aboutsummaryrefslogtreecommitdiffstats
path: root/src/gallium
diff options
context:
space:
mode:
authorPierre-Eric Pelloux-Prayer <[email protected]>2020-05-29 11:59:50 +0200
committerPierre-Eric Pelloux-Prayer <[email protected]>2020-06-30 09:23:14 +0200
commit5a05f9714bb5c78c01de39d2e6dc1ac5b351f8f1 (patch)
tree72eec38659db52936d706f7ae2d11be59a10af26 /src/gallium
parent7e8cfc0add8e6b28ca4babacd6263b23ad87b2fe (diff)
radeonsi: bump SI_NUM_SHADER_BUFFERS to 32
Some app uses more than 8 SSBOs (https://gitlab.freedesktop.org/mesa/mesa/-/issues/2946), so increase SI_NUM_SHADER_BUFFERS to 32 (which allows 16 SSBOs). Since we're now using a 64 bits number to track buffers, we could bump SI_NUM_SHADER_BUFFERS to 48 but that would conflict with Mesa's MAX_COMBINED_ATOMIC_BUFFERS limit (= 90). Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/2122 Reviewed-by: Marek Olšák <[email protected]> Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5632>
Diffstat (limited to 'src/gallium')
-rw-r--r--src/gallium/drivers/radeonsi/si_debug.c9
-rw-r--r--src/gallium/drivers/radeonsi/si_descriptors.c48
-rw-r--r--src/gallium/drivers/radeonsi/si_shader.h2
-rw-r--r--src/gallium/drivers/radeonsi/si_state.h12
-rw-r--r--src/gallium/drivers/radeonsi/si_state_shaders.c4
5 files changed, 40 insertions, 35 deletions
diff --git a/src/gallium/drivers/radeonsi/si_debug.c b/src/gallium/drivers/radeonsi/si_debug.c
index f596d51e0f6..4b22f4a7aed 100644
--- a/src/gallium/drivers/radeonsi/si_debug.c
+++ b/src/gallium/drivers/radeonsi/si_debug.c
@@ -796,8 +796,13 @@ static void si_dump_descriptors(struct si_context *sctx, enum pipe_shader_type p
enabled_constbuf =
sctx->const_and_shader_buffers[processor].enabled_mask >> SI_NUM_SHADER_BUFFERS;
enabled_shaderbuf = sctx->const_and_shader_buffers[processor].enabled_mask &
- u_bit_consecutive(0, SI_NUM_SHADER_BUFFERS);
- enabled_shaderbuf = util_bitreverse(enabled_shaderbuf) >> (32 - SI_NUM_SHADER_BUFFERS);
+ u_bit_consecutive64(0, SI_NUM_SHADER_BUFFERS);
+ enabled_shaderbuf = 0;
+ for (int i = 0; i < SI_NUM_SHADER_BUFFERS; i++) {
+ enabled_shaderbuf |=
+ (sctx->const_and_shader_buffers[processor].enabled_mask &
+ 1llu << (SI_NUM_SHADER_BUFFERS - i - 1)) << i;
+ }
enabled_samplers = sctx->samplers[processor].enabled_mask;
enabled_images = sctx->images[processor].enabled_mask;
}
diff --git a/src/gallium/drivers/radeonsi/si_descriptors.c b/src/gallium/drivers/radeonsi/si_descriptors.c
index 865f0b1b771..33d5d5fe063 100644
--- a/src/gallium/drivers/radeonsi/si_descriptors.c
+++ b/src/gallium/drivers/radeonsi/si_descriptors.c
@@ -897,12 +897,12 @@ void si_update_ps_colorbuf0_slot(struct si_context *sctx)
pipe_resource_reference(&buffers->buffers[slot], &tex->buffer.b.b);
radeon_add_to_buffer_list(sctx, sctx->gfx_cs, &tex->buffer, RADEON_USAGE_READ,
RADEON_PRIO_SHADER_RW_IMAGE);
- buffers->enabled_mask |= 1u << slot;
+ buffers->enabled_mask |= 1llu << slot;
} else {
/* Clear the descriptor. */
memset(descs->list + slot * 4, 0, 8 * 4);
pipe_resource_reference(&buffers->buffers[slot], NULL);
- buffers->enabled_mask &= ~(1u << slot);
+ buffers->enabled_mask &= ~(1llu << slot);
}
sctx->descriptors_dirty |= 1u << SI_DESCS_RW_BUFFERS;
@@ -985,15 +985,15 @@ static void si_release_buffer_resources(struct si_buffer_resources *buffers,
static void si_buffer_resources_begin_new_cs(struct si_context *sctx,
struct si_buffer_resources *buffers)
{
- unsigned mask = buffers->enabled_mask;
+ uint64_t mask = buffers->enabled_mask;
/* Add buffers to the CS. */
while (mask) {
- int i = u_bit_scan(&mask);
+ int i = u_bit_scan64(&mask);
radeon_add_to_buffer_list(
sctx, sctx->gfx_cs, si_resource(buffers->buffers[i]),
- buffers->writable_mask & (1u << i) ? RADEON_USAGE_READWRITE : RADEON_USAGE_READ,
+ buffers->writable_mask & (1llu << i) ? RADEON_USAGE_READWRITE : RADEON_USAGE_READ,
i < SI_NUM_SHADER_BUFFERS ? buffers->priority : buffers->priority_constbuf);
}
}
@@ -1001,13 +1001,13 @@ static void si_buffer_resources_begin_new_cs(struct si_context *sctx,
static bool si_buffer_resources_check_encrypted(struct si_context *sctx,
struct si_buffer_resources *buffers)
{
- unsigned mask = buffers->enabled_mask;
+ uint64_t mask = buffers->enabled_mask;
while (mask) {
- int i = u_bit_scan(&mask);
+ int i = u_bit_scan64(&mask);
/* only check for reads */
- if ((buffers->writable_mask & (1u << i)) == 0 &&
+ if ((buffers->writable_mask & (1llu << i)) == 0 &&
(si_resource(buffers->buffers[i])->flags & RADEON_FLAG_ENCRYPTED))
return true;
}
@@ -1238,11 +1238,11 @@ static void si_set_constant_buffer(struct si_context *sctx, struct si_buffer_res
buffers->offsets[slot] = buffer_offset;
radeon_add_to_gfx_buffer_list_check_mem(sctx, si_resource(buffer), RADEON_USAGE_READ,
buffers->priority_constbuf, true);
- buffers->enabled_mask |= 1u << slot;
+ buffers->enabled_mask |= 1llu << slot;
} else {
/* Clear the descriptor. */
memset(descs->list + slot * 4, 0, sizeof(uint32_t) * 4);
- buffers->enabled_mask &= ~(1u << slot);
+ buffers->enabled_mask &= ~(1llu << slot);
}
sctx->descriptors_dirty |= 1u << descriptors_idx;
@@ -1292,8 +1292,8 @@ static void si_set_shader_buffer(struct si_context *sctx, struct si_buffer_resou
if (!sbuffer || !sbuffer->buffer) {
pipe_resource_reference(&buffers->buffers[slot], NULL);
memset(desc, 0, sizeof(uint32_t) * 4);
- buffers->enabled_mask &= ~(1u << slot);
- buffers->writable_mask &= ~(1u << slot);
+ buffers->enabled_mask &= ~(1llu << slot);
+ buffers->writable_mask &= ~(1llu << slot);
sctx->descriptors_dirty |= 1u << descriptors_idx;
return;
}
@@ -1320,12 +1320,12 @@ static void si_set_shader_buffer(struct si_context *sctx, struct si_buffer_resou
radeon_add_to_gfx_buffer_list_check_mem(
sctx, buf, writable ? RADEON_USAGE_READWRITE : RADEON_USAGE_READ, priority, true);
if (writable)
- buffers->writable_mask |= 1u << slot;
+ buffers->writable_mask |= 1llu << slot;
else
- buffers->writable_mask &= ~(1u << slot);
+ buffers->writable_mask &= ~(1llu << slot);
- buffers->enabled_mask |= 1u << slot;
- sctx->descriptors_dirty |= 1u << descriptors_idx;
+ buffers->enabled_mask |= 1llu << slot;
+ sctx->descriptors_dirty |= 1lu << descriptors_idx;
util_range_add(&buf->b.b, &buf->valid_buffer_range, sbuffer->buffer_offset,
sbuffer->buffer_offset + sbuffer->buffer_size);
@@ -1469,11 +1469,11 @@ void si_set_ring_buffer(struct si_context *sctx, uint slot, struct pipe_resource
pipe_resource_reference(&buffers->buffers[slot], buffer);
radeon_add_to_buffer_list(sctx, sctx->gfx_cs, si_resource(buffer), RADEON_USAGE_READWRITE,
buffers->priority);
- buffers->enabled_mask |= 1u << slot;
+ buffers->enabled_mask |= 1llu << slot;
} else {
/* Clear the descriptor. */
memset(descs->list + slot * 4, 0, sizeof(uint32_t) * 4);
- buffers->enabled_mask &= ~(1u << slot);
+ buffers->enabled_mask &= ~(1llu << slot);
}
sctx->descriptors_dirty |= 1u << SI_DESCS_RW_BUFFERS;
@@ -1557,14 +1557,14 @@ void si_update_needs_color_decompress_masks(struct si_context *sctx)
* If buf == NULL, reset all descriptors.
*/
static void si_reset_buffer_resources(struct si_context *sctx, struct si_buffer_resources *buffers,
- unsigned descriptors_idx, unsigned slot_mask,
+ unsigned descriptors_idx, uint64_t slot_mask,
struct pipe_resource *buf, enum radeon_bo_priority priority)
{
struct si_descriptors *descs = &sctx->descriptors[descriptors_idx];
- unsigned mask = buffers->enabled_mask & slot_mask;
+ uint64_t mask = buffers->enabled_mask & slot_mask;
while (mask) {
- unsigned i = u_bit_scan(&mask);
+ unsigned i = u_bit_scan64(&mask);
struct pipe_resource *buffer = buffers->buffers[i];
if (buffer && (!buf || buffer == buf)) {
@@ -1573,7 +1573,7 @@ static void si_reset_buffer_resources(struct si_context *sctx, struct si_buffer_
radeon_add_to_gfx_buffer_list_check_mem(
sctx, si_resource(buffer),
- buffers->writable_mask & (1u << i) ? RADEON_USAGE_READWRITE : RADEON_USAGE_READ,
+ buffers->writable_mask & (1llu << i) ? RADEON_USAGE_READWRITE : RADEON_USAGE_READ,
priority, true);
}
}
@@ -1646,7 +1646,7 @@ void si_rebind_buffer(struct si_context *sctx, struct pipe_resource *buf)
for (shader = 0; shader < SI_NUM_SHADERS; shader++)
si_reset_buffer_resources(sctx, &sctx->const_and_shader_buffers[shader],
si_const_and_shader_buffer_descriptors_idx(shader),
- u_bit_consecutive(SI_NUM_SHADER_BUFFERS, SI_NUM_CONST_BUFFERS),
+ u_bit_consecutive64(SI_NUM_SHADER_BUFFERS, SI_NUM_CONST_BUFFERS),
buf, sctx->const_and_shader_buffers[shader].priority_constbuf);
}
@@ -1654,7 +1654,7 @@ void si_rebind_buffer(struct si_context *sctx, struct pipe_resource *buf)
for (shader = 0; shader < SI_NUM_SHADERS; shader++)
si_reset_buffer_resources(sctx, &sctx->const_and_shader_buffers[shader],
si_const_and_shader_buffer_descriptors_idx(shader),
- u_bit_consecutive(0, SI_NUM_SHADER_BUFFERS), buf,
+ u_bit_consecutive64(0, SI_NUM_SHADER_BUFFERS), buf,
sctx->const_and_shader_buffers[shader].priority);
}
diff --git a/src/gallium/drivers/radeonsi/si_shader.h b/src/gallium/drivers/radeonsi/si_shader.h
index 756bcd5a90e..b46a48380e0 100644
--- a/src/gallium/drivers/radeonsi/si_shader.h
+++ b/src/gallium/drivers/radeonsi/si_shader.h
@@ -482,7 +482,7 @@ struct si_shader_selector {
uint64_t inputs_read; /* "get_unique_index" bits */
/* bitmasks of used descriptor slots */
- uint32_t active_const_and_shader_buffers;
+ uint64_t active_const_and_shader_buffers;
uint64_t active_samplers_and_images;
};
diff --git a/src/gallium/drivers/radeonsi/si_state.h b/src/gallium/drivers/radeonsi/si_state.h
index 7e1c81d1dad..a45b18c827c 100644
--- a/src/gallium/drivers/radeonsi/si_state.h
+++ b/src/gallium/drivers/radeonsi/si_state.h
@@ -37,7 +37,7 @@
#define SI_NUM_CONST_BUFFERS 16
#define SI_NUM_IMAGES 16
#define SI_NUM_IMAGE_SLOTS (SI_NUM_IMAGES * 2) /* the second half are FMASK slots */
-#define SI_NUM_SHADER_BUFFERS 16
+#define SI_NUM_SHADER_BUFFERS 32
struct si_screen;
struct si_shader;
@@ -457,8 +457,8 @@ struct si_buffer_resources {
enum radeon_bo_priority priority_constbuf : 6;
/* The i-th bit is set if that element is enabled (non-NULL resource). */
- unsigned enabled_mask;
- unsigned writable_mask;
+ uint64_t enabled_mask;
+ uint64_t writable_mask;
};
#define si_pm4_state_changed(sctx, member) \
@@ -571,7 +571,7 @@ void si_schedule_initial_compile(struct si_context *sctx, unsigned processor,
struct util_queue_fence *ready_fence,
struct si_compiler_ctx_state *compiler_ctx_state, void *job,
util_queue_execute_func execute);
-void si_get_active_slot_masks(const struct si_shader_info *info, uint32_t *const_and_shader_buffers,
+void si_get_active_slot_masks(const struct si_shader_info *info, uint64_t *const_and_shader_buffers,
uint64_t *samplers_and_images);
int si_shader_select_with_key(struct si_screen *sscreen, struct si_shader_ctx_state *state,
struct si_compiler_ctx_state *compiler_state,
@@ -602,13 +602,13 @@ void si_init_streamout_functions(struct si_context *sctx);
static inline unsigned si_get_constbuf_slot(unsigned slot)
{
- /* Constant buffers are in slots [16..31], ascending */
+ /* Constant buffers are in slots [32..47], ascending */
return SI_NUM_SHADER_BUFFERS + slot;
}
static inline unsigned si_get_shaderbuf_slot(unsigned slot)
{
- /* shader buffers are in slots [15..0], descending */
+ /* shader buffers are in slots [31..0], descending */
return SI_NUM_SHADER_BUFFERS - 1 - slot;
}
diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c b/src/gallium/drivers/radeonsi/si_state_shaders.c
index cd14fdc741e..5945a47b167 100644
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -2542,7 +2542,7 @@ void si_schedule_initial_compile(struct si_context *sctx, unsigned processor,
}
/* Return descriptor slot usage masks from the given shader info. */
-void si_get_active_slot_masks(const struct si_shader_info *info, uint32_t *const_and_shader_buffers,
+void si_get_active_slot_masks(const struct si_shader_info *info, uint64_t *const_and_shader_buffers,
uint64_t *samplers_and_images)
{
unsigned start, num_shaderbufs, num_constbufs, num_images, num_msaa_images, num_samplers;
@@ -2556,7 +2556,7 @@ void si_get_active_slot_masks(const struct si_shader_info *info, uint32_t *const
/* The layout is: sb[last] ... sb[0], cb[0] ... cb[last] */
start = si_get_shaderbuf_slot(num_shaderbufs - 1);
- *const_and_shader_buffers = u_bit_consecutive(start, num_shaderbufs + num_constbufs);
+ *const_and_shader_buffers = u_bit_consecutive64(start, num_shaderbufs + num_constbufs);
/* The layout is:
* - fmask[last] ... fmask[0] go to [15-last .. 15]