wgblas 0.1.2 → 1.0.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +3 -0
- package/dist/wgblas.browser.js +1078 -37
- package/index.d.mts +6 -0
- package/index.mjs +6 -0
- package/package.json +32 -1
- package/src/classes/GpuMatrix.d.mts +85 -0
- package/src/classes/GpuMatrix.mjs +91 -0
- package/src/classes/GpuVector.d.mts +11 -7
- package/src/classes/GpuVector.mjs +31 -5
- package/src/dasum/dasum.d.mts +48 -0
- package/src/dasum/dasum.mjs +121 -0
- package/src/init.mjs +3 -1
- package/src/isamax/isamax.mjs +66 -67
- package/src/random/random.d.mts +37 -0
- package/src/random/random.mjs +17 -0
- package/src/sasum/sasum.mjs +55 -51
- package/src/saxpy/saxpy.mjs +48 -35
- package/src/scopy/scopy.mjs +43 -32
- package/src/sdot/sdot.mjs +60 -55
- package/src/sgemv/sgemv.d.mts +118 -0
- package/src/sgemv/sgemv.mjs +141 -0
- package/src/shaders/browser-shaders.mjs +20 -0
- package/src/shaders/dasum.wgsl +98 -0
- package/src/shaders/f64add.wgsl +281 -0
- package/src/shaders/isamax.wgsl +32 -9
- package/src/shaders/reduction/sumF64.wgsl +49 -0
- package/src/shaders/sasum.wgsl +18 -4
- package/src/shaders/sdot.wgsl +18 -4
- package/src/shaders/sgemv_n.wgsl +75 -0
- package/src/shaders/sgemv_t.wgsl +65 -0
- package/src/shaders/snrm2.wgsl +22 -4
- package/src/shaders/ssymv.wgsl +69 -0
- package/src/shaders/strmv.wgsl +103 -0
- package/src/shaders/strsv_apply_inverse.wgsl +47 -0
- package/src/shaders/strsv_invert_block.wgsl +109 -0
- package/src/shaders/strsv_update.wgsl +75 -0
- package/src/snrm2/snrm2.mjs +56 -52
- package/src/srot/srot.mjs +57 -41
- package/src/srotm/srotm.mjs +54 -38
- package/src/sscal/sscal.mjs +43 -32
- package/src/sswap/sswap.mjs +49 -34
- package/src/ssymv/ssymv.d.mts +109 -0
- package/src/ssymv/ssymv.mjs +130 -0
- package/src/strmv/strmv.d.mts +109 -0
- package/src/strmv/strmv.mjs +132 -0
- package/src/strsv/strsv.d.mts +98 -0
- package/src/strsv/strsv.mjs +212 -0
- package/src/util/benchmark.mjs +1 -1
- package/src/util/bindgroup.mjs +14 -10
- package/src/util/buffer.mjs +7 -2
- package/src/util/compute.mjs +41 -15
- package/src/util/f64pack.mjs +152 -0
- package/src/util/pipeline.mjs +32 -17
- package/src/util/result.mjs +8 -4
- package/src/util/workgroup.mjs +10 -10
package/src/util/workgroup.mjs
CHANGED
|
@@ -9,23 +9,23 @@ const WORKGROUP_SIZE_2D = 8;
|
|
|
9
9
|
* Calculates the number of workgroups to dispatch, clamped to the device's
|
|
10
10
|
* `maxComputeWorkgroupsPerDimension` limit (default 65535 across most devices).
|
|
11
11
|
*
|
|
12
|
-
* - 1D (pass only `
|
|
13
|
-
* - 2D (pass both `
|
|
14
|
-
* `
|
|
12
|
+
* - 1D (pass only `rows`): returns a single count for `dispatchWorkgroups(n)`.
|
|
13
|
+
* - 2D (pass both `rows` and `cols`): returns `{ x, y }` for `dispatchWorkgroups(x, y)`.
|
|
14
|
+
* `rows` maps to the y dimension and `cols` maps to the x dimension.
|
|
15
15
|
*
|
|
16
|
-
* @param {number}
|
|
17
|
-
* @param {number} [
|
|
16
|
+
* @param {number} rows - row count (1D: element count)
|
|
17
|
+
* @param {number} [cols] - column count; omit for a 1D dispatch
|
|
18
18
|
* @returns {number | { x: number, y: number }}
|
|
19
19
|
* @see {@link https://developer.mozilla.org/en-US/docs/Web/API/GPUSupportedLimits GPUSupportedLimits} (`maxComputeWorkgroupsPerDimension`)
|
|
20
20
|
*/
|
|
21
|
-
export function calcWorkgroups(
|
|
21
|
+
export function calcWorkgroups(rows, cols) {
|
|
22
22
|
const max = getDevice().limits.maxComputeWorkgroupsPerDimension;
|
|
23
|
-
if (
|
|
24
|
-
return Math.min(Math.ceil(
|
|
23
|
+
if (cols === undefined) {
|
|
24
|
+
return Math.min(Math.ceil(rows / WORKGROUP_SIZE_1D), max);
|
|
25
25
|
} else {
|
|
26
26
|
return {
|
|
27
|
-
x: Math.min(Math.ceil(
|
|
28
|
-
y: Math.min(Math.ceil(
|
|
27
|
+
x: Math.min(Math.ceil(cols / WORKGROUP_SIZE_2D), max),
|
|
28
|
+
y: Math.min(Math.ceil(rows / WORKGROUP_SIZE_2D), max),
|
|
29
29
|
};
|
|
30
30
|
}
|
|
31
31
|
}
|