On 11/18/2016 01:43 PM, Xiang, Haihao wrote:
User can set sampler entry size and the number of sampler entries now.
sampler.bo is
always set even if the sampler state is a part of the dynamic state buffer,
hence we
can use the corresponding settings no matter sampler state is a part of the
dynamic
state buffer or not.
This looks good to me.
Add: Reviewed-by: Zhao Yakui <yakui.z...@intel.com>
Signed-off-by: Xiang, Haihao<haihao.xi...@intel.com>
---
src/gen8_mfc.c | 3 ++-
src/gen8_vme.c | 4 ++--
src/gen9_post_processing.c | 13 +++++++------
src/gen9_vme.c | 4 ++--
src/gen9_vp9_encoder.c | 15 +++++++++------
src/i965_gpe_utils.c | 21 ++++++++++++++++-----
src/i965_gpe_utils.h | 11 +++++++----
7 files changed, 45 insertions(+), 26 deletions(-)
diff --git a/src/gen8_mfc.c b/src/gen8_mfc.c
index c4e46fb..b0ee6fb 100644
--- a/src/gen8_mfc.c
+++ b/src/gen8_mfc.c
@@ -4611,7 +4611,8 @@ Bool gen8_mfc_context_init(VADriverContextP ctx, struct
intel_encoder_context *e
mfc_context->gpe_context.idrt.entry_size = ALIGN(sizeof(struct
gen8_interface_descriptor_data), 64);
mfc_context->gpe_context.idrt.max_entries = MAX_INTERFACE_DESC_GEN6;
mfc_context->gpe_context.curbe.length = 32 * 4;
- mfc_context->gpe_context.sampler_size = 0;
+ mfc_context->gpe_context.sampler.entry_size = 0;
+ mfc_context->gpe_context.sampler.max_entries = 0;
mfc_context->gpe_context.vfe_state.max_num_threads = 60 - 1;
mfc_context->gpe_context.vfe_state.num_urb_entries = 16;
diff --git a/src/gen8_vme.c b/src/gen8_vme.c
index 5184ef1..b14d60a 100644
--- a/src/gen8_vme.c
+++ b/src/gen8_vme.c
@@ -1382,8 +1382,8 @@ Bool gen8_vme_context_init(VADriverContextP ctx, struct
intel_encoder_context *e
vme_context->gpe_context.idrt.max_entries = MAX_INTERFACE_DESC_GEN6;
vme_context->gpe_context.curbe.length = CURBE_TOTAL_DATA_LENGTH;
- vme_context->gpe_context.sampler_size = 0;
-
+ vme_context->gpe_context.sampler.entry_size = 0;
+ vme_context->gpe_context.sampler.max_entries = 0;
vme_context->gpe_context.vfe_state.max_num_threads = 60 - 1;
vme_context->gpe_context.vfe_state.num_urb_entries = 64;
diff --git a/src/gen9_post_processing.c b/src/gen9_post_processing.c
index 2473803..c0ae791 100644
--- a/src/gen9_post_processing.c
+++ b/src/gen9_post_processing.c
@@ -488,15 +488,15 @@ gen9_p010_scaling_sample_state(VADriverContextP ctx,
if (gpe_context == NULL || !src_rect || !dst_rect)
return;
- dri_bo_map(gpe_context->dynamic_state.bo, 1);
+ dri_bo_map(gpe_context->sampler.bo, 1);
- if (gpe_context->dynamic_state.bo->virtual == NULL)
+ if (gpe_context->sampler.bo->virtual == NULL)
return;
- assert(gpe_context->dynamic_state.bo->virtual);
+ assert(gpe_context->sampler.bo->virtual);
sampler_state = (struct gen8_sampler_state *)
- (gpe_context->dynamic_state.bo->virtual + gpe_context->sampler_offset);
+ (gpe_context->sampler.bo->virtual + gpe_context->sampler.offset);
memset(sampler_state, 0, sizeof(*sampler_state));
@@ -513,7 +513,7 @@ gen9_p010_scaling_sample_state(VADriverContextP ctx,
sampler_state->ss3.s_wrap_mode = I965_TEXCOORDMODE_CLAMP;
sampler_state->ss3.t_wrap_mode = I965_TEXCOORDMODE_CLAMP;
- dri_bo_unmap(gpe_context->dynamic_state.bo);
+ dri_bo_unmap(gpe_context->sampler.bo);
}
void
@@ -538,7 +538,8 @@ gen9_post_processing_context_init(VADriverContextP ctx,
gen8_gpe_load_kernels(ctx, gpe_context,&scaling_kernel, 1);
gpe_context->idrt.entry_size = ALIGN(sizeof(struct
gen8_interface_descriptor_data), 64);
gpe_context->idrt.max_entries = 1;
- gpe_context->sampler_size = ALIGN(sizeof(struct gen8_sampler_state), 64);
+ gpe_context->sampler.entry_size = ALIGN(sizeof(struct gen8_sampler_state),
64);
+ gpe_context->sampler.max_entries = 1;
gpe_context->curbe.length = ALIGN(sizeof(struct scaling_input_parameter),
64);
gpe_context->surface_state_binding_table.max_entries =
MAX_SCALING_SURFACES;
diff --git a/src/gen9_vme.c b/src/gen9_vme.c
index 33bf8aa..fab80ce 100644
--- a/src/gen9_vme.c
+++ b/src/gen9_vme.c
@@ -2034,8 +2034,8 @@ Bool gen9_vme_context_init(VADriverContextP ctx, struct
intel_encoder_context *e
vme_context->gpe_context.idrt.entry_size = ALIGN(sizeof(struct
gen8_interface_descriptor_data), 64);
vme_context->gpe_context.idrt.max_entries = MAX_INTERFACE_DESC_GEN6;
vme_context->gpe_context.curbe.length = CURBE_TOTAL_DATA_LENGTH;
- vme_context->gpe_context.sampler_size = 0;
-
+ vme_context->gpe_context.sampler.entry_size = 0;
+ vme_context->gpe_context.sampler.max_entries = 0;
vme_context->gpe_context.vfe_state.max_num_threads = 60 - 1;
vme_context->gpe_context.vfe_state.num_urb_entries = 64;
diff --git a/src/gen9_vp9_encoder.c b/src/gen9_vp9_encoder.c
index 5d4a4a8..1badd88 100644
--- a/src/gen9_vp9_encoder.c
+++ b/src/gen9_vp9_encoder.c
@@ -2594,13 +2594,13 @@ gen9_vp9_dys_set_sampler_state(struct i965_gpe_context
*gpe_context)
if (!gpe_context)
return;
- dri_bo_map(gpe_context->dynamic_state.bo, 1);
+ dri_bo_map(gpe_context->sampler.bo, 1);
- if (!gpe_context->dynamic_state.bo->virtual)
+ if (!gpe_context->sampler.bo->virtual)
return;
sampler_cmd = (struct gen9_sampler_8x8_avs *)
- (gpe_context->dynamic_state.bo->virtual + gpe_context->sampler_offset);
+ (gpe_context->sampler.bo->virtual + gpe_context->sampler.offset);
memset(sampler_cmd, 0, sizeof(struct gen9_sampler_8x8_avs));
@@ -2658,7 +2658,7 @@ gen9_vp9_dys_set_sampler_state(struct i965_gpe_context
*gpe_context)
&gen9_vp9_avs_coeffs[17 * 8],
15 * sizeof(struct gen8_sampler_8x8_avs_coefficients));
- dri_bo_unmap(gpe_context->dynamic_state.bo);
+ dri_bo_unmap(gpe_context->sampler.bo);
}
static void
@@ -3684,9 +3684,12 @@ gen9_init_gpe_context_vp9(struct i965_gpe_context
*gpe_context,
{
gpe_context->curbe.length = kernel_param->curbe_size; // in bytes
- gpe_context->sampler_size = 0;
+ gpe_context->sampler.entry_size = 0;
+ gpe_context->sampler.max_entries = 0;
+
if (kernel_param->sampler_size) {
- gpe_context->sampler_size = ALIGN(kernel_param->sampler_size, 64);
+ gpe_context->sampler.entry_size = ALIGN(kernel_param->sampler_size,
64);
+ gpe_context->sampler.max_entries = 1;
}
gpe_context->idrt.entry_size = ALIGN(sizeof(struct
gen8_interface_descriptor_data), 64); // 8 dws, 1 register
diff --git a/src/i965_gpe_utils.c b/src/i965_gpe_utils.c
index 2d7cfaf..d7286fc 100644
--- a/src/i965_gpe_utils.c
+++ b/src/i965_gpe_utils.c
@@ -1122,7 +1122,8 @@ gen8_gpe_context_init(VADriverContextP ctx,
assert(bo);
gpe_context->surface_state_binding_table.bo = bo;
- bo_size = gpe_context->idrt.max_entries * gpe_context->idrt.entry_size +
gpe_context->curbe.length + gpe_context->sampler_size + 192;
+ bo_size = gpe_context->idrt.max_entries * gpe_context->idrt.entry_size +
gpe_context->curbe.length +
+ gpe_context->sampler.max_entries * gpe_context->sampler.entry_size +
192;
dri_bo_unreference(gpe_context->dynamic_state.bo);
bo = dri_bo_alloc(i965->intel.bufmgr,
"surface state& binding table",
@@ -1153,8 +1154,11 @@ gen8_gpe_context_init(VADriverContextP ctx,
/* Sampler state offset */
start_offset = ALIGN(end_offset, 64);
- gpe_context->sampler_offset = start_offset;
- end_offset = start_offset + gpe_context->sampler_size;
+ dri_bo_unreference(gpe_context->sampler.bo);
+ gpe_context->sampler.bo = bo;
+ dri_bo_reference(gpe_context->sampler.bo);
+ gpe_context->sampler.offset = start_offset;
+ end_offset = start_offset + gpe_context->sampler.entry_size *
gpe_context->sampler.max_entries;
/* update the end offset of dynamic_state */
gpe_context->dynamic_state.end_offset = end_offset;
@@ -1181,6 +1185,9 @@ gen8_gpe_context_destroy(struct i965_gpe_context
*gpe_context)
dri_bo_unreference(gpe_context->idrt.bo);
gpe_context->idrt.bo = NULL;
+
+ dri_bo_unreference(gpe_context->sampler.bo);
+ gpe_context->sampler.bo = NULL;
}
@@ -1642,7 +1649,11 @@ gen8_gpe_context_set_dynamic_buffer(VADriverContextP ctx,
dri_bo_reference(gpe_context->idrt.bo);
gpe_context->idrt.offset = ds->idrt_offset;
- gpe_context->sampler_offset = ds->sampler_offset;
+ /* sampler buffer is a part of the dynamic buffer */
+ dri_bo_unreference(gpe_context->sampler.bo);
+ gpe_context->sampler.bo = ds->bo;
+ dri_bo_reference(gpe_context->sampler.bo);
+ gpe_context->sampler.offset = ds->sampler_offset;
return;
}
@@ -1704,7 +1715,7 @@ gen8_gpe_setup_interface_data(VADriverContextP ctx,
memset(desc, 0, sizeof(*desc));
desc->desc0.kernel_start_pointer = kernel->kernel_offset>> 6;
desc->desc3.sampler_count = 0;
- desc->desc3.sampler_state_pointer = (gpe_context->sampler_offset>> 5);
+ desc->desc3.sampler_state_pointer = (gpe_context->sampler.offset>> 5);
desc->desc4.binding_table_entry_count = 0;
desc->desc4.binding_table_pointer =
(gpe_context->surface_state_binding_table.binding_table_offset>> 5);
desc->desc5.constant_urb_entry_read_offset = 0;
diff --git a/src/i965_gpe_utils.h b/src/i965_gpe_utils.h
index c3b8c79..e19e107 100644
--- a/src/i965_gpe_utils.h
+++ b/src/i965_gpe_utils.h
@@ -97,6 +97,13 @@ struct i965_gpe_context
} curbe;
struct {
+ dri_bo *bo;
+ unsigned int max_entries;
+ unsigned int entry_size; /* in bytes */
+ unsigned int offset;
+ } sampler;
+
+ struct {
unsigned int gpgpu_mode : 1;
unsigned int pad0 : 7;
unsigned int max_num_threads : 16;
@@ -164,10 +171,6 @@ struct i965_gpe_context
int bo_size;
unsigned int end_offset;
} dynamic_state;
-
- unsigned int sampler_offset;
- int sampler_entries;
- int sampler_size;
};
struct gpe_mi_flush_dw_parameter
_______________________________________________
Libva mailing list
Libva@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/libva
