static nir_shader *
build_resolve_fragment_shader(struct radv_device *dev, bool is_integer, int samples)
{
nir_builder b;
char name[64];
const struct glsl_type *vec2 = glsl_vector_type(GLSL_TYPE_FLOAT, 2);
const struct glsl_type *vec4 = glsl_vec4_type();
const struct glsl_type *sampler_type = glsl_sampler_type(GLSL_SAMPLER_DIM_MS,
false,
false,
GLSL_TYPE_FLOAT);
snprintf(name, 64, "meta_resolve_fs-%d-%s", samples, is_integer ? "int" : "float");
nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_FRAGMENT, NULL);
b.shader->info.name = ralloc_strdup(b.shader, name);
nir_variable *input_img = nir_variable_create(b.shader, nir_var_uniform,
sampler_type, "s_tex");
input_img->data.descriptor_set = 0;
input_img->data.binding = 0;
nir_variable *fs_pos_in = nir_variable_create(b.shader, nir_var_shader_in, vec2, "fs_pos_in");
fs_pos_in->data.location = VARYING_SLOT_POS;
nir_variable *color_out = nir_variable_create(b.shader, nir_var_shader_out,
vec4, "f_color");
color_out->data.location = FRAG_RESULT_DATA0;
nir_ssa_def *pos_in = nir_load_var(&b, fs_pos_in);
nir_intrinsic_instr *src_offset = nir_intrinsic_instr_create(b.shader, nir_intrinsic_load_push_constant);
nir_intrinsic_set_base(src_offset, 0);
nir_intrinsic_set_range(src_offset, 8);
src_offset->src[0] = nir_src_for_ssa(nir_imm_int(&b, 0));
src_offset->num_components = 2;
nir_ssa_dest_init(&src_offset->instr, &src_offset->dest, 2, 32, "src_offset");
nir_builder_instr_insert(&b, &src_offset->instr);
nir_ssa_def *pos_int = nir_f2i32(&b, pos_in);
nir_ssa_def *img_coord = nir_channels(&b, nir_iadd(&b, pos_int, &src_offset->dest.ssa), 0x3);
nir_variable *color = nir_local_variable_create(b.impl, glsl_vec4_type(), "color");
radv_meta_build_resolve_shader_core(&b, is_integer, samples, input_img,
color, img_coord);
nir_ssa_def *outval = nir_load_var(&b, color);
nir_store_var(&b, color_out, outval, 0xf);
return b.shader;
}
static nir_ssa_def *
get_transform(lower_wpos_ytransform_state *state)
{
if (state->transform == NULL) {
/* NOTE: name must be prefixed w/ "gl_" to trigger slot based
* special handling in uniform setup:
*/
nir_variable *var = nir_variable_create(state->shader,
nir_var_uniform,
glsl_vec4_type(),
"gl_FbWposYTransform");
var->num_state_slots = 1;
var->state_slots = ralloc_array(var, nir_state_slot, 1);
var->state_slots[0].swizzle = SWIZZLE_XYZW;
memcpy(var->state_slots[0].tokens, state->options->state_tokens,
sizeof(var->state_slots[0].tokens));
state->transform = var;
}
return nir_load_var(&state->b, state->transform);
}
static nir_shader *
build_resolve_compute_shader(struct radv_device *dev, bool is_integer, int samples)
{
nir_builder b;
char name[64];
nir_if *outer_if = NULL;
const struct glsl_type *sampler_type = glsl_sampler_type(GLSL_SAMPLER_DIM_MS,
false,
false,
GLSL_TYPE_FLOAT);
const struct glsl_type *img_type = glsl_sampler_type(GLSL_SAMPLER_DIM_2D,
false,
false,
GLSL_TYPE_FLOAT);
snprintf(name, 64, "meta_resolve_cs-%d-%s", samples, is_integer ? "int" : "float");
nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_COMPUTE, NULL);
b.shader->info->name = ralloc_strdup(b.shader, name);
b.shader->info->cs.local_size[0] = 16;
b.shader->info->cs.local_size[1] = 16;
b.shader->info->cs.local_size[2] = 1;
nir_variable *input_img = nir_variable_create(b.shader, nir_var_uniform,
sampler_type, "s_tex");
input_img->data.descriptor_set = 0;
input_img->data.binding = 0;
nir_variable *output_img = nir_variable_create(b.shader, nir_var_uniform,
img_type, "out_img");
output_img->data.descriptor_set = 0;
output_img->data.binding = 1;
nir_ssa_def *invoc_id = nir_load_system_value(&b, nir_intrinsic_load_local_invocation_id, 0);
nir_ssa_def *wg_id = nir_load_system_value(&b, nir_intrinsic_load_work_group_id, 0);
nir_ssa_def *block_size = nir_imm_ivec4(&b,
b.shader->info->cs.local_size[0],
b.shader->info->cs.local_size[1],
b.shader->info->cs.local_size[2], 0);
nir_ssa_def *global_id = nir_iadd(&b, nir_imul(&b, wg_id, block_size), invoc_id);
nir_intrinsic_instr *src_offset = nir_intrinsic_instr_create(b.shader, nir_intrinsic_load_push_constant);
src_offset->src[0] = nir_src_for_ssa(nir_imm_int(&b, 0));
src_offset->num_components = 2;
nir_ssa_dest_init(&src_offset->instr, &src_offset->dest, 2, 32, "src_offset");
nir_builder_instr_insert(&b, &src_offset->instr);
nir_intrinsic_instr *dst_offset = nir_intrinsic_instr_create(b.shader, nir_intrinsic_load_push_constant);
dst_offset->src[0] = nir_src_for_ssa(nir_imm_int(&b, 8));
dst_offset->num_components = 2;
nir_ssa_dest_init(&dst_offset->instr, &dst_offset->dest, 2, 32, "dst_offset");
nir_builder_instr_insert(&b, &dst_offset->instr);
nir_ssa_def *img_coord = nir_iadd(&b, global_id, &src_offset->dest.ssa);
/* do a txf_ms on each sample */
nir_ssa_def *tmp;
nir_tex_instr *tex = nir_tex_instr_create(b.shader, 2);
tex->sampler_dim = GLSL_SAMPLER_DIM_MS;
tex->op = nir_texop_txf_ms;
tex->src[0].src_type = nir_tex_src_coord;
tex->src[0].src = nir_src_for_ssa(img_coord);
tex->src[1].src_type = nir_tex_src_ms_index;
tex->src[1].src = nir_src_for_ssa(nir_imm_int(&b, 0));
tex->dest_type = nir_type_float;
tex->is_array = false;
tex->coord_components = 2;
tex->texture = nir_deref_var_create(tex, input_img);
tex->sampler = NULL;
nir_ssa_dest_init(&tex->instr, &tex->dest, 4, 32, "tex");
nir_builder_instr_insert(&b, &tex->instr);
tmp = &tex->dest.ssa;
nir_variable *color =
nir_local_variable_create(b.impl, glsl_vec4_type(), "color");
if (!is_integer && samples > 1) {
nir_tex_instr *tex_all_same = nir_tex_instr_create(b.shader, 1);
tex_all_same->sampler_dim = GLSL_SAMPLER_DIM_MS;
tex_all_same->op = nir_texop_samples_identical;
tex_all_same->src[0].src_type = nir_tex_src_coord;
tex_all_same->src[0].src = nir_src_for_ssa(img_coord);
tex_all_same->dest_type = nir_type_float;
tex_all_same->is_array = false;
tex_all_same->coord_components = 2;
tex_all_same->texture = nir_deref_var_create(tex_all_same, input_img);
tex_all_same->sampler = NULL;
nir_ssa_dest_init(&tex_all_same->instr, &tex_all_same->dest, 1, 32, "tex");
nir_builder_instr_insert(&b, &tex_all_same->instr);
nir_ssa_def *all_same = nir_ine(&b, &tex_all_same->dest.ssa, nir_imm_int(&b, 0));
nir_if *if_stmt = nir_if_create(b.shader);
if_stmt->condition = nir_src_for_ssa(all_same);
nir_cf_node_insert(b.cursor, &if_stmt->cf_node);
b.cursor = nir_after_cf_list(&if_stmt->then_list);
for (int i = 1; i < samples; i++) {
nir_tex_instr *tex_add = nir_tex_instr_create(b.shader, 2);
tex_add->sampler_dim = GLSL_SAMPLER_DIM_MS;
tex_add->op = nir_texop_txf_ms;
tex_add->src[0].src_type = nir_tex_src_coord;
tex_add->src[0].src = nir_src_for_ssa(img_coord);
tex_add->src[1].src_type = nir_tex_src_ms_index;
tex_add->src[1].src = nir_src_for_ssa(nir_imm_int(&b, i));
tex_add->dest_type = nir_type_float;
tex_add->is_array = false;
tex_add->coord_components = 2;
tex_add->texture = nir_deref_var_create(tex_add, input_img);
tex_add->sampler = NULL;
nir_ssa_dest_init(&tex_add->instr, &tex_add->dest, 4, 32, "tex");
nir_builder_instr_insert(&b, &tex_add->instr);
tmp = nir_fadd(&b, tmp, &tex_add->dest.ssa);
}
tmp = nir_fdiv(&b, tmp, nir_imm_float(&b, samples));
nir_store_var(&b, color, tmp, 0xf);
b.cursor = nir_after_cf_list(&if_stmt->else_list);
outer_if = if_stmt;
}
nir_store_var(&b, color, &tex->dest.ssa, 0xf);
if (outer_if)
b.cursor = nir_after_cf_node(&outer_if->cf_node);
nir_ssa_def *newv = nir_load_var(&b, color);
nir_ssa_def *coord = nir_iadd(&b, global_id, &dst_offset->dest.ssa);
nir_intrinsic_instr *store = nir_intrinsic_instr_create(b.shader, nir_intrinsic_image_store);
store->src[0] = nir_src_for_ssa(coord);
store->src[1] = nir_src_for_ssa(nir_ssa_undef(&b, 1, 32));
store->src[2] = nir_src_for_ssa(newv);
store->variables[0] = nir_deref_var_create(store, output_img);
nir_builder_instr_insert(&b, &store->instr);
return b.shader;
}
/* see emit_wpos_adjustment() in st_mesa_to_tgsi.c */
static void
emit_wpos_adjustment(lower_wpos_ytransform_state *state,
nir_intrinsic_instr *intr,
bool invert, float adjX, float adjY[2])
{
nir_builder *b = &state->b;
nir_variable *fragcoord = intr->variables[0]->var;
nir_ssa_def *wpostrans, *wpos_temp, *wpos_temp_y, *wpos_input;
assert(intr->dest.is_ssa);
b->cursor = nir_before_instr(&intr->instr);
wpostrans = get_transform(state);
wpos_input = nir_load_var(b, fragcoord);
/* First, apply the coordinate shift: */
if (adjX || adjY[0] || adjY[1]) {
if (adjY[0] != adjY[1]) {
/* Adjust the y coordinate by adjY[1] or adjY[0] respectively
* depending on whether inversion is actually going to be applied
* or not, which is determined by testing against the inversion
* state variable used below, which will be either +1 or -1.
*/
nir_ssa_def *adj_temp;
adj_temp = nir_cmp(b,
nir_channel(b, wpostrans, invert ? 2 : 0),
nir_imm_vec4(b, adjX, adjY[0], 0.0f, 0.0f),
nir_imm_vec4(b, adjX, adjY[1], 0.0f, 0.0f));
wpos_temp = nir_fadd(b, wpos_input, adj_temp);
} else {
wpos_temp = nir_fadd(b,
wpos_input,
nir_imm_vec4(b, adjX, adjY[0], 0.0f, 0.0f));
}
wpos_input = wpos_temp;
} else {
/* MOV wpos_temp, input[wpos]
*/
wpos_temp = wpos_input;
}
/* Now the conditional y flip: STATE_FB_WPOS_Y_TRANSFORM.xy/zw will be
* inversion/identity, or the other way around if we're drawing to an FBO.
*/
if (invert) {
/* wpos_temp.y = wpos_input * wpostrans.xxxx + wpostrans.yyyy */
wpos_temp_y = nir_fadd(b, nir_fmul(b, nir_channel(b, wpos_temp, 1),
nir_channel(b, wpostrans, 0)),
nir_channel(b, wpostrans, 1));
} else {
/* wpos_temp.y = wpos_input * wpostrans.zzzz + wpostrans.wwww */
wpos_temp_y = nir_fadd(b, nir_fmul(b, nir_channel(b, wpos_temp, 1),
nir_channel(b, wpostrans, 2)),
nir_channel(b, wpostrans, 3));
}
wpos_temp = nir_vec4(b,
nir_channel(b, wpos_temp, 0),
wpos_temp_y,
nir_channel(b, wpos_temp, 2),
nir_channel(b, wpos_temp, 3));
nir_ssa_def_rewrite_uses(&intr->dest.ssa, nir_src_for_ssa(wpos_temp));
}
static void *vc4_get_yuv_vs(struct pipe_context *pctx)
{
struct vc4_context *vc4 = vc4_context(pctx);
struct pipe_screen *pscreen = pctx->screen;
if (vc4->yuv_linear_blit_vs)
return vc4->yuv_linear_blit_vs;
const struct nir_shader_compiler_options *options =
pscreen->get_compiler_options(pscreen,
PIPE_SHADER_IR_NIR,
PIPE_SHADER_VERTEX);
nir_builder b;
nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_VERTEX, options);
b.shader->info.name = ralloc_strdup(b.shader, "linear_blit_vs");
const struct glsl_type *vec4 = glsl_vec4_type();
nir_variable *pos_in = nir_variable_create(b.shader, nir_var_shader_in,
vec4, "pos");
nir_variable *pos_out = nir_variable_create(b.shader, nir_var_shader_out,
vec4, "gl_Position");
pos_out->data.location = VARYING_SLOT_POS;
nir_store_var(&b, pos_out, nir_load_var(&b, pos_in), 0xf);
struct pipe_shader_state shader_tmpl = {
.type = PIPE_SHADER_IR_NIR,
.ir.nir = b.shader,
};
vc4->yuv_linear_blit_vs = pctx->create_vs_state(pctx, &shader_tmpl);
return vc4->yuv_linear_blit_vs;
}
static void *vc4_get_yuv_fs(struct pipe_context *pctx, int cpp)
{
struct vc4_context *vc4 = vc4_context(pctx);
struct pipe_screen *pscreen = pctx->screen;
struct pipe_shader_state **cached_shader;
const char *name;
if (cpp == 1) {
cached_shader = &vc4->yuv_linear_blit_fs_8bit;
name = "linear_blit_8bit_fs";
} else {
cached_shader = &vc4->yuv_linear_blit_fs_16bit;
name = "linear_blit_16bit_fs";
}
if (*cached_shader)
return *cached_shader;
const struct nir_shader_compiler_options *options =
pscreen->get_compiler_options(pscreen,
PIPE_SHADER_IR_NIR,
PIPE_SHADER_FRAGMENT);
nir_builder b;
nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_FRAGMENT, options);
b.shader->info.name = ralloc_strdup(b.shader, name);
const struct glsl_type *vec4 = glsl_vec4_type();
const struct glsl_type *glsl_int = glsl_int_type();
nir_variable *color_out = nir_variable_create(b.shader, nir_var_shader_out,
vec4, "f_color");
color_out->data.location = FRAG_RESULT_COLOR;
nir_variable *pos_in = nir_variable_create(b.shader, nir_var_shader_in,
vec4, "pos");
pos_in->data.location = VARYING_SLOT_POS;
nir_ssa_def *pos = nir_load_var(&b, pos_in);
nir_ssa_def *one = nir_imm_int(&b, 1);
nir_ssa_def *two = nir_imm_int(&b, 2);
nir_ssa_def *x = nir_f2i32(&b, nir_channel(&b, pos, 0));
nir_ssa_def *y = nir_f2i32(&b, nir_channel(&b, pos, 1));
nir_variable *stride_in = nir_variable_create(b.shader, nir_var_uniform,
glsl_int, "stride");
nir_ssa_def *stride = nir_load_var(&b, stride_in);
nir_ssa_def *x_offset;
nir_ssa_def *y_offset;
if (cpp == 1) {
nir_ssa_def *intra_utile_x_offset =
nir_ishl(&b, nir_iand(&b, x, one), two);
nir_ssa_def *inter_utile_x_offset =
nir_ishl(&b, nir_iand(&b, x, nir_imm_int(&b, ~3)), one);
x_offset = nir_iadd(&b,
intra_utile_x_offset,
inter_utile_x_offset);
y_offset = nir_imul(&b,
nir_iadd(&b,
nir_ishl(&b, y, one),
nir_ushr(&b, nir_iand(&b, x, two), one)),
stride);
} else {
x_offset = nir_ishl(&b, x, two);
y_offset = nir_imul(&b, y, stride);
}
nir_intrinsic_instr *load =
nir_intrinsic_instr_create(b.shader, nir_intrinsic_load_ubo);
load->num_components = 1;
nir_ssa_dest_init(&load->instr, &load->dest, load->num_components, 32, NULL);
load->src[0] = nir_src_for_ssa(one);
load->src[1] = nir_src_for_ssa(nir_iadd(&b, x_offset, y_offset));
nir_builder_instr_insert(&b, &load->instr);
nir_store_var(&b, color_out,
nir_unpack_unorm_4x8(&b, &load->dest.ssa),
0xf);
struct pipe_shader_state shader_tmpl = {
.type = PIPE_SHADER_IR_NIR,
.ir.nir = b.shader,
};
*cached_shader = pctx->create_fs_state(pctx, &shader_tmpl);
return *cached_shader;
}
static bool
vc4_yuv_blit(struct pipe_context *pctx, const struct pipe_blit_info *info)
{
struct vc4_context *vc4 = vc4_context(pctx);
struct vc4_resource *src = vc4_resource(info->src.resource);
struct vc4_resource *dst = vc4_resource(info->dst.resource);
bool ok;
if (src->tiled)
return false;
if (src->base.format != PIPE_FORMAT_R8_UNORM &&
src->base.format != PIPE_FORMAT_R8G8_UNORM)
return false;
/* YUV blits always turn raster-order to tiled */
assert(dst->base.format == src->base.format);
assert(dst->tiled);
/* Always 1:1 and at the origin */
assert(info->src.box.x == 0 && info->dst.box.x == 0);
assert(info->src.box.y == 0 && info->dst.box.y == 0);
assert(info->src.box.width == info->dst.box.width);
assert(info->src.box.height == info->dst.box.height);
if ((src->slices[info->src.level].offset & 3) ||
(src->slices[info->src.level].stride & 3)) {
perf_debug("YUV-blit src texture offset/stride misaligned: 0x%08x/%d\n",
src->slices[info->src.level].offset,
src->slices[info->src.level].stride);
goto fallback;
}
vc4_blitter_save(vc4);
/* Create a renderable surface mapping the T-tiled shadow buffer.
*/
struct pipe_surface dst_tmpl;
util_blitter_default_dst_texture(&dst_tmpl, info->dst.resource,
info->dst.level, info->dst.box.z);
dst_tmpl.format = PIPE_FORMAT_RGBA8888_UNORM;
struct pipe_surface *dst_surf =
pctx->create_surface(pctx, info->dst.resource, &dst_tmpl);
if (!dst_surf) {
fprintf(stderr, "Failed to create YUV dst surface\n");
util_blitter_unset_running_flag(vc4->blitter);
return false;
}
dst_surf->width /= 2;
if (dst->cpp == 1)
dst_surf->height /= 2;
/* Set the constant buffer. */
uint32_t stride = src->slices[info->src.level].stride;
struct pipe_constant_buffer cb_uniforms = {
.user_buffer = &stride,
.buffer_size = sizeof(stride),
};
pctx->set_constant_buffer(pctx, PIPE_SHADER_FRAGMENT, 0, &cb_uniforms);
struct pipe_constant_buffer cb_src = {
.buffer = info->src.resource,
.buffer_offset = src->slices[info->src.level].offset,
.buffer_size = (src->bo->size -
src->slices[info->src.level].offset),
};
pctx->set_constant_buffer(pctx, PIPE_SHADER_FRAGMENT, 1, &cb_src);
/* Unbind the textures, to make sure we don't try to recurse into the
* shadow blit.
*/
pctx->set_sampler_views(pctx, PIPE_SHADER_FRAGMENT, 0, 0, NULL);
pctx->bind_sampler_states(pctx, PIPE_SHADER_FRAGMENT, 0, 0, NULL);
util_blitter_custom_shader(vc4->blitter, dst_surf,
vc4_get_yuv_vs(pctx),
vc4_get_yuv_fs(pctx, src->cpp));
util_blitter_restore_textures(vc4->blitter);
util_blitter_restore_constant_buffer_state(vc4->blitter);
/* Restore cb1 (util_blitter doesn't handle this one). */
struct pipe_constant_buffer cb_disabled = { 0 };
pctx->set_constant_buffer(pctx, PIPE_SHADER_FRAGMENT, 1, &cb_disabled);
pipe_surface_reference(&dst_surf, NULL);
return true;
fallback:
/* Do an immediate SW fallback, since the render blit path
* would just recurse.
*/
ok = util_try_blit_via_copy_region(pctx, info);
assert(ok); (void)ok;
return true;
}
static bool
vc4_render_blit(struct pipe_context *ctx, struct pipe_blit_info *info)
{
struct vc4_context *vc4 = vc4_context(ctx);
if (!util_blitter_is_blit_supported(vc4->blitter, info)) {
fprintf(stderr, "blit unsupported %s -> %s\n",
util_format_short_name(info->src.resource->format),
util_format_short_name(info->dst.resource->format));
return false;
}
/* Enable the scissor, so we get a minimal set of tiles rendered. */
if (!info->scissor_enable) {
info->scissor_enable = true;
info->scissor.minx = info->dst.box.x;
info->scissor.miny = info->dst.box.y;
info->scissor.maxx = info->dst.box.x + info->dst.box.width;
info->scissor.maxy = info->dst.box.y + info->dst.box.height;
}
vc4_blitter_save(vc4);
util_blitter_blit(vc4->blitter, info);
return true;
}
/* Optimal hardware path for blitting pixels.
* Scaling, format conversion, up- and downsampling (resolve) are allowed.
*/
void
vc4_blit(struct pipe_context *pctx, const struct pipe_blit_info *blit_info)
{
struct pipe_blit_info info = *blit_info;
if (vc4_yuv_blit(pctx, blit_info))
return;
if (vc4_tile_blit(pctx, blit_info))
return;
if (info.mask & PIPE_MASK_S) {
if (util_try_blit_via_copy_region(pctx, &info))
return;
info.mask &= ~PIPE_MASK_S;
fprintf(stderr, "cannot blit stencil, skipping\n");
}
if (vc4_render_blit(pctx, &info))
return;
fprintf(stderr, "Unsupported blit\n");
}
