Hugging Face's logo

Kernels:
webgpu-kernels
/
ai.onnx.ReduceProd

Kernel card Files
xet
 Community
Xenova HF Staff commited on
Commit
1a53346
·
verified ·
1 Parent(s): f62a28c

sync c928d21e6cc1

Browse files
Files changed (3) hide show
  1. build/webgpu/manifest.json +1 -1
  2. build/webgpu/metadata.json +4 -4
  3. build/webgpu/reduce-row-tree.wgsl.jinja +2 -1
build/webgpu/manifest.json CHANGED
@@ -615,7 +615,7 @@
615
  "id": "tree_last_axis_vec4",
616
  "priority": 23,
617
  "demoteWhen": ["rowSerialPreferred"],
618
- "when": ["not flatParallelCovered", "(dtypes.T == \"f32\" or dtypes.T == \"f16\")", "f16Ok(dtypes.T)", "ranks.data >= 1", "reduceAxis == ranks.data - 1", "numel(shapes.reduced) == rows(shapes.data, ranks.data - 1)", "attrs.noop_with_empty_axes == 0", "lastAxisCols >= tunables.ROW_PARALLEL_MIN_COLS", "lastAxisCols % tunables.VECTOR_WIDTH == 0", "treeWorkgroupOk"],
619
  "constants": {
620
  "scalar": "dtypes.T",
621
  "vectorScalar": "\"vec4<\" ~ dtypes.T ~ \">\"",
 
615
  "id": "tree_last_axis_vec4",
616
  "priority": 23,
617
  "demoteWhen": ["rowSerialPreferred"],
618
+ "when": ["not flatParallelCovered", "(dtypes.T == \"f32\" or dtypes.T == \"f16\" or dtypes.T == \"i32\")", "f16Ok(dtypes.T)", "ranks.data >= 1", "reduceAxis == ranks.data - 1", "numel(shapes.reduced) == rows(shapes.data, ranks.data - 1)", "attrs.noop_with_empty_axes == 0", "lastAxisCols >= tunables.ROW_PARALLEL_MIN_COLS", "lastAxisCols % tunables.VECTOR_WIDTH == 0", "treeWorkgroupOk"],
619
  "constants": {
620
  "scalar": "dtypes.T",
621
  "vectorScalar": "\"vec4<\" ~ dtypes.T ~ \">\"",
build/webgpu/metadata.json CHANGED
@@ -1,6 +1,6 @@
1
  {
2
  "name": "ai.onnx.ReduceProd",
3
- "id": "_ai_onnx_reduceprod_webgpu_c042834",
4
  "version": 1,
5
  "license": "Apache-2.0",
6
  "backend": { "type": "webgpu" },
@@ -8,7 +8,7 @@
8
  "algorithm": "sha256",
9
  "files": {
10
  "bench.json": "VznixYylucv/AFi4jPmrJwhC0TMoDKr+7rMdIVHsW+s=",
11
- "manifest.json": "UKVOISWKPO15+ThgFhIhsXDjqZGFvCjQbwA2sZXuzdM=",
12
  "reduce-axis-split-reduce.wgsl.jinja": "sKtJglLU0XOIB5OZ3ayiQ3hyBozHELXbS2bRxPdDBtQ=",
13
  "reduce-axis0-splitk-combine.wgsl.jinja": "CP4QmClXgzXKKmy/ynbQdaWsMEkI8ETCiaHU0P1mW1s=",
14
  "reduce-axis0-splitk-reduce.wgsl.jinja": "qArZcLmfHk+cK8BAtnBAhfUlN85EljIXgqfGotutV0g=",
@@ -17,11 +17,11 @@
17
  "reduce-i32-axes02.wgsl.jinja": "lyV7HWjK6vjRO2ekdha9khkcBkeSUa7xPPOA4YyByHI=",
18
  "reduce-noop-empty-axes.wgsl.jinja": "IO2MEzyBGJr3Z8AVP7ELJXqfhwcG/vtBfa0wCdkbuwo=",
19
  "reduce-row-subgroup.wgsl.jinja": "tCCSXf8DaUkcXj2vUD4bwfhvaYxXMN3bEE8q8L90ZrE=",
20
- "reduce-row-tree.wgsl.jinja": "3EVtK/xLb5Qsr9sFQMxSes8/XVdUIJSnIwboAG5hobQ=",
21
  "reduce-serial-axis.wgsl.jinja": "1GJeXytXvJYFn3JhcEk8sbpze8q8+nFVHQrf0lbSbXI=",
22
  "test.json": "w6OEmrn5Bdf1M23l36SbchwXIui/PnEzFjHVeHJUH1I="
23
  }
24
  },
25
- "provenance": { "kernel": { "sha": "2e7068faf55e7f43df740015f6d1ee49391a41c5", "dirty": false } },
26
  "webgpu": { "manifestSpec": "1.0", "specialized": true, "opPath": "ops/ai.onnx.ReduceProd" }
27
  }
 
1
  {
2
  "name": "ai.onnx.ReduceProd",
3
+ "id": "_ai_onnx_reduceprod_webgpu_af6578f",
4
  "version": 1,
5
  "license": "Apache-2.0",
6
  "backend": { "type": "webgpu" },
 
8
  "algorithm": "sha256",
9
  "files": {
10
  "bench.json": "VznixYylucv/AFi4jPmrJwhC0TMoDKr+7rMdIVHsW+s=",
11
+ "manifest.json": "RVeo8qczKuuOGfvWE9Qf77fp9IgC4/Vx/tCYmwo86+E=",
12
  "reduce-axis-split-reduce.wgsl.jinja": "sKtJglLU0XOIB5OZ3ayiQ3hyBozHELXbS2bRxPdDBtQ=",
13
  "reduce-axis0-splitk-combine.wgsl.jinja": "CP4QmClXgzXKKmy/ynbQdaWsMEkI8ETCiaHU0P1mW1s=",
14
  "reduce-axis0-splitk-reduce.wgsl.jinja": "qArZcLmfHk+cK8BAtnBAhfUlN85EljIXgqfGotutV0g=",
 
17
  "reduce-i32-axes02.wgsl.jinja": "lyV7HWjK6vjRO2ekdha9khkcBkeSUa7xPPOA4YyByHI=",
18
  "reduce-noop-empty-axes.wgsl.jinja": "IO2MEzyBGJr3Z8AVP7ELJXqfhwcG/vtBfa0wCdkbuwo=",
19
  "reduce-row-subgroup.wgsl.jinja": "tCCSXf8DaUkcXj2vUD4bwfhvaYxXMN3bEE8q8L90ZrE=",
20
+ "reduce-row-tree.wgsl.jinja": "cLIGeHfVYpn8YH3HHQg84DuMNe/3brWgwLUi2zCFmOw=",
21
  "reduce-serial-axis.wgsl.jinja": "1GJeXytXvJYFn3JhcEk8sbpze8q8+nFVHQrf0lbSbXI=",
22
  "test.json": "w6OEmrn5Bdf1M23l36SbchwXIui/PnEzFjHVeHJUH1I="
23
  }
24
  },
25
+ "provenance": { "kernel": { "sha": "c928d21e6cc1310861cba3bafb75f5f679ecf5f3", "dirty": false } },
26
  "webgpu": { "manifestSpec": "1.0", "specialized": true, "opPath": "ops/ai.onnx.ReduceProd" }
27
  }
build/webgpu/reduce-row-tree.wgsl.jinja CHANGED
@@ -26,7 +26,8 @@ enable f16;
26
  const WG: u32 = {{ workgroupSize }}u;
27
 
28
  {% set is_int = scalar == "i32" or scalar == "u32" %}
29
- {% set intAdditive = is_int and (source.op == "sum" or source.op == "l1" or source.op == "sumsquare") %}
 
30
  {% set accType = scalar if source.op == "max" or source.op == "min" or source.op == "prod" or intAdditive else "f32" %}
31
  {% if source.op == "prod" and accType == "i32" %}
32
  fn identity() -> i32 { return 1i; }
 
26
  const WG: u32 = {{ workgroupSize }}u;
27
 
28
  {% set is_int = scalar == "i32" or scalar == "u32" %}
29
+ {% set intAdditive = is_int and (source.op == "sum" or source.op == "l1" or source.op == "sumsquare"
30
+ or source.op == "l2" or source.op == "mean") %}
31
  {% set accType = scalar if source.op == "max" or source.op == "min" or source.op == "prod" or intAdditive else "f32" %}
32
  {% if source.op == "prod" and accType == "i32" %}
33
  fn identity() -> i32 { return 1i; }