Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

hexagon: support for op-trace (fine-grain tracing of HVX/HMX/DMA events) (#24592)

Browse Source 
* hex-optrace: add support for optrace and instrument matmul and flash-atten code

* hex-trace: improve trace event and prefetto generator

* hex-trace: add new script dedicated to handling traces, specifically perfetto traces

* hex-trace: add --head/--tail options to profile and trace tools

* hex-trace: fix whitespaces

* hex-trace: fix flake8 warnings

* hex-trace: fix flake8 warnings

* hmx-fa: restore q_tiles clearing

* hex-profile: remove circular dep in includes

* hex-trace: simplify trace sizing check

* hex-profile: sort events in the summary by name
This commit is contained in:
Max Krasnyansky
2026-06-18 08:35:02 -07:00
committed by GitHub
parent 7b6c5a2aed
commit d2c67959b3
17 changed files with 1121 additions and 224 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-hexagon/htp/htp-ops.h
+31 -4
View File
@@ -146,10 +146,36 @@ struct htp_op_desc {
uint16_t dst; // Output tensor index
};
#ifndef HTP_MAX_NTHREADS
#define HTP_MAX_NTHREADS 10
#endif
#define HTP_TRACE_MAX_EVENTS 256
enum htp_profiler_mode {
HTP_PROF_DISABLED = 0,
HTP_PROF_BASIC = 1,
HTP_PROF_PMU = 2,
HTP_PROF_TRACE = 3,
};
enum htp_trace_event_id {
HTP_TRACE_EVT_DMA = 0,
HTP_TRACE_EVT_HVX_COMP = 20,
HTP_TRACE_EVT_HVX_A_QUANT = 21,
HTP_TRACE_EVT_HVX_A_PREP = 22,
HTP_TRACE_EVT_HVX_W_DEQUANT = 23,
HTP_TRACE_EVT_HVX_W_PREP = 24,
HTP_TRACE_EVT_HVX_O_PROC = 25,
HTP_TRACE_EVT_HMX_COMP = 40,
};
struct htp_trace_desc {
uint32_t cycles; // lower 32-bits of cycle counter
uint16_t id; // Event ID
uint16_t info; // bit 15: is_stop. bits 14-0: tile/chunk index or other metadata.
};
#define HTP_PROF_PMU_NCNT 8
@@ -158,8 +184,8 @@ enum htp_profiler_mode {
struct htp_prof_desc {
uint32_t opcode; // GGML/HTP Op
uint32_t usecs; // Number of usec
uint32_t cycles; // Number of cycles
uint32_t pad; // Unused
uint32_t cycles_start; // Start cycle counter
uint32_t cycles_stop; // Stop cycle counter
uint32_t pmu[HTP_PROF_PMU_NCNT]; // PMU counters
};
@@ -168,7 +194,7 @@ struct htp_opbatch_req {
uint32_t n_bufs; // Number of buffers
uint32_t n_tensors; // Number of tensors
uint32_t n_ops; // Number of ops
uint32_t flags; // unused
uint32_t n_traces; // Number of trace descriptors per thread
uint32_t pad; // unused
// struct htp_buf_desc bufs[]; -- dspqueue buf 0
// struct htp_tensor tensors[]; -- dspqueue buf 0
@@ -181,7 +207,8 @@ struct htp_opbatch_rsp {
uint32_t n_bufs; // Number of buffers
uint32_t n_tensors; // Number of tensors
uint32_t n_ops; // Number of op profile descriptors
uint32_t pad; // unused
uint32_t n_traces[HTP_MAX_NTHREADS + 1];
uint8_t pad[8]; // align to 8 bytes
// struct htp_prof_desc profs[]; -- dspqueue buf 0
};