panfrost: Start tracking inter-batch dependencies
[mesa.git] / src / gallium / auxiliary / vl / vl_zscan.c
index fde27f396c526215fe042ab7f7b9a96245646d76..769fc71ff4243219efd1ffd21fa54303ac81ef12 100644 (file)
@@ -18,7 +18,7 @@
  * 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 TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * 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 <assert.h>
 
-#include <pipe/p_screen.h>
-#include <pipe/p_context.h>
+#include "pipe/p_screen.h"
+#include "pipe/p_context.h"
 
-#include <util/u_draw.h>
-#include <util/u_sampler.h>
-#include <util/u_inlines.h>
-#include <util/u_memory.h>
+#include "util/u_draw.h"
+#include "util/u_sampler.h"
+#include "util/u_inlines.h"
+#include "util/u_memory.h"
 
-#include <tgsi/tgsi_ureg.h>
+#include "tgsi/tgsi_ureg.h"
 
-#include <vl/vl_defines.h>
-#include <vl/vl_types.h>
+#include "vl_defines.h"
+#include "vl_types.h"
 
 #include "vl_zscan.h"
 #include "vl_vertex_buffers.h"
 
 enum VS_OUTPUT
 {
-   VS_O_VPOS,
-   VS_O_VTEX
+   VS_O_VPOS = 0,
+   VS_O_VTEX = 0
+};
+
+const int vl_zscan_normal_16[] =
+{
+   /* Zig-Zag scan pattern */
+    0, 1, 4, 8, 5, 2, 3, 6,
+    9,12,13,10, 7,11,14,15
 };
 
 const int vl_zscan_linear[] =
@@ -88,29 +95,47 @@ const int vl_zscan_alternate[] =
    38,46,54,62,39,47,55,63
 };
 
+const int vl_zscan_h265_up_right_diagonal_16[] =
+{
+   /* Up-right diagonal scan order for 4x4 blocks - see H.265 section 6.5.3. */
+    0,  4,  1,  8,  5,  2, 12,  9,
+    6,  3, 13, 10,  7, 14, 11, 15,
+};
+
+const int vl_zscan_h265_up_right_diagonal[] =
+{
+   /* Up-right diagonal scan order for 8x8 blocks - see H.265 section 6.5.3. */
+    0,  8,  1, 16,  9,  2, 24, 17,
+   10,  3, 32, 25, 18, 11,  4, 40,
+   33, 26, 19, 12,  5, 48, 41, 34,
+   27, 20, 13,  6, 56, 49, 42, 35,
+   28, 21, 14,  7, 57, 50, 43, 36,
+   29, 22, 15, 58, 51, 44, 37, 30,
+   23, 59, 52, 45, 38, 31, 60, 53,
+   46, 39, 61, 54, 47, 62, 55, 63,
+};
+
+
 static void *
 create_vert_shader(struct vl_zscan *zscan)
 {
    struct ureg_program *shader;
-
    struct ureg_src scale;
    struct ureg_src vrect, vpos, block_num;
-
    struct ureg_dst tmp;
    struct ureg_dst o_vpos;
    struct ureg_dst *o_vtex;
+   unsigned i;
 
-   signed i;
-
-   shader = ureg_create(TGSI_PROCESSOR_VERTEX);
+   shader = ureg_create(PIPE_SHADER_VERTEX);
    if (!shader)
       return NULL;
 
    o_vtex = MALLOC(zscan->num_channels * sizeof(struct ureg_dst));
 
    scale = ureg_imm2f(shader,
-      (float)BLOCK_WIDTH / zscan->buffer_width,
-      (float)BLOCK_HEIGHT / zscan->buffer_height);
+      (float)VL_BLOCK_WIDTH / zscan->buffer_width,
+      (float)VL_BLOCK_HEIGHT / zscan->buffer_height);
 
    vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
    vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
@@ -147,7 +172,8 @@ create_vert_shader(struct vl_zscan *zscan)
 
    for (i = 0; i < zscan->num_channels; ++i) {
       ureg_ADD(shader, ureg_writemask(tmp, TGSI_WRITEMASK_X), ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_Y),
-               ureg_imm1f(shader, 1.0f / (zscan->blocks_per_line * BLOCK_WIDTH) * (i - (signed)zscan->num_channels / 2)));
+               ureg_imm1f(shader, 1.0f / (zscan->blocks_per_line * VL_BLOCK_WIDTH)
+                * ((signed)i - (signed)zscan->num_channels / 2)));
 
       ureg_MAD(shader, ureg_writemask(o_vtex[i], TGSI_WRITEMASK_X), vrect,
                ureg_imm1f(shader, 1.0f / zscan->blocks_per_line), ureg_src(tmp));
@@ -178,7 +204,7 @@ create_frag_shader(struct vl_zscan *zscan)
 
    unsigned i;
 
-   shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
+   shader = ureg_create(PIPE_SHADER_FRAGMENT);
    if (!shader)
       return NULL;
 
@@ -269,7 +295,11 @@ init_state(struct vl_zscan *zscan)
    assert(zscan);
 
    memset(&rs_state, 0, sizeof(rs_state));
-   rs_state.gl_rasterization_rules = true;
+   rs_state.half_pixel_center = true;
+   rs_state.bottom_edge_rule = true;
+   rs_state.depth_clip_near = 1;
+   rs_state.depth_clip_far = 1;
+
    zscan->rs_state = zscan->pipe->create_rasterizer_state(zscan->pipe, &rs_state);
    if (!zscan->rs_state)
       goto error_rs_state;
@@ -342,7 +372,7 @@ cleanup_state(struct vl_zscan *zscan)
 struct pipe_sampler_view *
 vl_zscan_layout(struct pipe_context *pipe, const int layout[64], unsigned blocks_per_line)
 {
-   const unsigned total_size = blocks_per_line * BLOCK_WIDTH * BLOCK_HEIGHT;
+   const unsigned total_size = blocks_per_line * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT;
 
    int patched_layout[64];
 
@@ -355,8 +385,8 @@ vl_zscan_layout(struct pipe_context *pipe, const int layout[64], unsigned blocks
    struct pipe_box rect =
    {
       0, 0, 0,
-      BLOCK_WIDTH * blocks_per_line,
-      BLOCK_HEIGHT,
+      VL_BLOCK_WIDTH * blocks_per_line,
+      VL_BLOCK_HEIGHT,
       1
    };
 
@@ -368,8 +398,8 @@ vl_zscan_layout(struct pipe_context *pipe, const int layout[64], unsigned blocks
    memset(&res_tmpl, 0, sizeof(res_tmpl));
    res_tmpl.target = PIPE_TEXTURE_2D;
    res_tmpl.format = PIPE_FORMAT_R32_FLOAT;
-   res_tmpl.width0 = BLOCK_WIDTH * blocks_per_line;
-   res_tmpl.height0 = BLOCK_HEIGHT;
+   res_tmpl.width0 = VL_BLOCK_WIDTH * blocks_per_line;
+   res_tmpl.height0 = VL_BLOCK_HEIGHT;
    res_tmpl.depth0 = 1;
    res_tmpl.array_size = 1;
    res_tmpl.usage = PIPE_USAGE_IMMUTABLE;
@@ -379,34 +409,26 @@ vl_zscan_layout(struct pipe_context *pipe, const int layout[64], unsigned blocks
    if (!res)
       goto error_resource;
 
-   buf_transfer = pipe->get_transfer
-   (
-      pipe, res,
-      0, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
-      &rect
-   );
-   if (!buf_transfer)
-      goto error_transfer;
-
-   pitch = buf_transfer->stride / sizeof(float);
-
-   f = pipe->transfer_map(pipe, buf_transfer);
+   f = pipe->transfer_map(pipe, res,
+                          0, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD_RANGE,
+                          &rect, &buf_transfer);
    if (!f)
       goto error_map;
 
+   pitch = buf_transfer->stride / sizeof(float);
+
    for (i = 0; i < blocks_per_line; ++i)
-      for (y = 0; y < BLOCK_HEIGHT; ++y)
-         for (x = 0; x < BLOCK_WIDTH; ++x) {
-            float addr = patched_layout[x + y * BLOCK_WIDTH] +
-               i * BLOCK_WIDTH * BLOCK_HEIGHT;
+      for (y = 0; y < VL_BLOCK_HEIGHT; ++y)
+         for (x = 0; x < VL_BLOCK_WIDTH; ++x) {
+            float addr = patched_layout[x + y * VL_BLOCK_WIDTH] +
+               i * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT;
 
             addr /= total_size;
 
-            f[i * BLOCK_WIDTH + y * pitch + x] = addr;
+            f[i * VL_BLOCK_WIDTH + y * pitch + x] = addr;
          }
 
    pipe->transfer_unmap(pipe, buf_transfer);
-   pipe->transfer_destroy(pipe, buf_transfer);
 
    memset(&sv_tmpl, 0, sizeof(sv_tmpl));
    u_sampler_view_default_template(&sv_tmpl, res, res->format);
@@ -418,9 +440,6 @@ vl_zscan_layout(struct pipe_context *pipe, const int layout[64], unsigned blocks
    return sv;
 
 error_map:
-   pipe->transfer_destroy(pipe, buf_transfer);
-
-error_transfer:
    pipe_resource_reference(&res, NULL);
 
 error_resource:
@@ -473,18 +492,14 @@ vl_zscan_init_buffer(struct vl_zscan *zscan, struct vl_zscan_buffer *buffer,
 
    memset(buffer, 0, sizeof(struct vl_zscan_buffer));
 
-   buffer->zscan = zscan;
-
    pipe_sampler_view_reference(&buffer->src, src);
 
    buffer->viewport.scale[0] = dst->width;
    buffer->viewport.scale[1] = dst->height;
    buffer->viewport.scale[2] = 1;
-   buffer->viewport.scale[3] = 1;
    buffer->viewport.translate[0] = 0;
    buffer->viewport.translate[1] = 0;
    buffer->viewport.translate[2] = 0;
-   buffer->viewport.translate[3] = 0;
 
    buffer->fb_state.width = dst->width;
    buffer->fb_state.height = dst->height;
@@ -494,8 +509,8 @@ vl_zscan_init_buffer(struct vl_zscan *zscan, struct vl_zscan_buffer *buffer,
    memset(&res_tmpl, 0, sizeof(res_tmpl));
    res_tmpl.target = PIPE_TEXTURE_3D;
    res_tmpl.format = PIPE_FORMAT_R8_UNORM;
-   res_tmpl.width0 = BLOCK_WIDTH * zscan->blocks_per_line;
-   res_tmpl.height0 = BLOCK_HEIGHT;
+   res_tmpl.width0 = VL_BLOCK_WIDTH * zscan->blocks_per_line;
+   res_tmpl.height0 = VL_BLOCK_HEIGHT;
    res_tmpl.depth0 = 2;
    res_tmpl.array_size = 1;
    res_tmpl.usage = PIPE_USAGE_IMMUTABLE;
@@ -537,7 +552,8 @@ vl_zscan_set_layout(struct vl_zscan_buffer *buffer, struct pipe_sampler_view *la
 }
 
 void
-vl_zscan_upload_quant(struct vl_zscan_buffer *buffer, const uint8_t matrix[64], bool intra)
+vl_zscan_upload_quant(struct vl_zscan *zscan, struct vl_zscan_buffer *buffer,
+                      const uint8_t matrix[64], bool intra)
 {
    struct pipe_context *pipe;
    struct pipe_transfer *buf_transfer;
@@ -547,62 +563,48 @@ vl_zscan_upload_quant(struct vl_zscan_buffer *buffer, const uint8_t matrix[64],
    struct pipe_box rect =
    {
       0, 0, intra ? 1 : 0,
-      BLOCK_WIDTH,
-      BLOCK_HEIGHT,
+      VL_BLOCK_WIDTH,
+      VL_BLOCK_HEIGHT,
       1
    };
 
    assert(buffer);
    assert(matrix);
 
-   pipe = buffer->zscan->pipe;
+   pipe = zscan->pipe;
 
-   rect.width *= buffer->zscan->blocks_per_line;
+   rect.width *= zscan->blocks_per_line;
 
-   buf_transfer = pipe->get_transfer
-   (
-      pipe, buffer->quant->texture,
-      0, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
-      &rect
-   );
-   if (!buf_transfer)
-      goto error_transfer;
+   data = pipe->transfer_map(pipe, buffer->quant->texture,
+                             0, PIPE_TRANSFER_WRITE |
+                             PIPE_TRANSFER_DISCARD_RANGE,
+                             &rect, &buf_transfer);
+   if (!data)
+      return;
 
    pitch = buf_transfer->stride;
 
-   data = pipe->transfer_map(pipe, buf_transfer);
-   if (!data)
-      goto error_map;
-
-   for (i = 0; i < buffer->zscan->blocks_per_line; ++i)
-      for (y = 0; y < BLOCK_HEIGHT; ++y)
-         for (x = 0; x < BLOCK_WIDTH; ++x)
-            data[i * BLOCK_WIDTH + y * pitch + x] = matrix[x + y * BLOCK_WIDTH];
+   for (i = 0; i < zscan->blocks_per_line; ++i)
+      for (y = 0; y < VL_BLOCK_HEIGHT; ++y)
+         for (x = 0; x < VL_BLOCK_WIDTH; ++x)
+            data[i * VL_BLOCK_WIDTH + y * pitch + x] = matrix[x + y * VL_BLOCK_WIDTH];
 
    pipe->transfer_unmap(pipe, buf_transfer);
-
-error_map:
-   pipe->transfer_destroy(pipe, buf_transfer);
-
-error_transfer:
-   return;
 }
 
 void
-vl_zscan_render(struct vl_zscan_buffer *buffer, unsigned num_instances)
+vl_zscan_render(struct vl_zscan *zscan, struct vl_zscan_buffer *buffer, unsigned num_instances)
 {
-   struct vl_zscan *zscan;
-
    assert(buffer);
 
-   zscan = buffer->zscan;
-
    zscan->pipe->bind_rasterizer_state(zscan->pipe, zscan->rs_state);
    zscan->pipe->bind_blend_state(zscan->pipe, zscan->blend);
-   zscan->pipe->bind_fragment_sampler_states(zscan->pipe, 3, zscan->samplers);
+   zscan->pipe->bind_sampler_states(zscan->pipe, PIPE_SHADER_FRAGMENT,
+                                    0, 3, zscan->samplers);
    zscan->pipe->set_framebuffer_state(zscan->pipe, &buffer->fb_state);
-   zscan->pipe->set_viewport_state(zscan->pipe, &buffer->viewport);
-   zscan->pipe->set_fragment_sampler_views(zscan->pipe, 3, &buffer->src);
+   zscan->pipe->set_viewport_states(zscan->pipe, 0, 1, &buffer->viewport);
+   zscan->pipe->set_sampler_views(zscan->pipe, PIPE_SHADER_FRAGMENT,
+                                  0, 3, &buffer->src);
    zscan->pipe->bind_vs_state(zscan->pipe, zscan->vs);
    zscan->pipe->bind_fs_state(zscan->pipe, zscan->fs);
    util_draw_arrays_instanced(zscan->pipe, PIPE_PRIM_QUADS, 0, 4, 0, num_instances);