batching_dimensions = [2], | ||
| lhs_contracting_dimensions = [0], | ||
| rhs_contracting_dimensions = [1], | ||
| >, | ||
| precision_config = [#stablehlo<precision DEFAULT>, #stablehlo<precision DEFAULT>] | ||
| } : (tensor<2x3x8xf32>, tensor<4x2x8xf32>) -> tensor<8x3x4xf32> | ||
| func.return %c : tensor<8x3x4xf32> | ||
| } | ||
|
|
||
| // FORWARD-BATCH: func.func @dot_general_batch(%arg0: tensor<2x3x8xf32>, %arg1: tensor<2x3x8xf32>, %arg2: tensor<4x2x8xf32>, %arg3: tensor<4x2x8xf32>) -> (tensor<8x3x4xf32>, tensor<8x3x4xf32>) { | ||
| // FORWARD-BATCH-NEXT: %0 = stablehlo.dot_general %arg1, %arg2, contracting_dims = [0] x [1], precision = [DEFAULT, DEFAULT] : (tensor<2x3x8xf32>, tensor<4x2x8xf32>) -> tensor<8x3x4xf32> | ||
| // FORWARD-BATCH-NEXT: %1 = stablehlo.dot_general %arg0, %arg3, contracting_dims = [0] x [1], precision = [DEFAULT, DEFAULT] : (tensor<2x3x8xf32>, tensor<4x2x8xf32>) -> tensor<8x3x4xf32> | ||
| // FORWARD-BATCH-NEXT: %2 = stablehlo.add %0, %1 : tensor<8x3x4xf32> | ||
| // FORWARD-BATCH-NEXT: %3 = stablehlo.dot_general %arg0, %arg2, contracting_dims = [0] x [1], precision = [DEFAULT, DEFAULT] : (tensor<2x3x8xf32>, tensor<4x2x8xf32>) -> tensor<8x3x4xf32> | ||
| // FORWARD-BATCH-NEXT: return %3, %2 : tensor<8x3x4xf32>, tensor<8x3x4xf32> | ||
| // FORWARD-BATCH-NEXT: } | ||
|
|
||
| // REVERSE-BATCH: func.func @dot_general_batch(%arg0: tensor<2x3x8xf32>, %arg1: tensor<4x2x8xf32>, %arg2: tensor<8x3x4xf32>) -> (tensor<8x2x3xf32>, tensor<8x4x2xf32>) { | ||
| // REVERSE-BATCH-NEXT: %cst = arith.constant dense<0.000000e+00> : tensor<8x3x4xf32> | ||
| // REVERSE-BATCH-NEXT: %cst_0 = arith.constant dense<0.000000e+00> : tensor<2x3x8xf32> | ||
| // REVERSE-BATCH-NEXT: %cst_1 = arith.constant dense<0.000000e+00> : tensor<4x2x8xf32> | ||
| // REVERSE-BATCH-NEXT: %0 = arith.addf %arg2, %cst : tensor<8x3x4xf32> | ||
| // REVERSE-BATCH-NEXT: %1 = stablehlo.dot_general %0, %arg1, batching_dims = [0] x [2], contracting_dims = [1] x [0], precision = [DEFAULT, DEFAULT] : (tensor<8x3x4xf32>, tensor<4x2x8xf32>) -> tensor<8x3x2xf32> | ||
| // REVERSE-BATCH-NEXT: %2 = stablehlo.transpose %1, dims = [1, 0] : (tensor<8x3x2xf32>) -> tensor<2x3x8xf32> | ||
| // REVERSE-BATCH-NEXT: %3 = arith.addf %2, %cst_0 : tensor<2x3x8xf32> | ||
| // REVERSE-BATCH-NEXT: %4 = stablehlo.dot_general %arg0, %0, batching_dims = [2] x [0], contracting_dims = [1] x [0], precision = [DEFAULT, DEFAULT] : (tensor<2x3x8xf32>, tensor<8x3x4xf32>) -> tensor<8x2x4xf32> | ||
| // REVERSE-BATCH-NEXT: %5 = stablehlo.transpose %4, dims = [1, 0] : (tensor<8x2x4xf32>) -> tensor<4x2x8xf32> | ||
| // REVERSE-BATCH-NEXT: %6 = arith.addf %5, %cst_1 : tensor<4x2x8xf32> | ||
| // REVERSE-BATCH-NEXT: return %3, %6 : tensor<2x3x8xf32>, tensor<4x2x8xf32> | ||
| // REVERSE-BATCH-NEXT: } |