diff --git a/src/gallium/drivers/panfrost/pan_fragment.c b/src/gallium/drivers/panfrost/pan_fragment.c
index 9c8f31fc96c..3c7d983a620 100644
--- a/src/gallium/drivers/panfrost/pan_fragment.c
+++ b/src/gallium/drivers/panfrost/pan_fragment.c
@@ -68,12 +68,6 @@ panfrost_fragment_job(struct panfrost_batch *batch, bool has_draws)
if (fb->zsbuf)
panfrost_initialize_surface(batch, fb->zsbuf);
- struct mali_job_descriptor_header header = {
- .job_type = MALI_JOB_TYPE_FRAGMENT,
- .job_index = 1,
- .job_descriptor_size = 1
- };
-
/* The passed tile coords can be out of range in some cases, so we need
* to clamp them to the framebuffer size to avoid a TILE_RANGE_FAULT.
* Theoretically we also need to clamp the coordinates positive, but we
@@ -100,8 +94,16 @@ panfrost_fragment_job(struct panfrost_batch *batch, bool has_draws)
.framebuffer = framebuffer,
};
- struct panfrost_transfer transfer = panfrost_pool_alloc_aligned(&batch->pool, sizeof(header) + sizeof(payload), 64);
- memcpy(transfer.cpu, &header, sizeof(header));
- memcpy(transfer.cpu + sizeof(header), &payload, sizeof(payload));
+ struct panfrost_transfer transfer =
+ panfrost_pool_alloc_aligned(&batch->pool,
+ MALI_JOB_HEADER_LENGTH + sizeof(payload),
+ 64);
+
+ pan_pack(transfer.cpu, JOB_HEADER, header) {
+ header.type = MALI_JOB_TYPE_FRAGMENT;
+ header.index = 1;
+ }
+
+ memcpy(transfer.cpu + MALI_JOB_HEADER_LENGTH, &payload, sizeof(payload));
return transfer.gpu;
}
diff --git a/src/panfrost/bifrost/test/bi_submit.c b/src/panfrost/bifrost/test/bi_submit.c
index 408171f3cd0..d131427de24 100644
--- a/src/panfrost/bifrost/test/bi_submit.c
+++ b/src/panfrost/bifrost/test/bi_submit.c
@@ -63,15 +63,12 @@ bit_submit(struct panfrost_device *dev,
void *payload, size_t payload_size,
struct panfrost_bo **bos, size_t bo_count, enum bit_debug debug)
{
- struct mali_job_descriptor_header header = {
- .job_descriptor_size = MALI_JOB_64,
- .job_type = T,
- .job_index = 1
- };
-
struct panfrost_bo *job = bit_bo_create(dev, 4096);
- memcpy(job->cpu, &header, sizeof(header));
- memcpy(job->cpu + sizeof(header), payload, payload_size);
+ pan_pack(job->cpu, JOB_HEADER, cfg) {
+ cfg.type = T;
+ cfg.index = 1;
+ }
+ memcpy(job->cpu + MALI_JOB_HEADER_LENGTH, payload, payload_size);
uint32_t *bo_handles = calloc(sizeof(uint32_t), bo_count);
diff --git a/src/panfrost/include/panfrost-job.h b/src/panfrost/include/panfrost-job.h
index 4ecf9c3c9cc..27bd04df981 100644
--- a/src/panfrost/include/panfrost-job.h
+++ b/src/panfrost/include/panfrost-job.h
@@ -238,25 +238,6 @@ struct bifrost_blend_rt {
};
} __attribute__((packed));
-/* Possible values for job_descriptor_size */
-
-#define MALI_JOB_32 0
-#define MALI_JOB_64 1
-
-struct mali_job_descriptor_header {
- u32 exception_status;
- u32 first_incomplete_task;
- u64 fault_pointer;
- u8 job_descriptor_size : 1;
- enum mali_job_type job_type : 7;
- u8 job_barrier : 1;
- u8 unknown_flags : 7;
- u16 job_index;
- u16 job_dependency_index_1;
- u16 job_dependency_index_2;
- u64 next_job;
-} __attribute__((packed));
-
/* Details about write_value from panfrost igt tests which use it as a generic
* dword write primitive */
diff --git a/src/panfrost/lib/decode.c b/src/panfrost/lib/decode.c
index 58765f19df0..ca6f0441eaa 100644
--- a/src/panfrost/lib/decode.c
+++ b/src/panfrost/lib/decode.c
@@ -1324,16 +1324,16 @@ pandecode_primitive_size(union midgard_primitive_size u, bool constant)
}
static int
-pandecode_vertex_job_bfr(const struct mali_job_descriptor_header *h,
- const struct pandecode_mapped_memory *mem,
- mali_ptr payload, int job_no, unsigned gpu_id)
+pandecode_vertex_job_bfr(const struct MALI_JOB_HEADER *h,
+ const struct pandecode_mapped_memory *mem,
+ mali_ptr payload, int job_no, unsigned gpu_id)
{
struct bifrost_payload_vertex *PANDECODE_PTR_VAR(v, mem, payload);
struct mali_draw_packed draw_packed;
memcpy(&draw_packed, &v->postfix, sizeof(draw_packed));
pan_unpack(&draw_packed, DRAW, draw);
- pandecode_vertex_tiler_postfix_pre(&draw, job_no, h->job_type, "", true, gpu_id);
+ pandecode_vertex_tiler_postfix_pre(&draw, job_no, h->type, "", true, gpu_id);
pandecode_vertex_tiler_prefix(&v->prefix, job_no, false);
DUMP_CL(DRAW, &draw_packed, "Draw:\n");
@@ -1342,16 +1342,16 @@ pandecode_vertex_job_bfr(const struct mali_job_descriptor_header *h,
}
static int
-pandecode_tiler_job_bfr(const struct mali_job_descriptor_header *h,
- const struct pandecode_mapped_memory *mem,
- mali_ptr payload, int job_no, unsigned gpu_id)
+pandecode_tiler_job_bfr(const struct MALI_JOB_HEADER *h,
+ const struct pandecode_mapped_memory *mem,
+ mali_ptr payload, int job_no, unsigned gpu_id)
{
struct bifrost_payload_tiler *PANDECODE_PTR_VAR(t, mem, payload);
struct mali_draw_packed draw_packed;
memcpy(&draw_packed, &t->postfix, sizeof(draw_packed));
pan_unpack(&draw_packed, DRAW, draw);
- pandecode_vertex_tiler_postfix_pre(&draw, job_no, h->job_type, "", true, gpu_id);
+ pandecode_vertex_tiler_postfix_pre(&draw, job_no, h->type, "", true, gpu_id);
pandecode_bifrost_tiler(t->tiler_meta, job_no);
pandecode_vertex_tiler_prefix(&t->prefix, job_no, false);
@@ -1376,17 +1376,17 @@ pandecode_tiler_job_bfr(const struct mali_job_descriptor_header *h,
}
static int
-pandecode_vertex_or_tiler_job_mdg(const struct mali_job_descriptor_header *h,
- const struct pandecode_mapped_memory *mem,
- mali_ptr payload, int job_no, unsigned gpu_id)
+pandecode_vertex_or_tiler_job_mdg(const struct MALI_JOB_HEADER *h,
+ const struct pandecode_mapped_memory *mem,
+ mali_ptr payload, int job_no, unsigned gpu_id)
{
struct midgard_payload_vertex_tiler *PANDECODE_PTR_VAR(v, mem, payload);
- bool is_graphics = (h->job_type == MALI_JOB_TYPE_VERTEX) || (h->job_type == MALI_JOB_TYPE_TILER);
+ bool is_graphics = (h->type == MALI_JOB_TYPE_VERTEX) || (h->type == MALI_JOB_TYPE_TILER);
struct mali_draw_packed draw_packed;
memcpy(&draw_packed, &v->postfix, sizeof(draw_packed));
pan_unpack(&draw_packed, DRAW, draw);
- pandecode_vertex_tiler_postfix_pre(&draw, job_no, h->job_type, "", false, gpu_id);
+ pandecode_vertex_tiler_postfix_pre(&draw, job_no, h->type, "", false, gpu_id);
pandecode_vertex_tiler_prefix(&v->prefix, job_no, is_graphics);
DUMP_CL(DRAW, &draw_packed, "Draw:\n");
@@ -1504,65 +1504,30 @@ pandecode_jc(mali_ptr jc_gpu_va, bool bifrost, unsigned gpu_id, bool minimal)
{
pandecode_dump_file_open();
- struct mali_job_descriptor_header *h;
unsigned job_descriptor_number = 0;
+ mali_ptr next_job = 0;
do {
struct pandecode_mapped_memory *mem =
pandecode_find_mapped_gpu_mem_containing(jc_gpu_va);
-
void *payload;
- h = PANDECODE_PTR(mem, jc_gpu_va, struct mali_job_descriptor_header);
-
- mali_ptr payload_ptr = jc_gpu_va + sizeof(*h);
+ pan_unpack(PANDECODE_PTR(mem, jc_gpu_va, struct mali_job_header_packed),
+ JOB_HEADER, h);
+ next_job = h.next;
+ mali_ptr payload_ptr = jc_gpu_va + MALI_JOB_HEADER_LENGTH;
payload = pandecode_fetch_gpu_mem(mem, payload_ptr, 64);
int job_no = job_descriptor_number++;
/* If the job is good to go, skip it in minimal mode */
- if (minimal && (h->exception_status == 0x0 || h->exception_status == 0x1))
+ if (minimal && (h.exception_status == 0x0 || h.exception_status == 0x1))
continue;
- pandecode_log("struct mali_job_descriptor_header job_%"PRIx64"_%d = {\n", jc_gpu_va, job_no);
- pandecode_indent++;
+ DUMP_UNPACKED(JOB_HEADER, h, "Job Header:\n");
+ pandecode_log("\n");
- pandecode_prop("job_type = %s", mali_job_type_as_str(h->job_type));
-
- if (h->job_descriptor_size)
- pandecode_prop("job_descriptor_size = %d", h->job_descriptor_size);
-
- if (h->exception_status && h->exception_status != 0x1)
- pandecode_prop("exception_status = %x (source ID: 0x%x access: %s exception: 0x%x)",
- h->exception_status,
- (h->exception_status >> 16) & 0xFFFF,
- mali_exception_access_as_str((h->exception_status >> 8) & 0x3),
- h->exception_status & 0xFF);
-
- if (h->first_incomplete_task)
- pandecode_prop("first_incomplete_task = %d", h->first_incomplete_task);
-
- if (h->fault_pointer)
- pandecode_prop("fault_pointer = 0x%" PRIx64, h->fault_pointer);
-
- if (h->job_barrier)
- pandecode_prop("job_barrier = %d", h->job_barrier);
-
- pandecode_prop("job_index = %d", h->job_index);
-
- if (h->unknown_flags)
- pandecode_prop("unknown_flags = %d", h->unknown_flags);
-
- if (h->job_dependency_index_1)
- pandecode_prop("job_dependency_index_1 = %d", h->job_dependency_index_1);
-
- if (h->job_dependency_index_2)
- pandecode_prop("job_dependency_index_2 = %d", h->job_dependency_index_2);
-
- pandecode_indent--;
- pandecode_log("};\n");
-
- switch (h->job_type) {
+ switch (h.type) {
case MALI_JOB_TYPE_WRITE_VALUE: {
struct mali_payload_write_value *s = payload;
pandecode_log("struct mali_payload_write_value payload_%"PRIx64"_%d = {\n", payload_ptr, job_no);
@@ -1590,12 +1555,12 @@ pandecode_jc(mali_ptr jc_gpu_va, bool bifrost, unsigned gpu_id, bool minimal)
case MALI_JOB_TYPE_VERTEX:
case MALI_JOB_TYPE_COMPUTE:
if (bifrost) {
- if (h->job_type == MALI_JOB_TYPE_TILER)
- pandecode_tiler_job_bfr(h, mem, payload_ptr, job_no, gpu_id);
+ if (h.type == MALI_JOB_TYPE_TILER)
+ pandecode_tiler_job_bfr(&h, mem, payload_ptr, job_no, gpu_id);
else
- pandecode_vertex_job_bfr(h, mem, payload_ptr, job_no, gpu_id);
+ pandecode_vertex_job_bfr(&h, mem, payload_ptr, job_no, gpu_id);
} else
- pandecode_vertex_or_tiler_job_mdg(h, mem, payload_ptr, job_no, gpu_id);
+ pandecode_vertex_or_tiler_job_mdg(&h, mem, payload_ptr, job_no, gpu_id);
break;
@@ -1606,7 +1571,7 @@ pandecode_jc(mali_ptr jc_gpu_va, bool bifrost, unsigned gpu_id, bool minimal)
default:
break;
}
- } while ((jc_gpu_va = h->next_job));
+ } while ((jc_gpu_va = next_job));
pandecode_map_read_write();
}
diff --git a/src/panfrost/lib/midgard.xml b/src/panfrost/lib/midgard.xml
index 53aadda80e8..6f90e2867d5 100644
--- a/src/panfrost/lib/midgard.xml
+++ b/src/panfrost/lib/midgard.xml
@@ -1041,4 +1041,22 @@
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
diff --git a/src/panfrost/lib/pan_scoreboard.c b/src/panfrost/lib/pan_scoreboard.c
index c885dbae09d..6d564c957c8 100644
--- a/src/panfrost/lib/pan_scoreboard.c
+++ b/src/panfrost/lib/pan_scoreboard.c
@@ -133,22 +133,21 @@ panfrost_new_job(
/* Assign the index */
unsigned index = ++scoreboard->job_index;
- struct mali_job_descriptor_header job = {
- .job_descriptor_size = 1,
- .job_type = type,
- .job_barrier = barrier,
- .job_index = index,
- .job_dependency_index_1 = local_dep,
- .job_dependency_index_2 = global_dep,
- };
-
- if (inject)
- job.next_job = scoreboard->first_job;
-
struct panfrost_transfer transfer =
- panfrost_pool_alloc_aligned(pool, sizeof(job) + payload_size, 64);
- memcpy(transfer.cpu, &job, sizeof(job));
- memcpy(transfer.cpu + sizeof(job), payload, payload_size);
+ panfrost_pool_alloc_aligned(pool, MALI_JOB_HEADER_LENGTH + payload_size, 64);
+
+ pan_pack(transfer.cpu, JOB_HEADER, job) {
+ job.type = type;
+ job.barrier = barrier;
+ job.index = index;
+ job.dependency_1 = local_dep;
+ job.dependency_2 = global_dep;
+
+ if (inject)
+ job.next = scoreboard->first_job;
+ }
+
+ memcpy(transfer.cpu + MALI_JOB_HEADER_LENGTH, payload, payload_size);
if (inject) {
scoreboard->first_job = transfer.gpu;
@@ -159,12 +158,19 @@ panfrost_new_job(
if (type == MALI_JOB_TYPE_TILER)
scoreboard->tiler_dep = index;
- if (scoreboard->prev_job)
- scoreboard->prev_job->next_job = transfer.gpu;
- else
+ if (scoreboard->prev_job) {
+ /* Manual update of the next pointer. This is bad, don't copy
+ * this pattern.
+ * TODO: Find a way to defer last job header emission until we
+ * have a new job to queue or the batch is ready for execution.
+ */
+ scoreboard->prev_job->opaque[6] = transfer.gpu;
+ scoreboard->prev_job->opaque[7] = transfer.gpu >> 32;
+ } else {
scoreboard->first_job = transfer.gpu;
+ }
- scoreboard->prev_job = (struct mali_job_descriptor_header *) transfer.cpu;
+ scoreboard->prev_job = (struct mali_job_header_packed *)transfer.cpu;
return index;
}
@@ -183,21 +189,23 @@ panfrost_scoreboard_initialize_tiler(struct pan_pool *pool,
/* Okay, we do. Let's generate it. We'll need the job's polygon list
* regardless of size. */
- struct mali_job_descriptor_header job = {
- .job_type = MALI_JOB_TYPE_WRITE_VALUE,
- .job_index = scoreboard->write_value_index,
- .job_descriptor_size = 1,
- .next_job = scoreboard->first_job
- };
-
struct mali_payload_write_value payload = {
.address = polygon_list,
.value_descriptor = MALI_WRITE_VALUE_ZERO,
};
- struct panfrost_transfer transfer = panfrost_pool_alloc_aligned(pool, sizeof(job) + sizeof(payload), 64);
- memcpy(transfer.cpu, &job, sizeof(job));
- memcpy(transfer.cpu + sizeof(job), &payload, sizeof(payload));
+ struct panfrost_transfer transfer =
+ panfrost_pool_alloc_aligned(pool,
+ MALI_JOB_HEADER_LENGTH + sizeof(payload),
+ 64);
+
+ pan_pack(transfer.cpu, JOB_HEADER, job) {
+ job.type = MALI_JOB_TYPE_WRITE_VALUE;
+ job.index = scoreboard->write_value_index;
+ job.next = scoreboard->first_job;
+ }
+
+ memcpy(transfer.cpu + MALI_JOB_HEADER_LENGTH, &payload, sizeof(payload));
scoreboard->first_job = transfer.gpu;
}
diff --git a/src/panfrost/lib/pan_scoreboard.h b/src/panfrost/lib/pan_scoreboard.h
index dcf12410ede..53c65d43764 100644
--- a/src/panfrost/lib/pan_scoreboard.h
+++ b/src/panfrost/lib/pan_scoreboard.h
@@ -38,7 +38,7 @@ struct pan_scoreboard {
unsigned job_index;
/* A CPU-side pointer to the previous job for next_job linking */
- struct mali_job_descriptor_header *prev_job;
+ struct mali_job_header_packed *prev_job;
/* The dependency for tiler jobs (i.e. the index of the last emitted
* tiler job, or zero if none have been emitted) */