[Mesa-dev] [PATCH 01/14] ac/gpu_info: add kernel_flushes_hdp_before_ib
Marek Olšák
maraeo at gmail.com
Thu May 3 00:19:39 UTC 2018
From: Marek Olšák <marek.olsak at amd.com>
---
src/amd/common/ac_gpu_info.c | 2 ++
src/amd/common/ac_gpu_info.h | 1 +
src/gallium/drivers/radeonsi/si_buffer.c | 6 ++----
src/gallium/winsys/radeon/drm/radeon_drm_winsys.c | 1 +
4 files changed, 6 insertions(+), 4 deletions(-)
diff --git a/src/amd/common/ac_gpu_info.c b/src/amd/common/ac_gpu_info.c
index fd49dbefd58..b1022ef75de 100644
--- a/src/amd/common/ac_gpu_info.c
+++ b/src/amd/common/ac_gpu_info.c
@@ -309,20 +309,21 @@ bool ac_query_gpu_info(int fd, amdgpu_device_handle dev,
info->uvd_enc_supported =
uvd_enc.available_rings ? true : false;
info->has_userptr = true;
info->has_syncobj = has_syncobj(fd);
info->has_syncobj_wait_for_submit = info->has_syncobj && info->drm_minor >= 20;
info->has_fence_to_handle = info->has_syncobj && info->drm_minor >= 21;
info->has_ctx_priority = info->drm_minor >= 22;
/* TODO: Enable this once the kernel handles it efficiently. */
info->has_local_buffers = info->drm_minor >= 20 &&
!info->has_dedicated_vram;
+ info->kernel_flushes_hdp_before_ib = true;
info->num_render_backends = amdinfo->rb_pipes;
/* The value returned by the kernel driver was wrong. */
if (info->family == CHIP_KAVERI)
info->num_render_backends = 2;
info->clock_crystal_freq = amdinfo->gpu_counter_freq;
if (!info->clock_crystal_freq) {
fprintf(stderr, "amdgpu: clock crystal frequency is 0, timestamps will be wrong\n");
info->clock_crystal_freq = 1;
@@ -456,20 +457,21 @@ void ac_print_gpu_info(struct radeon_info *info)
printf("Kernel info:\n");
printf(" drm = %i.%i.%i\n", info->drm_major,
info->drm_minor, info->drm_patchlevel);
printf(" has_userptr = %i\n", info->has_userptr);
printf(" has_syncobj = %u\n", info->has_syncobj);
printf(" has_syncobj_wait_for_submit = %u\n", info->has_syncobj_wait_for_submit);
printf(" has_fence_to_handle = %u\n", info->has_fence_to_handle);
printf(" has_ctx_priority = %u\n", info->has_ctx_priority);
printf(" has_local_buffers = %u\n", info->has_local_buffers);
+ printf(" kernel_flushes_hdp_before_ib = %u\n", info->kernel_flushes_hdp_before_ib);
printf("Shader core info:\n");
printf(" max_shader_clock = %i\n", info->max_shader_clock);
printf(" num_good_compute_units = %i\n", info->num_good_compute_units);
printf(" max_se = %i\n", info->max_se);
printf(" max_sh_per_se = %i\n", info->max_sh_per_se);
printf("Render backend info:\n");
printf(" num_render_backends = %i\n", info->num_render_backends);
printf(" num_tile_pipes = %i\n", info->num_tile_pipes);
diff --git a/src/amd/common/ac_gpu_info.h b/src/amd/common/ac_gpu_info.h
index 7af6fbfca97..8a9721750a6 100644
--- a/src/amd/common/ac_gpu_info.h
+++ b/src/amd/common/ac_gpu_info.h
@@ -89,20 +89,21 @@ struct radeon_info {
/* Kernel info. */
uint32_t drm_major; /* version */
uint32_t drm_minor;
uint32_t drm_patchlevel;
bool has_userptr;
bool has_syncobj;
bool has_syncobj_wait_for_submit;
bool has_fence_to_handle;
bool has_ctx_priority;
bool has_local_buffers;
+ bool kernel_flushes_hdp_before_ib;
/* Shader cores. */
uint32_t r600_max_quad_pipes; /* wave size / 16 */
uint32_t max_shader_clock;
uint32_t num_good_compute_units;
uint32_t max_se; /* shader engines */
uint32_t max_sh_per_se; /* shader arrays per shader engine */
/* Render backends (color + depth blocks). */
uint32_t r300_num_gb_pipes;
diff --git a/src/gallium/drivers/radeonsi/si_buffer.c b/src/gallium/drivers/radeonsi/si_buffer.c
index 504e0c723dc..2d68edc3404 100644
--- a/src/gallium/drivers/radeonsi/si_buffer.c
+++ b/src/gallium/drivers/radeonsi/si_buffer.c
@@ -118,22 +118,21 @@ void si_init_resource_fields(struct si_screen *sscreen,
/* fall through */
case PIPE_USAGE_STAGING:
/* Transfers are likely to occur more often with these
* resources. */
res->domains = RADEON_DOMAIN_GTT;
break;
case PIPE_USAGE_DYNAMIC:
/* Older kernels didn't always flush the HDP cache before
* CS execution
*/
- if (sscreen->info.drm_major == 2 &&
- sscreen->info.drm_minor < 40) {
+ if (!sscreen->info.kernel_flushes_hdp_before_ib) {
res->domains = RADEON_DOMAIN_GTT;
res->flags |= RADEON_FLAG_GTT_WC;
break;
}
/* fall through */
case PIPE_USAGE_DEFAULT:
case PIPE_USAGE_IMMUTABLE:
default:
/* Not listing GTT here improves performance in some
* apps. */
@@ -146,22 +145,21 @@ void si_init_resource_fields(struct si_screen *sscreen,
res->b.b.flags & (PIPE_RESOURCE_FLAG_MAP_PERSISTENT |
PIPE_RESOURCE_FLAG_MAP_COHERENT)) {
/* Use GTT for all persistent mappings with older
* kernels, because they didn't always flush the HDP
* cache before CS execution.
*
* Write-combined CPU mappings are fine, the kernel
* ensures all CPU writes finish before the GPU
* executes a command stream.
*/
- if (sscreen->info.drm_major == 2 &&
- sscreen->info.drm_minor < 40)
+ if (!sscreen->info.kernel_flushes_hdp_before_ib)
res->domains = RADEON_DOMAIN_GTT;
}
/* Tiled textures are unmappable. Always put them in VRAM. */
if ((res->b.b.target != PIPE_BUFFER && !rtex->surface.is_linear) ||
res->b.b.flags & SI_RESOURCE_FLAG_UNMAPPABLE) {
res->domains = RADEON_DOMAIN_VRAM;
res->flags |= RADEON_FLAG_NO_CPU_ACCESS |
RADEON_FLAG_GTT_WC;
}
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c b/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c
index 28811c959fe..ab6cc60597d 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c
@@ -525,20 +525,21 @@ static bool do_winsys_init(struct radeon_drm_winsys *ws)
}
/* Hawaii with old firmware needs type2 nop packet.
* accel_working2 with value 3 indicates the new firmware.
*/
ws->info.gfx_ib_pad_with_type2 = ws->info.chip_class <= SI ||
(ws->info.family == CHIP_HAWAII &&
ws->accel_working2 < 3);
ws->info.tcc_cache_line_size = 64; /* TC L2 line size on GCN */
ws->info.ib_start_alignment = 4096;
+ ws->info.kernel_flushes_hdp_before_ib = ws->info.drm_minor >= 40;
ws->check_vm = strstr(debug_get_option("R600_DEBUG", ""), "check_vm") != NULL;
return true;
}
static void radeon_winsys_destroy(struct radeon_winsys *rws)
{
struct radeon_drm_winsys *ws = (struct radeon_drm_winsys*)rws;
--
2.17.0
More information about the mesa-dev
mailing list