Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-18 16:55:14 +02:00
Code Issues Packages Projects Releases Wiki Activity

hexagon: rope updates (#28628)

Browse Source 
* hexagon: vectorize RoPE theta cache on v75

* hexagon: vectorize MROPE/IMROPE theta pick

* hexagon: tighten NEOX RoPE rotate and aligned tail copy

* hex-rope: use inplace rope for all scenarios

* hex-rope: remove ctx->spad usage and legacy timers

* hex-rope: add kernel params and enforce vtcm reqs at the host

* hex-rope: cleanup unused params and tighten the mode checks

* hex-rope: add  missing ops header

---------

Co-authored-by: Max Krasnyansky <maxk@qti.qualcomm.com>
This commit is contained in:
Todor Boinovski
2026-09-09 08:40:24 -07:00
committed by GitHub
parent 22397c31a0
commit 6d9c82ea2b
5 changed files with 565 additions and 391 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/test-backend-ops.cpp
+12
View File
@@ -10282,6 +10282,12 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
test_cases.emplace_back(new test_rope(type, {128, 32, 2, 1}, 32, GGML_ROPE_TYPE_NEOX, 512, 1.4245f, 0.7465f, 1.4245f, false, 0, true, true, 32));
}
// Real-model RoPE: F32 forward, packed Q, 512-token prefill.
test_cases.emplace_back(new test_rope(GGML_TYPE_F32, {256, 8, 512, 1}, 64, GGML_ROPE_TYPE_IMROPE, 512, 1.0f, 0.0f, 1.0f, false, 0, true)); // qwen3.5 0.8B
test_cases.emplace_back(new test_rope(GGML_TYPE_F32, {256, 16, 512, 1}, 64, GGML_ROPE_TYPE_IMROPE, 512, 1.0f, 0.0f, 1.0f, false, 0, true)); // qwen3.5 4B
test_cases.emplace_back(new test_rope(GGML_TYPE_F32, {256, 8, 512, 1}, 256, GGML_ROPE_TYPE_NEOX, 512, 1.0f, 0.0f, 1.0f, false, 0, true)); // gemma4 E2B sliding
test_cases.emplace_back(new test_rope(GGML_TYPE_F32, {512, 8, 512, 1}, 128, GGML_ROPE_TYPE_NEOX, 512, 1.0f, 0.0f, 1.0f, true, 0, true)); // gemma4 E4B global
for (int v : { 0, 1, 2, 3 }) {
for (int dim : { 0, 1, 2, 3, }) {
test_cases.emplace_back(new test_concat(GGML_TYPE_F32, {11, 12, 13, 14}, 7, dim, v));
@@ -11178,6 +11184,12 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_perf() {
}
}
// Real-model RoPE: F32 forward, packed Q, 512-token prefill.
test_cases.emplace_back(new test_rope(GGML_TYPE_F32, {256, 8, 512, 1}, 64, GGML_ROPE_TYPE_IMROPE, 512, 1.0f, 0.0f, 1.0f, false, 0, true)); // qwen3.5 0.8B
test_cases.emplace_back(new test_rope(GGML_TYPE_F32, {256, 16, 512, 1}, 64, GGML_ROPE_TYPE_IMROPE, 512, 1.0f, 0.0f, 1.0f, false, 0, true)); // qwen3.5 4B
test_cases.emplace_back(new test_rope(GGML_TYPE_F32, {256, 8, 512, 1}, 256, GGML_ROPE_TYPE_NEOX, 512, 1.0f, 0.0f, 1.0f, false, 0, true)); // gemma4 E2B sliding
test_cases.emplace_back(new test_rope(GGML_TYPE_F32, {512, 8, 512, 1}, 128, GGML_ROPE_TYPE_NEOX, 512, 1.0f, 0.0f, 1.0f, true, 0, true)); // gemma4 E4B global
std::vector<std::array<int64_t, 4>> reduce_rows_cases = {
{ 8192, 1, 1, 1 },
{ 8192, 8192, 1, 1 },