mesa gallium: use compute shaders for vaapi blit

Submitted by Jiang, Sonny on April 2, 2019, 5:44 p.m.

Details

Message ID 20190402174321.19349-1-sonny.jiang@amd.com
State New
Headers show
Series "mesa gallium: use compute shaders for vaapi blit" ( rev: 1 ) in Mesa

Not browsing as part of any series.

Commit Message

Jiang, Sonny April 2, 2019, 5:44 p.m.
Signed-off-by: Sonny Jiang <sonny.jiang@amd.com>
---
 src/gallium/auxiliary/Makefile.sources     |   2 +
 src/gallium/auxiliary/util/u_compute.c     | 165 +++++++++++++++++++++
 src/gallium/auxiliary/util/u_compute.h     |  44 ++++++
 src/gallium/state_trackers/va/context.c    |   2 +
 src/gallium/state_trackers/va/postproc.c   |   6 +-
 src/gallium/state_trackers/va/va_private.h |   1 +
 6 files changed, 219 insertions(+), 1 deletion(-)
 create mode 100644 src/gallium/auxiliary/util/u_compute.c
 create mode 100644 src/gallium/auxiliary/util/u_compute.h

Patch hide | download patch | download mbox

diff --git a/src/gallium/auxiliary/Makefile.sources b/src/gallium/auxiliary/Makefile.sources
index 02cc5df70a7..6f5266fe273 100644
--- a/src/gallium/auxiliary/Makefile.sources
+++ b/src/gallium/auxiliary/Makefile.sources
@@ -218,6 +218,8 @@  C_SOURCES := \
 	util/u_box.h \
 	util/u_cache.c \
 	util/u_cache.h \
+	util/u_compute.c \
+	util/u_compute.h \
 	util/u_debug_gallium.h \
 	util/u_debug_gallium.c \
 	util/u_debug_describe.c \
diff --git a/src/gallium/auxiliary/util/u_compute.c b/src/gallium/auxiliary/util/u_compute.c
new file mode 100644
index 00000000000..e2e39227206
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_compute.c
@@ -0,0 +1,165 @@ 
+/**************************************************************************
+ *
+ * Copyright 2019 Sonny Jiang <sonnyj608@gmail.com>
+ * Copyright 2019 Advanced Micro Devices, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#include "pipe/p_context.h"
+#include "pipe/p_state.h"
+
+#include "u_bitcast.h"
+#include "u_format.h"
+#include "u_sampler.h"
+#include "tgsi/tgsi_text.h"
+#include "tgsi/tgsi_ureg.h"
+#include "u_inlines.h"
+
+void *blit_compute_shader(struct pipe_context *ctx)
+{
+	static const char text[] =
+		"COMP\n"
+		"PROPERTY CS_FIXED_BLOCK_WIDTH 64\n"
+		"PROPERTY CS_FIXED_BLOCK_HEIGHT 1\n"
+		"PROPERTY CS_FIXED_BLOCK_DEPTH 1\n"
+		"DCL SV[0], THREAD_ID\n"
+		"DCL SV[1], BLOCK_ID\n"
+		"DCL IMAGE[0], 2D_ARRAY, PIPE_FORMAT_R32G32B32A32_FLOAT, WR\n"
+		"DCL SAMP[0]\n"
+		"DCL SVIEW[0], 2D_ARRAY, FLOAT\n"
+		"DCL CONST[0][0..2]\n" // 0:xyzw 1:xyzw
+		"DCL TEMP[0..4], LOCAL\n"
+		"IMM[0] UINT32 {64, 1, 0, 0}\n"
+
+		"UMAD TEMP[0].xyz, SV[1].xyzz, IMM[0].xyyy, SV[0].xyzz\n"
+		"U2F TEMP[1].xyz, TEMP[0]\n"
+		"MAD TEMP[2].xyz, TEMP[1], CONST[0][1], CONST[0][0]\n"
+		"TEX_LZ TEMP[3], TEMP[2], SAMP[0], 2D_ARRAY\n"
+		"UADD TEMP[4].xyz, TEMP[0], CONST[0][2]\n"
+		"STORE IMAGE[0], TEMP[4], TEMP[3], 2D_ARRAY, PIPE_FORMAT_R32G32B32A32_FLOAT\n"
+		"END\n";
+
+	struct tgsi_token tokens[1024];
+	struct pipe_compute_state state = {0};
+
+	if (!tgsi_text_translate(text, tokens, ARRAY_SIZE(tokens))) {
+		assert(false);
+		return NULL;
+	}
+
+	state.ir_type = PIPE_SHADER_IR_TGSI;
+	state.prog = tokens;
+
+	return ctx->create_compute_state(ctx, &state);
+}
+
+void util_compute_blit(struct pipe_context *ctx, struct pipe_blit_info *blit_info,
+		       void **compute_state)
+{
+	if (blit_info->src.box.width == 0 || blit_info->src.box.height == 0 ||
+	    blit_info->dst.box.width == 0 || blit_info->dst.box.height == 0)
+		return;
+
+	struct pipe_resource *src = blit_info->src.resource;
+	struct pipe_resource *dst = blit_info->dst.resource;
+	struct pipe_sampler_view src_templ = {0}, *src_view;
+	void *sampler_state_p;
+	unsigned width = blit_info->dst.box.width;
+	unsigned height = blit_info->dst.box.height;
+	float x_scale = blit_info->src.box.width / (float)blit_info->dst.box.width;
+	float y_scale = blit_info->src.box.height / (float)blit_info->dst.box.height;
+	float z_scale = blit_info->src.box.depth / (float)blit_info->dst.box.depth;
+
+	unsigned data[] = {u_bitcast_f2u((blit_info->src.box.x + 0.5) / (float)src->width0),
+			   u_bitcast_f2u((blit_info->src.box.y + 0.5) / (float)src->height0),
+			   u_bitcast_f2u(blit_info->src.box.z),
+			   u_bitcast_f2u(0),
+			   u_bitcast_f2u(x_scale / src->width0),
+			   u_bitcast_f2u(y_scale / src->height0),
+			   u_bitcast_f2u(z_scale),
+			   u_bitcast_f2u(0),
+			   blit_info->dst.box.x,
+			   blit_info->dst.box.y,
+			   blit_info->dst.box.z,
+			   0};
+
+	struct pipe_constant_buffer cb = {};
+	cb.buffer_size = sizeof(data);
+	cb.user_buffer = data;
+	ctx->set_constant_buffer(ctx, PIPE_SHADER_COMPUTE, 0, &cb);
+
+	struct pipe_image_view image = {0};
+	image.resource = dst;
+	image.shader_access = image.access = PIPE_IMAGE_ACCESS_WRITE;
+	image.format = util_format_linear(blit_info->dst.format);
+	image.u.tex.level = blit_info->dst.level;
+	image.u.tex.first_layer = 0;
+	image.u.tex.last_layer = (unsigned)(dst->array_size - 1);
+
+	ctx->set_shader_images(ctx, PIPE_SHADER_COMPUTE, 0, 1, &image);
+
+	/* Initialize the sampler view. */
+	u_sampler_view_default_template(&src_templ, src, src->format);
+	src_templ.format = util_format_linear(blit_info->src.format);
+	src_view = ctx->create_sampler_view(ctx, src, &src_templ);
+	ctx->set_sampler_views(ctx, PIPE_SHADER_COMPUTE, 0, 1, &src_view);
+
+	struct pipe_sampler_state sampler_state={0};
+	sampler_state.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+	sampler_state.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+	sampler_state.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+	sampler_state.normalized_coords = 1;
+
+	if (blit_info->filter == PIPE_TEX_FILTER_LINEAR) {
+		sampler_state.min_img_filter = PIPE_TEX_FILTER_LINEAR;
+		sampler_state.mag_img_filter = PIPE_TEX_FILTER_LINEAR;
+	}
+
+	sampler_state_p = ctx->create_sampler_state(ctx, &sampler_state);
+	ctx->bind_sampler_states(ctx, PIPE_SHADER_COMPUTE, 0, 1, &sampler_state_p);
+
+	if (!*compute_state)
+		*compute_state = blit_compute_shader(ctx);
+	ctx->bind_compute_state(ctx, *compute_state);
+
+	struct pipe_grid_info grid_info = {0};
+	grid_info.block[0] = 64;
+	grid_info.last_block[0] = width % 64;
+	grid_info.block[1] = 1;
+	grid_info.block[2] = 1;
+	grid_info.grid[0] = DIV_ROUND_UP(width, 64);
+	grid_info.grid[1] = height;
+	grid_info.grid[2] = 1;
+
+	ctx->launch_grid(ctx, &grid_info);
+
+	ctx->memory_barrier(ctx, PIPE_BARRIER_ALL);
+
+	ctx->set_shader_images(ctx, PIPE_SHADER_COMPUTE, 0, 1, NULL);
+	ctx->set_constant_buffer(ctx, PIPE_SHADER_COMPUTE, 0, NULL);
+	ctx->set_sampler_views(ctx, PIPE_SHADER_COMPUTE, 0, 1, NULL);
+	pipe_sampler_view_reference(&src_view, NULL);
+	ctx->delete_sampler_state(ctx, sampler_state_p);
+	ctx->bind_compute_state(ctx, NULL);
+}
diff --git a/src/gallium/auxiliary/util/u_compute.h b/src/gallium/auxiliary/util/u_compute.h
new file mode 100644
index 00000000000..78dfcfec1c0
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_compute.h
@@ -0,0 +1,44 @@ 
+/**************************************************************************
+ * Copyright 2019 Sonny Jiang <sonnyj608@gmail.com>
+ * Copyright 2019 Advanced Micro Devices, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#ifndef U_COMPUTE_H
+#define U_COMPUTE_H
+
+#include "pipe/p_context.h"
+#include "pipe/p_state.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void util_compute_blit(struct pipe_context *ctx, struct pipe_blit_info *blit_info,
+                       void **compute_state);
+#ifdef __cplusplus
+}
+#endif
+
+#endif
diff --git a/src/gallium/state_trackers/va/context.c b/src/gallium/state_trackers/va/context.c
index 47a5e7be230..ac408102718 100644
--- a/src/gallium/state_trackers/va/context.c
+++ b/src/gallium/state_trackers/va/context.c
@@ -364,6 +364,8 @@  vlVaDestroyContext(VADriverContextP ctx, VAContextID context_id)
       }
       context->decoder->destroy(context->decoder);
    }
+   if (context->blit_cs)
+           drv->pipe->delete_compute_state(drv->pipe, context->blit_cs);
    if (context->deint) {
       vl_deint_filter_cleanup(context->deint);
       FREE(context->deint);
diff --git a/src/gallium/state_trackers/va/postproc.c b/src/gallium/state_trackers/va/postproc.c
index 21d316c7e97..cd05ab361ac 100644
--- a/src/gallium/state_trackers/va/postproc.c
+++ b/src/gallium/state_trackers/va/postproc.c
@@ -27,6 +27,7 @@ 
 
 #include "util/u_handle_table.h"
 #include "util/u_memory.h"
+#include "util/u_compute.h"
 
 #include "vl/vl_defines.h"
 #include "vl/vl_video_buffer.h"
@@ -220,7 +221,10 @@  static VAStatus vlVaPostProcBlit(vlVaDriver *drv, vlVaContext *context,
       blit.mask = PIPE_MASK_RGBA;
       blit.filter = PIPE_TEX_MIPFILTER_LINEAR;
 
-      drv->pipe->blit(drv->pipe, &blit);
+      if (drv->pipe->screen->get_param(drv->pipe->screen, PIPE_CAP_COMPUTE))
+              util_compute_blit(drv->pipe, &blit, &context->blit_cs);
+      else
+              drv->pipe->blit(drv->pipe, &blit);
    }
 
    // TODO: figure out why this is necessary for DMA-buf sharing
diff --git a/src/gallium/state_trackers/va/va_private.h b/src/gallium/state_trackers/va/va_private.h
index c4b49e30509..b2b997d4799 100644
--- a/src/gallium/state_trackers/va/va_private.h
+++ b/src/gallium/state_trackers/va/va_private.h
@@ -312,6 +312,7 @@  typedef struct {
    bool first_single_submitted;
    int gop_coeff;
    bool needs_begin_frame;
+   void *blit_cs;
 } vlVaContext;
 
 typedef struct {

Comments

Shouldn't this sort of decision be left up to the driver? If the
driver would like to use CS for blits, fine, but why not let it blit
in the most optimal way possible and force it to use a compute shader?

On Tue, Apr 2, 2019 at 1:44 PM Jiang, Sonny <Sonny.Jiang@amd.com> wrote:
>
> Signed-off-by: Sonny Jiang <sonny.jiang@amd.com>
> ---
>  src/gallium/auxiliary/Makefile.sources     |   2 +
>  src/gallium/auxiliary/util/u_compute.c     | 165 +++++++++++++++++++++
>  src/gallium/auxiliary/util/u_compute.h     |  44 ++++++
>  src/gallium/state_trackers/va/context.c    |   2 +
>  src/gallium/state_trackers/va/postproc.c   |   6 +-
>  src/gallium/state_trackers/va/va_private.h |   1 +
>  6 files changed, 219 insertions(+), 1 deletion(-)
>  create mode 100644 src/gallium/auxiliary/util/u_compute.c
>  create mode 100644 src/gallium/auxiliary/util/u_compute.h
>
> diff --git a/src/gallium/auxiliary/Makefile.sources b/src/gallium/auxiliary/Makefile.sources
> index 02cc5df70a7..6f5266fe273 100644
> --- a/src/gallium/auxiliary/Makefile.sources
> +++ b/src/gallium/auxiliary/Makefile.sources
> @@ -218,6 +218,8 @@ C_SOURCES := \
>         util/u_box.h \
>         util/u_cache.c \
>         util/u_cache.h \
> +       util/u_compute.c \
> +       util/u_compute.h \
>         util/u_debug_gallium.h \
>         util/u_debug_gallium.c \
>         util/u_debug_describe.c \
> diff --git a/src/gallium/auxiliary/util/u_compute.c b/src/gallium/auxiliary/util/u_compute.c
> new file mode 100644
> index 00000000000..e2e39227206
> --- /dev/null
> +++ b/src/gallium/auxiliary/util/u_compute.c
> @@ -0,0 +1,165 @@
> +/**************************************************************************
> + *
> + * Copyright 2019 Sonny Jiang <sonnyj608@gmail.com>
> + * Copyright 2019 Advanced Micro Devices, Inc.
> + * All Rights Reserved.
> + *
> + * Permission is hereby granted, free of charge, to any person obtaining a
> + * copy of this software and associated documentation files (the
> + * "Software"), to deal in the Software without restriction, including
> + * without limitation the rights to use, copy, modify, merge, publish,
> + * distribute, sub license, and/or sell copies of the Software, and to
> + * permit persons to whom the Software is furnished to do so, subject to
> + * the following conditions:
> + *
> + * The above copyright notice and this permission notice (including the
> + * next paragraph) shall be included in all copies or substantial portions
> + * of the Software.
> + *
> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
> + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
> + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
> + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
> + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
> + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
> + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
> + *
> + **************************************************************************/
> +
> +#include "pipe/p_context.h"
> +#include "pipe/p_state.h"
> +
> +#include "u_bitcast.h"
> +#include "u_format.h"
> +#include "u_sampler.h"
> +#include "tgsi/tgsi_text.h"
> +#include "tgsi/tgsi_ureg.h"
> +#include "u_inlines.h"
> +
> +void *blit_compute_shader(struct pipe_context *ctx)
> +{
> +       static const char text[] =
> +               "COMP\n"
> +               "PROPERTY CS_FIXED_BLOCK_WIDTH 64\n"
> +               "PROPERTY CS_FIXED_BLOCK_HEIGHT 1\n"
> +               "PROPERTY CS_FIXED_BLOCK_DEPTH 1\n"
> +               "DCL SV[0], THREAD_ID\n"
> +               "DCL SV[1], BLOCK_ID\n"
> +               "DCL IMAGE[0], 2D_ARRAY, PIPE_FORMAT_R32G32B32A32_FLOAT, WR\n"
> +               "DCL SAMP[0]\n"
> +               "DCL SVIEW[0], 2D_ARRAY, FLOAT\n"
> +               "DCL CONST[0][0..2]\n" // 0:xyzw 1:xyzw
> +               "DCL TEMP[0..4], LOCAL\n"
> +               "IMM[0] UINT32 {64, 1, 0, 0}\n"
> +
> +               "UMAD TEMP[0].xyz, SV[1].xyzz, IMM[0].xyyy, SV[0].xyzz\n"
> +               "U2F TEMP[1].xyz, TEMP[0]\n"
> +               "MAD TEMP[2].xyz, TEMP[1], CONST[0][1], CONST[0][0]\n"
> +               "TEX_LZ TEMP[3], TEMP[2], SAMP[0], 2D_ARRAY\n"
> +               "UADD TEMP[4].xyz, TEMP[0], CONST[0][2]\n"
> +               "STORE IMAGE[0], TEMP[4], TEMP[3], 2D_ARRAY, PIPE_FORMAT_R32G32B32A32_FLOAT\n"
> +               "END\n";
> +
> +       struct tgsi_token tokens[1024];
> +       struct pipe_compute_state state = {0};
> +
> +       if (!tgsi_text_translate(text, tokens, ARRAY_SIZE(tokens))) {
> +               assert(false);
> +               return NULL;
> +       }
> +
> +       state.ir_type = PIPE_SHADER_IR_TGSI;
> +       state.prog = tokens;
> +
> +       return ctx->create_compute_state(ctx, &state);
> +}
> +
> +void util_compute_blit(struct pipe_context *ctx, struct pipe_blit_info *blit_info,
> +                      void **compute_state)
> +{
> +       if (blit_info->src.box.width == 0 || blit_info->src.box.height == 0 ||
> +           blit_info->dst.box.width == 0 || blit_info->dst.box.height == 0)
> +               return;
> +
> +       struct pipe_resource *src = blit_info->src.resource;
> +       struct pipe_resource *dst = blit_info->dst.resource;
> +       struct pipe_sampler_view src_templ = {0}, *src_view;
> +       void *sampler_state_p;
> +       unsigned width = blit_info->dst.box.width;
> +       unsigned height = blit_info->dst.box.height;
> +       float x_scale = blit_info->src.box.width / (float)blit_info->dst.box.width;
> +       float y_scale = blit_info->src.box.height / (float)blit_info->dst.box.height;
> +       float z_scale = blit_info->src.box.depth / (float)blit_info->dst.box.depth;
> +
> +       unsigned data[] = {u_bitcast_f2u((blit_info->src.box.x + 0.5) / (float)src->width0),
> +                          u_bitcast_f2u((blit_info->src.box.y + 0.5) / (float)src->height0),
> +                          u_bitcast_f2u(blit_info->src.box.z),
> +                          u_bitcast_f2u(0),
> +                          u_bitcast_f2u(x_scale / src->width0),
> +                          u_bitcast_f2u(y_scale / src->height0),
> +                          u_bitcast_f2u(z_scale),
> +                          u_bitcast_f2u(0),
> +                          blit_info->dst.box.x,
> +                          blit_info->dst.box.y,
> +                          blit_info->dst.box.z,
> +                          0};
> +
> +       struct pipe_constant_buffer cb = {};
> +       cb.buffer_size = sizeof(data);
> +       cb.user_buffer = data;
> +       ctx->set_constant_buffer(ctx, PIPE_SHADER_COMPUTE, 0, &cb);
> +
> +       struct pipe_image_view image = {0};
> +       image.resource = dst;
> +       image.shader_access = image.access = PIPE_IMAGE_ACCESS_WRITE;
> +       image.format = util_format_linear(blit_info->dst.format);
> +       image.u.tex.level = blit_info->dst.level;
> +       image.u.tex.first_layer = 0;
> +       image.u.tex.last_layer = (unsigned)(dst->array_size - 1);
> +
> +       ctx->set_shader_images(ctx, PIPE_SHADER_COMPUTE, 0, 1, &image);
> +
> +       /* Initialize the sampler view. */
> +       u_sampler_view_default_template(&src_templ, src, src->format);
> +       src_templ.format = util_format_linear(blit_info->src.format);
> +       src_view = ctx->create_sampler_view(ctx, src, &src_templ);
> +       ctx->set_sampler_views(ctx, PIPE_SHADER_COMPUTE, 0, 1, &src_view);
> +
> +       struct pipe_sampler_state sampler_state={0};
> +       sampler_state.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
> +       sampler_state.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
> +       sampler_state.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
> +       sampler_state.normalized_coords = 1;
> +
> +       if (blit_info->filter == PIPE_TEX_FILTER_LINEAR) {
> +               sampler_state.min_img_filter = PIPE_TEX_FILTER_LINEAR;
> +               sampler_state.mag_img_filter = PIPE_TEX_FILTER_LINEAR;
> +       }
> +
> +       sampler_state_p = ctx->create_sampler_state(ctx, &sampler_state);
> +       ctx->bind_sampler_states(ctx, PIPE_SHADER_COMPUTE, 0, 1, &sampler_state_p);
> +
> +       if (!*compute_state)
> +               *compute_state = blit_compute_shader(ctx);
> +       ctx->bind_compute_state(ctx, *compute_state);
> +
> +       struct pipe_grid_info grid_info = {0};
> +       grid_info.block[0] = 64;
> +       grid_info.last_block[0] = width % 64;
> +       grid_info.block[1] = 1;
> +       grid_info.block[2] = 1;
> +       grid_info.grid[0] = DIV_ROUND_UP(width, 64);
> +       grid_info.grid[1] = height;
> +       grid_info.grid[2] = 1;
> +
> +       ctx->launch_grid(ctx, &grid_info);
> +
> +       ctx->memory_barrier(ctx, PIPE_BARRIER_ALL);
> +
> +       ctx->set_shader_images(ctx, PIPE_SHADER_COMPUTE, 0, 1, NULL);
> +       ctx->set_constant_buffer(ctx, PIPE_SHADER_COMPUTE, 0, NULL);
> +       ctx->set_sampler_views(ctx, PIPE_SHADER_COMPUTE, 0, 1, NULL);
> +       pipe_sampler_view_reference(&src_view, NULL);
> +       ctx->delete_sampler_state(ctx, sampler_state_p);
> +       ctx->bind_compute_state(ctx, NULL);
> +}
> diff --git a/src/gallium/auxiliary/util/u_compute.h b/src/gallium/auxiliary/util/u_compute.h
> new file mode 100644
> index 00000000000..78dfcfec1c0
> --- /dev/null
> +++ b/src/gallium/auxiliary/util/u_compute.h
> @@ -0,0 +1,44 @@
> +/**************************************************************************
> + * Copyright 2019 Sonny Jiang <sonnyj608@gmail.com>
> + * Copyright 2019 Advanced Micro Devices, Inc.
> + * All Rights Reserved.
> + *
> + * Permission is hereby granted, free of charge, to any person obtaining a
> + * copy of this software and associated documentation files (the
> + * "Software"), to deal in the Software without restriction, including
> + * without limitation the rights to use, copy, modify, merge, publish,
> + * distribute, sub license, and/or sell copies of the Software, and to
> + * permit persons to whom the Software is furnished to do so, subject to
> + * the following conditions:
> + *
> + * The above copyright notice and this permission notice (including the
> + * next paragraph) shall be included in all copies or substantial portions
> + * of the Software.
> + *
> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
> + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
> + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
> + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
> + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
> + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
> + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
> + *
> + **************************************************************************/
> +
> +#ifndef U_COMPUTE_H
> +#define U_COMPUTE_H
> +
> +#include "pipe/p_context.h"
> +#include "pipe/p_state.h"
> +
> +#ifdef __cplusplus
> +extern "C" {
> +#endif
> +
> +void util_compute_blit(struct pipe_context *ctx, struct pipe_blit_info *blit_info,
> +                       void **compute_state);
> +#ifdef __cplusplus
> +}
> +#endif
> +
> +#endif
> diff --git a/src/gallium/state_trackers/va/context.c b/src/gallium/state_trackers/va/context.c
> index 47a5e7be230..ac408102718 100644
> --- a/src/gallium/state_trackers/va/context.c
> +++ b/src/gallium/state_trackers/va/context.c
> @@ -364,6 +364,8 @@ vlVaDestroyContext(VADriverContextP ctx, VAContextID context_id)
>        }
>        context->decoder->destroy(context->decoder);
>     }
> +   if (context->blit_cs)
> +           drv->pipe->delete_compute_state(drv->pipe, context->blit_cs);
>     if (context->deint) {
>        vl_deint_filter_cleanup(context->deint);
>        FREE(context->deint);
> diff --git a/src/gallium/state_trackers/va/postproc.c b/src/gallium/state_trackers/va/postproc.c
> index 21d316c7e97..cd05ab361ac 100644
> --- a/src/gallium/state_trackers/va/postproc.c
> +++ b/src/gallium/state_trackers/va/postproc.c
> @@ -27,6 +27,7 @@
>
>  #include "util/u_handle_table.h"
>  #include "util/u_memory.h"
> +#include "util/u_compute.h"
>
>  #include "vl/vl_defines.h"
>  #include "vl/vl_video_buffer.h"
> @@ -220,7 +221,10 @@ static VAStatus vlVaPostProcBlit(vlVaDriver *drv, vlVaContext *context,
>        blit.mask = PIPE_MASK_RGBA;
>        blit.filter = PIPE_TEX_MIPFILTER_LINEAR;
>
> -      drv->pipe->blit(drv->pipe, &blit);
> +      if (drv->pipe->screen->get_param(drv->pipe->screen, PIPE_CAP_COMPUTE))
> +              util_compute_blit(drv->pipe, &blit, &context->blit_cs);
> +      else
> +              drv->pipe->blit(drv->pipe, &blit);
>     }
>
>     // TODO: figure out why this is necessary for DMA-buf sharing
> diff --git a/src/gallium/state_trackers/va/va_private.h b/src/gallium/state_trackers/va/va_private.h
> index c4b49e30509..b2b997d4799 100644
> --- a/src/gallium/state_trackers/va/va_private.h
> +++ b/src/gallium/state_trackers/va/va_private.h
> @@ -312,6 +312,7 @@ typedef struct {
>     bool first_single_submitted;
>     int gop_coeff;
>     bool needs_begin_frame;
> +   void *blit_cs;
>  } vlVaContext;
>
>  typedef struct {
> --
> 2.17.1
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev


On Wed, Apr 3, 2019 at 8:38 AM Marek Olšák <maraeo@gmail.com> wrote:
>
> On Tue, Apr 2, 2019 at 2:14 PM Eric Anholt <eric@anholt.net> wrote:
>>
>> Ilia Mirkin <imirkin@alum.mit.edu> writes:
>>
>> > Shouldn't this sort of decision be left up to the driver? If the
>> > driver would like to use CS for blits, fine, but why not let it blit
>> > in the most optimal way possible and force it to use a compute shader?
>>
>> Yeah, commit messages require an explanation of why a change is being
>> made.
>
>
> We plan to create vaapi contexts with PIPE_CONTEXT_COMPUTE_ONLY for better GPU multitasking.
>
> RadeonSI uses async compute queues if PIPE_CONTEXT_COMPUTE_ONLY is set, so it can't do any graphics stuff, not even blit. (pipe_context::blit is NULL)

Makes sense. Sounds like one of those would be a better condition than
the mere existence of compute support then?

  -ilia

On Wed, Apr 3, 2019 at 9:36 AM Marek Olšák <maraeo@gmail.com> wrote:
>
> On Wed, Apr 3, 2019 at 9:06 AM Ilia Mirkin <imirkin@alum.mit.edu> wrote:
>>
>> On Wed, Apr 3, 2019 at 8:38 AM Marek Olšák <maraeo@gmail.com> wrote:
>> >
>> > On Tue, Apr 2, 2019 at 2:14 PM Eric Anholt <eric@anholt.net> wrote:
>> >>
>> >> Ilia Mirkin <imirkin@alum.mit.edu> writes:
>> >>
>> >> > Shouldn't this sort of decision be left up to the driver? If the
>> >> > driver would like to use CS for blits, fine, but why not let it blit
>> >> > in the most optimal way possible and force it to use a compute shader?
>> >>
>> >> Yeah, commit messages require an explanation of why a change is being
>> >> made.
>> >
>> >
>> > We plan to create vaapi contexts with PIPE_CONTEXT_COMPUTE_ONLY for better GPU multitasking.
>> >
>> > RadeonSI uses async compute queues if PIPE_CONTEXT_COMPUTE_ONLY is set, so it can't do any graphics stuff, not even blit. (pipe_context::blit is NULL)
>>
>> Makes sense. Sounds like one of those would be a better condition than
>> the mere existence of compute support then?
>
>
> Or we can add PIPE_CAP_PREFER_COMPUTE_BLIT as a performance hint.

When would a driver set that, and when would a state tracker respect it?

As I see it, if the driver prefers compute blits, it can just do that
in its ->blit impl. If the state tracker created a
PIPE_CONTEXT_COMPUTE_ONLY context, then it can also decide to not use
->blit(). I don't see what the CAP adds, but perhaps I'm missing
something.

  -ilia