#include "vl_idct.h"
#include "vl_vertex_buffers.h"
+#include "vl_defines.h"
#include "util/u_draw.h"
#include <assert.h>
#include <pipe/p_context.h>
#include <tgsi/tgsi_ureg.h>
#include "vl_types.h"
-#define BLOCK_WIDTH 8
-#define BLOCK_HEIGHT 8
-
#define SCALE_FACTOR_16_TO_9 (32768.0f / 256.0f)
#define NR_RENDER_TARGETS 4
-enum VS_INPUT
-{
- VS_I_RECT,
- VS_I_VPOS,
-
- NUM_VS_INPUTS
-};
-
enum VS_OUTPUT
{
VS_O_VPOS,
}
static void *
-create_vert_shader(struct vl_idct *idct, bool matrix_stage)
+create_vert_shader(struct vl_idct *idct, bool matrix_stage, int color_swizzle)
{
struct ureg_program *shader;
- struct ureg_src scale;
- struct ureg_src vrect, vpos;
+ struct ureg_src vrect, vpos, vblock, eb[4];
+ struct ureg_src scale, blocks_xy, t_eb;
struct ureg_dst t_tex, t_start;
struct ureg_dst o_vpos, o_l_addr[2], o_r_addr[2];
+ unsigned label;
shader = ureg_create(TGSI_PROCESSOR_VERTEX);
if (!shader)
vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
+ vblock = ureg_swizzle(vrect, TGSI_SWIZZLE_Z, TGSI_SWIZZLE_W, TGSI_SWIZZLE_X, TGSI_SWIZZLE_X);
o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
+ eb[0] = ureg_DECL_vs_input(shader, VS_I_EB_0_0);
+ eb[1] = ureg_DECL_vs_input(shader, VS_I_EB_1_0);
+ eb[2] = ureg_DECL_vs_input(shader, VS_I_EB_0_1);
+ eb[3] = ureg_DECL_vs_input(shader, VS_I_EB_1_1);
+
o_l_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0);
o_l_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1);
/*
* scale = (BLOCK_WIDTH, BLOCK_HEIGHT) / (dst.width, dst.height)
+ * blocks_xy = (blocks_x, blocks_y)
*
- * t_vpos = vpos + vrect
- * o_vpos.xy = t_vpos * scale
- * o_vpos.zw = vpos
+ * ar = vblock.y * blocks.x + vblock.x
+ * if eb[ar].(color_swizzle)
+ * o_vpos.xy = -1
+ * else
+ * t_tex = vpos * blocks_xy + vblock
+ * t_start = t_tex * scale
+ * t_tex = t_tex + vrect
+ * o_vpos.xy = t_tex * scale
*
- * o_l_addr = calc_addr(...)
- * o_r_addr = calc_addr(...)
+ * o_l_addr = calc_addr(...)
+ * o_r_addr = calc_addr(...)
+ * endif
+ * o_vpos.zw = vpos
*
*/
+
scale = ureg_imm2f(shader,
(float)BLOCK_WIDTH / idct->buffer_width,
(float)BLOCK_HEIGHT / idct->buffer_height);
- ureg_ADD(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), vpos, vrect);
- ureg_MUL(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), ureg_src(t_tex), scale);
- ureg_MUL(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_Z),
- ureg_scalar(vrect, TGSI_SWIZZLE_X),
- ureg_imm1f(shader, BLOCK_WIDTH / NR_RENDER_TARGETS));
+ blocks_xy = ureg_imm2f(shader, idct->blocks_x, idct->blocks_y);
- ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), ureg_src(t_tex));
- ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), vpos);
+ if (idct->blocks_x > 1 || idct->blocks_y > 1) {
+ struct ureg_dst ar = ureg_DECL_address(shader);
- ureg_MUL(shader, ureg_writemask(t_start, TGSI_WRITEMASK_XY), vpos, scale);
+ ureg_MAD(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_X),
+ ureg_scalar(vblock, TGSI_SWIZZLE_Y), blocks_xy, vblock);
- if(matrix_stage) {
- calc_addr(shader, o_l_addr, ureg_src(t_tex), ureg_src(t_start), false, false, idct->buffer_width / 4);
- calc_addr(shader, o_r_addr, vrect, ureg_imm1f(shader, 0.0f), true, true, BLOCK_WIDTH / 4);
+ ureg_ARL(shader, ureg_writemask(ar, TGSI_WRITEMASK_X), ureg_src(t_tex));
+ t_eb = ureg_src_indirect(eb[0], ureg_src(ar));
} else {
- calc_addr(shader, o_l_addr, vrect, ureg_imm1f(shader, 0.0f), false, false, BLOCK_WIDTH / 4);
- calc_addr(shader, o_r_addr, ureg_src(t_tex), ureg_src(t_start), true, false, idct->buffer_height / 4);
+ t_eb = eb[0];
}
+ ureg_IF(shader, ureg_scalar(t_eb, color_swizzle), &label);
+
+ ureg_MOV(shader, o_vpos, ureg_imm1f(shader, -1.0f));
+
+ ureg_fixup_label(shader, label, ureg_get_instruction_number(shader));
+ ureg_ELSE(shader, &label);
+
+ ureg_MAD(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), vpos, blocks_xy, vblock);
+ ureg_MUL(shader, ureg_writemask(t_start, TGSI_WRITEMASK_XY), ureg_src(t_tex), scale);
+
+ ureg_ADD(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), ureg_src(t_tex), vrect);
+
+ ureg_MUL(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), ureg_src(t_tex), scale);
+ ureg_MUL(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_Z),
+ ureg_scalar(vrect, TGSI_SWIZZLE_X),
+ ureg_imm1f(shader, BLOCK_WIDTH / NR_RENDER_TARGETS));
+
+ ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), ureg_src(t_tex));
+
+ if(matrix_stage) {
+ calc_addr(shader, o_l_addr, ureg_src(t_tex), ureg_src(t_start), false, false, idct->buffer_width / 4);
+ calc_addr(shader, o_r_addr, vrect, ureg_imm1f(shader, 0.0f), true, true, BLOCK_WIDTH / 4);
+ } else {
+ calc_addr(shader, o_l_addr, vrect, ureg_imm1f(shader, 0.0f), false, false, BLOCK_WIDTH / 4);
+ calc_addr(shader, o_r_addr, ureg_src(t_tex), ureg_src(t_start), true, false, idct->buffer_height / 4);
+ }
+
+ ureg_fixup_label(shader, label, ureg_get_instruction_number(shader));
+ ureg_ENDIF(shader);
+
+ ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), vpos);
+
ureg_release_temporary(shader, t_tex);
ureg_release_temporary(shader, t_start);
unsigned wm_start = (right_side == transposed) ? TGSI_WRITEMASK_X : TGSI_WRITEMASK_Y;
unsigned wm_tc = (right_side == transposed) ? TGSI_WRITEMASK_Y : TGSI_WRITEMASK_X;
- /*
- * daddr[0..1].(start) = saddr[0..1].(start)
- * daddr[0..1].(tc) = saddr[0..1].(tc)
+ /*
+ * daddr[0..1].(start) = saddr[0..1].(start)
+ * daddr[0..1].(tc) = saddr[0..1].(tc)
*/
-
+
ureg_MOV(shader, ureg_writemask(daddr[0], wm_start), saddr[0]);
ureg_ADD(shader, ureg_writemask(daddr[0], wm_tc), saddr[0], ureg_imm1f(shader, pos / size));
ureg_MOV(shader, ureg_writemask(daddr[1], wm_start), saddr[1]);
s_addr[1] = i == 0 ? l_addr[1] : ureg_src(l[i][1]);
fetch_four(shader, l[i], s_addr, ureg_DECL_sampler(shader, 1));
}
-
+
for (i = 0; i < NR_RENDER_TARGETS; ++i) {
if(i > 0)
increment_addr(shader, r, r_addr, true, true, i, BLOCK_HEIGHT);
}
static bool
-init_shaders(struct vl_idct *idct)
+init_shaders(struct vl_idct *idct, int color_swizzle)
{
- idct->matrix_vs = create_vert_shader(idct, true);
+ idct->matrix_vs = create_vert_shader(idct, true, color_swizzle);
idct->matrix_fs = create_matrix_frag_shader(idct);
- idct->transpose_vs = create_vert_shader(idct, false);
+ idct->transpose_vs = create_vert_shader(idct, false, color_swizzle);
idct->transpose_fs = create_transpose_frag_shader(idct);
- return
+ return
idct->matrix_vs != NULL &&
idct->matrix_fs != NULL &&
idct->transpose_vs != NULL &&
static bool
init_state(struct vl_idct *idct)
{
- struct pipe_vertex_element vertex_elems[NUM_VS_INPUTS];
struct pipe_sampler_state sampler;
struct pipe_rasterizer_state rs_state;
unsigned i;
assert(idct);
- idct->quad = vl_vb_upload_quads(idct->pipe, idct->max_blocks);
-
- if(idct->quad.buffer == NULL)
- return false;
-
for (i = 0; i < 4; ++i) {
memset(&sampler, 0, sizeof(sampler));
sampler.wrap_s = PIPE_TEX_WRAP_REPEAT;
rs_state.gl_rasterization_rules = false;
idct->rs_state = idct->pipe->create_rasterizer_state(idct->pipe, &rs_state);
- vertex_elems[VS_I_RECT] = vl_vb_get_quad_vertex_element();
-
- /* Pos element */
- vertex_elems[VS_I_VPOS].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
- idct->vertex_buffer_stride = vl_vb_element_helper(&vertex_elems[VS_I_VPOS], 1, 1);
- idct->vertex_elems_state = idct->pipe->create_vertex_elements_state(idct->pipe, 2, vertex_elems);
-
return true;
}
idct->pipe->delete_sampler_state(idct->pipe, idct->samplers.all[i]);
idct->pipe->delete_rasterizer_state(idct->pipe, idct->rs_state);
- idct->pipe->delete_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
}
static bool
/* create textures */
memset(&template, 0, sizeof(struct pipe_resource));
template.last_level = 0;
- template.depth0 = 1;
template.bind = PIPE_BIND_SAMPLER_VIEW;
template.flags = 0;
template.width0 = idct->buffer_width / 4;
template.height0 = idct->buffer_height;
template.depth0 = 1;
+ template.array_size = 1;
template.usage = PIPE_USAGE_STREAM;
buffer->textures.individual.source = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
buffer->sampler_views.all[i] = idct->pipe->create_sampler_view(idct->pipe, buffer->textures.all[i], &sampler_view);
}
+ template.target = PIPE_TEXTURE_2D;
+ /* TODO: Accomodate HW that can't do this and also for cases when this isn't precise enough */
+ template.format = PIPE_FORMAT_R16_SNORM;
+ template.width0 = idct->buffer_width;
+ template.height0 = idct->buffer_height;
+ template.depth0 = 1;
+
+ buffer->destination = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
+
return true;
}
}
}
-static bool
-init_vertex_buffers(struct vl_idct *idct, struct vl_idct_buffer *buffer)
-{
- assert(idct && buffer);
-
- buffer->vertex_bufs.individual.quad.stride = idct->quad.stride;
- buffer->vertex_bufs.individual.quad.max_index = idct->quad.max_index;
- buffer->vertex_bufs.individual.quad.buffer_offset = idct->quad.buffer_offset;
- pipe_resource_reference(&buffer->vertex_bufs.individual.quad.buffer, idct->quad.buffer);
-
- buffer->vertex_bufs.individual.pos = vl_vb_init(
- &buffer->blocks, idct->pipe, idct->max_blocks, 2,
- idct->vertex_buffer_stride);
-
- if(buffer->vertex_bufs.individual.pos.buffer == NULL)
- return false;
-
- return true;
-}
-
-static void
-cleanup_vertex_buffers(struct vl_idct *idct, struct vl_idct_buffer *buffer)
-{
- assert(idct && buffer);
-
- pipe_resource_reference(&buffer->vertex_bufs.individual.quad.buffer, NULL);
- pipe_resource_reference(&buffer->vertex_bufs.individual.pos.buffer, NULL);
-
- vl_vb_cleanup(&buffer->blocks);
-}
-
struct pipe_resource *
vl_idct_upload_matrix(struct pipe_context *pipe)
{
+ const float scale = sqrtf(SCALE_FACTOR_16_TO_9);
+
struct pipe_resource template, *matrix;
struct pipe_transfer *buf_transfer;
unsigned i, j, pitch;
template.width0 = 2;
template.height0 = 8;
template.depth0 = 1;
+ template.array_size = 1;
template.usage = PIPE_USAGE_IMMUTABLE;
template.bind = PIPE_BIND_SAMPLER_VIEW;
template.flags = 0;
for(i = 0; i < BLOCK_HEIGHT; ++i)
for(j = 0; j < BLOCK_WIDTH; ++j)
// transpose and scale
- f[i * pitch + j] = const_matrix[j][i] * sqrtf(SCALE_FACTOR_16_TO_9);
+ f[i * pitch + j] = const_matrix[j][i] * scale;
pipe->transfer_unmap(pipe, buf_transfer);
pipe->transfer_destroy(pipe, buf_transfer);
return matrix;
}
-bool vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe,
+bool vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe,
unsigned buffer_width, unsigned buffer_height,
- struct pipe_resource *matrix)
+ unsigned blocks_x, unsigned blocks_y,
+ int color_swizzle, struct pipe_resource *matrix)
{
assert(idct && pipe && matrix);
idct->pipe = pipe;
idct->buffer_width = buffer_width;
idct->buffer_height = buffer_height;
+ idct->blocks_x = blocks_x;
+ idct->blocks_y = blocks_y;
pipe_resource_reference(&idct->matrix, matrix);
- idct->max_blocks =
- align(buffer_width, BLOCK_WIDTH) / BLOCK_WIDTH *
- align(buffer_height, BLOCK_HEIGHT) / BLOCK_HEIGHT;
-
- if(!init_shaders(idct))
+ if(!init_shaders(idct, color_swizzle))
return false;
if(!init_state(idct)) {
pipe_resource_reference(&idct->matrix, NULL);
}
-bool
-vl_idct_init_buffer(struct vl_idct *idct, struct vl_idct_buffer *buffer, struct pipe_resource *dst)
+struct pipe_resource *
+vl_idct_init_buffer(struct vl_idct *idct, struct vl_idct_buffer *buffer)
{
struct pipe_surface template;
assert(buffer);
assert(idct);
- assert(dst);
pipe_resource_reference(&buffer->textures.individual.matrix, idct->matrix);
pipe_resource_reference(&buffer->textures.individual.transpose, idct->matrix);
- pipe_resource_reference(&buffer->destination, dst);
if (!init_textures(idct, buffer))
- return false;
-
- if (!init_vertex_buffers(idct, buffer))
- return false;
+ return NULL;
/* init state */
buffer->viewport[0].scale[0] = buffer->textures.individual.intermediate->width0;
buffer->fb_state[i].zsbuf = NULL;
}
- return true;
+ return buffer->destination;
}
void
idct->pipe->surface_destroy(idct->pipe, buffer->fb_state[1].cbufs[0]);
cleanup_textures(idct, buffer);
- cleanup_vertex_buffers(idct, buffer);
}
void
);
buffer->texels = idct->pipe->transfer_map(idct->pipe, buffer->tex_transfer);
-
- vl_vb_map(&buffer->blocks, idct->pipe);
}
void
vl_idct_add_block(struct vl_idct_buffer *buffer, unsigned x, unsigned y, short *block)
{
- struct vertex2f v;
unsigned tex_pitch;
short *texels;
for (i = 0; i < BLOCK_HEIGHT; ++i)
memcpy(texels + i * tex_pitch, block + i * BLOCK_WIDTH, BLOCK_WIDTH * sizeof(short));
-
- v.x = x;
- v.y = y;
- vl_vb_add_block(&buffer->blocks, (float*)&v);
}
void
idct->pipe->transfer_unmap(idct->pipe, buffer->tex_transfer);
idct->pipe->transfer_destroy(idct->pipe, buffer->tex_transfer);
- vl_vb_unmap(&buffer->blocks, idct->pipe);
}
void
-vl_idct_flush(struct vl_idct *idct, struct vl_idct_buffer *buffer)
+vl_idct_flush(struct vl_idct *idct, struct vl_idct_buffer *buffer, unsigned num_instances)
{
unsigned num_verts;
assert(idct);
+ assert(buffer);
- num_verts = vl_vb_restart(&buffer->blocks);
-
- if(num_verts > 0) {
+ if(num_instances > 0) {
+ num_verts = idct->blocks_x * idct->blocks_y * 4;
idct->pipe->bind_rasterizer_state(idct->pipe, idct->rs_state);
- idct->pipe->set_vertex_buffers(idct->pipe, 2, buffer->vertex_bufs.all);
- idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
/* first stage */
idct->pipe->set_framebuffer_state(idct->pipe, &buffer->fb_state[0]);
idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[0]);
idct->pipe->bind_vs_state(idct->pipe, idct->matrix_vs);
idct->pipe->bind_fs_state(idct->pipe, idct->matrix_fs);
- util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, num_verts);
+ util_draw_arrays_instanced(idct->pipe, PIPE_PRIM_QUADS, 0, num_verts, 0, num_instances);
/* second stage */
idct->pipe->set_framebuffer_state(idct->pipe, &buffer->fb_state[1]);
idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[1]);
idct->pipe->bind_vs_state(idct->pipe, idct->transpose_vs);
idct->pipe->bind_fs_state(idct->pipe, idct->transpose_fs);
- util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, num_verts);
+ util_draw_arrays_instanced(idct->pipe, PIPE_PRIM_QUADS, 0, num_verts, 0, num_instances);
}
}