webgpu-bench 0.1.1 → 0.3.0

package/dist/shaders/flopsI8Dp4a.js

@@ -2,14 +2,17 @@
  * int8 packed-dot-product FLOPS probe: uses the `packed_4x8_integer_dot_product`
  * extension's `dot4I8Packed` builtin directly (two u32s, each four packed
  * signed int8 lanes, one instruction) in a tight accumulation loop, to
- * isolate the instruction's peak throughput. `a` is fed back from the
- * running accumulator every iteration (not just `idx`-derived and fixed),
- * so `dot4I8Packed(a, b)` is genuinely loop-carried — a compiler that
+ * isolate the instruction's peak throughput. Eight independent accumulators
+ * per thread so the number is throughput, not the latency of one dependent
+ * chain. Each `a` is fed back from its own running accumulator every
+ * iteration (not just `idx`-derived and fixed), so `dot4I8Packed(a, b)` is
+ * genuinely loop-carried — a compiler that
  * noticed `a`/`b` were loop-invariant could otherwise hoist the call out of
  * the loop and replace all `iterations` of it with one call + a multiply,
  * which is exactly what happened before this fix (the loop collapsed to
  * ~free, and the dispatch finished faster than the timestamp-query clock's
- * resolution, reading back as a literal 0ns elapsed).
+ * resolution, reading back as a literal 0ns elapsed). 8 dots x 4 MACs =
+ * 64 ops per loop iteration.
  *
  * `packed_4x8_integer_dot_product` is a WGSL *language extension* (gated by
  * `navigator.gpu.wgslLanguageFeatures`, checked as `ctx.info.supportsI8Dot`
@@ -37,13 +40,41 @@ fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
     return;
   }
   let b: u32 = 0x05060708u ^ (idx * 0x85ebca6bu);
-  var a: u32 = 0x01020304u ^ (idx * 0x9e3779b1u);
-  var acc: i32 = i32(idx & 255u);
+  var a0: u32 = 0x01020300u ^ (idx * 0x9e3779b1u);
+  var acc0: i32 = i32(idx & 255u) + 0;
+  var a1: u32 = 0x01020301u ^ (idx * 0x9e3779b1u);
+  var acc1: i32 = i32(idx & 255u) + 1;
+  var a2: u32 = 0x01020302u ^ (idx * 0x9e3779b1u);
+  var acc2: i32 = i32(idx & 255u) + 2;
+  var a3: u32 = 0x01020303u ^ (idx * 0x9e3779b1u);
+  var acc3: i32 = i32(idx & 255u) + 3;
+  var a4: u32 = 0x01020304u ^ (idx * 0x9e3779b1u);
+  var acc4: i32 = i32(idx & 255u) + 4;
+  var a5: u32 = 0x01020305u ^ (idx * 0x9e3779b1u);
+  var acc5: i32 = i32(idx & 255u) + 5;
+  var a6: u32 = 0x01020306u ^ (idx * 0x9e3779b1u);
+  var acc6: i32 = i32(idx & 255u) + 6;
+  var a7: u32 = 0x01020307u ^ (idx * 0x9e3779b1u);
+  var acc7: i32 = i32(idx & 255u) + 7;
   for (var i: u32 = 0u; i < params.iterations; i = i + 1u) {
-    acc = acc + dot4I8Packed(a, b);
-    a = a ^ u32(acc);
+    acc0 = acc0 + dot4I8Packed(a0, b);
+    a0 = a0 ^ u32(acc0);
+    acc1 = acc1 + dot4I8Packed(a1, b);
+    a1 = a1 ^ u32(acc1);
+    acc2 = acc2 + dot4I8Packed(a2, b);
+    a2 = a2 ^ u32(acc2);
+    acc3 = acc3 + dot4I8Packed(a3, b);
+    a3 = a3 ^ u32(acc3);
+    acc4 = acc4 + dot4I8Packed(a4, b);
+    a4 = a4 ^ u32(acc4);
+    acc5 = acc5 + dot4I8Packed(a5, b);
+    a5 = a5 ^ u32(acc5);
+    acc6 = acc6 + dot4I8Packed(a6, b);
+    a6 = a6 ^ u32(acc6);
+    acc7 = acc7 + dot4I8Packed(a7, b);
+    a7 = a7 ^ u32(acc7);
   }
-  out[idx] = f32(acc);
+  out[idx] = f32(acc0 + acc1 + acc2 + acc3 + acc4 + acc5 + acc6 + acc7);
 }
 `;
 //# sourceMappingURL=flopsI8Dp4a.js.map

package/dist/shaders/flopsI8Dp4a.js.map

@@ -1 +1 @@
-{"version":3,"file":"flopsI8Dp4a.js","sourceRoot":"","sources":["../../src/shaders/flopsI8Dp4a.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;GAoBG;AACH,MAAM,CAAC,MAAM,eAAe,GAAG,UAAU,CAAC;;;;;;;;;;;;;;;;;;;;;;;;;;CA0BzC,CAAC"}
+{"version":3,"file":"flopsI8Dp4a.js","sourceRoot":"","sources":["../../src/shaders/flopsI8Dp4a.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AACH,MAAM,CAAC,MAAM,eAAe,GAAG,UAAU,CAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAsDzC,CAAC"}

package/dist/shaders/flopsI8Vec4.d.ts

@@ -1,7 +1,7 @@
 /**
- * int8-range vec4 FLOPS probe: a single integer FMA chain held in a
- * `vec4<i32>` register (4 independent lanes per step; see the fp32 vec4
- * probe for why this is ILP, not SIMD, on most GPUs).
+ * int8-range vec4 ops probe — {@link flopsF32Vec4Wgsl}'s eight independent,
+ * 4x-unrolled `vec4` multiply-add chains, on `vec4<i32>`. Overflow wraps, so
+ * the chains stay bounded. 256 integer ops per loop iteration.
  */
-export declare const flopsI8Vec4Wgsl = "\nstruct Params {\n  threads: u32,\n  iterations: u32,\n};\n\n@group(0) @binding(0) var<uniform> params: Params;\n@group(0) @binding(1) var<storage, read_write> out: array<f32>;\n\n@compute @workgroup_size(64)\nfn main(@builtin(global_invocation_id) gid: vec3<u32>) {\n  let idx = gid.x;\n  if (idx >= params.threads) {\n    return;\n  }\n  let s: i32 = i32(idx & 15u);\n  let a: vec4<i32> = vec4<i32>(3 + s, 5 + s, 7 + s, 9 + s);\n  let b: vec4<i32> = vec4<i32>(11 - s, 13 - s, 17 - s, 19 - s);\n  var x: vec4<i32> = vec4<i32>(i32(idx & 255u));\n  for (var i: u32 = 0u; i < params.iterations; i = i + 1u) {\n    x = x * a + b;\n  }\n  out[idx] = f32(x.x + x.y + x.z + x.w);\n}\n";
+export declare const flopsI8Vec4Wgsl = "\nstruct Params {\n  threads: u32,\n  iterations: u32,\n};\n\n@group(0) @binding(0) var<uniform> params: Params;\n@group(0) @binding(1) var<storage, read_write> out: array<f32>;\n\n@compute @workgroup_size(64)\nfn main(@builtin(global_invocation_id) gid: vec3<u32>) {\n  let idx = gid.x;\n  if (idx >= params.threads) {\n    return;\n  }\n  let s: i32 = i32(idx & 15u);\n  let a: vec4<i32> = vec4<i32>(3 + s, 5 + s, 7 + s, 9 + s);\n  let b: vec4<i32> = vec4<i32>(11 - s, 13 - s, 17 - s, 19 - s);\n  var x0: vec4<i32> = vec4<i32>(i32(idx & 255u) + 1);\n  var x1: vec4<i32> = vec4<i32>(i32(idx & 255u) + 2);\n  var x2: vec4<i32> = vec4<i32>(i32(idx & 255u) + 3);\n  var x3: vec4<i32> = vec4<i32>(i32(idx & 255u) + 4);\n  var x4: vec4<i32> = vec4<i32>(i32(idx & 255u) + 5);\n  var x5: vec4<i32> = vec4<i32>(i32(idx & 255u) + 6);\n  var x6: vec4<i32> = vec4<i32>(i32(idx & 255u) + 7);\n  var x7: vec4<i32> = vec4<i32>(i32(idx & 255u) + 8);\n  for (var i: u32 = 0u; i < params.iterations; i = i + 1u) {\n    x0 = x0 * a + b;\n    x1 = x1 * a + b;\n    x2 = x2 * a + b;\n    x3 = x3 * a + b;\n    x4 = x4 * a + b;\n    x5 = x5 * a + b;\n    x6 = x6 * a + b;\n    x7 = x7 * a + b;\n    x0 = x0 * a + b;\n    x1 = x1 * a + b;\n    x2 = x2 * a + b;\n    x3 = x3 * a + b;\n    x4 = x4 * a + b;\n    x5 = x5 * a + b;\n    x6 = x6 * a + b;\n    x7 = x7 * a + b;\n    x0 = x0 * a + b;\n    x1 = x1 * a + b;\n    x2 = x2 * a + b;\n    x3 = x3 * a + b;\n    x4 = x4 * a + b;\n    x5 = x5 * a + b;\n    x6 = x6 * a + b;\n    x7 = x7 * a + b;\n    x0 = x0 * a + b;\n    x1 = x1 * a + b;\n    x2 = x2 * a + b;\n    x3 = x3 * a + b;\n    x4 = x4 * a + b;\n    x5 = x5 * a + b;\n    x6 = x6 * a + b;\n    x7 = x7 * a + b;\n  }\n  let t = x0 + x1 + x2 + x3 + x4 + x5 + x6 + x7;\n  out[idx] = f32(t.x + t.y + t.z + t.w);\n}\n";
 //# sourceMappingURL=flopsI8Vec4.d.ts.map

package/dist/shaders/flopsI8Vec4.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"flopsI8Vec4.d.ts","sourceRoot":"","sources":["../../src/shaders/flopsI8Vec4.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AACH,eAAO,MAAM,eAAe,4qBAwB3B,CAAC"}
+{"version":3,"file":"flopsI8Vec4.d.ts","sourceRoot":"","sources":["../../src/shaders/flopsI8Vec4.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AACH,eAAO,MAAM,eAAe,gxDA+D3B,CAAC"}

package/dist/shaders/flopsI8Vec4.js

@@ -1,7 +1,7 @@
 /**
- * int8-range vec4 FLOPS probe: a single integer FMA chain held in a
- * `vec4<i32>` register (4 independent lanes per step; see the fp32 vec4
- * probe for why this is ILP, not SIMD, on most GPUs).
+ * int8-range vec4 ops probe — {@link flopsF32Vec4Wgsl}'s eight independent,
+ * 4x-unrolled `vec4` multiply-add chains, on `vec4<i32>`. Overflow wraps, so
+ * the chains stay bounded. 256 integer ops per loop iteration.
  */
 export const flopsI8Vec4Wgsl = /* wgsl */ `
 struct Params {
@@ -21,11 +21,50 @@ fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
   let s: i32 = i32(idx & 15u);
   let a: vec4<i32> = vec4<i32>(3 + s, 5 + s, 7 + s, 9 + s);
   let b: vec4<i32> = vec4<i32>(11 - s, 13 - s, 17 - s, 19 - s);
-  var x: vec4<i32> = vec4<i32>(i32(idx & 255u));
+  var x0: vec4<i32> = vec4<i32>(i32(idx & 255u) + 1);
+  var x1: vec4<i32> = vec4<i32>(i32(idx & 255u) + 2);
+  var x2: vec4<i32> = vec4<i32>(i32(idx & 255u) + 3);
+  var x3: vec4<i32> = vec4<i32>(i32(idx & 255u) + 4);
+  var x4: vec4<i32> = vec4<i32>(i32(idx & 255u) + 5);
+  var x5: vec4<i32> = vec4<i32>(i32(idx & 255u) + 6);
+  var x6: vec4<i32> = vec4<i32>(i32(idx & 255u) + 7);
+  var x7: vec4<i32> = vec4<i32>(i32(idx & 255u) + 8);
   for (var i: u32 = 0u; i < params.iterations; i = i + 1u) {
-    x = x * a + b;
+    x0 = x0 * a + b;
+    x1 = x1 * a + b;
+    x2 = x2 * a + b;
+    x3 = x3 * a + b;
+    x4 = x4 * a + b;
+    x5 = x5 * a + b;
+    x6 = x6 * a + b;
+    x7 = x7 * a + b;
+    x0 = x0 * a + b;
+    x1 = x1 * a + b;
+    x2 = x2 * a + b;
+    x3 = x3 * a + b;
+    x4 = x4 * a + b;
+    x5 = x5 * a + b;
+    x6 = x6 * a + b;
+    x7 = x7 * a + b;
+    x0 = x0 * a + b;
+    x1 = x1 * a + b;
+    x2 = x2 * a + b;
+    x3 = x3 * a + b;
+    x4 = x4 * a + b;
+    x5 = x5 * a + b;
+    x6 = x6 * a + b;
+    x7 = x7 * a + b;
+    x0 = x0 * a + b;
+    x1 = x1 * a + b;
+    x2 = x2 * a + b;
+    x3 = x3 * a + b;
+    x4 = x4 * a + b;
+    x5 = x5 * a + b;
+    x6 = x6 * a + b;
+    x7 = x7 * a + b;
   }
-  out[idx] = f32(x.x + x.y + x.z + x.w);
+  let t = x0 + x1 + x2 + x3 + x4 + x5 + x6 + x7;
+  out[idx] = f32(t.x + t.y + t.z + t.w);
 }
 `;
 //# sourceMappingURL=flopsI8Vec4.js.map

package/dist/shaders/flopsI8Vec4.js.map

@@ -1 +1 @@
-{"version":3,"file":"flopsI8Vec4.js","sourceRoot":"","sources":["../../src/shaders/flopsI8Vec4.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AACH,MAAM,CAAC,MAAM,eAAe,GAAG,UAAU,CAAC;;;;;;;;;;;;;;;;;;;;;;;;CAwBzC,CAAC"}
+{"version":3,"file":"flopsI8Vec4.js","sourceRoot":"","sources":["../../src/shaders/flopsI8Vec4.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AACH,MAAM,CAAC,MAAM,eAAe,GAAG,UAAU,CAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CA+DzC,CAAC"}

package/dist/shaders/flopsU32PackUnpack.d.ts

@@ -1,11 +1,13 @@
 /**
- * Manual byte pack/unpack ops probe: no `pack4x*` (or `unpack4x*`) builtin, just
- * the shift+mask bit-twiddling those builtins compile to, so this measures
- * the raw integer ALU/shifter cost of doing it by hand. Eight independent
- * u32 lanes, each per step: unpack 4 bytes (4 mask + 3 shift = 7 ops),
- * increment+wrap each byte (4 add + 4 mask = 8 ops), repack (4 mask + 3
- * shift + 3 or = 10 ops) = 25 ops/lane/step, no unrolling (each step is
- * already many instructions deep). 200 ops per loop iteration.
+ * Manual byte pack/unpack probe: no `pack4x*` (or `unpack4x*`) builtin, just
+ * the shift+mask bit-twiddling those builtins compile to. Eight independent
+ * u32 lanes, unrolled 4x; each step unpacks 4 bytes, increments each, and
+ * repacks. Written with no redundant ops (the top byte needs no mask after
+ * `>> 24u`, and `<< 24u` discards the bits a mask would) so the source is as
+ * close as possible to what executes — but the compiler is still free to
+ * reassociate shifts and masks, so each lane step counts as 2 ops (one
+ * unpack + one pack of a whole u32), not an instruction count. 64 ops per
+ * loop iteration.
  */
-export declare const flopsU32PackUnpackWgsl = "\nstruct Params {\n  threads: u32,\n  iterations: u32,\n};\n\n@group(0) @binding(0) var<uniform> params: Params;\n@group(0) @binding(1) var<storage, read_write> out: array<f32>;\n\nfn unpack4(v: u32) -> vec4<u32> {\n  return vec4<u32>(v & 0xFFu, (v >> 8u) & 0xFFu, (v >> 16u) & 0xFFu, (v >> 24u) & 0xFFu);\n}\n\nfn pack4(b: vec4<u32>) -> u32 {\n  return (b.x & 0xFFu) | ((b.y & 0xFFu) << 8u) | ((b.z & 0xFFu) << 16u) | ((b.w & 0xFFu) << 24u);\n}\n\n@compute @workgroup_size(64)\nfn main(@builtin(global_invocation_id) gid: vec3<u32>) {\n  let idx = gid.x;\n  if (idx >= params.threads) {\n    return;\n  }\n  var v0: u32 = idx * 2654435761u;\n  var v1: u32 = (idx + 1u) * 2654435761u;\n  var v2: u32 = (idx + 2u) * 2654435761u;\n  var v3: u32 = (idx + 3u) * 2654435761u;\n  var v4: u32 = (idx + 4u) * 2654435761u;\n  var v5: u32 = (idx + 5u) * 2654435761u;\n  var v6: u32 = (idx + 6u) * 2654435761u;\n  var v7: u32 = (idx + 7u) * 2654435761u;\n  let one = vec4<u32>(1u);\n  for (var i: u32 = 0u; i < params.iterations; i = i + 1u) {\n    v0 = pack4((unpack4(v0) + one) & vec4<u32>(0xFFu));\n    v1 = pack4((unpack4(v1) + one) & vec4<u32>(0xFFu));\n    v2 = pack4((unpack4(v2) + one) & vec4<u32>(0xFFu));\n    v3 = pack4((unpack4(v3) + one) & vec4<u32>(0xFFu));\n    v4 = pack4((unpack4(v4) + one) & vec4<u32>(0xFFu));\n    v5 = pack4((unpack4(v5) + one) & vec4<u32>(0xFFu));\n    v6 = pack4((unpack4(v6) + one) & vec4<u32>(0xFFu));\n    v7 = pack4((unpack4(v7) + one) & vec4<u32>(0xFFu));\n  }\n  out[idx] = f32(v0 ^ v1 ^ v2 ^ v3 ^ v4 ^ v5 ^ v6 ^ v7);\n}\n";
+export declare const flopsU32PackUnpackWgsl = "\nstruct Params {\n  threads: u32,\n  iterations: u32,\n};\n\n@group(0) @binding(0) var<uniform> params: Params;\n@group(0) @binding(1) var<storage, read_write> out: array<f32>;\n\n// Unpack 4 bytes, add 1 to each (wrapping), repack.\nfn step(v: u32) -> u32 {\n  let b0 = (v & 0xFFu) + 1u;\n  let b1 = ((v >> 8u) & 0xFFu) + 1u;\n  let b2 = ((v >> 16u) & 0xFFu) + 1u;\n  let b3 = (v >> 24u) + 1u;\n  return (b0 & 0xFFu) | ((b1 & 0xFFu) << 8u) | ((b2 & 0xFFu) << 16u) | (b3 << 24u);\n}\n\n@compute @workgroup_size(64)\nfn main(@builtin(global_invocation_id) gid: vec3<u32>) {\n  let idx = gid.x;\n  if (idx >= params.threads) {\n    return;\n  }\n  var v0: u32 = (idx + 0u) * 2654435761u;\n  var v1: u32 = (idx + 1u) * 2654435761u;\n  var v2: u32 = (idx + 2u) * 2654435761u;\n  var v3: u32 = (idx + 3u) * 2654435761u;\n  var v4: u32 = (idx + 4u) * 2654435761u;\n  var v5: u32 = (idx + 5u) * 2654435761u;\n  var v6: u32 = (idx + 6u) * 2654435761u;\n  var v7: u32 = (idx + 7u) * 2654435761u;\n  for (var i: u32 = 0u; i < params.iterations; i = i + 1u) {\n    v0 = step(v0);\n    v1 = step(v1);\n    v2 = step(v2);\n    v3 = step(v3);\n    v4 = step(v4);\n    v5 = step(v5);\n    v6 = step(v6);\n    v7 = step(v7);\n    v0 = step(v0);\n    v1 = step(v1);\n    v2 = step(v2);\n    v3 = step(v3);\n    v4 = step(v4);\n    v5 = step(v5);\n    v6 = step(v6);\n    v7 = step(v7);\n    v0 = step(v0);\n    v1 = step(v1);\n    v2 = step(v2);\n    v3 = step(v3);\n    v4 = step(v4);\n    v5 = step(v5);\n    v6 = step(v6);\n    v7 = step(v7);\n    v0 = step(v0);\n    v1 = step(v1);\n    v2 = step(v2);\n    v3 = step(v3);\n    v4 = step(v4);\n    v5 = step(v5);\n    v6 = step(v6);\n    v7 = step(v7);\n  }\n  out[idx] = f32(v0 ^ v1 ^ v2 ^ v3 ^ v4 ^ v5 ^ v6 ^ v7);\n}\n";
 //# sourceMappingURL=flopsU32PackUnpack.d.ts.map

package/dist/shaders/flopsU32PackUnpack.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"flopsU32PackUnpack.d.ts","sourceRoot":"","sources":["../../src/shaders/flopsU32PackUnpack.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AACH,eAAO,MAAM,sBAAsB,yhDA4ClC,CAAC"}
+{"version":3,"file":"flopsU32PackUnpack.d.ts","sourceRoot":"","sources":["../../src/shaders/flopsU32PackUnpack.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AACH,eAAO,MAAM,sBAAsB,kuDAoElC,CAAC"}

package/dist/shaders/flopsU32PackUnpack.js

@@ -1,11 +1,13 @@
 /**
- * Manual byte pack/unpack ops probe: no `pack4x*` (or `unpack4x*`) builtin, just
- * the shift+mask bit-twiddling those builtins compile to, so this measures
- * the raw integer ALU/shifter cost of doing it by hand. Eight independent
- * u32 lanes, each per step: unpack 4 bytes (4 mask + 3 shift = 7 ops),
- * increment+wrap each byte (4 add + 4 mask = 8 ops), repack (4 mask + 3
- * shift + 3 or = 10 ops) = 25 ops/lane/step, no unrolling (each step is
- * already many instructions deep). 200 ops per loop iteration.
+ * Manual byte pack/unpack probe: no `pack4x*` (or `unpack4x*`) builtin, just
+ * the shift+mask bit-twiddling those builtins compile to. Eight independent
+ * u32 lanes, unrolled 4x; each step unpacks 4 bytes, increments each, and
+ * repacks. Written with no redundant ops (the top byte needs no mask after
+ * `>> 24u`, and `<< 24u` discards the bits a mask would) so the source is as
+ * close as possible to what executes — but the compiler is still free to
+ * reassociate shifts and masks, so each lane step counts as 2 ops (one
+ * unpack + one pack of a whole u32), not an instruction count. 64 ops per
+ * loop iteration.
  */
 export const flopsU32PackUnpackWgsl = /* wgsl */ `
 struct Params {
@@ -16,12 +18,13 @@ struct Params {
 @group(0) @binding(0) var<uniform> params: Params;
 @group(0) @binding(1) var<storage, read_write> out: array<f32>;
 
-fn unpack4(v: u32) -> vec4<u32> {
-  return vec4<u32>(v & 0xFFu, (v >> 8u) & 0xFFu, (v >> 16u) & 0xFFu, (v >> 24u) & 0xFFu);
-}
-
-fn pack4(b: vec4<u32>) -> u32 {
-  return (b.x & 0xFFu) | ((b.y & 0xFFu) << 8u) | ((b.z & 0xFFu) << 16u) | ((b.w & 0xFFu) << 24u);
+// Unpack 4 bytes, add 1 to each (wrapping), repack.
+fn step(v: u32) -> u32 {
+  let b0 = (v & 0xFFu) + 1u;
+  let b1 = ((v >> 8u) & 0xFFu) + 1u;
+  let b2 = ((v >> 16u) & 0xFFu) + 1u;
+  let b3 = (v >> 24u) + 1u;
+  return (b0 & 0xFFu) | ((b1 & 0xFFu) << 8u) | ((b2 & 0xFFu) << 16u) | (b3 << 24u);
 }
 
 @compute @workgroup_size(64)
@@ -30,7 +33,7 @@ fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
   if (idx >= params.threads) {
     return;
   }
-  var v0: u32 = idx * 2654435761u;
+  var v0: u32 = (idx + 0u) * 2654435761u;
   var v1: u32 = (idx + 1u) * 2654435761u;
   var v2: u32 = (idx + 2u) * 2654435761u;
   var v3: u32 = (idx + 3u) * 2654435761u;
@@ -38,16 +41,39 @@ fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
   var v5: u32 = (idx + 5u) * 2654435761u;
   var v6: u32 = (idx + 6u) * 2654435761u;
   var v7: u32 = (idx + 7u) * 2654435761u;
-  let one = vec4<u32>(1u);
   for (var i: u32 = 0u; i < params.iterations; i = i + 1u) {
-    v0 = pack4((unpack4(v0) + one) & vec4<u32>(0xFFu));
-    v1 = pack4((unpack4(v1) + one) & vec4<u32>(0xFFu));
-    v2 = pack4((unpack4(v2) + one) & vec4<u32>(0xFFu));
-    v3 = pack4((unpack4(v3) + one) & vec4<u32>(0xFFu));
-    v4 = pack4((unpack4(v4) + one) & vec4<u32>(0xFFu));
-    v5 = pack4((unpack4(v5) + one) & vec4<u32>(0xFFu));
-    v6 = pack4((unpack4(v6) + one) & vec4<u32>(0xFFu));
-    v7 = pack4((unpack4(v7) + one) & vec4<u32>(0xFFu));
+    v0 = step(v0);
+    v1 = step(v1);
+    v2 = step(v2);
+    v3 = step(v3);
+    v4 = step(v4);
+    v5 = step(v5);
+    v6 = step(v6);
+    v7 = step(v7);
+    v0 = step(v0);
+    v1 = step(v1);
+    v2 = step(v2);
+    v3 = step(v3);
+    v4 = step(v4);
+    v5 = step(v5);
+    v6 = step(v6);
+    v7 = step(v7);
+    v0 = step(v0);
+    v1 = step(v1);
+    v2 = step(v2);
+    v3 = step(v3);
+    v4 = step(v4);
+    v5 = step(v5);
+    v6 = step(v6);
+    v7 = step(v7);
+    v0 = step(v0);
+    v1 = step(v1);
+    v2 = step(v2);
+    v3 = step(v3);
+    v4 = step(v4);
+    v5 = step(v5);
+    v6 = step(v6);
+    v7 = step(v7);
   }
   out[idx] = f32(v0 ^ v1 ^ v2 ^ v3 ^ v4 ^ v5 ^ v6 ^ v7);
 }

package/dist/shaders/flopsU32PackUnpack.js.map

@@ -1 +1 @@
-{"version":3,"file":"flopsU32PackUnpack.js","sourceRoot":"","sources":["../../src/shaders/flopsU32PackUnpack.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AACH,MAAM,CAAC,MAAM,sBAAsB,GAAG,UAAU,CAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CA4ChD,CAAC"}
+{"version":3,"file":"flopsU32PackUnpack.js","sourceRoot":"","sources":["../../src/shaders/flopsU32PackUnpack.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AACH,MAAM,CAAC,MAAM,sBAAsB,GAAG,UAAU,CAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAoEhD,CAAC"}

package/dist/shaders/streamRead.d.ts

@@ -1,10 +1,12 @@
 /**
- * Read-bandwidth probe: each thread streams a run of `vec4<f32>` values out
- * of a large storage buffer and folds them together with plain addition (no
- * second buffer, no multiply) — the cheapest possible "prove you read it"
- * op — then writes a single scalar. Reads dominate; writes are negligible
- * (one f32 per thread vs. `cols4` vec4s read), so throughput here is close
- * to the device's peak storage-buffer read bandwidth.
+ * Read-bandwidth probe. Grid-stride loop: on step `i` every thread `t` loads
+ * `data[t + i * threads]`, so adjacent threads in a SIMD group always touch
+ * adjacent `vec4<f32>`s (fully coalesced) and the loads within a thread are
+ * independent (the compiler can keep several in flight). Folds with plain
+ * addition — the cheapest "prove you read it" op — and writes one scalar per
+ * thread, so reads dominate. The previous one-row-per-thread layout had
+ * adjacent threads 16 KB apart and only 4096 threads in total: fine on an
+ * M3, badly under-reports on a wide discrete GPU.
  */
-export declare const streamReadWgsl = "\nstruct Params {\n  rows: u32,\n  cols4: u32,\n};\n\n@group(0) @binding(0) var<uniform> params: Params;\n@group(0) @binding(1) var<storage, read> data: array<vec4<f32>>;\n@group(0) @binding(2) var<storage, read_write> out: array<f32>;\n\n@compute @workgroup_size(64)\nfn main(@builtin(global_invocation_id) gid: vec3<u32>) {\n  let row = gid.x;\n  if (row >= params.rows) {\n    return;\n  }\n  var sum: vec4<f32> = vec4<f32>(0.0, 0.0, 0.0, 0.0);\n  let base = row * params.cols4;\n  for (var c: u32 = 0u; c < params.cols4; c = c + 1u) {\n    sum = sum + data[base + c];\n  }\n  out[row] = sum.x + sum.y + sum.z + sum.w;\n}\n";
+export declare const streamReadWgsl = "\nstruct Params {\n  count: u32,\n  threads: u32,\n};\n\n@group(0) @binding(0) var<uniform> params: Params;\n@group(0) @binding(1) var<storage, read> data: array<vec4<f32>>;\n@group(0) @binding(2) var<storage, read_write> out: array<f32>;\n\n@compute @workgroup_size(256)\nfn main(@builtin(global_invocation_id) gid: vec3<u32>) {\n  let t = gid.x;\n  var sum: vec4<f32> = vec4<f32>(0.0, 0.0, 0.0, 0.0);\n  for (var i: u32 = t; i < params.count; i = i + params.threads) {\n    sum = sum + data[i];\n  }\n  out[t] = sum.x + sum.y + sum.z + sum.w;\n}\n";
 //# sourceMappingURL=streamRead.d.ts.map

package/dist/shaders/streamRead.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"streamRead.d.ts","sourceRoot":"","sources":["../../src/shaders/streamRead.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AACH,eAAO,MAAM,cAAc,unBAuB1B,CAAC"}
+{"version":3,"file":"streamRead.d.ts","sourceRoot":"","sources":["../../src/shaders/streamRead.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AACH,eAAO,MAAM,cAAc,0iBAmB1B,CAAC"}

package/dist/shaders/streamRead.js

@@ -1,33 +1,31 @@
 /**
- * Read-bandwidth probe: each thread streams a run of `vec4<f32>` values out
- * of a large storage buffer and folds them together with plain addition (no
- * second buffer, no multiply) — the cheapest possible "prove you read it"
- * op — then writes a single scalar. Reads dominate; writes are negligible
- * (one f32 per thread vs. `cols4` vec4s read), so throughput here is close
- * to the device's peak storage-buffer read bandwidth.
+ * Read-bandwidth probe. Grid-stride loop: on step `i` every thread `t` loads
+ * `data[t + i * threads]`, so adjacent threads in a SIMD group always touch
+ * adjacent `vec4<f32>`s (fully coalesced) and the loads within a thread are
+ * independent (the compiler can keep several in flight). Folds with plain
+ * addition — the cheapest "prove you read it" op — and writes one scalar per
+ * thread, so reads dominate. The previous one-row-per-thread layout had
+ * adjacent threads 16 KB apart and only 4096 threads in total: fine on an
+ * M3, badly under-reports on a wide discrete GPU.
  */
 export const streamReadWgsl = /* wgsl */ `
 struct Params {
-  rows: u32,
-  cols4: u32,
+  count: u32,
+  threads: u32,
 };
 
 @group(0) @binding(0) var<uniform> params: Params;
 @group(0) @binding(1) var<storage, read> data: array<vec4<f32>>;
 @group(0) @binding(2) var<storage, read_write> out: array<f32>;
 
-@compute @workgroup_size(64)
+@compute @workgroup_size(256)
 fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
-  let row = gid.x;
-  if (row >= params.rows) {
-    return;
-  }
+  let t = gid.x;
   var sum: vec4<f32> = vec4<f32>(0.0, 0.0, 0.0, 0.0);
-  let base = row * params.cols4;
-  for (var c: u32 = 0u; c < params.cols4; c = c + 1u) {
-    sum = sum + data[base + c];
+  for (var i: u32 = t; i < params.count; i = i + params.threads) {
+    sum = sum + data[i];
   }
-  out[row] = sum.x + sum.y + sum.z + sum.w;
+  out[t] = sum.x + sum.y + sum.z + sum.w;
 }
 `;
 //# sourceMappingURL=streamRead.js.map

package/dist/shaders/streamRead.js.map

@@ -1 +1 @@
-{"version":3,"file":"streamRead.js","sourceRoot":"","sources":["../../src/shaders/streamRead.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AACH,MAAM,CAAC,MAAM,cAAc,GAAG,UAAU,CAAC;;;;;;;;;;;;;;;;;;;;;;;CAuBxC,CAAC"}
+{"version":3,"file":"streamRead.js","sourceRoot":"","sources":["../../src/shaders/streamRead.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AACH,MAAM,CAAC,MAAM,cAAc,GAAG,UAAU,CAAC;;;;;;;;;;;;;;;;;;;CAmBxC,CAAC"}

package/dist/shaders/streamWrite.d.ts

@@ -1,8 +1,7 @@
 /**
- * Write-bandwidth probe: each thread stores a run of `vec4<f32>` values into
- * a large storage buffer. No buffer reads at all; the only input is the
- * thread's own id, so throughput here is close to the device's peak
- * storage-buffer write bandwidth.
+ * Write-bandwidth probe: the grid-stride layout of {@link streamReadWgsl}
+ * with the loads replaced by stores. No buffer reads at all; the only input
+ * is the thread's own id.
  */
-export declare const streamWriteWgsl = "\nstruct Params {\n  rows: u32,\n  cols4: u32,\n};\n\n@group(0) @binding(0) var<uniform> params: Params;\n@group(0) @binding(1) var<storage, read_write> out: array<vec4<f32>>;\n\n@compute @workgroup_size(64)\nfn main(@builtin(global_invocation_id) gid: vec3<u32>) {\n  let row = gid.x;\n  if (row >= params.rows) {\n    return;\n  }\n  let base = row * params.cols4;\n  let v = vec4<f32>(f32(row), f32(row), f32(row), f32(row));\n  for (var c: u32 = 0u; c < params.cols4; c = c + 1u) {\n    out[base + c] = v;\n  }\n}\n";
+export declare const streamWriteWgsl = "\nstruct Params {\n  count: u32,\n  threads: u32,\n};\n\n@group(0) @binding(0) var<uniform> params: Params;\n@group(0) @binding(1) var<storage, read_write> out: array<vec4<f32>>;\n\n@compute @workgroup_size(256)\nfn main(@builtin(global_invocation_id) gid: vec3<u32>) {\n  let t = gid.x;\n  let v = vec4<f32>(f32(t), f32(t), f32(t), f32(t));\n  for (var i: u32 = t; i < params.count; i = i + params.threads) {\n    out[i] = v;\n  }\n}\n";
 //# sourceMappingURL=streamWrite.d.ts.map

package/dist/shaders/streamWrite.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"streamWrite.d.ts","sourceRoot":"","sources":["../../src/shaders/streamWrite.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AACH,eAAO,MAAM,eAAe,4gBAqB3B,CAAC"}
+{"version":3,"file":"streamWrite.d.ts","sourceRoot":"","sources":["../../src/shaders/streamWrite.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AACH,eAAO,MAAM,eAAe,ybAiB3B,CAAC"}

package/dist/shaders/streamWrite.js

@@ -1,28 +1,23 @@
 /**
- * Write-bandwidth probe: each thread stores a run of `vec4<f32>` values into
- * a large storage buffer. No buffer reads at all; the only input is the
- * thread's own id, so throughput here is close to the device's peak
- * storage-buffer write bandwidth.
+ * Write-bandwidth probe: the grid-stride layout of {@link streamReadWgsl}
+ * with the loads replaced by stores. No buffer reads at all; the only input
+ * is the thread's own id.
  */
 export const streamWriteWgsl = /* wgsl */ `
 struct Params {
-  rows: u32,
-  cols4: u32,
+  count: u32,
+  threads: u32,
 };
 
 @group(0) @binding(0) var<uniform> params: Params;
 @group(0) @binding(1) var<storage, read_write> out: array<vec4<f32>>;
 
-@compute @workgroup_size(64)
+@compute @workgroup_size(256)
 fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
-  let row = gid.x;
-  if (row >= params.rows) {
-    return;
-  }
-  let base = row * params.cols4;
-  let v = vec4<f32>(f32(row), f32(row), f32(row), f32(row));
-  for (var c: u32 = 0u; c < params.cols4; c = c + 1u) {
-    out[base + c] = v;
+  let t = gid.x;
+  let v = vec4<f32>(f32(t), f32(t), f32(t), f32(t));
+  for (var i: u32 = t; i < params.count; i = i + params.threads) {
+    out[i] = v;
   }
 }
 `;

package/dist/shaders/streamWrite.js.map

@@ -1 +1 @@
-{"version":3,"file":"streamWrite.js","sourceRoot":"","sources":["../../src/shaders/streamWrite.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AACH,MAAM,CAAC,MAAM,eAAe,GAAG,UAAU,CAAC;;;;;;;;;;;;;;;;;;;;;CAqBzC,CAAC"}
+{"version":3,"file":"streamWrite.js","sourceRoot":"","sources":["../../src/shaders/streamWrite.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AACH,MAAM,CAAC,MAAM,eAAe,GAAG,UAAU,CAAC;;;;;;;;;;;;;;;;;CAiBzC,CAAC"}

package/dist/suite.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"suite.d.ts","sourceRoot":"","sources":["../src/suite.ts"],"names":[],"mappings":"AAAA,OAAO,EAAqB,KAAK,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAkDtE,OAAO,KAAK,EAAqB,eAAe,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAEnF,YAAY,EACV,eAAe,EACf,YAAY,EACZ,UAAU,EACV,KAAK,EACL,iBAAiB,EACjB,eAAe,EACf,kBAAkB,EAClB,kBAAkB,EAClB,SAAS,GACV,MAAM,YAAY,CAAC;AACpB,OAAO,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAgB1C;;;;;;;;;;;;;;;;;GAiBG;AACH,wBAAuB,QAAQ,CAAC,OAAO,GAAE,YAAiB,GAAG,cAAc,CAAC,eAAe,CAAC,CAiI3F;AAyCD,YAAY,EAAE,UAAU,EAAE,CAAC"}
+{"version":3,"file":"suite.d.ts","sourceRoot":"","sources":["../src/suite.ts"],"names":[],"mappings":"AAAA,OAAO,EAAqB,KAAK,UAAU,EAAE,MAAM,kBAAkB,CAAC;AA+CtE,OAAO,KAAK,EAAqB,eAAe,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAEnF,YAAY,EACV,eAAe,EACf,YAAY,EACZ,UAAU,EACV,KAAK,EACL,iBAAiB,EACjB,eAAe,EACf,kBAAkB,EAClB,kBAAkB,EAClB,SAAS,GACV,MAAM,YAAY,CAAC;AACpB,OAAO,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAgB1C;;;;;;;;;;;;;;;;;GAiBG;AACH,wBAAuB,QAAQ,CAAC,OAAO,GAAE,YAAiB,GAAG,cAAc,CAAC,eAAe,CAAC,CA4H3F;AAyCD,YAAY,EAAE,UAAU,EAAE,CAAC"}

package/dist/suite.js

@@ -7,8 +7,8 @@ import { prepareFlopsF32Scalar, prepareFlopsF32Vec4, prepareFlopsF32Mat4, prepar
 import { prepareFlopsF16Scalar, prepareFlopsF16Vec4, prepareFlopsF16Mat4, prepareFlopsF16Matvec, } from "./benchmarks/flopsF16.js";
 import { prepareFlopsI8Scalar, prepareFlopsI8Vec4, prepareFlopsI8Mat4, prepareFlopsI8Matvec, prepareFlopsI8MatvecDp4a, prepareFlopsI8Dp4a, } from "./benchmarks/flopsI8.js";
 import { prepareFlopsF32Div, prepareFlopsI32Div, prepareFlopsF32Sqrt, prepareFlopsF32Rsqrt, prepareFlopsF32Pow, prepareFlopsF32Sincos, prepareFlopsF32Log, } from "./benchmarks/flopsMath.js";
-import { prepareFlopsU32PackUnpack, prepareFlopsI32F32Convert, prepareFlopsF32F16Convert, prepareFlopsI32F16Convert, } from "./benchmarks/flopsConvert.js";
-import { errorResult, metricPerSecond, rowFromMeta, FLOPS_METRIC, BYTES_METRIC, } from "./benchmarks/common.js";
+import { prepareFlopsU32PackUnpack, prepareFlopsI32F32Convert, prepareFlopsF32F16Convert, } from "./benchmarks/flopsConvert.js";
+import { errorResult, metricPerSecond, rowFromMeta, } from "./benchmarks/common.js";
 export { computeStats } from "./stats.js";
 const DEFAULT_ROWS = 4096;
 const DEFAULT_COLS = 4096;
@@ -71,7 +71,6 @@ export async function* runSuite(options = {}) {
         compute('flops-u32-packunpack', () => prepareFlopsU32PackUnpack(ctx, flopsHarness)),
         compute('flops-i32-f32-convert', () => prepareFlopsI32F32Convert(ctx, flopsHarness)),
         compute('flops-f32-f16-convert', () => prepareFlopsF32F16Convert(ctx, flopsHarness)),
-        compute('flops-i32-f16-convert', () => prepareFlopsI32F16Convert(ctx, flopsHarness)),
     ];
     // Phase 1: build every benchmark's GPU resources up front. Rows that can't
     // run at all (missing feature, failed shader compile) resolve right here;
@@ -81,13 +80,9 @@ export async function* runSuite(options = {}) {
     for (const [id, category, prepare] of benchmarks) {
         const fallbackMeta = {
             id,
-            label: id,
-            description: '',
-            source: '',
             category,
             rows,
             cols,
-            metric: category === 'bandwidth' ? BYTES_METRIC : FLOPS_METRIC,
             amountPerOp: 0,
         };
         let prepared;

package/dist/suite.js.map

@@ -1 +1 @@
-{"version":3,"file":"suite.js","sourceRoot":"","sources":["../src/suite.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,iBAAiB,EAAmB,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAE,kBAAkB,EAAE,gBAAgB,EAAE,MAAM,oBAAoB,CAAC;AAC1E,OAAO,EAAE,aAAa,EAAE,MAAM,0BAA0B,CAAC;AACzD,OAAO,EAAE,WAAW,EAA6C,MAAM,eAAe,CAAC;AACvF,OAAO,EAAE,oBAAoB,EAAE,qBAAqB,EAAE,MAAM,iCAAiC,CAAC;AAC9F,OAAO,EACL,qBAAqB,EACrB,mBAAmB,EACnB,mBAAmB,EACnB,qBAAqB,GACtB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,qBAAqB,EACrB,mBAAmB,EACnB,mBAAmB,EACnB,qBAAqB,GACtB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,oBAAoB,EACpB,kBAAkB,EAClB,kBAAkB,EAClB,oBAAoB,EACpB,wBAAwB,EACxB,kBAAkB,GACnB,MAAM,yBAAyB,CAAC;AACjC,OAAO,EACL,kBAAkB,EAClB,kBAAkB,EAClB,mBAAmB,EACnB,oBAAoB,EACpB,kBAAkB,EAClB,qBAAqB,EACrB,kBAAkB,GACnB,MAAM,2BAA2B,CAAC;AACnC,OAAO,EACL,yBAAyB,EACzB,yBAAyB,EACzB,yBAAyB,EACzB,yBAAyB,GAC1B,MAAM,8BAA8B,CAAC;AACtC,OAAO,EACL,WAAW,EACX,eAAe,EACf,WAAW,EACX,YAAY,EACZ,YAAY,GAIb,MAAM,wBAAwB,CAAC;AAchC,OAAO,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAE1C,MAAM,YAAY,GAAG,IAAI,CAAC;AAC1B,MAAM,YAAY,GAAG,IAAI,CAAC;AAG1B,MAAM,SAAS,GAAG,CAAC,EAAU,EAAE,OAAyC,EAAkB,EAAE,CAAC,CAAC,EAAE,EAAE,WAAW,EAAE,OAAO,CAAC,CAAC;AACxH,MAAM,OAAO,GAAG,CAAC,EAAU,EAAE,OAAyC,EAAkB,EAAE,CAAC,CAAC,EAAE,EAAE,SAAS,EAAE,OAAO,CAAC,CAAC;AASpH;;;;;;;;;;;;;;;;;GAiBG;AACH,MAAM,CAAC,KAAK,SAAS,CAAC,CAAC,QAAQ,CAAC,UAAwB,EAAE;IACxD,MAAM,IAAI,GAAG,gBAAgB,CAAC,OAAO,CAAC,IAAI,IAAI,YAAY,CAAC,CAAC;IAC5D,MAAM,IAAI,GAAG,gBAAgB,CAAC,OAAO,CAAC,IAAI,IAAI,YAAY,CAAC,CAAC;IAE5D,MAAM,GAAG,GAAG,MAAM,iBAAiB,EAAE,CAAC;IACtC,OAAO,CAAC,YAAY,EAAE,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IAEjC,MAAM,IAAI,GAAG,kBAAkB,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;IAC5C,MAAM,OAAO,GAAkB;QAC7B,QAAQ,EAAE,OAAO,CAAC,QAAQ;QAC1B,gBAAgB,EAAE,OAAO,CAAC,gBAAgB;QAC1C,OAAO,EAAE,OAAO,CAAC,OAAO;KACzB,CAAC;IACF,MAAM,YAAY,GAAG,EAAE,GAAG,OAAO,EAAE,OAAO,EAAE,OAAO,CAAC,cAAc,EAAE,UAAU,EAAE,OAAO,CAAC,iBAAiB,EAAE,CAAC;IAE5G,MAAM,UAAU,GAA8B;QAC5C,SAAS,CAAC,gBAAgB,EAAE,GAAG,EAAE,CAAC,oBAAoB,CAAC,GAAG,EAAE,IAAI,EAAE,OAAO,CAAC,CAAC;QAC3E,SAAS,CAAC,iBAAiB,EAAE,GAAG,EAAE,CAAC,qBAAqB,CAAC,GAAG,EAAE,IAAI,EAAE,OAAO,CAAC,CAAC;QAC7E,OAAO,CAAC,kBAAkB,EAAE,GAAG,EAAE,CAAC,qBAAqB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QAC3E,OAAO,CAAC,gBAAgB,EAAE,GAAG,EAAE,CAAC,mBAAmB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACvE,OAAO,CAAC,gBAAgB,EAAE,GAAG,EAAE,CAAC,mBAAmB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACvE,OAAO,CAAC,kBAAkB,EAAE,GAAG,EAAE,CAAC,qBAAqB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QAC3E,OAAO,CAAC,kBAAkB,EAAE,GAAG,EAAE,CAAC,qBAAqB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QAC3E,OAAO,CAAC,gBAAgB,EAAE,GAAG,EAAE,CAAC,mBAAmB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACvE,OAAO,CAAC,gBAAgB,EAAE,GAAG,EAAE,CAAC,mBAAmB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACvE,OAAO,CAAC,kBAAkB,EAAE,GAAG,EAAE,CAAC,qBAAqB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QAC3E,OAAO,CAAC,iBAAiB,EAAE,GAAG,EAAE,CAAC,oBAAoB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACzE,OAAO,CAAC,eAAe,EAAE,GAAG,EAAE,CAAC,kBAAkB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACrE,OAAO,CAAC,eAAe,EAAE,GAAG,EAAE,CAAC,kBAAkB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACrE,OAAO,CAAC,iBAAiB,EAAE,GAAG,EAAE,CAAC,oBAAoB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACzE,OAAO,CAAC,sBAAsB,EAAE,GAAG,EAAE,CAAC,wBAAwB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QAClF,OAAO,CAAC,eAAe,EAAE,GAAG,EAAE,CAAC,kBAAkB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACrE,OAAO,CAAC,eAAe,EAAE,GAAG,EAAE,CAAC,kBAAkB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACrE,OAAO,CAAC,eAAe,EAAE,GAAG,EAAE,CAAC,kBAAkB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACrE,OAAO,CAAC,gBAAgB,EAAE,GAAG,EAAE,CAAC,mBAAmB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACvE,OAAO,CAAC,iBAAiB,EAAE,GAAG,EAAE,CAAC,oBAAoB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACzE,OAAO,CAAC,eAAe,EAAE,GAAG,EAAE,CAAC,kBAAkB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACrE,OAAO,CAAC,kBAAkB,EAAE,GAAG,EAAE,CAAC,qBAAqB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QAC3E,OAAO,CAAC,eAAe,EAAE,GAAG,EAAE,CAAC,kBAAkB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACrE,OAAO,CAAC,sBAAsB,EAAE,GAAG,EAAE,CAAC,yBAAyB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACnF,OAAO,CAAC,uBAAuB,EAAE,GAAG,EAAE,CAAC,yBAAyB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACpF,OAAO,CAAC,uBAAuB,EAAE,GAAG,EAAE,CAAC,yBAAyB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACpF,OAAO,CAAC,uBAAuB,EAAE,GAAG,EAAE,CAAC,yBAAyB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;KACrF,CAAC;IAEF,2EAA2E;IAC3E,0EAA0E;IAC1E,wEAAwE;IACxE,+BAA+B;IAC/B,MAAM,SAAS,GAAsB,EAAE,CAAC;IACxC,KAAK,MAAM,CAAC,EAAE,EAAE,QAAQ,EAAE,OAAO,CAAC,IAAI,UAAU,EAAE,CAAC;QACjD,MAAM,YAAY,GAAkB;YAClC,EAAE;YACF,KAAK,EAAE,EAAE;YACT,WAAW,EAAE,EAAE;YACf,MAAM,EAAE,EAAE;YACV,QAAQ;YACR,IAAI;YACJ,IAAI;YACJ,MAAM,EAAE,QAAQ,KAAK,WAAW,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,YAAY;YAC9D,WAAW,EAAE,CAAC;SACf,CAAC;QACF,IAAI,QAA2B,CAAC;QAChC,IAAI,CAAC;YACH,QAAQ,GAAG,MAAM,OAAO,EAAE,CAAC;QAC7B,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,WAAW,CAAC,YAAY,EAAE,KAAK,CAAC,CAAC;YACvC,SAAS;QACX,CAAC;QACD,IAAI,QAAQ,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;YAChC,MAAM,QAAQ,CAAC,MAAM,CAAC;YACtB,SAAS;QACX,CAAC;QACD,SAAS,CAAC,IAAI,CAAC;YACb,IAAI,EAAE,QAAQ,CAAC,IAAI;YACnB,OAAO,EAAE,IAAI,aAAa,CAAC,QAAQ,CAAC,OAAO,CAAC;YAC5C,UAAU,EAAE,QAAQ,CAAC,UAAU;SAChC,CAAC,CAAC;QACH,MAAM,WAAW,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;IACnC,CAAC;IAED,uEAAuE;IACvE,mEAAmE;IACnE,MAAM,IAAI,GAAG,IAAI,GAAG,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IAC3D,MAAM,KAAK,GAAsB,EAAE,CAAC;IACpC,IAAI,IAAI,GAAwB,IAAI,CAAC;IACrC,MAAM,IAAI,GAAG,CAAC,CAAkB,EAAE,EAAE;QAClC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACd,IAAI,EAAE,EAAE,CAAC;QACT,IAAI,GAAG,IAAI,CAAC;IACd,CAAC,CAAC;IAEF,MAAM,WAAW,GAAiB,SAAS,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,EAAE,EAAE,CAAC,CAAC;QACtE,EAAE,EAAE,IAAI,CAAC,EAAE;QACX,SAAS,EAAE,GAAG,EAAE,CAAC,OAAO,CAAC,SAAS,EAAE;QACpC,MAAM,EAAE,GAAG,EAAE,CAAC,OAAO,CAAC,MAAM,EAAE;KAC/B,CAAC,CAAC,CAAC;IAEJ,MAAM,GAAG,GAAG,WAAW,CAAC,WAAW,EAAE;QACnC,GAAG,mBAAmB,CAAC,OAAO,CAAC;QAC/B,UAAU,EAAE,OAAO,CAAC,UAAU;QAC9B,QAAQ,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,eAAe,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAE,EAAE,KAAK,CAAC,CAAC;KACvE,CAAC,CAAC,OAAO,CAAC,GAAG,EAAE;QACd,KAAK,MAAM,CAAC,IAAI,SAAS;YAAE,CAAC,CAAC,OAAO,CAAC,OAAO,EAAE,CAAC;IACjD,CAAC,CAAC,CAAC;IACH,IAAI,IAAI,GAAG,KAAK,CAAC;IACjB,KAAK,GAAG,CAAC,IAAI,CACX,GAAG,EAAE;QACH,IAAI,GAAG,IAAI,CAAC;QACZ,IAAI,EAAE,EAAE,CAAC;IACX,CAAC,EACD,GAAG,EAAE;QACH,IAAI,GAAG,IAAI,CAAC;QACZ,IAAI,EAAE,EAAE,CAAC;IACX,CAAC,CACF,CAAC;IAEF,IAAI,CAAC;QACH,OAAO,IAAI,EAAE,CAAC;YACZ,OAAO,KAAK,CAAC,MAAM,GAAG,CAAC;gBAAE,MAAM,KAAK,CAAC,KAAK,EAAG,CAAC;YAC9C,IAAI,IAAI;gBAAE,MAAM;YAChB,MAAM,IAAI,OAAO,CAAO,CAAC,OAAO,EAAE,EAAE;gBAClC,IAAI,GAAG,OAAO,CAAC;YACjB,CAAC,CAAC,CAAC;QACL,CAAC;QACD,MAAM,GAAG,CAAC;IACZ,CAAC;YAAS,CAAC;QACT,GAAG,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;IACvB,CAAC;AACH,CAAC;AAED,SAAS,mBAAmB,CAAC,CAAe;IAC1C,OAAO;QACL,SAAS,EAAE,CAAC,CAAC,SAAS;QACtB,SAAS,EAAE,CAAC,CAAC,SAAS;QACtB,YAAY,EAAE,CAAC,CAAC,YAAY;QAC5B,oBAAoB,EAAE,CAAC,CAAC,oBAAoB;QAC5C,iBAAiB,EAAE,CAAC,CAAC,iBAAiB;QACtC,MAAM,EAAE,CAAC,CAAC,MAAM;QAChB,UAAU,EAAE,CAAC,CAAC,UAAU;QACxB,YAAY,EAAE,CAAC,CAAC,YAAY;QAC5B,iBAAiB,EAAE,CAAC,CAAC,iBAAiB;KACvC,CAAC;AACJ,CAAC;AAED,SAAS,eAAe,CAAC,MAAuB,EAAE,KAA0B;IAC1E,MAAM,EAAE,OAAO,EAAE,GAAG,MAAM,CAAC;IAC3B,8FAA8F;IAC9F,MAAM,IAAI,GAAG,MAAM,CAAC,UAAU,IAAI,OAAO,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,UAAU,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC;IAC7G,IAAI,KAAK,CAAC,KAAK,KAAK,SAAS;QAAE,OAAO,WAAW,CAAC,IAAI,EAAE,KAAK,CAAC,KAAK,CAAC,CAAC;IAErE,MAAM,GAAG,GAAG,WAAW,CAAC,IAAI,CAAC,CAAC;IAC9B,GAAG,CAAC,MAAM,GAAG,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,CAAC;IACjD,GAAG,CAAC,eAAe,GAAG,OAAO,CAAC,eAAe,CAAC;IAC9C,GAAG,CAAC,YAAY,GAAG,OAAO,CAAC,YAAY,CAAC;IACxC,GAAG,CAAC,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC;IAC5B,GAAG,CAAC,WAAW,GAAG,KAAK,CAAC,WAAW,CAAC;IACpC,GAAG,CAAC,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC;IACxB,GAAG,CAAC,UAAU,GAAG,KAAK,CAAC,UAAU,CAAC;IAClC,IAAI,KAAK,CAAC,KAAK,EAAE,CAAC;QAChB,wEAAwE;QACxE,uEAAuE;QACvE,GAAG,CAAC,WAAW,GAAG,eAAe,CAAC,IAAI,CAAC,WAAW,EAAE,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IACvE,CAAC;IACD,IAAI,KAAK,CAAC,UAAU,KAAK,WAAW,EAAE,CAAC;QACrC,GAAG,CAAC,OAAO,GAAG,uFAAuF,CAAC;IACxG,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC"}
+{"version":3,"file":"suite.js","sourceRoot":"","sources":["../src/suite.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,iBAAiB,EAAmB,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAE,kBAAkB,EAAE,gBAAgB,EAAE,MAAM,oBAAoB,CAAC;AAC1E,OAAO,EAAE,aAAa,EAAE,MAAM,0BAA0B,CAAC;AACzD,OAAO,EAAE,WAAW,EAA6C,MAAM,eAAe,CAAC;AACvF,OAAO,EAAE,oBAAoB,EAAE,qBAAqB,EAAE,MAAM,iCAAiC,CAAC;AAC9F,OAAO,EACL,qBAAqB,EACrB,mBAAmB,EACnB,mBAAmB,EACnB,qBAAqB,GACtB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,qBAAqB,EACrB,mBAAmB,EACnB,mBAAmB,EACnB,qBAAqB,GACtB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,oBAAoB,EACpB,kBAAkB,EAClB,kBAAkB,EAClB,oBAAoB,EACpB,wBAAwB,EACxB,kBAAkB,GACnB,MAAM,yBAAyB,CAAC;AACjC,OAAO,EACL,kBAAkB,EAClB,kBAAkB,EAClB,mBAAmB,EACnB,oBAAoB,EACpB,kBAAkB,EAClB,qBAAqB,EACrB,kBAAkB,GACnB,MAAM,2BAA2B,CAAC;AACnC,OAAO,EACL,yBAAyB,EACzB,yBAAyB,EACzB,yBAAyB,GAC1B,MAAM,8BAA8B,CAAC;AACtC,OAAO,EACL,WAAW,EACX,eAAe,EACf,WAAW,GAIZ,MAAM,wBAAwB,CAAC;AAchC,OAAO,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAE1C,MAAM,YAAY,GAAG,IAAI,CAAC;AAC1B,MAAM,YAAY,GAAG,IAAI,CAAC;AAG1B,MAAM,SAAS,GAAG,CAAC,EAAU,EAAE,OAAyC,EAAkB,EAAE,CAAC,CAAC,EAAE,EAAE,WAAW,EAAE,OAAO,CAAC,CAAC;AACxH,MAAM,OAAO,GAAG,CAAC,EAAU,EAAE,OAAyC,EAAkB,EAAE,CAAC,CAAC,EAAE,EAAE,SAAS,EAAE,OAAO,CAAC,CAAC;AASpH;;;;;;;;;;;;;;;;;GAiBG;AACH,MAAM,CAAC,KAAK,SAAS,CAAC,CAAC,QAAQ,CAAC,UAAwB,EAAE;IACxD,MAAM,IAAI,GAAG,gBAAgB,CAAC,OAAO,CAAC,IAAI,IAAI,YAAY,CAAC,CAAC;IAC5D,MAAM,IAAI,GAAG,gBAAgB,CAAC,OAAO,CAAC,IAAI,IAAI,YAAY,CAAC,CAAC;IAE5D,MAAM,GAAG,GAAG,MAAM,iBAAiB,EAAE,CAAC;IACtC,OAAO,CAAC,YAAY,EAAE,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IAEjC,MAAM,IAAI,GAAG,kBAAkB,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;IAC5C,MAAM,OAAO,GAAkB;QAC7B,QAAQ,EAAE,OAAO,CAAC,QAAQ;QAC1B,gBAAgB,EAAE,OAAO,CAAC,gBAAgB;QAC1C,OAAO,EAAE,OAAO,CAAC,OAAO;KACzB,CAAC;IACF,MAAM,YAAY,GAAG,EAAE,GAAG,OAAO,EAAE,OAAO,EAAE,OAAO,CAAC,cAAc,EAAE,UAAU,EAAE,OAAO,CAAC,iBAAiB,EAAE,CAAC;IAE5G,MAAM,UAAU,GAA8B;QAC5C,SAAS,CAAC,gBAAgB,EAAE,GAAG,EAAE,CAAC,oBAAoB,CAAC,GAAG,EAAE,IAAI,EAAE,OAAO,CAAC,CAAC;QAC3E,SAAS,CAAC,iBAAiB,EAAE,GAAG,EAAE,CAAC,qBAAqB,CAAC,GAAG,EAAE,IAAI,EAAE,OAAO,CAAC,CAAC;QAC7E,OAAO,CAAC,kBAAkB,EAAE,GAAG,EAAE,CAAC,qBAAqB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QAC3E,OAAO,CAAC,gBAAgB,EAAE,GAAG,EAAE,CAAC,mBAAmB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACvE,OAAO,CAAC,gBAAgB,EAAE,GAAG,EAAE,CAAC,mBAAmB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACvE,OAAO,CAAC,kBAAkB,EAAE,GAAG,EAAE,CAAC,qBAAqB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QAC3E,OAAO,CAAC,kBAAkB,EAAE,GAAG,EAAE,CAAC,qBAAqB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QAC3E,OAAO,CAAC,gBAAgB,EAAE,GAAG,EAAE,CAAC,mBAAmB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACvE,OAAO,CAAC,gBAAgB,EAAE,GAAG,EAAE,CAAC,mBAAmB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACvE,OAAO,CAAC,kBAAkB,EAAE,GAAG,EAAE,CAAC,qBAAqB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QAC3E,OAAO,CAAC,iBAAiB,EAAE,GAAG,EAAE,CAAC,oBAAoB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACzE,OAAO,CAAC,eAAe,EAAE,GAAG,EAAE,CAAC,kBAAkB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACrE,OAAO,CAAC,eAAe,EAAE,GAAG,EAAE,CAAC,kBAAkB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACrE,OAAO,CAAC,iBAAiB,EAAE,GAAG,EAAE,CAAC,oBAAoB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACzE,OAAO,CAAC,sBAAsB,EAAE,GAAG,EAAE,CAAC,wBAAwB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QAClF,OAAO,CAAC,eAAe,EAAE,GAAG,EAAE,CAAC,kBAAkB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACrE,OAAO,CAAC,eAAe,EAAE,GAAG,EAAE,CAAC,kBAAkB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACrE,OAAO,CAAC,eAAe,EAAE,GAAG,EAAE,CAAC,kBAAkB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACrE,OAAO,CAAC,gBAAgB,EAAE,GAAG,EAAE,CAAC,mBAAmB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACvE,OAAO,CAAC,iBAAiB,EAAE,GAAG,EAAE,CAAC,oBAAoB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACzE,OAAO,CAAC,eAAe,EAAE,GAAG,EAAE,CAAC,kBAAkB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACrE,OAAO,CAAC,kBAAkB,EAAE,GAAG,EAAE,CAAC,qBAAqB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QAC3E,OAAO,CAAC,eAAe,EAAE,GAAG,EAAE,CAAC,kBAAkB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACrE,OAAO,CAAC,sBAAsB,EAAE,GAAG,EAAE,CAAC,yBAAyB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACnF,OAAO,CAAC,uBAAuB,EAAE,GAAG,EAAE,CAAC,yBAAyB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACpF,OAAO,CAAC,uBAAuB,EAAE,GAAG,EAAE,CAAC,yBAAyB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;KACrF,CAAC;IAEF,2EAA2E;IAC3E,0EAA0E;IAC1E,wEAAwE;IACxE,+BAA+B;IAC/B,MAAM,SAAS,GAAsB,EAAE,CAAC;IACxC,KAAK,MAAM,CAAC,EAAE,EAAE,QAAQ,EAAE,OAAO,CAAC,IAAI,UAAU,EAAE,CAAC;QACjD,MAAM,YAAY,GAAkB;YAClC,EAAE;YACF,QAAQ;YACR,IAAI;YACJ,IAAI;YACJ,WAAW,EAAE,CAAC;SACf,CAAC;QACF,IAAI,QAA2B,CAAC;QAChC,IAAI,CAAC;YACH,QAAQ,GAAG,MAAM,OAAO,EAAE,CAAC;QAC7B,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,WAAW,CAAC,YAAY,EAAE,KAAK,CAAC,CAAC;YACvC,SAAS;QACX,CAAC;QACD,IAAI,QAAQ,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;YAChC,MAAM,QAAQ,CAAC,MAAM,CAAC;YACtB,SAAS;QACX,CAAC;QACD,SAAS,CAAC,IAAI,CAAC;YACb,IAAI,EAAE,QAAQ,CAAC,IAAI;YACnB,OAAO,EAAE,IAAI,aAAa,CAAC,QAAQ,CAAC,OAAO,CAAC;YAC5C,UAAU,EAAE,QAAQ,CAAC,UAAU;SAChC,CAAC,CAAC;QACH,MAAM,WAAW,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;IACnC,CAAC;IAED,uEAAuE;IACvE,mEAAmE;IACnE,MAAM,IAAI,GAAG,IAAI,GAAG,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IAC3D,MAAM,KAAK,GAAsB,EAAE,CAAC;IACpC,IAAI,IAAI,GAAwB,IAAI,CAAC;IACrC,MAAM,IAAI,GAAG,CAAC,CAAkB,EAAE,EAAE;QAClC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACd,IAAI,EAAE,EAAE,CAAC;QACT,IAAI,GAAG,IAAI,CAAC;IACd,CAAC,CAAC;IAEF,MAAM,WAAW,GAAiB,SAAS,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,EAAE,EAAE,CAAC,CAAC;QACtE,EAAE,EAAE,IAAI,CAAC,EAAE;QACX,SAAS,EAAE,GAAG,EAAE,CAAC,OAAO,CAAC,SAAS,EAAE;QACpC,MAAM,EAAE,GAAG,EAAE,CAAC,OAAO,CAAC,MAAM,EAAE;KAC/B,CAAC,CAAC,CAAC;IAEJ,MAAM,GAAG,GAAG,WAAW,CAAC,WAAW,EAAE;QACnC,GAAG,mBAAmB,CAAC,OAAO,CAAC;QAC/B,UAAU,EAAE,OAAO,CAAC,UAAU;QAC9B,QAAQ,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,eAAe,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAE,EAAE,KAAK,CAAC,CAAC;KACvE,CAAC,CAAC,OAAO,CAAC,GAAG,EAAE;QACd,KAAK,MAAM,CAAC,IAAI,SAAS;YAAE,CAAC,CAAC,OAAO,CAAC,OAAO,EAAE,CAAC;IACjD,CAAC,CAAC,CAAC;IACH,IAAI,IAAI,GAAG,KAAK,CAAC;IACjB,KAAK,GAAG,CAAC,IAAI,CACX,GAAG,EAAE;QACH,IAAI,GAAG,IAAI,CAAC;QACZ,IAAI,EAAE,EAAE,CAAC;IACX,CAAC,EACD,GAAG,EAAE;QACH,IAAI,GAAG,IAAI,CAAC;QACZ,IAAI,EAAE,EAAE,CAAC;IACX,CAAC,CACF,CAAC;IAEF,IAAI,CAAC;QACH,OAAO,IAAI,EAAE,CAAC;YACZ,OAAO,KAAK,CAAC,MAAM,GAAG,CAAC;gBAAE,MAAM,KAAK,CAAC,KAAK,EAAG,CAAC;YAC9C,IAAI,IAAI;gBAAE,MAAM;YAChB,MAAM,IAAI,OAAO,CAAO,CAAC,OAAO,EAAE,EAAE;gBAClC,IAAI,GAAG,OAAO,CAAC;YACjB,CAAC,CAAC,CAAC;QACL,CAAC;QACD,MAAM,GAAG,CAAC;IACZ,CAAC;YAAS,CAAC;QACT,GAAG,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;IACvB,CAAC;AACH,CAAC;AAED,SAAS,mBAAmB,CAAC,CAAe;IAC1C,OAAO;QACL,SAAS,EAAE,CAAC,CAAC,SAAS;QACtB,SAAS,EAAE,CAAC,CAAC,SAAS;QACtB,YAAY,EAAE,CAAC,CAAC,YAAY;QAC5B,oBAAoB,EAAE,CAAC,CAAC,oBAAoB;QAC5C,iBAAiB,EAAE,CAAC,CAAC,iBAAiB;QACtC,MAAM,EAAE,CAAC,CAAC,MAAM;QAChB,UAAU,EAAE,CAAC,CAAC,UAAU;QACxB,YAAY,EAAE,CAAC,CAAC,YAAY;QAC5B,iBAAiB,EAAE,CAAC,CAAC,iBAAiB;KACvC,CAAC;AACJ,CAAC;AAED,SAAS,eAAe,CAAC,MAAuB,EAAE,KAA0B;IAC1E,MAAM,EAAE,OAAO,EAAE,GAAG,MAAM,CAAC;IAC3B,8FAA8F;IAC9F,MAAM,IAAI,GAAG,MAAM,CAAC,UAAU,IAAI,OAAO,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,UAAU,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC;IAC7G,IAAI,KAAK,CAAC,KAAK,KAAK,SAAS;QAAE,OAAO,WAAW,CAAC,IAAI,EAAE,KAAK,CAAC,KAAK,CAAC,CAAC;IAErE,MAAM,GAAG,GAAG,WAAW,CAAC,IAAI,CAAC,CAAC;IAC9B,GAAG,CAAC,MAAM,GAAG,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,CAAC;IACjD,GAAG,CAAC,eAAe,GAAG,OAAO,CAAC,eAAe,CAAC;IAC9C,GAAG,CAAC,YAAY,GAAG,OAAO,CAAC,YAAY,CAAC;IACxC,GAAG,CAAC,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC;IAC5B,GAAG,CAAC,WAAW,GAAG,KAAK,CAAC,WAAW,CAAC;IACpC,GAAG,CAAC,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC;IACxB,GAAG,CAAC,UAAU,GAAG,KAAK,CAAC,UAAU,CAAC;IAClC,IAAI,KAAK,CAAC,KAAK,EAAE,CAAC;QAChB,wEAAwE;QACxE,uEAAuE;QACvE,GAAG,CAAC,WAAW,GAAG,eAAe,CAAC,IAAI,CAAC,WAAW,EAAE,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IACvE,CAAC;IACD,IAAI,KAAK,CAAC,UAAU,KAAK,WAAW,EAAE,CAAC;QACrC,GAAG,CAAC,OAAO,GAAG,uFAAuF,CAAC;IACxG,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC"}

package/dist/types.d.ts

@@ -40,13 +40,14 @@ export interface MetricDef {
  * after every sample it takes (still `running`, with the partial stats so
  * far), and when it finishes (ok/skipped/error) — so the UI can render
  * progress incrementally.
+ *
+ * Pure measurement, no identity/display metadata: label, description, WGSL
+ * source, and the metric definition all live on the matching `id` in
+ * `BENCHMARK_CATALOG` instead, so a UI joins a result onto its catalog row
+ * rather than getting them handed twice.
  */
 export interface BenchmarkResult {
     id: string;
-    label: string;
-    description: string;
-    /** WGSL source of the kernel that was run, for display alongside the result. */
-    source: string;
     category: BenchmarkCategory;
     status: BenchmarkStatus;
     message?: string;
@@ -71,8 +72,6 @@ export interface BenchmarkResult {
     stats?: Stats;
     /** Set for `ok` rows: see `SamplingStopReason`. */
     stopReason?: SamplingStopReason;
-    /** What `metricValue` counts. Set from the start, before any measurement lands. */
-    metric: MetricDef;
     /** Peak `<metric.unit>/s`, computed from the best run (`stats.min`). */
     metricValue?: number;
     timingMethod: TimingMethod;

package/dist/types.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA,oFAAoF;AACpF,MAAM,WAAW,KAAK;IACpB,IAAI,EAAE,MAAM,CAAC;IACb,oGAAoG;IACpG,GAAG,EAAE,MAAM,CAAC;IACZ,GAAG,EAAE,MAAM,CAAC;IACZ,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,wGAAwG;IACxG,IAAI,EAAE,MAAM,CAAC;CACd;AAED;;;;;;;GAOG;AACH,MAAM,MAAM,kBAAkB,GAAG,WAAW,GAAG,YAAY,GAAG,WAAW,CAAC;AAE1E,MAAM,MAAM,YAAY,GAAG,eAAe,GAAG,eAAe,CAAC;AAE7D,iGAAiG;AACjG,MAAM,MAAM,iBAAiB,GAAG,WAAW,GAAG,SAAS,CAAC;AAExD,MAAM,MAAM,eAAe,GAAG,SAAS,GAAG,IAAI,GAAG,SAAS,GAAG,OAAO,CAAC;AAErE;;;;;;GAMG;AACH,MAAM,WAAW,SAAS;IACxB,GAAG,EAAE,MAAM,CAAC;IACZ,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;CACd;AAED;;;;;;GAMG;AACH,MAAM,WAAW,eAAe;IAC9B,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,EAAE,MAAM,CAAC;IACd,WAAW,EAAE,MAAM,CAAC;IACpB,gFAAgF;IAChF,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,iBAAiB,CAAC;IAC5B,MAAM,EAAE,eAAe,CAAC;IACxB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB;;;;;OAKG;IACH,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,oEAAoE;IACpE,eAAe,EAAE,MAAM,CAAC;IACxB,iHAAiH;IACjH,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB;;;OAGG;IACH,WAAW,EAAE,MAAM,EAAE,CAAC;IACtB,2DAA2D;IAC3D,KAAK,CAAC,EAAE,KAAK,CAAC;IACd,mDAAmD;IACnD,UAAU,CAAC,EAAE,kBAAkB,CAAC;IAChC,mFAAmF;IACnF,MAAM,EAAE,SAAS,CAAC;IAClB,wEAAwE;IACxE,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,YAAY,EAAE,YAAY,CAAC;CAC5B;AAED,kFAAkF;AAClF,MAAM,MAAM,kBAAkB,GAC1B;IAAE,IAAI,EAAE,OAAO,CAAC;IAAC,KAAK,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,CAAA;CAAE,GAChD;IAAE,IAAI,EAAE,UAAU,CAAC;IAAC,OAAO,EAAE,MAAM,CAAC;IAAC,WAAW,EAAE,MAAM,CAAC;IAAC,EAAE,EAAE,MAAM,CAAC;IAAC,YAAY,EAAE,MAAM,EAAE,CAAA;CAAE,GAC9F;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,EAAE,CAAA;CAAE,CAAC;AAEvD,MAAM,WAAW,YAAY;IAC3B,uGAAuG;IACvG,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,uGAAuG;IACvG,IAAI,CAAC,EAAE,MAAM,CAAC;IACd;;;;OAIG;IACH,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB;;;;;;OAMG;IACH,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,6IAA6I;IAC7I,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,6FAA6F;IAC7F,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,sGAAsG;IACtG,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;;OAIG;IACH,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,mHAAmH;IACnH,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAC9B;;;;;OAKG;IACH,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,yHAAyH;IACzH,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,uFAAuF;IACvF,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,qIAAqI;IACrI,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB;;;;OAIG;IACH,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,2GAA2G;IAC3G,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB;;;;;OAKG;IACH,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,qFAAqF;IACrF,YAAY,CAAC,EAAE,CAAC,IAAI,EAAE,UAAU,KAAK,IAAI,CAAC;IAC1C,mFAAmF;IACnF,UAAU,CAAC,EAAE,CAAC,KAAK,EAAE,kBAAkB,KAAK,IAAI,CAAC;CAClD;AAED,MAAM,WAAW,UAAU;IACzB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC/B,WAAW,EAAE,OAAO,CAAC;IACrB,aAAa,EAAE,OAAO,CAAC;IACvB,sBAAsB,EAAE,OAAO,CAAC;CACjC"}
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA,oFAAoF;AACpF,MAAM,WAAW,KAAK;IACpB,IAAI,EAAE,MAAM,CAAC;IACb,oGAAoG;IACpG,GAAG,EAAE,MAAM,CAAC;IACZ,GAAG,EAAE,MAAM,CAAC;IACZ,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,wGAAwG;IACxG,IAAI,EAAE,MAAM,CAAC;CACd;AAED;;;;;;;GAOG;AACH,MAAM,MAAM,kBAAkB,GAAG,WAAW,GAAG,YAAY,GAAG,WAAW,CAAC;AAE1E,MAAM,MAAM,YAAY,GAAG,eAAe,GAAG,eAAe,CAAC;AAE7D,iGAAiG;AACjG,MAAM,MAAM,iBAAiB,GAAG,WAAW,GAAG,SAAS,CAAC;AAExD,MAAM,MAAM,eAAe,GAAG,SAAS,GAAG,IAAI,GAAG,SAAS,GAAG,OAAO,CAAC;AAErE;;;;;;GAMG;AACH,MAAM,WAAW,SAAS;IACxB,GAAG,EAAE,MAAM,CAAC;IACZ,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;CACd;AAED;;;;;;;;;;;GAWG;AACH,MAAM,WAAW,eAAe;IAC9B,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,iBAAiB,CAAC;IAC5B,MAAM,EAAE,eAAe,CAAC;IACxB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB;;;;;OAKG;IACH,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,oEAAoE;IACpE,eAAe,EAAE,MAAM,CAAC;IACxB,iHAAiH;IACjH,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB;;;OAGG;IACH,WAAW,EAAE,MAAM,EAAE,CAAC;IACtB,2DAA2D;IAC3D,KAAK,CAAC,EAAE,KAAK,CAAC;IACd,mDAAmD;IACnD,UAAU,CAAC,EAAE,kBAAkB,CAAC;IAChC,wEAAwE;IACxE,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,YAAY,EAAE,YAAY,CAAC;CAC5B;AAED,kFAAkF;AAClF,MAAM,MAAM,kBAAkB,GAC1B;IAAE,IAAI,EAAE,OAAO,CAAC;IAAC,KAAK,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,CAAA;CAAE,GAChD;IAAE,IAAI,EAAE,UAAU,CAAC;IAAC,OAAO,EAAE,MAAM,CAAC;IAAC,WAAW,EAAE,MAAM,CAAC;IAAC,EAAE,EAAE,MAAM,CAAC;IAAC,YAAY,EAAE,MAAM,EAAE,CAAA;CAAE,GAC9F;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,EAAE,CAAA;CAAE,CAAC;AAEvD,MAAM,WAAW,YAAY;IAC3B,uGAAuG;IACvG,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,uGAAuG;IACvG,IAAI,CAAC,EAAE,MAAM,CAAC;IACd;;;;OAIG;IACH,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB;;;;;;OAMG;IACH,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,6IAA6I;IAC7I,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,6FAA6F;IAC7F,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,sGAAsG;IACtG,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;;OAIG;IACH,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,mHAAmH;IACnH,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAC9B;;;;;OAKG;IACH,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,yHAAyH;IACzH,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,uFAAuF;IACvF,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,qIAAqI;IACrI,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB;;;;OAIG;IACH,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,2GAA2G;IAC3G,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB;;;;;OAKG;IACH,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,qFAAqF;IACrF,YAAY,CAAC,EAAE,CAAC,IAAI,EAAE,UAAU,KAAK,IAAI,CAAC;IAC1C,mFAAmF;IACnF,UAAU,CAAC,EAAE,CAAC,KAAK,EAAE,kBAAkB,KAAK,IAAI,CAAC;CAClD;AAED,MAAM,WAAW,UAAU;IACzB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC/B,WAAW,EAAE,OAAO,CAAC;IACrB,aAAa,EAAE,OAAO,CAAC;IACvB,sBAAsB,EAAE,OAAO,CAAC;CACjC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "webgpu-bench",
-  "version": "0.1.1",
+  "version": "0.3.0",
   "description": "Benchmark suite for WebGPU's raw ceilings: memory bandwidth and fp32/fp16/int8 FLOPS.",
   "keywords": [
     "webgpu",