-
Notifications
You must be signed in to change notification settings - Fork 16
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
#2022: Added data parallel and model parallel tests for all multichip…
… configurations
- Loading branch information
Showing
8 changed files
with
610 additions
and
0 deletions.
There are no files selected for viewing
334 changes: 334 additions & 0 deletions
334
test/ttmlir/Conversion/StableHLOToTTIR/model_parallel/model_parallel_gspmd.mlir
Large diffs are not rendered by default.
Oops, something went wrong.
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
54 changes: 54 additions & 0 deletions
54
test/ttmlir/Silicon/TTNN/llmbox/model_parallel/model_parallel_1x8.mlir
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 |
|---|---|---|
| @@ -0,0 +1,54 @@ | ||
| // RUN: ttmlir-opt --ttir-to-ttnn-backend-pipeline="system-desc-path=%system_desc_path% mesh-shape=1,8" %s > %t.mlir | ||
| // RUN: FileCheck %s --input-file=%t.mlir | ||
| // RUN: ttmlir-translate --ttnn-to-flatbuffer %t.mlir > %t.ttnn | ||
|
|
||
| func.func public @jit_data_parallel_t3000(%arg0: tensor<64x1x1024x2048xf32>, %arg1: tensor<1x1x2048x512xf32>) -> (tensor<64x1x1024x512xf32> {jax.result_info = ""}) { | ||
| %0 = tensor.empty() : tensor<8x1x1024x2048xf32> | ||
| %1 = "ttir.mesh_shard"(%arg0, %0) <{shard_dims = array<i64: -1, 0>, shard_direction = #tt.shard_direction<full_to_shard>, shard_shape = array<i64: 8, 1, 1, 1>, shard_type = #tt.shard_type<devices>}> : (tensor<64x1x1024x2048xf32>, tensor<8x1x1024x2048xf32>) -> tensor<8x1x1024x2048xf32> | ||
| // CHECK: "ttnn.mesh_shard" | ||
| %2 = tensor.empty() : tensor<1x1x2048x512xf32> | ||
| %3 = "ttir.mesh_shard"(%arg1, %2) <{shard_dims = array<i64: -1>, shard_direction = #tt.shard_direction<full_to_shard>, shard_shape = array<i64: 1>, shard_type = #tt.shard_type<replicate>}> : (tensor<1x1x2048x512xf32>, tensor<1x1x2048x512xf32>) -> tensor<1x1x2048x512xf32> | ||
| // CHECK: "ttnn.mesh_shard" | ||
| %4 = tensor.empty() : tensor<8x1024x2048xf32> | ||
| %5 = "ttir.reshape"(%1, %4) <{shape = [8 : i32, 1024 : i32, 2048 : i32]}> : (tensor<8x1x1024x2048xf32>, tensor<8x1024x2048xf32>) -> tensor<8x1024x2048xf32> | ||
| // CHECK: = "ttnn.reshape" | ||
| %6 = tensor.empty() : tensor<1x2048x512xf32> | ||
| %7 = "ttir.reshape"(%3, %6) <{shape = [1 : i32, 2048 : i32, 512 : i32]}> : (tensor<1x1x2048x512xf32>, tensor<1x2048x512xf32>) -> tensor<1x2048x512xf32> | ||
| // CHECK: = "ttnn.reshape" | ||
| %8 = "ttir.dot_general"(%5, %7) <{batch_dims_lhs = array<i64>, batch_dims_rhs = array<i64>, contract_dims_lhs = array<i64: 2>, contract_dims_rhs = array<i64: 1>}> : (tensor<8x1024x2048xf32>, tensor<1x2048x512xf32>) -> tensor<8x1024x1x512xf32> | ||
| // CHECK: "ttir.matmul" | ||
| %9 = tensor.empty() : tensor<8x1x1024x512xf32> | ||
| %10 = "ttir.permute"(%8, %9) <{permutation = array<i64: 0, 2, 1, 3>}> : (tensor<8x1024x1x512xf32>, tensor<8x1x1024x512xf32>) -> tensor<8x1x1024x512xf32> | ||
| // CHECK: "ttnn.permute" | ||
| %11 = tensor.empty() : tensor<64x1x1024x512xf32> | ||
| %12 = "ttir.mesh_shard"(%10, %11) <{shard_dims = array<i64: -1, 0>, shard_direction = #tt.shard_direction<shard_to_full>, shard_shape = array<i64: 8, 1, 1, 1>, shard_type = #tt.shard_type<devices>}> : (tensor<8x1x1024x512xf32>, tensor<64x1x1024x512xf32>) -> tensor<64x1x1024x512xf32> | ||
| // CHECK: "ttnn.mesh_shard" | ||
| return %12 : tensor<64x1x1024x512xf32> | ||
| } | ||
|
|
||
| func.func public @jit_tensor_parallel_t3000(%arg0: tensor<64x1x1024x2048xf32>, %arg1: tensor<1x1x2048x512xf32>) -> (tensor<64x1x1024x512xf32> {jax.result_info = ""}) { | ||
| %0 = tensor.empty() : tensor<64x1x1024x256xf32> | ||
| %1 = "ttir.mesh_shard"(%arg0, %0) <{shard_dims = array<i64: -1, 3>, shard_direction = #tt.shard_direction<full_to_shard>, shard_shape = array<i64: 1, 1, 1, 8>, shard_type = #tt.shard_type<devices>}> : (tensor<64x1x1024x2048xf32>, tensor<64x1x1024x256xf32>) -> tensor<64x1x1024x256xf32> | ||
| // CHECK: "ttnn.mesh_shard" | ||
| %2 = tensor.empty() : tensor<1x1x256x512xf32> | ||
| %3 = "ttir.mesh_shard"(%arg1, %2) <{shard_dims = array<i64: -1, 2>, shard_direction = #tt.shard_direction<full_to_shard>, shard_shape = array<i64: 1, 1, 8, 1>, shard_type = #tt.shard_type<devices>}> : (tensor<1x1x2048x512xf32>, tensor<1x1x256x512xf32>) -> tensor<1x1x256x512xf32> | ||
| // CHECK: "ttnn.mesh_shard" | ||
| %4 = tensor.empty() : tensor<64x1024x256xf32> | ||
| %5 = "ttir.reshape"(%1, %4) <{shape = [64 : i32, 1024 : i32, 256 : i32]}> : (tensor<64x1x1024x256xf32>, tensor<64x1024x256xf32>) -> tensor<64x1024x256xf32> | ||
| // CHECK: = "ttnn.reshape" | ||
| %6 = tensor.empty() : tensor<1x256x512xf32> | ||
| %7 = "ttir.reshape"(%3, %6) <{shape = [1 : i32, 256 : i32, 512 : i32]}> : (tensor<1x1x256x512xf32>, tensor<1x256x512xf32>) -> tensor<1x256x512xf32> | ||
| // CHECK: = "ttnn.reshape" | ||
| %8 = "ttir.dot_general"(%5, %7) <{batch_dims_lhs = array<i64>, batch_dims_rhs = array<i64>, contract_dims_lhs = array<i64: 2>, contract_dims_rhs = array<i64: 1>}> : (tensor<64x1024x256xf32>, tensor<1x256x512xf32>) -> tensor<64x1024x1x512xf32> | ||
| // CHECK: "ttir.matmul" | ||
| %9 = tensor.empty() : tensor<64x1x1024x512xf32> | ||
| %10 = "ttir.permute"(%8, %9) <{permutation = array<i64: 0, 2, 1, 3>}> : (tensor<64x1024x1x512xf32>, tensor<64x1x1024x512xf32>) -> tensor<64x1x1024x512xf32> | ||
| // CHECK: "ttnn.permute" | ||
| %11 = tensor.empty() : tensor<64x1x128x512xf32> | ||
| %12 = "ttir.reduce_scatter"(%10, %11) <{cluster_axis = 1 : ui32, reduce_type = #tt.reduce_type<sum>, scatter_dim = 2 : si32}> : (tensor<64x1x1024x512xf32>, tensor<64x1x128x512xf32>) -> tensor<64x1x128x512xf32> | ||
| // CHECK: "ttnn.reduce_scatter" | ||
| %13 = tensor.empty() : tensor<64x1x1024x512xf32> | ||
| %14 = "ttir.mesh_shard"(%12, %13) <{shard_dims = array<i64: -1, 2>, shard_direction = #tt.shard_direction<shard_to_full>, shard_shape = array<i64: 1, 1, 8, 1>, shard_type = #tt.shard_type<devices>}> : (tensor<64x1x128x512xf32>, tensor<64x1x1024x512xf32>) -> tensor<64x1x1024x512xf32> | ||
| // CHECK: "ttnn.mesh_shard" | ||
| return %14 : tensor<64x1x1024x512xf32> | ||
| } |
30 changes: 30 additions & 0 deletions
30
test/ttmlir/Silicon/TTNN/llmbox/model_parallel/model_parallel_2x4.mlir
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 |
|---|---|---|
| @@ -0,0 +1,30 @@ | ||
| // RUN: ttmlir-opt --ttir-to-ttnn-backend-pipeline="system-desc-path=%system_desc_path% mesh-shape=2,4" %s > %t.mlir | ||
| // RUN: FileCheck %s --input-file=%t.mlir | ||
| // RUN: ttmlir-translate --ttnn-to-flatbuffer %t.mlir > %t.ttnn | ||
|
|
||
| func.func public @jit_data_tensor_parallel_t3000(%arg0: tensor<64x1x1024x2048xf32>, %arg1: tensor<1x1x2048x512xf32>) -> (tensor<64x1x1024x512xf32> {jax.result_info = ""}) { | ||
| %0 = tensor.empty() : tensor<32x1x1024x512xf32> | ||
| %1 = "ttir.mesh_shard"(%arg0, %0) <{shard_dims = array<i64: 0, 3>, shard_direction = #tt.shard_direction<full_to_shard>, shard_shape = array<i64: 2, 1, 1, 4>, shard_type = #tt.shard_type<devices>}> : (tensor<64x1x1024x2048xf32>, tensor<32x1x1024x512xf32>) -> tensor<32x1x1024x512xf32> | ||
| // CHECK: "ttnn.mesh_shard" | ||
| %2 = tensor.empty() : tensor<1x1x512x512xf32> | ||
| %3 = "ttir.mesh_shard"(%arg1, %2) <{shard_dims = array<i64: -1, 2>, shard_direction = #tt.shard_direction<full_to_shard>, shard_shape = array<i64: 1, 1, 4, 1>, shard_type = #tt.shard_type<devices>}> : (tensor<1x1x2048x512xf32>, tensor<1x1x512x512xf32>) -> tensor<1x1x512x512xf32> | ||
| // CHECK: "ttnn.mesh_shard" | ||
| %4 = tensor.empty() : tensor<32x1024x512xf32> | ||
| %5 = "ttir.reshape"(%1, %4) <{shape = [32 : i32, 1024 : i32, 512 : i32]}> : (tensor<32x1x1024x512xf32>, tensor<32x1024x512xf32>) -> tensor<32x1024x512xf32> | ||
| // CHECK: = "ttnn.reshape" | ||
| %6 = tensor.empty() : tensor<1x512x512xf32> | ||
| %7 = "ttir.reshape"(%3, %6) <{shape = [1 : i32, 512 : i32, 512 : i32]}> : (tensor<1x1x512x512xf32>, tensor<1x512x512xf32>) -> tensor<1x512x512xf32> | ||
| // CHECK: = "ttnn.reshape" | ||
| %8 = "ttir.dot_general"(%5, %7) <{batch_dims_lhs = array<i64>, batch_dims_rhs = array<i64>, contract_dims_lhs = array<i64: 2>, contract_dims_rhs = array<i64: 1>}> : (tensor<32x1024x512xf32>, tensor<1x512x512xf32>) -> tensor<32x1024x1x512xf32> | ||
| // CHECK: "ttir.matmul" | ||
| %9 = tensor.empty() : tensor<32x1x1024x512xf32> | ||
| %10 = "ttir.permute"(%8, %9) <{permutation = array<i64: 0, 2, 1, 3>}> : (tensor<32x1024x1x512xf32>, tensor<32x1x1024x512xf32>) -> tensor<32x1x1024x512xf32> | ||
| // CHECK: "ttnn.permute" | ||
| %11 = tensor.empty() : tensor<32x1x256x512xf32> | ||
| %12 = "ttir.reduce_scatter"(%10, %11) <{cluster_axis = 1 : ui32, reduce_type = #tt.reduce_type<sum>, scatter_dim = 2 : si32}> : (tensor<32x1x1024x512xf32>, tensor<32x1x256x512xf32>) -> tensor<32x1x256x512xf32> | ||
| // CHECK: "ttnn.reduce_scatter" | ||
| %13 = tensor.empty() : tensor<64x1x1024x512xf32> | ||
| %14 = "ttir.mesh_shard"(%12, %13) <{shard_dims = array<i64: 0, 2>, shard_direction = #tt.shard_direction<shard_to_full>, shard_shape = array<i64: 2, 1, 4, 1>, shard_type = #tt.shard_type<devices>}> : (tensor<32x1x256x512xf32>, tensor<64x1x1024x512xf32>) -> tensor<64x1x1024x512xf32> | ||
| // CHECK: "ttnn.mesh_shard" | ||
| return %14 : tensor<64x1x1024x512xf32> | ||
| } |
54 changes: 54 additions & 0 deletions
54
test/ttmlir/Silicon/TTNN/n300/model_parallel/model_parallel_1x2.mlir
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 |
|---|---|---|
| @@ -0,0 +1,54 @@ | ||
| // RUN: ttmlir-opt --ttir-to-ttnn-backend-pipeline="system-desc-path=%system_desc_path% mesh-shape=1,2" %s > %t.mlir | ||
| // RUN: FileCheck %s --input-file=%t.mlir | ||
| // RUN: ttmlir-translate --ttnn-to-flatbuffer %t.mlir > %t.ttnn | ||
|
|
||
| func.func public @jit_data_parallel_n300(%arg0: tensor<64x1x1024x2048xf32>, %arg1: tensor<1x1x2048x512xf32>) -> (tensor<64x1x1024x512xf32> {jax.result_info = ""}) { | ||
| %0 = tensor.empty() : tensor<32x1x1024x2048xf32> | ||
| %1 = "ttir.mesh_shard"(%arg0, %0) <{shard_dims = array<i64: -1, 0>, shard_direction = #tt.shard_direction<full_to_shard>, shard_shape = array<i64: 2, 1, 1, 1>, shard_type = #tt.shard_type<devices>}> : (tensor<64x1x1024x2048xf32>, tensor<32x1x1024x2048xf32>) -> tensor<32x1x1024x2048xf32> | ||
| // CHECK: "ttnn.mesh_shard" | ||
| %2 = tensor.empty() : tensor<1x1x2048x512xf32> | ||
| %3 = "ttir.mesh_shard"(%arg1, %2) <{shard_dims = array<i64: -1>, shard_direction = #tt.shard_direction<full_to_shard>, shard_shape = array<i64: 1>, shard_type = #tt.shard_type<replicate>}> : (tensor<1x1x2048x512xf32>, tensor<1x1x2048x512xf32>) -> tensor<1x1x2048x512xf32> | ||
| // CHECK: "ttnn.mesh_shard" | ||
| %4 = tensor.empty() : tensor<32x1024x2048xf32> | ||
| %5 = "ttir.reshape"(%1, %4) <{shape = [32 : i32, 1024 : i32, 2048 : i32]}> : (tensor<32x1x1024x2048xf32>, tensor<32x1024x2048xf32>) -> tensor<32x1024x2048xf32> | ||
| // CHECK: = "ttnn.reshape" | ||
| %6 = tensor.empty() : tensor<1x2048x512xf32> | ||
| %7 = "ttir.reshape"(%3, %6) <{shape = [1 : i32, 2048 : i32, 512 : i32]}> : (tensor<1x1x2048x512xf32>, tensor<1x2048x512xf32>) -> tensor<1x2048x512xf32> | ||
| // CHECK: = "ttnn.reshape" | ||
| %8 = "ttir.dot_general"(%5, %7) <{batch_dims_lhs = array<i64>, batch_dims_rhs = array<i64>, contract_dims_lhs = array<i64: 2>, contract_dims_rhs = array<i64: 1>}> : (tensor<32x1024x2048xf32>, tensor<1x2048x512xf32>) -> tensor<32x1024x1x512xf32> | ||
| // CHECK: "ttir.matmul" | ||
| %9 = tensor.empty() : tensor<32x1x1024x512xf32> | ||
| %10 = "ttir.permute"(%8, %9) <{permutation = array<i64: 0, 2, 1, 3>}> : (tensor<32x1024x1x512xf32>, tensor<32x1x1024x512xf32>) -> tensor<32x1x1024x512xf32> | ||
| // CHECK: "ttnn.permute" | ||
| %11 = tensor.empty() : tensor<64x1x1024x512xf32> | ||
| %12 = "ttir.mesh_shard"(%10, %11) <{shard_dims = array<i64: -1, 0>, shard_direction = #tt.shard_direction<shard_to_full>, shard_shape = array<i64: 2, 1, 1, 1>, shard_type = #tt.shard_type<devices>}> : (tensor<32x1x1024x512xf32>, tensor<64x1x1024x512xf32>) -> tensor<64x1x1024x512xf32> | ||
| // CHECK: "ttnn.mesh_shard" | ||
| return %12 : tensor<64x1x1024x512xf32> | ||
| } | ||
|
|
||
| func.func public @jit_tensor_parallel_n300(%arg0: tensor<64x1x1024x2048xf32>, %arg1: tensor<1x1x2048x512xf32>) -> (tensor<64x1x1024x512xf32> {jax.result_info = ""}) { | ||
| %0 = tensor.empty() : tensor<64x1x1024x1024xf32> | ||
| %1 = "ttir.mesh_shard"(%arg0, %0) <{shard_dims = array<i64: -1, 3>, shard_direction = #tt.shard_direction<full_to_shard>, shard_shape = array<i64: 1, 1, 1, 2>, shard_type = #tt.shard_type<devices>}> : (tensor<64x1x1024x2048xf32>, tensor<64x1x1024x1024xf32>) -> tensor<64x1x1024x1024xf32> | ||
| // CHECK: "ttnn.mesh_shard" | ||
| %2 = tensor.empty() : tensor<1x1x1024x512xf32> | ||
| %3 = "ttir.mesh_shard"(%arg1, %2) <{shard_dims = array<i64: -1, 2>, shard_direction = #tt.shard_direction<full_to_shard>, shard_shape = array<i64: 1, 1, 2, 1>, shard_type = #tt.shard_type<devices>}> : (tensor<1x1x2048x512xf32>, tensor<1x1x1024x512xf32>) -> tensor<1x1x1024x512xf32> | ||
| // CHECK: "ttnn.mesh_shard" | ||
| %4 = tensor.empty() : tensor<64x1024x1024xf32> | ||
| %5 = "ttir.reshape"(%1, %4) <{shape = [64 : i32, 1024 : i32, 1024 : i32]}> : (tensor<64x1x1024x1024xf32>, tensor<64x1024x1024xf32>) -> tensor<64x1024x1024xf32> | ||
| // CHECK: = "ttnn.reshape" | ||
| %6 = tensor.empty() : tensor<1x1024x512xf32> | ||
| %7 = "ttir.reshape"(%3, %6) <{shape = [1 : i32, 1024 : i32, 512 : i32]}> : (tensor<1x1x1024x512xf32>, tensor<1x1024x512xf32>) -> tensor<1x1024x512xf32> | ||
| // CHECK: = "ttnn.reshape" | ||
| %8 = "ttir.dot_general"(%5, %7) <{batch_dims_lhs = array<i64>, batch_dims_rhs = array<i64>, contract_dims_lhs = array<i64: 2>, contract_dims_rhs = array<i64: 1>}> : (tensor<64x1024x1024xf32>, tensor<1x1024x512xf32>) -> tensor<64x1024x1x512xf32> | ||
| // CHECK: "ttir.matmul" | ||
| %9 = tensor.empty() : tensor<64x1x1024x512xf32> | ||
| %10 = "ttir.permute"(%8, %9) <{permutation = array<i64: 0, 2, 1, 3>}> : (tensor<64x1024x1x512xf32>, tensor<64x1x1024x512xf32>) -> tensor<64x1x1024x512xf32> | ||
| // CHECK: "ttnn.permute" | ||
| %11 = tensor.empty() : tensor<64x1x512x512xf32> | ||
| %12 = "ttir.reduce_scatter"(%10, %11) <{cluster_axis = 1 : ui32, reduce_type = #tt.reduce_type<sum>, scatter_dim = 2 : si32}> : (tensor<64x1x1024x512xf32>, tensor<64x1x512x512xf32>) -> tensor<64x1x512x512xf32> | ||
| // CHECK: "ttnn.reduce_scatter" | ||
| %13 = tensor.empty() : tensor<64x1x1024x512xf32> | ||
| %14 = "ttir.mesh_shard"(%12, %13) <{shard_dims = array<i64: -1, 2>, shard_direction = #tt.shard_direction<shard_to_full>, shard_shape = array<i64: 1, 1, 2, 1>, shard_type = #tt.shard_type<devices>}> : (tensor<64x1x512x512xf32>, tensor<64x1x1024x512xf32>) -> tensor<64x1x1024x512xf32> | ||
| // CHECK: "ttnn.mesh_shard" | ||
| return %14 : tensor<64x1x1024x512xf32> | ||
| } |
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
Oops, something went wrong.