Skip to content

Actions: yhirose/cpp-tensorlib

Actions

All workflows

Actions

Loading...
Loading

Showing runs from all workflows
145 workflow runs
145 workflow runs

Filter by Workflow

⌥ + click/return to exclude

Filter by Event

⌥ + click/return to exclude

Filter by Status

⌥ + click/return to exclude

Filter by Branch

⌥ + click/return to exclude

Filter by Actor

⌥ + click/return to exclude
clone copies on the device, and a CPU read waits per buffer CI #95: Commit 663a240 pushed by yhirose
4m 21s master
master
4m 21s
cuda: the wave plan's single-fill layer count is computed as long, so… CI #94: Commit c2bbdab pushed by yhirose
6m 22s master
master
6m 22s
cleanup: a fused dot bias is owed until the gemm takes it and the one… CI #93: Commit a9bea53 pushed by yhirose
6m 43s master
master
6m 43s
cleanup: the one-input GPU dispatches share their gate and encode, de… CI #92: Commit 7fd4a8c pushed by yhirose
4m 59s master
master
4m 59s
cuda: device_read_ keeps its blocking copy inside a graph capture CI #91: Commit 1bae52e pushed by yhirose
6m 1s master
master
6m 1s
cleanup: the two sgemm cores share their tiling, the tile table owns … CI #90: Commit b622dec pushed by yhirose
5m 26s master
master
5m 26s
array: the eval epilogue's affine runs on the GPU before falling back… CI #89: Commit da74013 pushed by yhirose
5m 37s master
master
5m 37s
cuda: a cp.async-pipelined 128x128 sgemm replaces the register-staged… CI #88: Commit f48790d pushed by yhirose
4m 59s master
master
4m 59s
cpu: gemm splits N for short activations, the pool balances dynamical… CI #87: Commit 901989c pushed by yhirose
6m 26s master
master
6m 26s
array: one pool-parallel elementwise driver; sgemm_ packs stop re-zer… CI #86: Commit a7f31b6 pushed by yhirose
5m 33s master
master
5m 33s
array: softmax on the CPU spreads its rows across the thread pool CI #85: Commit 909ca35 pushed by yhirose
5m 1s master
master
5m 1s
cuda: split-K plan as one function; census toggles the launch mode in… CI #84: Commit 79171ce pushed by yhirose
5m 1s master
master
5m 1s
cuda: split-K sized by the grid's fill, and through the affine epilogue CI #83: Commit 2f97474 pushed by yhirose
5m 7s master
master
5m 7s
array: causal prefill attention on the CPU runs as parallel query tiles CI #82: Commit 851ec53 pushed by yhirose
5m 1s master
master
5m 1s
cuda: batched dot in one launch, the batch folded into the gemm grid CI #81: Commit 6809b6e pushed by yhirose
7m 15s master
master
7m 15s
array: batched dot keeps its epilogue on-device; causal prefill gets … CI #80: Commit a803f9e pushed by yhirose
4m 26s master
master
4m 26s
array: batched dot takes transposed slices and any batch strides CI #79: Commit 21b67b9 pushed by yhirose
4m 30s master
master
4m 30s
array: dedup the prefill attention paths against their decode twins CI #78: Commit 01ebdfd pushed by yhirose
4m 38s master
master
4m 38s
array: causal prefill attention as an op, fused on CUDA CI #77: Commit 3e175ac pushed by yhirose
4m 39s master
master
4m 39s
auto: derive the matmul crossover and the thread floor on the host CI #76: Commit 2f0cc4a pushed by yhirose
4m 53s master
master
4m 53s
cuda: take transposed operands on the register-blocked sgemm CI #75: Commit 21bcc6a pushed by yhirose
4m 31s master
master
4m 31s
misc: census the pool's round trip and the gemm operand layouts CI #74: Commit dbed412 pushed by yhirose
3m 42s master
master
3m 42s
cuda: re-census the auto thresholds against the unburdened CPU gemm CI #73: Commit d6fcb9b pushed by yhirose
4m 16s master
master
4m 16s
index ops: check a label before either backend indexes with it CI #72: Commit d120269 pushed by yhirose
4m 30s master
master
4m 30s
tools: add a CUDA/Metal/WebGPU backend parity checker CI #71: Commit e6971d5 pushed by yhirose
4m 9s master
master
4m 9s