[Mesa-dev] [PATCH] radv: merge tess rings into a single bo
Dave Airlie
airlied at gmail.com
Sun Feb 25 23:28:26 UTC 2018
From: Dave Airlie <airlied at redhat.com>
Inspired by a passing commit to radeonsi.
---
src/amd/vulkan/radv_device.c | 93 ++++++++++++++++++-------------------------
src/amd/vulkan/radv_private.h | 3 +-
2 files changed, 40 insertions(+), 56 deletions(-)
diff --git a/src/amd/vulkan/radv_device.c b/src/amd/vulkan/radv_device.c
index 98cfd9b58a9..9cf4564b115 100644
--- a/src/amd/vulkan/radv_device.c
+++ b/src/amd/vulkan/radv_device.c
@@ -1049,10 +1049,8 @@ radv_queue_finish(struct radv_queue *queue)
queue->device->ws->buffer_destroy(queue->esgs_ring_bo);
if (queue->gsvs_ring_bo)
queue->device->ws->buffer_destroy(queue->gsvs_ring_bo);
- if (queue->tess_factor_ring_bo)
- queue->device->ws->buffer_destroy(queue->tess_factor_ring_bo);
- if (queue->tess_offchip_ring_bo)
- queue->device->ws->buffer_destroy(queue->tess_offchip_ring_bo);
+ if (queue->tess_rings_bo)
+ queue->device->ws->buffer_destroy(queue->tess_rings_bo);
if (queue->compute_scratch_bo)
queue->device->ws->buffer_destroy(queue->compute_scratch_bo);
}
@@ -1366,22 +1364,22 @@ fill_geom_tess_rings(struct radv_queue *queue,
uint32_t gsvs_ring_size,
struct radeon_winsys_bo *gsvs_ring_bo,
uint32_t tess_factor_ring_size,
- struct radeon_winsys_bo *tess_factor_ring_bo,
+ uint32_t tess_offchip_ring_offset,
uint32_t tess_offchip_ring_size,
- struct radeon_winsys_bo *tess_offchip_ring_bo)
+ struct radeon_winsys_bo *tess_rings_bo)
{
uint64_t esgs_va = 0, gsvs_va = 0;
- uint64_t tess_factor_va = 0, tess_offchip_va = 0;
+ uint64_t tess_va = 0, tess_offchip_va = 0;
uint32_t *desc = &map[4];
if (esgs_ring_bo)
esgs_va = radv_buffer_get_va(esgs_ring_bo);
if (gsvs_ring_bo)
gsvs_va = radv_buffer_get_va(gsvs_ring_bo);
- if (tess_factor_ring_bo)
- tess_factor_va = radv_buffer_get_va(tess_factor_ring_bo);
- if (tess_offchip_ring_bo)
- tess_offchip_va = radv_buffer_get_va(tess_offchip_ring_bo);
+ if (tess_rings_bo) {
+ tess_va = radv_buffer_get_va(tess_rings_bo);
+ tess_offchip_va = tess_va + tess_offchip_ring_offset;
+ }
/* stride 0, num records - size, add tid, swizzle, elsize4,
index stride 64 */
@@ -1458,8 +1456,8 @@ fill_geom_tess_rings(struct radv_queue *queue,
S_008F0C_ADD_TID_ENABLE(true);
desc += 4;
- desc[0] = tess_factor_va;
- desc[1] = S_008F04_BASE_ADDRESS_HI(tess_factor_va >> 32) |
+ desc[0] = tess_va;
+ desc[1] = S_008F04_BASE_ADDRESS_HI(tess_va >> 32) |
S_008F04_STRIDE(0) |
S_008F04_SWIZZLE_ENABLE(false);
desc[2] = tess_factor_ring_size;
@@ -1568,13 +1566,13 @@ radv_get_preamble_cs(struct radv_queue *queue,
struct radeon_winsys_bo *compute_scratch_bo = NULL;
struct radeon_winsys_bo *esgs_ring_bo = NULL;
struct radeon_winsys_bo *gsvs_ring_bo = NULL;
- struct radeon_winsys_bo *tess_factor_ring_bo = NULL;
- struct radeon_winsys_bo *tess_offchip_ring_bo = NULL;
+ struct radeon_winsys_bo *tess_rings_bo = NULL;
struct radeon_winsys_cs *dest_cs[3] = {0};
bool add_tess_rings = false, add_sample_positions = false;
unsigned tess_factor_ring_size = 0, tess_offchip_ring_size = 0;
unsigned max_offchip_buffers;
unsigned hs_offchip_param = 0;
+ unsigned tess_offchip_ring_offset;
uint32_t ring_bo_flags = RADEON_FLAG_NO_CPU_ACCESS | RADEON_FLAG_NO_INTERPROCESS_SHARING;
if (!queue->has_tess_rings) {
if (needs_tess_rings)
@@ -1587,6 +1585,7 @@ radv_get_preamble_cs(struct radv_queue *queue,
tess_factor_ring_size = 32768 * queue->device->physical_device->rad_info.max_se;
hs_offchip_param = radv_get_hs_offchip_param(queue->device,
&max_offchip_buffers);
+ tess_offchip_ring_offset = align(tess_factor_ring_size, 64 * 1024);
tess_offchip_ring_size = max_offchip_buffers *
queue->device->tess_offchip_block_dw_size * 4;
@@ -1654,33 +1653,25 @@ radv_get_preamble_cs(struct radv_queue *queue,
}
if (add_tess_rings) {
- tess_factor_ring_bo = queue->device->ws->buffer_create(queue->device->ws,
- tess_factor_ring_size,
- 256,
- RADEON_DOMAIN_VRAM,
- ring_bo_flags);
- if (!tess_factor_ring_bo)
- goto fail;
- tess_offchip_ring_bo = queue->device->ws->buffer_create(queue->device->ws,
- tess_offchip_ring_size,
- 256,
- RADEON_DOMAIN_VRAM,
- ring_bo_flags);
- if (!tess_offchip_ring_bo)
+ tess_rings_bo = queue->device->ws->buffer_create(queue->device->ws,
+ tess_offchip_ring_offset + tess_offchip_ring_size,
+ 256,
+ RADEON_DOMAIN_VRAM,
+ ring_bo_flags);
+ if (!tess_rings_bo)
goto fail;
} else {
- tess_factor_ring_bo = queue->tess_factor_ring_bo;
- tess_offchip_ring_bo = queue->tess_offchip_ring_bo;
+ tess_rings_bo = queue->tess_rings_bo;
}
if (scratch_bo != queue->scratch_bo ||
esgs_ring_bo != queue->esgs_ring_bo ||
gsvs_ring_bo != queue->gsvs_ring_bo ||
- tess_factor_ring_bo != queue->tess_factor_ring_bo ||
- tess_offchip_ring_bo != queue->tess_offchip_ring_bo || add_sample_positions) {
+ tess_rings_bo != queue->tess_rings_bo ||
+ add_sample_positions) {
uint32_t size = 0;
if (gsvs_ring_bo || esgs_ring_bo ||
- tess_factor_ring_bo || tess_offchip_ring_bo || add_sample_positions) {
+ tess_rings_bo || add_sample_positions) {
size = 112; /* 2 dword + 2 padding + 4 dword * 6 */
if (add_sample_positions)
size += 256; /* 32+16+8+4+2+1 samples * 4 * 2 = 248 bytes. */
@@ -1718,11 +1709,8 @@ radv_get_preamble_cs(struct radv_queue *queue,
if (gsvs_ring_bo)
radv_cs_add_buffer(queue->device->ws, cs, gsvs_ring_bo, 8);
- if (tess_factor_ring_bo)
- radv_cs_add_buffer(queue->device->ws, cs, tess_factor_ring_bo, 8);
-
- if (tess_offchip_ring_bo)
- radv_cs_add_buffer(queue->device->ws, cs, tess_offchip_ring_bo, 8);
+ if (tess_rings_bo)
+ radv_cs_add_buffer(queue->device->ws, cs, tess_rings_bo, 8);
if (descriptor_bo)
radv_cs_add_buffer(queue->device->ws, cs, descriptor_bo, 8);
@@ -1738,18 +1726,20 @@ radv_get_preamble_cs(struct radv_queue *queue,
map[1] = rsrc1;
}
- if (esgs_ring_bo || gsvs_ring_bo || tess_factor_ring_bo || tess_offchip_ring_bo ||
+ if (esgs_ring_bo || gsvs_ring_bo || tess_rings_bo ||
add_sample_positions)
fill_geom_tess_rings(queue, map, add_sample_positions,
esgs_ring_size, esgs_ring_bo,
gsvs_ring_size, gsvs_ring_bo,
- tess_factor_ring_size, tess_factor_ring_bo,
- tess_offchip_ring_size, tess_offchip_ring_bo);
+ tess_factor_ring_size,
+ tess_offchip_ring_offset,
+ tess_offchip_ring_size,
+ tess_rings_bo);
queue->device->ws->buffer_unmap(descriptor_bo);
}
- if (esgs_ring_bo || gsvs_ring_bo || tess_factor_ring_bo || tess_offchip_ring_bo) {
+ if (esgs_ring_bo || gsvs_ring_bo || tess_rings_bo) {
radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
radeon_emit(cs, EVENT_TYPE(V_028A90_VS_PARTIAL_FLUSH) | EVENT_INDEX(4));
radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
@@ -1768,8 +1758,8 @@ radv_get_preamble_cs(struct radv_queue *queue,
}
}
- if (tess_factor_ring_bo) {
- uint64_t tf_va = radv_buffer_get_va(tess_factor_ring_bo);
+ if (tess_rings_bo) {
+ uint64_t tf_va = radv_buffer_get_va(tess_rings_bo);
if (queue->device->physical_device->rad_info.chip_class >= CIK) {
radeon_set_uconfig_reg(cs, R_030938_VGT_TF_RING_SIZE,
S_030938_SIZE(tess_factor_ring_size / 4));
@@ -1899,15 +1889,12 @@ radv_get_preamble_cs(struct radv_queue *queue,
queue->gsvs_ring_size = gsvs_ring_size;
}
- if (tess_factor_ring_bo != queue->tess_factor_ring_bo) {
- queue->tess_factor_ring_bo = tess_factor_ring_bo;
- }
-
- if (tess_offchip_ring_bo != queue->tess_offchip_ring_bo) {
- queue->tess_offchip_ring_bo = tess_offchip_ring_bo;
+ if (tess_rings_bo != queue->tess_rings_bo) {
+ queue->tess_rings_bo = tess_rings_bo;
queue->has_tess_rings = true;
}
+
if (descriptor_bo != queue->descriptor_bo) {
if (queue->descriptor_bo)
queue->device->ws->buffer_destroy(queue->descriptor_bo);
@@ -1938,10 +1925,8 @@ fail:
queue->device->ws->buffer_destroy(esgs_ring_bo);
if (gsvs_ring_bo && gsvs_ring_bo != queue->gsvs_ring_bo)
queue->device->ws->buffer_destroy(gsvs_ring_bo);
- if (tess_factor_ring_bo && tess_factor_ring_bo != queue->tess_factor_ring_bo)
- queue->device->ws->buffer_destroy(tess_factor_ring_bo);
- if (tess_offchip_ring_bo && tess_offchip_ring_bo != queue->tess_offchip_ring_bo)
- queue->device->ws->buffer_destroy(tess_offchip_ring_bo);
+ if (tess_rings_bo && tess_rings_bo != queue->tess_rings_bo)
+ queue->device->ws->buffer_destroy(tess_rings_bo);
return vk_error(VK_ERROR_OUT_OF_DEVICE_MEMORY);
}
diff --git a/src/amd/vulkan/radv_private.h b/src/amd/vulkan/radv_private.h
index be9e8f43964..00a15255fee 100644
--- a/src/amd/vulkan/radv_private.h
+++ b/src/amd/vulkan/radv_private.h
@@ -569,8 +569,7 @@ struct radv_queue {
struct radeon_winsys_bo *compute_scratch_bo;
struct radeon_winsys_bo *esgs_ring_bo;
struct radeon_winsys_bo *gsvs_ring_bo;
- struct radeon_winsys_bo *tess_factor_ring_bo;
- struct radeon_winsys_bo *tess_offchip_ring_bo;
+ struct radeon_winsys_bo *tess_rings_bo;
struct radeon_winsys_cs *initial_preamble_cs;
struct radeon_winsys_cs *initial_full_flush_preamble_cs;
struct radeon_winsys_cs *continue_preamble_cs;
--
2.14.3
More information about the mesa-dev
mailing list