Skip to content

Commit

Permalink
Merge branch 'main' into Cjian/dml-ng
Browse files Browse the repository at this point in the history
  • Loading branch information
@jchen351
jchen351 committed Oct 23, 2024
2 parents 09c1197 + fd8ee48 commit 50be1f0
Show file tree
Hide file tree
Showing 41 changed files with 2,026 additions and 1,106 deletions.
3 changes: 2 additions & 1 deletion docs/OperatorKernels.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -258,7 +258,8 @@ Do not modify directly.*
|||12|**T** = tensor(double), tensor(float), tensor(int32), tensor(int64)<br/> **T1** = tensor(double), tensor(float), tensor(int32), tensor(int64)|
|||[7, 11]|**T** = tensor(double), tensor(float)|
|QLinearConv|*in* x:**T1**<br> *in* x_scale:**tensor(float)**<br> *in* x_zero_point:**T1**<br> *in* w:**T2**<br> *in* w_scale:**tensor(float)**<br> *in* w_zero_point:**T2**<br> *in* y_scale:**tensor(float)**<br> *in* y_zero_point:**T3**<br> *in* B:**T4**<br> *out* y:**T3**|10+|**T1** = tensor(int8), tensor(uint8)<br/> **T2** = tensor(int8), tensor(uint8)<br/> **T3** = tensor(int8), tensor(uint8)<br/> **T4** = tensor(int32)|
|QLinearMatMul|*in* a:**T1**<br> *in* a_scale:**TS**<br> *in* a_zero_point:**T1**<br> *in* b:**T2**<br> *in* b_scale:**TS**<br> *in* b_zero_point:**T2**<br> *in* y_scale:**TS**<br> *in* y_zero_point:**T3**<br> *out* y:**T3**<br><br>or<br><br>*in* a:**T1**<br> *in* a_scale:**tensor(float)**<br> *in* a_zero_point:**T1**<br> *in* b:**T2**<br> *in* b_scale:**tensor(float)**<br> *in* b_zero_point:**T2**<br> *in* y_scale:**tensor(float)**<br> *in* y_zero_point:**T3**<br> *out* y:**T3**|10+|**T1** = tensor(int8), tensor(uint8)<br/> **T2** = tensor(int8), tensor(uint8)<br/> **T3** = tensor(int8), tensor(uint8)|
|QLinearMatMul|*in* a:**T1**<br> *in* a_scale:**TS**<br> *in* a_zero_point:**T1**<br> *in* b:**T2**<br> *in* b_scale:**TS**<br> *in* b_zero_point:**T2**<br> *in* y_scale:**TS**<br> *in* y_zero_point:**T3**<br> *out* y:**T3**<br><br>or<br><br>*in* a:**T1**<br> *in* a_scale:**tensor(float)**<br> *in* a_zero_point:**T1**<br> *in* b:**T2**<br> *in* b_scale:**tensor(float)**<br> *in* b_zero_point:**T2**<br> *in* y_scale:**tensor(float)**<br> *in* y_zero_point:**T3**<br> *out* y:**T3**|21+|**T1** = tensor(int8), tensor(uint8)<br/> **T2** = tensor(int8), tensor(uint8)<br/> **T3** = tensor(int8), tensor(uint8)<br/> **TS** = tensor(float)|
|||[10, 20]|**T1** = tensor(int8), tensor(uint8)<br/> **T2** = tensor(int8), tensor(uint8)<br/> **T3** = tensor(int8), tensor(uint8)|
|QuantizeLinear|*in* x:**T1**<br> *in* y_scale:**T1**<br> *in* y_zero_point:**T2**<br> *out* y:**T2**<br><br>or<br><br>*in* x:**T1**<br> *in* y_scale:**tensor(float)**<br> *in* y_zero_point:**T2**<br> *out* y:**T2**|21+|**T1** = tensor(float), tensor(float16)<br/> **T2** = tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(int16), tensor(int4), tensor(int8), tensor(uint16), tensor(uint4), tensor(uint8)|
|||[19, 20]|**T1** = tensor(float), tensor(float16)<br/> **T2** = tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(int8), tensor(uint8)|
|||[13, 18]|**T1** = tensor(float)<br/> **T2** = tensor(int8), tensor(uint8)|
Expand Down
2 changes: 1 addition & 1 deletion js/web/docs/webnn-operators.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -78,7 +78,7 @@ operators and the supported opset domain/versions in **WebNN EP** by ONNX Runtim
| ReduceSumSquare | ai.onnx(7-10, 11-12, 13-17, 18+) | reduceSumSquare | ✓ | ✓ | Input 'axes' if present should be a constant |
| Relu | ai.onnx(7-12, 13, 14+) | relu | ✓ | ✓ | |
| Reshape | ai.onnx(7-12, 13, 14-18, 19-20, 21+) | reshape | ✓ | ✓ | Input 'shape' should be a constant, 0 dimension value in 'shape' is not supported |
| Resize | ai.onnx(11-12, 13-17, 18, 19+) | resample2d | ✓ | ✓ | Only supports 4-D input, exclude_outside != 0, input 'scales' and 'sizes' if present must be a constant, 'linear' and 'nearest' modes |
| Resize | ai.onnx(11-12, 13-17, 18, 19+) | resample2d | ✓ | ✓ | Only supports 4-D input, antialias == 0, coordinate_transformation_mode == 'half_pixel', exclude_outside == 0, keep_aspect_ratio_policy == 'stretch', 'linear' and 'nearest' modes, input 'scales' and 'sizes' if present must be a constant |
| Shape | ai.onnx(7-12, 13-14, 15-18, 19-20, 21+) | slice | ✓ | ✓ | |
| Sigmoid | ai.onnx(7-12, 13+) | sigmoid | ✓ | ✓ | |
| Softplus | ai.onnx(7+) | softplus | ✓ | ✓ | |
Expand Down
63 changes: 63 additions & 0 deletions js/web/lib/wasm/jsep/backend-webnn.ts
View file
Original file line number Diff line number Diff line change
Expand Up @@ -163,6 +163,69 @@ export class WebNNBackend {
return id;
}

// Register WebNN Constant operands from external data.
public registerMLConstant(
externalFilePath: string,
dataOffset: number,
dataLength: number,
builder: MLGraphBuilder,
desc: MLOperandDescriptor,
mountedFiles: Map<string, Uint8Array> | undefined,
): MLOperand {
// If available, "Module.MountedFiles" is a Map for all preloaded files.
if (!mountedFiles) {
throw new Error('External mounted files are not available.');
}

let filePath = externalFilePath;
if (externalFilePath.startsWith('./')) {
filePath = externalFilePath.substring(2);
}
const fileData = mountedFiles.get(filePath);
if (!fileData) {
throw new Error(`File with name ${filePath} not found in preloaded files.`);
}

if (dataOffset + dataLength > fileData.byteLength) {
throw new Error('Out of bounds: data offset and length exceed the external file data size.');
}

const buffer = fileData.slice(dataOffset, dataOffset + dataLength).buffer;
let bufferView: ArrayBufferView;
switch (desc.dataType) {
case 'float32':
bufferView = new Float32Array(buffer);
break;
case 'float16':
bufferView = new Uint16Array(buffer);
break;
case 'int32':
bufferView = new Int32Array(buffer);
break;
case 'uint32':
bufferView = new Uint32Array(buffer);
break;
case 'int64':
bufferView = new BigInt64Array(buffer);
break;
case 'uint64':
bufferView = new BigUint64Array(buffer);
break;
case 'int8':
bufferView = new Int8Array(buffer);
break;
case 'uint8':
bufferView = new Uint8Array(buffer);
break;
default:
throw new Error(`Unsupported data type: ${desc.dataType} in creating WebNN Constant from external data.`);
}

LOG_DEBUG('verbose', () => `[WebNN] registerMLConstant {dataType: ${desc.dataType}, shape: ${desc.shape}}}`);

return builder.constant(desc, bufferView);
}

public flush(): void {
// Unlike the WebGPU backend, the WebNN backend does not need to flush any pending operations.
}
Expand Down
4 changes: 2 additions & 2 deletions js/web/lib/wasm/jsep/webgpu/op-resolve-rules.ts
View file
Original file line number Diff line number Diff line change
Expand Up @@ -19,7 +19,7 @@ import { gather, parseGatherAttributes } from './ops/gather';
import { gatherBlockQuantized, parseGatherBlockQuantizedAttributes } from './ops/gather-block-quantized';
import { gatherElements, parseGatherElementsAttributes } from './ops/gather-elements';
import { gemm, parseGemmAttributes } from './ops/gemm';
import { groupQueryAttention, parseGroupQueryAttentionAttributes } from './ops/group-query-attention';
import { groupQueryAttention } from './ops/group-query-attention';
import { instanceNorm } from './ops/instance-norm';
import { layerNorm } from './ops/layer-norm';
import { matMul } from './ops/matmul';
Expand Down Expand Up @@ -104,7 +104,7 @@ export const WEBGPU_OP_RESOLVE_RULES: Map<string, OperatorImplementation> = new
['GlobalMaxPool', [pool.globalMaxPool, pool.parseGlobalMaxPoolAttributes]],
['Greater', [binaryOps.greater]],
['GreaterOrEqual', [binaryOps.greaterOrEqual]],
['GroupQueryAttention', [groupQueryAttention, parseGroupQueryAttentionAttributes]],
['GroupQueryAttention', [groupQueryAttention]],
['HardSigmoid', [unaryOps.hardSigmoid, unaryOps.parseHardSigmoidAttributes]],
['InstanceNormalization', [instanceNorm]],
['LayerNormalization', [layerNorm]],
Expand Down
Loading

0 comments on commit 50be1f0

Please sign in to comment.