fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-07 09:18:04 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 4

panvk/utrace: Add sync32/64_add support

Browse source 
Add the ability to capture sync32/64_add during tracing.

Reviewed-by: Christoph Pillmayer <christoph.pillmayer@arm.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/36519>
This commit is contained in:
Lars-Ivar Hesselberg Simonsen 2025-07-31 17:30:59 +02:00 • committed by Marge Bot
parent 4709dcaf52
commit 2c8d77b94a
9 changed files with 133 additions and 18 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

20
src/panfrost/vulkan/csf/panvk_instr.h
View file

@ -14,6 +14,8 @@ enum panvk_instr_work_type {
PANVK_INSTR_WORK_TYPE_DISPATCH,
PANVK_INSTR_WORK_TYPE_DISPATCH_INDIRECT,
PANVK_INSTR_WORK_TYPE_BARRIER,
PANVK_INSTR_WORK_TYPE_SYNC32_ADD,
PANVK_INSTR_WORK_TYPE_SYNC64_ADD,
PANVK_INSTR_WORK_TYPE_SYNC32_WAIT,
PANVK_INSTR_WORK_TYPE_SYNC64_WAIT,
};
@ -86,6 +88,24 @@ void panvk_per_arch(panvk_instr_end_work_async)(
struct cs_async_op ts_async_op);
#define PANVK_INSTR_SYNC_OPS(__cnt_width) \
static inline void panvk_instr_sync##__cnt_width##_add( \
struct panvk_cmd_buffer *cmdbuf, enum panvk_subqueue_id id, \
bool propagate_error, enum mali_cs_sync_scope scope, \
struct cs_index val, struct cs_index addr, struct cs_async_op async) \
{ \
struct cs_builder *b = panvk_get_cs_builder(cmdbuf, id); \
panvk_per_arch(panvk_instr_begin_work)( \
id, cmdbuf, PANVK_INSTR_WORK_TYPE_SYNC##__cnt_width##_ADD); \
cs_sync##__cnt_width##_add(b, propagate_error, scope, val, addr, async); \
struct panvk_instr_end_args instr_info = { \
.sync = {.addr_regs = addr, .val_regs = val}, \
}; \
async.signal_slot = 0; \
panvk_per_arch(panvk_instr_end_work_async)( \
id, cmdbuf, PANVK_INSTR_WORK_TYPE_SYNC##__cnt_width##_ADD, \
&instr_info, async); \
} \
\
static inline void panvk_instr_sync##__cnt_width##_wait( \
struct panvk_cmd_buffer *cmdbuf, enum panvk_subqueue_id id, \
bool reject_error, enum mali_cs_condition cond, struct cs_index ref, \

4
src/panfrost/vulkan/csf/panvk_vX_cmd_buffer.c
View file

@ -638,8 +638,8 @@ panvk_per_arch(emit_barrier)(struct panvk_cmd_buffer *cmdbuf,
offsetof(struct panvk_cs_subqueue_context, syncobjs));
cs_add64(b, sync_addr, sync_addr, sizeof(struct panvk_cs_sync64) * i);
cs_move64_to(b, add_val, 1);
cs_sync64_add(b, true, MALI_CS_SYNC_SCOPE_CSG, add_val, sync_addr,
cs_now());
panvk_instr_sync64_add(cmdbuf, i, true, MALI_CS_SYNC_SCOPE_CSG,
add_val, sync_addr, cs_now());
++cs_state->relative_sync_point;
}
}

10
src/panfrost/vulkan/csf/panvk_vX_cmd_dispatch.c
View file

@ -313,8 +313,9 @@ cmd_dispatch(struct panvk_cmd_buffer *cmdbuf, struct panvk_dispatch_info *info)
cs_add64(b, sync_addr, sync_addr,
PANVK_SUBQUEUE_COMPUTE * sizeof(struct panvk_cs_sync64));
cs_move64_to(b, add_val, 1);
cs_sync64_add(b, true, MALI_CS_SYNC_SCOPE_CSG, add_val, sync_addr,
cs_defer_indirect());
panvk_instr_sync64_add(cmdbuf, PANVK_SUBQUEUE_COMPUTE, true,
MALI_CS_SYNC_SCOPE_CSG, add_val, sync_addr,
cs_defer_indirect());
#else
struct cs_index sync_addr = cs_scratch_reg64(b, 0);
struct cs_index iter_sb = cs_scratch_reg32(b, 2);
@ -332,8 +333,9 @@ cmd_dispatch(struct panvk_cmd_buffer *cmdbuf, struct panvk_dispatch_info *info)
cs_match(b, iter_sb, cmp_scratch) {
#define CASE(x) \
cs_case(b, SB_ITER(x)) { \
cs_sync64_add(b, true, MALI_CS_SYNC_SCOPE_CSG, add_val, sync_addr, \
cs_defer(SB_WAIT_ITER(x), SB_ID(DEFERRED_SYNC))); \
panvk_instr_sync64_add(cmdbuf, PANVK_SUBQUEUE_COMPUTE, true, \
MALI_CS_SYNC_SCOPE_CSG, add_val, sync_addr, \
cs_defer(SB_WAIT_ITER(x), SB_ID(DEFERRED_SYNC))); \
}
CASE(0)

20
src/panfrost/vulkan/csf/panvk_vX_cmd_draw.c
View file

@ -2950,8 +2950,9 @@ flush_tiling(struct panvk_cmd_buffer *cmdbuf)
cs_move64_to(b, add_val, 1);
cs_heap_operation(b, MALI_CS_HEAP_OPERATION_VERTEX_TILER_COMPLETED,
cs_defer_indirect());
cs_sync64_add(b, true, MALI_CS_SYNC_SCOPE_CSG, add_val, sync_addr,
cs_defer_indirect());
panvk_instr_sync64_add(cmdbuf, PANVK_SUBQUEUE_VERTEX_TILER, true,
MALI_CS_SYNC_SCOPE_CSG, add_val, sync_addr,
cs_defer_indirect());
#else
struct cs_index sync_addr = cs_scratch_reg64(b, 0);
struct cs_index iter_sb = cs_scratch_reg32(b, 2);
@ -2969,8 +2970,9 @@ flush_tiling(struct panvk_cmd_buffer *cmdbuf)
cs_case(b, SB_ITER(x)) { \
cs_heap_operation(b, MALI_CS_HEAP_OPERATION_VERTEX_TILER_COMPLETED, \
cs_defer(SB_WAIT_ITER(x), SB_ID(DEFERRED_SYNC))); \
cs_sync64_add(b, true, MALI_CS_SYNC_SCOPE_CSG, add_val, sync_addr, \
cs_defer(SB_WAIT_ITER(x), SB_ID(DEFERRED_SYNC))); \
panvk_instr_sync64_add(cmdbuf, PANVK_SUBQUEUE_VERTEX_TILER, true, \
MALI_CS_SYNC_SCOPE_CSG, add_val, sync_addr, \
cs_defer(SB_WAIT_ITER(x), SB_ID(DEFERRED_SYNC))); \
}
CASE(0)
@ -3331,8 +3333,9 @@ issue_fragment_jobs(struct panvk_cmd_buffer *cmdbuf)
}
}
cs_sync64_add(b, true, MALI_CS_SYNC_SCOPE_CSG, add_val, sync_addr,
cs_defer_indirect());
panvk_instr_sync64_add(cmdbuf, PANVK_SUBQUEUE_FRAGMENT, true,
MALI_CS_SYNC_SCOPE_CSG, add_val, sync_addr,
cs_defer_indirect());
#else
struct cs_index iter_sb = cs_scratch_reg32(b, 2);
struct cs_index cmp_scratch = cs_scratch_reg32(b, 3);
@ -3392,8 +3395,9 @@ issue_fragment_jobs(struct panvk_cmd_buffer *cmdbuf)
cs_defer(SB_MASK(DEFERRED_FLUSH), SB_ID(DEFERRED_SYNC))); \
} \
} \
cs_sync64_add(b, true, MALI_CS_SYNC_SCOPE_CSG, add_val, sync_addr, \
async); \
panvk_instr_sync64_add(cmdbuf, PANVK_SUBQUEUE_FRAGMENT, true, \
MALI_CS_SYNC_SCOPE_CSG, add_val, sync_addr, \
async); \
}
CASE(0)

11
src/panfrost/vulkan/csf/panvk_vX_cmd_precomp.c
View file

@ -10,6 +10,7 @@
#include "panvk_cmd_alloc.h"
#include "panvk_cmd_buffer.h"
#include "panvk_cmd_precomp.h"
#include "panvk_instr.h"
#include "panvk_macros.h"
#include "panvk_mempool.h"
#include "panvk_precomp_cache.h"
@ -125,8 +126,9 @@ panvk_per_arch(dispatch_precomp)(struct panvk_precomp_ctx *ctx,
cs_add64(b, sync_addr, sync_addr,
PANVK_SUBQUEUE_COMPUTE * sizeof(struct panvk_cs_sync64));
cs_move64_to(b, add_val, 1);
cs_sync64_add(b, true, MALI_CS_SYNC_SCOPE_CSG, add_val, sync_addr,
cs_defer_indirect());
panvk_instr_sync64_add(cmdbuf, PANVK_SUBQUEUE_COMPUTE, true,
MALI_CS_SYNC_SCOPE_CSG, add_val, sync_addr,
cs_defer_indirect());
#else
struct cs_index sync_addr = cs_scratch_reg64(b, 0);
struct cs_index iter_sb = cs_scratch_reg32(b, 2);
@ -144,8 +146,9 @@ panvk_per_arch(dispatch_precomp)(struct panvk_precomp_ctx *ctx,
cs_match(b, iter_sb, cmp_scratch) {
#define CASE(x) \
cs_case(b, SB_ITER(x)) { \
cs_sync64_add(b, true, MALI_CS_SYNC_SCOPE_CSG, add_val, sync_addr, \
cs_defer(SB_WAIT_ITER(x), SB_ID(DEFERRED_SYNC))); \
panvk_instr_sync64_add(cmdbuf, PANVK_SUBQUEUE_COMPUTE, true, \
MALI_CS_SYNC_SCOPE_CSG, add_val, sync_addr, \
cs_defer(SB_WAIT_ITER(x), SB_ID(DEFERRED_SYNC))); \
}
CASE(0)

46
src/panfrost/vulkan/csf/panvk_vX_instr.c
View file

@ -63,6 +63,40 @@ panvk_instr_end_dispatch_indirect(enum panvk_subqueue_id id,
});
}
static void
panvk_instr_end_sync32_add(enum panvk_subqueue_id id,
struct panvk_utrace_cs_info *cs_info,
const struct panvk_instr_end_args *const args)
{
assert(args->sync.val_regs.size == 1);
trace_end_sync32_add(&cs_info->cmdbuf->utrace.uts[id], cs_info,
(struct u_trace_address){
.bo = (void *)PANVK_UTRACE_CAPTURE_REGISTERS,
.offset = args->sync.addr_regs.reg,
},
(struct u_trace_address){
.bo = (void *)PANVK_UTRACE_CAPTURE_REGISTERS,
.offset = args->sync.val_regs.reg,
});
}
static void
panvk_instr_end_sync64_add(enum panvk_subqueue_id id,
struct panvk_utrace_cs_info *cs_info,
const struct panvk_instr_end_args *const args)
{
assert(args->sync.val_regs.size == 2);
trace_end_sync64_add(&cs_info->cmdbuf->utrace.uts[id], cs_info,
(struct u_trace_address){
.bo = (void *)PANVK_UTRACE_CAPTURE_REGISTERS,
.offset = args->sync.addr_regs.reg,
},
(struct u_trace_address){
.bo = (void *)PANVK_UTRACE_CAPTURE_REGISTERS,
.offset = args->sync.val_regs.reg,
});
}
static void
panvk_instr_end_sync32_wait(enum panvk_subqueue_id id,
struct panvk_utrace_cs_info *cs_info,
@ -131,6 +165,12 @@ panvk_per_arch(panvk_instr_begin_work)(enum panvk_subqueue_id id,
case PANVK_INSTR_WORK_TYPE_BARRIER:
trace_begin_barrier(&cmdbuf->utrace.uts[id], &cs_info);
break;
case PANVK_INSTR_WORK_TYPE_SYNC32_ADD:
trace_begin_sync32_add(&cmdbuf->utrace.uts[id], &cs_info);
break;
case PANVK_INSTR_WORK_TYPE_SYNC64_ADD:
trace_begin_sync64_add(&cmdbuf->utrace.uts[id], &cs_info);
break;
case PANVK_INSTR_WORK_TYPE_SYNC32_WAIT:
trace_begin_sync32_wait(&cmdbuf->utrace.uts[id], &cs_info);
break;
@ -183,6 +223,12 @@ panvk_per_arch(panvk_instr_end_work_async)(
case PANVK_INSTR_WORK_TYPE_BARRIER:
panvk_instr_end_barrier(id, &cs_info, args);
break;
case PANVK_INSTR_WORK_TYPE_SYNC32_ADD:
panvk_instr_end_sync32_add(id, &cs_info, args);
break;
case PANVK_INSTR_WORK_TYPE_SYNC64_ADD:
panvk_instr_end_sync64_add(id, &cs_info, args);
break;
case PANVK_INSTR_WORK_TYPE_SYNC32_WAIT:
panvk_instr_end_sync32_wait(id, &cs_info, args);
break;

36
src/panfrost/vulkan/panvk_tracepoints.py
View file

@ -222,6 +222,42 @@ def define_tracepoints():
# low-level tracepoints for CS commands
begin_end_tp(
'sync32_add',
args=[
Arg(
type='uint64_t',
var='sync_addr',
is_indirect=True,
c_format='0x%" PRIx64 "'
),
Arg(
type='uint32_t',
var='sync_val',
is_indirect=True,
c_format='%u',
),
],
)
begin_end_tp(
'sync64_add',
args=[
Arg(
type='uint64_t',
var='sync_addr',
is_indirect=True,
c_format='0x%" PRIx64 "'
),
Arg(
type='uint64_t',
var='sync_val',
is_indirect=True,
c_format='%" PRIu64 "',
),
],
)
begin_end_tp(
'sync32_wait',
args=[

3
src/panfrost/vulkan/panvk_utrace_perfetto.cc
View file

@ -43,6 +43,7 @@ get_stage_name(enum panvk_utrace_perfetto_stage stage)
CASE(RENDER);
CASE(DISPATCH);
CASE(BARRIER);
CASE(SYNC_ADD);
CASE(SYNC_WAIT);
#undef CASE
default:
@ -249,6 +250,8 @@ PANVK_UTRACE_PERFETTO_PROCESS_EVENT(render, RENDER)
PANVK_UTRACE_PERFETTO_PROCESS_EVENT(dispatch, DISPATCH)
PANVK_UTRACE_PERFETTO_PROCESS_EVENT(dispatch_indirect, DISPATCH)
PANVK_UTRACE_PERFETTO_PROCESS_EVENT(barrier, BARRIER)
PANVK_UTRACE_PERFETTO_PROCESS_EVENT(sync32_add, SYNC_ADD)
PANVK_UTRACE_PERFETTO_PROCESS_EVENT(sync64_add, SYNC_ADD)
PANVK_UTRACE_PERFETTO_PROCESS_EVENT(sync32_wait, SYNC_WAIT)
PANVK_UTRACE_PERFETTO_PROCESS_EVENT(sync64_wait, SYNC_WAIT)

1
src/panfrost/vulkan/panvk_utrace_perfetto.h
View file

@ -24,6 +24,7 @@ enum panvk_utrace_perfetto_stage {
PANVK_UTRACE_PERFETTO_STAGE_RENDER,
PANVK_UTRACE_PERFETTO_STAGE_DISPATCH,
PANVK_UTRACE_PERFETTO_STAGE_BARRIER,
PANVK_UTRACE_PERFETTO_STAGE_SYNC_ADD,
PANVK_UTRACE_PERFETTO_STAGE_SYNC_WAIT,
PANVK_UTRACE_PERFETTO_STAGE_COUNT,