hexagon: improved Op queuing, buffer and cache management (#21705)
* hexagon: introduce op request batching and rewrite buffer managment The host now prepares batches of requests and dispatches them via a single dspqueue message. Buffers are mapped explicitly by NPU while processing batches. * hex-dma: disable l2 bypass since to work around new issue due to no flushes between Ops * hex-utils: add explicit l2flush and l2clear helpers * hex-opreq: use fine-grain per tensor l2 management * hex-opreq: avoid redundant invalidates for tensors we already flushed * hex-opreq: update debug messages * htp-opreq: reuse ops_context * hex-opreq: do not flush or invalidate cache lines beyond buffer boundry * hex-opreq: fix errors in log message * Revert "hex-opreq: do not flush or invalidate cache lines beyond buffer boundry" This reverts commit 8b7f0a55a750a6430ce4eb1874c7feb3d720056d. * hexagon: limit l2 flushes to 1MB which covers l2 cache * hex-opreq: limit cache flush to 4MB Looks like 4MB cont. vitual space should cover the 1MB cache. * hexagon: drop cache flush size to 2MB * hex-opreq: start reworking opreq packing * hex-opreq: introduce new way of packing opbatch where tensors are stored separately * hex-opreq: add a simple fastrpc call to force unmap all buffers * hex-l2flush: somehow 2MB does not seem robust, also cleanup step size to use line-size * hex-opreq: bump opreq batch size to 256 * hex-mm: place src1 spad at the top of vtcm for easy reuse * hex-ops: introduce internal types and disable src1 reuse for now Nothing new just formalizing the repack / qyn.quant types we've been using. * htp-opreq: use tensor pointers instead of copies * hex-opreq: introduce more robust way for tracking vtcm/spad reuse This removes the SKIP_QUANTIZE flag that became fragile with the addition of HMX and other ops. * hex-cumsum: fix error post opreq merge * hex-opreq: move request batch handling into the session Prepping everything for using dspqueue buffers and doing that inside the session is much cleaner. * hex-mm: yet another fix for src1 reuse when we're mixing hmx/hvx * hex-bufs: introduce pinned mmapings and use non-pinned ones for model buffers * hex-buf: add support for allocating shared/pinned buffer for opreqs * hex-opbatch: make opbatches configurable * hex-naming: better name for ggml_hexagon_shared_buffer * hex-naming: add session->c_name() helper * hex-opbatch: start using shm but still copy for now * hex-opbatch: use shared buffer for packing opbatch * hex-opbatch: beter naming for opbatch related classes and code * hex-opbatch: reuse batched tensors with same data/dims/strides * hex-opbatch: update logging * hex-opbatch: add support for vmem limit for op batching * hex-opbatch: update htp side to properly support dynamic mmap/unmap * hex-opbatch: add OB and OQ params for run-completion script and fix the asserts in batch processing * hex-opbatch: fixed src1 handling in act ops * hex-act: fix empty src1 handling in swiglu and friends Simplify preamble macro while at it * hex-mm: minor fix vtcm and dma handling in matmul cleaning up some left-overs from merges * hex-opbatch: allocate extra 1KB for dspqueue overhead * hexagon: fix softmax for non-aligned tensors and cleanup vtcm alloc * hex-mm: properly handle hmx_disabled flag * hex-ops: update comments * hex-ops: add debug output for get/set-rows * hex-mmap: optimize un/mapping of buffers * hex-opreq: global cache flush and invalidate beyond 128KB threshold * hex-ops: add super simple opfilter regex for debugging If an Op matches the regex hex backend will reject it. * hex-opbatch: wireup newer ops missed in merge and update main switch to detect this in future * hexagon: improved vtcm acquision to remove inter-op overhead Fully compatible with QNN-HTP coex * hex-mm: fixed hvx fallback path * hex-mm: lower the vmem threshold a bit further to ~3GB * hexagon: update debug & error logs This also fixes an issue with newer llvm merging repack and non-repack functions. We use those pointer to distinguish between buffer types. * hexagon: move ops context into main context Just a cleanup. We don't need separate contexts at this point. * hex-opbatch: cleanup naming and headers for opbatch and related descriptors * hex-fa: it's now better to enable FA during TG to reduce graph splits * hexagon: remove GGML_HEXAGON_EXPERIMENTAL env var It's no longer useful. Please use more flexible GGML_HEXAGON_OPFILTER to disable Ops if needed for debugging or validation. * hexagon: fixed editorconfig check * Update ggml/src/ggml-hexagon/ggml-hexagon.cpp Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com> --------- Co-authored-by: Trivikram Reddy <tamarnat@qti.qualcomm.com> Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
This commit is contained in:
Max Krasnyansky
co-authored by
Sigbjørn Skjæret
Trivikram Reddy
Sigbjørn Skjæret
Trivikram Reddy
parent
3fc65063d9
commit
9aa2807769
@@ -20,7 +20,7 @@
|
||||
#include "hvx-dump.h"
|
||||
#include "worker-pool.h"
|
||||
#include "htp-ctx.h"
|
||||
#include "htp-msg.h"
|
||||
#include "htp-ops.h"
|
||||
|
||||
#include "hmx-utils.h"
|
||||
#include "hmx-ops.h"
|
||||
@@ -821,7 +821,7 @@ int hmx_mat_mul_permuted_w16a32_batched(struct htp_context *ctx, const hmx_matmu
|
||||
// and each q_head is computed individually to avoid tile-major packing
|
||||
// issues. m_chunk_n_rows is always a multiple of 32 (from
|
||||
// hmx_compute_chunks), so per-head tile arrays don't overlap.
|
||||
const size_t vtcm_budget = ctx->vtcm_scratch_size;
|
||||
const size_t vtcm_budget = ctx->vtcm_size;
|
||||
const size_t vec_dot_size = params->k * sizeof(__fp16);
|
||||
|
||||
// When the activation has a large stride (e.g. permuted Q tensor with
|
||||
@@ -998,7 +998,7 @@ int hmx_mat_mul_permuted_w16a32(struct htp_context *ctx, float *restrict dst, co
|
||||
}
|
||||
|
||||
// --- Dynamic VTCM layout ---
|
||||
const size_t vtcm_budget = ctx->vtcm_scratch_size;
|
||||
const size_t vtcm_budget = ctx->vtcm_size;
|
||||
const size_t vec_dot_size = k * sizeof(__fp16);
|
||||
|
||||
// DMA-based activation gather for strided tensors (see batched path comment).
|
||||
@@ -1182,7 +1182,7 @@ int hmx_mat_mul_permuted_qk_0_d16a32(struct htp_context *ctx, float *restrict ds
|
||||
FARF(MEDIUM, "hmx_matmul_qk: STANDARD path m=%d k=%d n=%d type=%d", m, k, n, weight_type);
|
||||
|
||||
// --- Dynamic VTCM layout ---
|
||||
const size_t vtcm_budget = ctx->vtcm_scratch_size;
|
||||
const size_t vtcm_budget = ctx->vtcm_size;
|
||||
const size_t vec_dot_size = k * sizeof(__fp16);
|
||||
const bool use_pipeline = (m >= 128) && (k <= n);
|
||||
|
||||
@@ -1273,9 +1273,6 @@ int hmx_mat_mul_permuted_qk_0_d16a32(struct htp_context *ctx, float *restrict ds
|
||||
void *buf_curr = vtcm_scratch0;
|
||||
void *buf_next = vtcm_scratch1;
|
||||
|
||||
// issue async DDR data transfer for the first weight chunk
|
||||
// NOTE: use 2D DMA (n_cols rows x row_stride bytes) instead of 1D
|
||||
// because UDMA roiwidth is 16-bit and total size can exceed 65535.
|
||||
{
|
||||
const size_t n_cols_first = hex_smin(n, n_chunk_n_cols);
|
||||
dma_queue_push(ctx->dma[0], dma_make_ptr(buf_curr, permuted_weight), row_stride, row_stride, row_stride, n_cols_first);
|
||||
@@ -1533,20 +1530,15 @@ void transfer_activation_chunk_threaded(struct htp_context *ctx, __fp16 *dst, co
|
||||
worker_pool_run_func(ctx->worker_pool, transfer_activation_chunk_worker_fn, &state, ctx->n_threads);
|
||||
}
|
||||
|
||||
int mat_mul_qk_0_d16a32_out_stationary(struct htp_context *ctx, float *restrict out, const float *restrict x, const uint8_t *restrict w, int m,
|
||||
int k, int n, int weight_type) {
|
||||
// Runtime check -- k >= 16384 exceeds 2D DMA limit
|
||||
if (k >= 16384) {
|
||||
FARF(HIGH, "%s: k=%d exceeds 2D DMA limit", __func__, k);
|
||||
return -1;
|
||||
}
|
||||
int mat_mul_qk_0_d16a32_out_stationary(struct htp_context *ctx, float *restrict out, const float *restrict x, const uint8_t *restrict w,
|
||||
int m, int k, int n, int weight_type) {
|
||||
// assume k % 32 == 0 && n % 32 == 0
|
||||
const size_t row_stride = get_x4x2_row_stride(weight_type, k);
|
||||
if (row_stride == 0) {
|
||||
return -1;
|
||||
}
|
||||
|
||||
const size_t vtcm_budget = ctx->vtcm_scratch_size;
|
||||
const size_t vtcm_budget = ctx->vtcm_size;
|
||||
|
||||
const size_t M_BLOCK_SIZE = 512;
|
||||
const size_t N_BLOCK_SIZE = 512;
|
||||
@@ -1576,8 +1568,7 @@ int mat_mul_qk_0_d16a32_out_stationary(struct htp_context *ctx, float *restrict
|
||||
__fp16 *vtcm_scales = (__fp16 *) vtcm_seq_alloc(&vtcm_ptr, 256);
|
||||
assert((size_t)(vtcm_ptr - (uint8_t *)ctx->vtcm_base) <= vtcm_budget);
|
||||
|
||||
FARF(MEDIUM, "%s: m=%d k=%d n=%d wtype=%d vtcm=%zu/%zu",
|
||||
__func__, m, k, n, weight_type,
|
||||
FARF(MEDIUM, "%s: m=%d k=%d n=%d wtype=%d vtcm=%zu/%zu", __func__, m, k, n, weight_type,
|
||||
(size_t)(vtcm_ptr - (uint8_t *)ctx->vtcm_base), vtcm_budget);
|
||||
|
||||
// initialize eye tile (32x32 identity matrix)
|
||||
|
||||
Reference in New Issue
Block a user