fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-04 20:38:06 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

panvk: Implement v11+ path for issue_fragment_jobs

Browse source 
Signed-off-by: Mary Guillemard <mary.guillemard@collabora.com>
Reviewed-by: Lars-Ivar Hesselberg Simonsen <lars-ivar.simonsen@arm.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35307>
This commit is contained in:
Mary Guillemard 2025-06-03 10:59:47 +02:00 • committed by Marge Bot
parent 6207426706
commit cf8c71c0a0
1 changed files with 80 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

86
src/panfrost/vulkan/csf/panvk_vX_cmd_draw.c
View file

@ -3099,8 +3099,6 @@ issue_fragment_jobs(struct panvk_cmd_buffer *cmdbuf)
}
struct cs_index sync_addr = cs_scratch_reg64(b, 0);
struct cs_index iter_sb = cs_scratch_reg32(b, 2);
struct cs_index cmp_scratch = cs_scratch_reg32(b, 3);
struct cs_index add_val = cs_scratch_reg64(b, 4);
struct cs_index add_val_lo = cs_scratch_reg32(b, 4);
struct cs_index ringbuf_sync_addr = cs_scratch_reg64(b, 6);
@ -3117,9 +3115,6 @@ issue_fragment_jobs(struct panvk_cmd_buffer *cmdbuf)
struct cs_index oq_syncobj = cs_scratch_reg64(b, 12);
cs_move64_to(b, add_val, 1);
cs_load_to(b, cs_scratch_reg_tuple(b, 0, 3), cs_subqueue_ctx_reg(b),
BITFIELD_MASK(3),
offsetof(struct panvk_cs_subqueue_context, syncobjs));
if (free_render_descs) {
cs_move32_to(b, release_sz, calc_render_descs_size(cmdbuf));
@ -3128,9 +3123,87 @@ issue_fragment_jobs(struct panvk_cmd_buffer *cmdbuf)
render.desc_ringbuf.syncobj));
}
cs_move32_to(b, tiler_count, td_count);
#if PAN_ARCH >= 11
cs_load64_to(b, sync_addr, cs_subqueue_ctx_reg(b),
offsetof(struct panvk_cs_subqueue_context, syncobjs));
cs_add64(b, sync_addr, sync_addr,
PANVK_SUBQUEUE_FRAGMENT * sizeof(struct panvk_cs_sync64));
if (td_count == 1) {
cs_load_to(b, completed, cur_tiler, BITFIELD_MASK(4), 40);
cs_finish_fragment(b, true, completed_top, completed_bottom,
cs_defer_indirect());
} else if (td_count > 1) {
cs_while(b, MALI_CS_CONDITION_GREATER, tiler_count) {
cs_load_to(b, completed, cur_tiler, BITFIELD_MASK(4), 40);
cs_finish_fragment(b, false, completed_top, completed_bottom,
cs_defer_indirect());
cs_update_frag_ctx(b)
cs_add64(b, cur_tiler, cur_tiler, pan_size(TILER_CONTEXT));
cs_add32(b, tiler_count, tiler_count, -1);
}
cs_frag_end(b, cs_defer_indirect());
}
if (free_render_descs) {
cs_sync32_add(b, true, MALI_CS_SYNC_SCOPE_CSG, release_sz,
ringbuf_sync_addr, cs_defer_indirect());
}
if (has_oq_chain) {
struct cs_index flush_id = oq_chain_lo;
cs_move32_to(b, flush_id, 0);
/* FLUSH_CACHE2 is part of the deferred group so we need to
* temporarily set DEFERRED_FLUSH here to use the right scoreboard in
* indirect mode */
cs_set_state_imm32(b, MALI_CS_SET_STATE_TYPE_SB_SEL_DEFERRED,
SB_ID(DEFERRED_FLUSH));
cs_flush_caches(b, MALI_CS_FLUSH_MODE_CLEAN, MALI_CS_FLUSH_MODE_CLEAN,
MALI_CS_OTHER_FLUSH_MODE_NONE, flush_id,
cs_defer_indirect());
cs_set_state_imm32(b, MALI_CS_SET_STATE_TYPE_SB_SEL_DEFERRED,
SB_ID(DEFERRED_SYNC));
cs_load64_to(b, oq_chain, cs_subqueue_ctx_reg(b),
offsetof(struct panvk_cs_subqueue_context, render.oq_chain));
/* For WAR dependency on subqueue_context.render.oq_chain. */
cs_flush_loads(b);
/* We use oq_syncobj as a placeholder to reset the oq_chain. */
cs_move64_to(b, oq_syncobj, 0);
cs_store64(b, oq_syncobj, cs_subqueue_ctx_reg(b),
offsetof(struct panvk_cs_subqueue_context, render.oq_chain));
cs_while(b, MALI_CS_CONDITION_ALWAYS, cs_undef()) {
cs_load64_to(b, oq_syncobj, oq_chain,
offsetof(struct panvk_cs_occlusion_query, syncobj));
cs_load64_to(b, oq_chain, oq_chain,
offsetof(struct panvk_cs_occlusion_query, next));
cs_sync32_set(b, true, MALI_CS_SYNC_SCOPE_CSG, add_val_lo, oq_syncobj,
cs_defer(SB_MASK(DEFERRED_FLUSH), SB_ID(DEFERRED_SYNC)));
cs_if(b, MALI_CS_CONDITION_NEQUAL, oq_chain_lo)
cs_continue(b);
cs_if(b, MALI_CS_CONDITION_NEQUAL, oq_chain_hi)
cs_continue(b);
cs_break(b);
}
}
cs_sync64_add(b, 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);
cs_load_to(b, cs_scratch_reg_tuple(b, 0, 3), cs_subqueue_ctx_reg(b),
BITFIELD_MASK(3),
offsetof(struct panvk_cs_subqueue_context, syncobjs));
cs_add64(b, sync_addr, sync_addr,
PANVK_SUBQUEUE_FRAGMENT * sizeof(struct panvk_cs_sync64));
cs_move32_to(b, tiler_count, td_count);
cs_match(b, iter_sb, cmp_scratch) {
#define CASE(x) \
@ -3198,6 +3271,7 @@ issue_fragment_jobs(struct panvk_cmd_buffer *cmdbuf)
CASE(4)
#undef CASE
}
#endif
/* Update the ring buffer position. */
if (free_render_descs) {