AlexIndustrial/mesa
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

panvk/utrace: Add sync32/64_wait support

Browse Source 
Add support for capturing sync32/64_wait including the addresses and
values they wait for.

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 15:09:04 +02:00
committed by Marge Bot
parent 5d71bf46a7
commit 4709dcaf52
7 changed files with 131 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/panfrost/vulkan/csf/panvk_instr.h
+29 -1
View File
@@ -14,7 +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_SYNC_WAIT,
PANVK_INSTR_WORK_TYPE_SYNC32_WAIT,
PANVK_INSTR_WORK_TYPE_SYNC64_WAIT,
};
struct panvk_instr_end_args {
@@ -52,6 +53,12 @@ struct panvk_instr_end_args {
struct {
uint64_t buffer_gpu;
} dispatch_indirect;
struct {
struct cs_index addr_regs;
struct cs_index val_regs;
enum mali_cs_condition cond;
} sync;
};
};
@@ -77,3 +84,24 @@ void panvk_per_arch(panvk_instr_end_work_async)(
enum panvk_instr_work_type work_type,
const struct panvk_instr_end_args *const args,
struct cs_async_op ts_async_op);
#define PANVK_INSTR_SYNC_OPS(__cnt_width) \
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, \
struct cs_index addr) \
{ \
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##_WAIT); \
cs_sync##__cnt_width##_wait(b, reject_error, cond, ref, addr); \
struct panvk_instr_end_args instr_info = { \
.sync = {.addr_regs = addr, .val_regs = ref, .cond = cond}, \
}; \
panvk_per_arch(panvk_instr_end_work)( \
id, cmdbuf, PANVK_INSTR_WORK_TYPE_SYNC##__cnt_width##_WAIT, \
&instr_info); \
}
PANVK_INSTR_SYNC_OPS(32)
PANVK_INSTR_SYNC_OPS(64)
src/panfrost/vulkan/csf/panvk_vX_cmd_buffer.c
+3 -1
View File
@@ -581,7 +581,9 @@ emit_barrier_insert_waits(struct cs_builder *b, struct panvk_cmd_buffer *cmdbuf,
cs_add64(b, wait_val, cs_progress_seqno_reg(b, j),
cs_state->relative_sync_point);
cs_sync64_wait(b, false, MALI_CS_CONDITION_GREATER, wait_val, sync_addr);
panvk_instr_sync64_wait(cmdbuf, i, false, MALI_CS_CONDITION_GREATER,
wait_val, sync_addr);
}
}
src/panfrost/vulkan/csf/panvk_vX_cmd_draw.c
+4 -6
View File
@@ -3001,8 +3001,10 @@ wait_finish_tiling(struct panvk_cmd_buffer *cmdbuf)
cs_add64(b, vt_sync_point,
cs_progress_seqno_reg(b, PANVK_SUBQUEUE_VERTEX_TILER),
rel_vt_sync_point);
cs_sync64_wait(b, false, MALI_CS_CONDITION_GREATER, vt_sync_point,
vt_sync_addr);
panvk_instr_sync64_wait(cmdbuf, PANVK_SUBQUEUE_FRAGMENT, false,
MALI_CS_CONDITION_GREATER, vt_sync_point,
vt_sync_addr);
}
static uint32_t
@@ -3117,11 +3119,7 @@ issue_fragment_jobs(struct panvk_cmd_buffer *cmdbuf)
length_reg);
/* Wait for the tiling to be done before submitting the fragment job. */
panvk_per_arch(panvk_instr_begin_work)(PANVK_SUBQUEUE_FRAGMENT, cmdbuf,
PANVK_INSTR_WORK_TYPE_SYNC_WAIT);
wait_finish_tiling(cmdbuf);
panvk_per_arch(panvk_instr_end_work)(PANVK_SUBQUEUE_FRAGMENT, cmdbuf,
PANVK_INSTR_WORK_TYPE_SYNC_WAIT, NULL);
/* Disable the oom handler once the vertex/tiler work has finished.
* We need to disable the handler at this point as the vertex/tiler subqueue
src/panfrost/vulkan/csf/panvk_vX_cmd_event.c
+3 -1
View File
@@ -6,6 +6,7 @@
#include "panvk_cmd_buffer.h"
#include "panvk_entrypoints.h"
#include "panvk_event.h"
#include "panvk_instr.h"
#include "util/bitscan.h"
@@ -124,7 +125,8 @@ cmd_wait_event(struct panvk_cmd_buffer *cmdbuf, struct panvk_event *event,
(j * sizeof(struct panvk_cs_sync32)));
cs_move32_to(b, seqno, 0);
cs_sync32_wait(b, false, MALI_CS_CONDITION_GREATER, seqno, sync_addr);
panvk_instr_sync32_wait(cmdbuf, i, false, MALI_CS_CONDITION_GREATER,
seqno, sync_addr);
}
}
}
src/panfrost/vulkan/csf/panvk_vX_instr.c
+46 -4
View File
@@ -63,6 +63,42 @@ panvk_instr_end_dispatch_indirect(enum panvk_subqueue_id id,
});
}
static void
panvk_instr_end_sync32_wait(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_wait(&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,
},
args->sync.cond);
}
static void
panvk_instr_end_sync64_wait(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_wait(&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,
},
args->sync.cond);
}
void
panvk_per_arch(panvk_instr_begin_work)(enum panvk_subqueue_id id,
struct panvk_cmd_buffer *cmdbuf,
@@ -95,8 +131,11 @@ 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_SYNC_WAIT:
trace_begin_sync_wait(&cmdbuf->utrace.uts[id], &cs_info);
case PANVK_INSTR_WORK_TYPE_SYNC32_WAIT:
trace_begin_sync32_wait(&cmdbuf->utrace.uts[id], &cs_info);
break;
case PANVK_INSTR_WORK_TYPE_SYNC64_WAIT:
trace_begin_sync64_wait(&cmdbuf->utrace.uts[id], &cs_info);
break;
default:
UNREACHABLE("unsupported panvk_instr_work_type");
@@ -144,8 +183,11 @@ 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_SYNC_WAIT:
trace_end_sync_wait(&cmdbuf->utrace.uts[id], &cs_info);
case PANVK_INSTR_WORK_TYPE_SYNC32_WAIT:
panvk_instr_end_sync32_wait(id, &cs_info, args);
break;
case PANVK_INSTR_WORK_TYPE_SYNC64_WAIT:
panvk_instr_end_sync64_wait(id, &cs_info, args);
break;
default:
UNREACHABLE("unsupported panvk_instr_work_type");
src/panfrost/vulkan/panvk_tracepoints.py
+44 -1
View File
@@ -222,8 +222,51 @@ def define_tracepoints():
# low-level tracepoints for CS commands
begin_end_tp('sync_wait')
begin_end_tp(
'sync32_wait',
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',
),
Arg(
type='unsigned',
var='cond',
c_format='%u',
),
],
)
begin_end_tp(
'sync64_wait',
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 "',
),
Arg(
type='unsigned',
var='cond',
c_format='%u',
),
],
)
def generate_code():
utrace_generate(
src/panfrost/vulkan/panvk_utrace_perfetto.cc
+2 -1
View File
@@ -249,7 +249,8 @@ 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(sync_wait, SYNC_WAIT)
PANVK_UTRACE_PERFETTO_PROCESS_EVENT(sync32_wait, SYNC_WAIT)
PANVK_UTRACE_PERFETTO_PROCESS_EVENT(sync64_wait, SYNC_WAIT)
static uint32_t
get_gpu_clock_id(void)