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

panvk/csf: factor out cs_match_iter_sb helper macro

Browse Source 
This simplifies cases where we need to match on all of the possible
iter_sb values, which occurs frequently.

Signed-off-by: Olivia Lee <olivia.lee@collabora.com>
Reviewed-by: Lars-Ivar Hesselberg Simonsen <lars-ivar.simonsen@arm.com>
Acked-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38496>
This commit is contained in:
Olivia Lee
2025-11-17 11:41:52 -08:00
committed by Marge Bot
parent 298ad17b81
commit 5c6c4cbebd
5 changed files with 77 additions and 118 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/panfrost/vulkan/csf/panvk_cmd_buffer.h
+9
View File
@@ -470,6 +470,15 @@ extern const struct vk_command_buffer_ops panvk_per_arch(cmd_buffer_ops);
void panvk_per_arch(cmd_flush_draws)(struct panvk_cmd_buffer *cmdbuf);
#if PAN_ARCH == 10
/* Match against all possible iter_sb values. The constant iter_sb value for
* the current match arm is in '__val'. */
#define cs_match_iter_sb(__b, __val, __iter_sb, __scratch) \
cs_match(__b, __iter_sb, __scratch) \
for (unsigned __val = 0; __val < PANVK_SB_ITER_COUNT; __val++) \
cs_case(__b, SB_ITER(__val))
#endif
void panvk_per_arch(cs_next_iter_sb)(struct panvk_cmd_buffer *cmdbuf,
enum panvk_subqueue_id subqueue,
struct cs_index scratch_regs);
src/panfrost/vulkan/csf/panvk_vX_cmd_buffer.c
+3 -13
View File
@@ -808,19 +808,9 @@ panvk_per_arch(cs_next_iter_sb)(struct panvk_cmd_buffer *cmdbuf,
cs_move32_to(b, iter_sb, SB_ITER(0));
}
cs_match(b, iter_sb, cmp_scratch) {
#define CASE(x) \
cs_case(b, SB_ITER(x)) { \
cs_wait_slot(b, SB_ITER(x)); \
cs_select_sb_entries_for_async_ops(b, SB_ITER(x)); \
}
CASE(0)
CASE(1)
CASE(2)
CASE(3)
CASE(4)
#undef CASE
cs_match_iter_sb(b, x, iter_sb, cmp_scratch) {
cs_wait_slot(b, SB_ITER(x));
cs_select_sb_entries_for_async_ops(b, SB_ITER(x));
}
cs_store32(b, iter_sb, cs_subqueue_ctx_reg(b),
src/panfrost/vulkan/csf/panvk_vX_cmd_dispatch.c
+4 -14
View File
@@ -330,20 +330,10 @@ cmd_dispatch(struct panvk_cmd_buffer *cmdbuf, struct panvk_dispatch_info *info)
PANVK_SUBQUEUE_COMPUTE * sizeof(struct panvk_cs_sync64));
cs_move64_to(b, add_val, 1);
cs_match(b, iter_sb, cmp_scratch) {
#define CASE(x) \
cs_case(b, SB_ITER(x)) { \
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)
CASE(1)
CASE(2)
CASE(3)
CASE(4)
#undef CASE
cs_match_iter_sb(b, x, iter_sb, cmp_scratch) {
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)));
}
#endif
src/panfrost/vulkan/csf/panvk_vX_cmd_draw.c
+56 -76
View File
@@ -2979,22 +2979,12 @@ flush_tiling(struct panvk_cmd_buffer *cmdbuf)
cs_move64_to(b, add_val, 1);
cs_match(b, iter_sb, cmp_scratch) {
#define CASE(x) \
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))); \
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)
CASE(1)
CASE(2)
CASE(3)
CASE(4)
#undef CASE
cs_match_iter_sb(b, x, iter_sb, cmp_scratch) {
cs_heap_operation(b, MALI_CS_HEAP_OPERATION_VERTEX_TILER_COMPLETED,
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)));
}
#endif
@@ -3358,66 +3348,56 @@ issue_fragment_jobs(struct panvk_cmd_buffer *cmdbuf)
cs_add64(b, sync_addr, sync_addr,
PANVK_SUBQUEUE_FRAGMENT * sizeof(struct panvk_cs_sync64));
cs_match(b, iter_sb, cmp_scratch) {
#define CASE(x) \
cs_case(b, SB_ITER(x)) { \
const struct cs_async_op async = \
cs_defer(SB_WAIT_ITER(x), SB_ID(DEFERRED_SYNC)); \
if (td_count == 1) { \
cs_load_to(b, completed, cur_tiler, BITFIELD_MASK(4), 40); \
cs_finish_fragment(b, true, completed_top, completed_bottom, async); \
} 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, \
async); \
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, async); \
} \
if (free_render_descs) { \
cs_sync32_add(b, true, MALI_CS_SYNC_SCOPE_CSG, release_sz, \
ringbuf_sync_addr, async); \
} \
if (has_oq_chain) { \
struct cs_index flush_id = oq_chain_lo; \
cs_move32_to(b, flush_id, 0); \
cs_flush_caches(b, MALI_CS_FLUSH_MODE_CLEAN, \
MALI_CS_FLUSH_MODE_CLEAN, \
MALI_CS_OTHER_FLUSH_MODE_NONE, flush_id, \
cs_defer(SB_WAIT_ITER(x), SB_ID(DEFERRED_FLUSH))); \
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_single_link_list_for_each_from( \
b, oq_chain, struct panvk_cs_occlusion_query, node) { \
cs_load64_to(b, oq_syncobj, oq_chain, \
offsetof(struct panvk_cs_occlusion_query, syncobj)); \
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))); \
} \
} \
panvk_instr_sync64_add(cmdbuf, PANVK_SUBQUEUE_FRAGMENT, true, \
MALI_CS_SYNC_SCOPE_CSG, add_val, sync_addr, \
async); \
}
CASE(0)
CASE(1)
CASE(2)
CASE(3)
CASE(4)
#undef CASE
cs_match_iter_sb(b, x, iter_sb, cmp_scratch) {
const struct cs_async_op async =
cs_defer(SB_WAIT_ITER(x), SB_ID(DEFERRED_SYNC));
if (td_count == 1) {
cs_load_to(b, completed, cur_tiler, BITFIELD_MASK(4), 40);
cs_finish_fragment(b, true, completed_top, completed_bottom, async);
} 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,
async);
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, async);
}
if (free_render_descs) {
cs_sync32_add(b, true, MALI_CS_SYNC_SCOPE_CSG, release_sz,
ringbuf_sync_addr, async);
}
if (has_oq_chain) {
struct cs_index flush_id = oq_chain_lo;
cs_move32_to(b, flush_id, 0);
cs_flush_caches(b, MALI_CS_FLUSH_MODE_CLEAN,
MALI_CS_FLUSH_MODE_CLEAN,
MALI_CS_OTHER_FLUSH_MODE_NONE, flush_id,
cs_defer(SB_WAIT_ITER(x), SB_ID(DEFERRED_FLUSH)));
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_single_link_list_for_each_from(
b, oq_chain, struct panvk_cs_occlusion_query, node) {
cs_load64_to(b, oq_syncobj, oq_chain,
offsetof(struct panvk_cs_occlusion_query, syncobj));
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)));
}
}
panvk_instr_sync64_add(cmdbuf, PANVK_SUBQUEUE_FRAGMENT, true,
MALI_CS_SYNC_SCOPE_CSG, add_val, sync_addr,
async);
}
#endif
src/panfrost/vulkan/csf/panvk_vX_cmd_precomp.c
+5 -15
View File
@@ -150,21 +150,11 @@ panvk_per_arch(dispatch_precomp)(struct panvk_precomp_ctx *ctx,
PANVK_SUBQUEUE_COMPUTE * sizeof(struct panvk_cs_sync64));
cs_move64_to(b, add_val, 1);
cs_match(b, iter_sb, cmp_scratch) {
#define CASE(x) \
cs_case(b, SB_ITER(x)) { \
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)
CASE(1)
CASE(2)
CASE(3)
CASE(4)
#undef CASE
cs_match_iter_sb(b, x, iter_sb, cmp_scratch) {
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)));
}
#endif