unsigned r300_get_swizzle_combined(const unsigned char *swizzle_format,
const unsigned char *swizzle_view,
boolean dxtc_swizzle)
{
unsigned i;
unsigned char swizzle[4];
unsigned result = 0;
const uint32_t swizzle_shift[4] = {
R300_TX_FORMAT_R_SHIFT,
R300_TX_FORMAT_G_SHIFT,
R300_TX_FORMAT_B_SHIFT,
R300_TX_FORMAT_A_SHIFT
};
uint32_t swizzle_bit[4] = {
dxtc_swizzle ? R300_TX_FORMAT_Z : R300_TX_FORMAT_X,
R300_TX_FORMAT_Y,
dxtc_swizzle ? R300_TX_FORMAT_X : R300_TX_FORMAT_Z,
R300_TX_FORMAT_W
};
if (swizzle_view) {
/* Combine two sets of swizzles. */
util_format_compose_swizzles(swizzle_format, swizzle_view, swizzle);
} else {
memcpy(swizzle, swizzle_format, 4);
}
/* Get swizzle. */
for (i = 0; i < 4; i++) {
switch (swizzle[i]) {
case UTIL_FORMAT_SWIZZLE_Y:
result |= swizzle_bit[1] << swizzle_shift[i];
break;
case UTIL_FORMAT_SWIZZLE_Z:
result |= swizzle_bit[2] << swizzle_shift[i];
break;
case UTIL_FORMAT_SWIZZLE_W:
result |= swizzle_bit[3] << swizzle_shift[i];
break;
case UTIL_FORMAT_SWIZZLE_0:
result |= R300_TX_FORMAT_ZERO << swizzle_shift[i];
break;
case UTIL_FORMAT_SWIZZLE_1:
result |= R300_TX_FORMAT_ONE << swizzle_shift[i];
break;
default: /* UTIL_FORMAT_SWIZZLE_X */
result |= swizzle_bit[0] << swizzle_shift[i];
}
}
return result;
}
uint32_t
fd4_tex_swiz(enum pipe_format format, unsigned swizzle_r, unsigned swizzle_g,
unsigned swizzle_b, unsigned swizzle_a)
{
const struct util_format_description *desc =
util_format_description(format);
unsigned char swiz[4] = {
swizzle_r, swizzle_g, swizzle_b, swizzle_a,
}, rswiz[4];
util_format_compose_swizzles(desc->swizzle, swiz, rswiz);
return A4XX_TEX_CONST_0_SWIZ_X(tex_swiz(rswiz[0])) |
A4XX_TEX_CONST_0_SWIZ_Y(tex_swiz(rswiz[1])) |
A4XX_TEX_CONST_0_SWIZ_Z(tex_swiz(rswiz[2])) |
A4XX_TEX_CONST_0_SWIZ_W(tex_swiz(rswiz[3]));
}
static void *
vc5_vertex_state_create(struct pipe_context *pctx, unsigned num_elements,
const struct pipe_vertex_element *elements)
{
struct vc5_context *vc5 = vc5_context(pctx);
struct vc5_vertex_stateobj *so = CALLOC_STRUCT(vc5_vertex_stateobj);
if (!so)
return NULL;
memcpy(so->pipe, elements, sizeof(*elements) * num_elements);
so->num_elements = num_elements;
for (int i = 0; i < so->num_elements; i++) {
const struct pipe_vertex_element *elem = &elements[i];
const struct util_format_description *desc =
util_format_description(elem->src_format);
uint32_t r_size = desc->channel[0].size;
struct V3D33_GL_SHADER_STATE_ATTRIBUTE_RECORD attr_unpacked = {
/* vec_size == 0 means 4 */
.vec_size = desc->nr_channels & 3,
.signed_int_type = (desc->channel[0].type ==
UTIL_FORMAT_TYPE_SIGNED),
.normalized_int_type = desc->channel[0].normalized,
.read_as_int_uint = desc->channel[0].pure_integer,
.instance_divisor = elem->instance_divisor,
};
switch (desc->channel[0].type) {
case UTIL_FORMAT_TYPE_FLOAT:
if (r_size == 32) {
attr_unpacked.type = ATTRIBUTE_FLOAT;
} else {
assert(r_size == 16);
attr_unpacked.type = ATTRIBUTE_HALF_FLOAT;
}
break;
case UTIL_FORMAT_TYPE_SIGNED:
case UTIL_FORMAT_TYPE_UNSIGNED:
switch (r_size) {
case 32:
attr_unpacked.type = ATTRIBUTE_INT;
break;
case 16:
attr_unpacked.type = ATTRIBUTE_SHORT;
break;
case 10:
attr_unpacked.type = ATTRIBUTE_INT2_10_10_10;
break;
case 8:
attr_unpacked.type = ATTRIBUTE_BYTE;
break;
default:
fprintf(stderr,
"format %s unsupported\n",
desc->name);
attr_unpacked.type = ATTRIBUTE_BYTE;
abort();
}
break;
default:
fprintf(stderr,
"format %s unsupported\n",
desc->name);
abort();
}
const uint32_t size =
cl_packet_length(GL_SHADER_STATE_ATTRIBUTE_RECORD);
V3D33_GL_SHADER_STATE_ATTRIBUTE_RECORD_pack(NULL,
(uint8_t *)&so->attrs[i * size],
&attr_unpacked);
}
/* Set up the default attribute values in case any of the vertex
* elements use them.
*/
so->default_attribute_values = vc5_bo_alloc(vc5->screen,
VC5_MAX_ATTRIBUTES *
4 * sizeof(float),
"default attributes");
uint32_t *attrs = vc5_bo_map(so->default_attribute_values);
for (int i = 0; i < VC5_MAX_ATTRIBUTES; i++) {
attrs[i * 4 + 0] = 0;
attrs[i * 4 + 1] = 0;
attrs[i * 4 + 2] = 0;
if (i < so->num_elements &&
util_format_is_pure_integer(so->pipe[i].src_format)) {
attrs[i * 4 + 3] = 1;
} else {
attrs[i * 4 + 3] = fui(1.0);
}
}
return so;
}
static void
vc5_vertex_state_bind(struct pipe_context *pctx, void *hwcso)
{
struct vc5_context *vc5 = vc5_context(pctx);
vc5->vtx = hwcso;
vc5->dirty |= VC5_DIRTY_VTXSTATE;
}
static void
vc5_set_constant_buffer(struct pipe_context *pctx, uint shader, uint index,
const struct pipe_constant_buffer *cb)
{
struct vc5_context *vc5 = vc5_context(pctx);
struct vc5_constbuf_stateobj *so = &vc5->constbuf[shader];
util_copy_constant_buffer(&so->cb[index], cb);
/* Note that the state tracker can unbind constant buffers by
* passing NULL here.
*/
if (unlikely(!cb)) {
so->enabled_mask &= ~(1 << index);
so->dirty_mask &= ~(1 << index);
return;
}
so->enabled_mask |= 1 << index;
so->dirty_mask |= 1 << index;
vc5->dirty |= VC5_DIRTY_CONSTBUF;
}
static void
vc5_set_framebuffer_state(struct pipe_context *pctx,
const struct pipe_framebuffer_state *framebuffer)
{
struct vc5_context *vc5 = vc5_context(pctx);
struct pipe_framebuffer_state *cso = &vc5->framebuffer;
unsigned i;
vc5->job = NULL;
for (i = 0; i < framebuffer->nr_cbufs; i++)
pipe_surface_reference(&cso->cbufs[i], framebuffer->cbufs[i]);
for (; i < vc5->framebuffer.nr_cbufs; i++)
pipe_surface_reference(&cso->cbufs[i], NULL);
cso->nr_cbufs = framebuffer->nr_cbufs;
pipe_surface_reference(&cso->zsbuf, framebuffer->zsbuf);
cso->width = framebuffer->width;
cso->height = framebuffer->height;
vc5->dirty |= VC5_DIRTY_FRAMEBUFFER;
}
static struct vc5_texture_stateobj *
vc5_get_stage_tex(struct vc5_context *vc5, enum pipe_shader_type shader)
{
switch (shader) {
case PIPE_SHADER_FRAGMENT:
vc5->dirty |= VC5_DIRTY_FRAGTEX;
return &vc5->fragtex;
break;
case PIPE_SHADER_VERTEX:
vc5->dirty |= VC5_DIRTY_VERTTEX;
return &vc5->verttex;
break;
default:
fprintf(stderr, "Unknown shader target %d\n", shader);
abort();
}
}
static uint32_t translate_wrap(uint32_t pipe_wrap, bool using_nearest)
{
switch (pipe_wrap) {
case PIPE_TEX_WRAP_REPEAT:
return 0;
case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
return 1;
case PIPE_TEX_WRAP_MIRROR_REPEAT:
return 2;
case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
return 3;
case PIPE_TEX_WRAP_CLAMP:
return (using_nearest ? 1 : 3);
default:
unreachable("Unknown wrap mode");
}
}
static void *
vc5_create_sampler_state(struct pipe_context *pctx,
const struct pipe_sampler_state *cso)
{
struct vc5_sampler_state *so = CALLOC_STRUCT(vc5_sampler_state);
if (!so)
return NULL;
memcpy(so, cso, sizeof(*cso));
bool either_nearest =
(cso->mag_img_filter == PIPE_TEX_MIPFILTER_NEAREST ||
cso->min_img_filter == PIPE_TEX_MIPFILTER_NEAREST);
struct V3D33_TEXTURE_UNIFORM_PARAMETER_0_CFG_MODE1 p0_unpacked = {
.s_wrap_mode = translate_wrap(cso->wrap_s, either_nearest),
.t_wrap_mode = translate_wrap(cso->wrap_t, either_nearest),
.r_wrap_mode = translate_wrap(cso->wrap_r, either_nearest),
};
V3D33_TEXTURE_UNIFORM_PARAMETER_0_CFG_MODE1_pack(NULL,
(uint8_t *)&so->p0,
&p0_unpacked);
struct V3D33_TEXTURE_SHADER_STATE state_unpacked = {
cl_packet_header(TEXTURE_SHADER_STATE),
.min_level_of_detail = MAX2(cso->min_lod, 0.0),
.depth_compare_function = cso->compare_func,
.fixed_bias = cso->lod_bias,
};
STATIC_ASSERT(ARRAY_SIZE(so->texture_shader_state) ==
cl_packet_length(TEXTURE_SHADER_STATE));
cl_packet_pack(TEXTURE_SHADER_STATE)(NULL, so->texture_shader_state,
&state_unpacked);
return so;
}
static void
vc5_sampler_states_bind(struct pipe_context *pctx,
enum pipe_shader_type shader, unsigned start,
unsigned nr, void **hwcso)
{
struct vc5_context *vc5 = vc5_context(pctx);
struct vc5_texture_stateobj *stage_tex = vc5_get_stage_tex(vc5, shader);
assert(start == 0);
unsigned i;
unsigned new_nr = 0;
for (i = 0; i < nr; i++) {
if (hwcso[i])
new_nr = i + 1;
stage_tex->samplers[i] = hwcso[i];
}
for (; i < stage_tex->num_samplers; i++) {
stage_tex->samplers[i] = NULL;
}
stage_tex->num_samplers = new_nr;
}
static uint32_t
translate_swizzle(unsigned char pipe_swizzle)
{
switch (pipe_swizzle) {
case PIPE_SWIZZLE_0:
return 0;
case PIPE_SWIZZLE_1:
return 1;
case PIPE_SWIZZLE_X:
case PIPE_SWIZZLE_Y:
case PIPE_SWIZZLE_Z:
case PIPE_SWIZZLE_W:
return 2 + pipe_swizzle;
default:
unreachable("unknown swizzle");
}
}
static struct pipe_sampler_view *
vc5_create_sampler_view(struct pipe_context *pctx, struct pipe_resource *prsc,
const struct pipe_sampler_view *cso)
{
struct vc5_sampler_view *so = CALLOC_STRUCT(vc5_sampler_view);
struct vc5_resource *rsc = vc5_resource(prsc);
if (!so)
return NULL;
so->base = *cso;
pipe_reference(NULL, &prsc->reference);
struct V3D33_TEXTURE_UNIFORM_PARAMETER_1_CFG_MODE1 unpacked = {
};
unpacked.return_word_0_of_texture_data = true;
if (vc5_get_tex_return_size(cso->format) == 16) {
unpacked.return_word_1_of_texture_data = true;
} else {
int chans = vc5_get_tex_return_channels(cso->format);
if (chans > 1)
unpacked.return_word_1_of_texture_data = true;
if (chans > 2)
unpacked.return_word_2_of_texture_data = true;
if (chans > 3)
unpacked.return_word_3_of_texture_data = true;
}
V3D33_TEXTURE_UNIFORM_PARAMETER_1_CFG_MODE1_pack(NULL,
(uint8_t *)&so->p1,
&unpacked);
/* Compute the sampler view's swizzle up front. This will be plugged
* into either the sampler (for 16-bit returns) or the shader's
* texture key (for 32)
*/
uint8_t view_swizzle[4] = {
cso->swizzle_r,
cso->swizzle_g,
cso->swizzle_b,
cso->swizzle_a
};
const uint8_t *fmt_swizzle = vc5_get_format_swizzle(so->base.format);
util_format_compose_swizzles(fmt_swizzle, view_swizzle, so->swizzle);
so->base.texture = prsc;
so->base.reference.count = 1;
so->base.context = pctx;
struct V3D33_TEXTURE_SHADER_STATE state_unpacked = {
cl_packet_header(TEXTURE_SHADER_STATE),
.image_width = prsc->width0,
.image_height = prsc->height0,
.image_depth = prsc->depth0,
.texture_type = rsc->tex_format,
.srgb = util_format_is_srgb(cso->format),
.base_level = cso->u.tex.first_level,
.array_stride_64_byte_aligned = rsc->cube_map_stride / 64,
};
/* Note: Contrary to the docs, the swizzle still applies even
* if the return size is 32. It's just that you probably want
* to swizzle in the shader, because you need the Y/Z/W
* channels to be defined.
*/
if (vc5_get_tex_return_size(cso->format) != 32) {
state_unpacked.swizzle_r = translate_swizzle(so->swizzle[0]);
state_unpacked.swizzle_g = translate_swizzle(so->swizzle[1]);
state_unpacked.swizzle_b = translate_swizzle(so->swizzle[2]);
state_unpacked.swizzle_a = translate_swizzle(so->swizzle[3]);
} else {
state_unpacked.swizzle_r = translate_swizzle(PIPE_SWIZZLE_X);
state_unpacked.swizzle_g = translate_swizzle(PIPE_SWIZZLE_Y);
state_unpacked.swizzle_b = translate_swizzle(PIPE_SWIZZLE_Z);
state_unpacked.swizzle_a = translate_swizzle(PIPE_SWIZZLE_W);
}
/* XXX: While we need to use this flag to enable tiled
* resource sharing (even a small shared buffer should be UIF,
* not UBLINEAR or raster), this is also at the moment
* patching up the fact that our resource layout's decisions
* about XOR don't quite match the HW's.
*/
switch (rsc->slices[0].tiling) {
case VC5_TILING_UIF_NO_XOR:
case VC5_TILING_UIF_XOR:
state_unpacked.level_0_is_strictly_uif = true;
state_unpacked.level_0_xor_enable = false;
break;
default:
break;
}
STATIC_ASSERT(ARRAY_SIZE(so->texture_shader_state) ==
cl_packet_length(TEXTURE_SHADER_STATE));
cl_packet_pack(TEXTURE_SHADER_STATE)(NULL, so->texture_shader_state,
&state_unpacked);
return &so->base;
}
static void
vc5_sampler_view_destroy(struct pipe_context *pctx,
struct pipe_sampler_view *view)
{
pipe_resource_reference(&view->texture, NULL);
free(view);
}
static void
vc5_set_sampler_views(struct pipe_context *pctx,
enum pipe_shader_type shader,
unsigned start, unsigned nr,
struct pipe_sampler_view **views)
{
struct vc5_context *vc5 = vc5_context(pctx);
struct vc5_texture_stateobj *stage_tex = vc5_get_stage_tex(vc5, shader);
unsigned i;
unsigned new_nr = 0;
assert(start == 0);
for (i = 0; i < nr; i++) {
if (views[i])
new_nr = i + 1;
pipe_sampler_view_reference(&stage_tex->textures[i], views[i]);
}
for (; i < stage_tex->num_textures; i++) {
pipe_sampler_view_reference(&stage_tex->textures[i], NULL);
}
stage_tex->num_textures = new_nr;
}
static struct pipe_stream_output_target *
vc5_create_stream_output_target(struct pipe_context *pctx,
struct pipe_resource *prsc,
unsigned buffer_offset,
unsigned buffer_size)
{
struct pipe_stream_output_target *target;
target = CALLOC_STRUCT(pipe_stream_output_target);
if (!target)
return NULL;
pipe_reference_init(&target->reference, 1);
pipe_resource_reference(&target->buffer, prsc);
target->context = pctx;
target->buffer_offset = buffer_offset;
target->buffer_size = buffer_size;
return target;
}
static void
vc5_stream_output_target_destroy(struct pipe_context *pctx,
struct pipe_stream_output_target *target)
{
pipe_resource_reference(&target->buffer, NULL);
free(target);
}
static void get_external_state(
struct r300_context* r300,
struct r300_fragment_program_external_state* state)
{
struct r300_textures_state *texstate = r300->textures_state.state;
struct r300_rs_state *rs = r300->rs_state.state;
unsigned i;
state->frag_clamp = rs ? rs->rs.clamp_fragment_color : 0;
for (i = 0; i < texstate->sampler_state_count; i++) {
struct r300_sampler_state *s = texstate->sampler_states[i];
struct r300_sampler_view *v = texstate->sampler_views[i];
struct r300_resource *t;
if (!s || !v) {
continue;
}
t = r300_resource(v->base.texture);
if (s->state.compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
state->unit[i].compare_mode_enabled = 1;
/* Fortunately, no need to translate this. */
state->unit[i].texture_compare_func = s->state.compare_func;
}
state->unit[i].non_normalized_coords = !s->state.normalized_coords;
state->unit[i].convert_unorm_to_snorm =
v->base.format == PIPE_FORMAT_RGTC1_SNORM ||
v->base.format == PIPE_FORMAT_LATC1_SNORM;
/* Pass texture swizzling to the compiler, some lowering passes need it. */
if (v->base.format == PIPE_FORMAT_RGTC1_SNORM ||
v->base.format == PIPE_FORMAT_LATC1_SNORM) {
unsigned char swizzle[4];
util_format_compose_swizzles(
util_format_description(v->base.format)->swizzle,
v->swizzle,
swizzle);
state->unit[i].texture_swizzle =
RC_MAKE_SWIZZLE(swizzle[0], swizzle[1],
swizzle[2], swizzle[3]);
} else if (state->unit[i].compare_mode_enabled) {
state->unit[i].texture_swizzle =
RC_MAKE_SWIZZLE(v->swizzle[0], v->swizzle[1],
v->swizzle[2], v->swizzle[3]);
}
/* XXX this should probably take into account STR, not just S. */
if (t->tex.is_npot) {
switch (s->state.wrap_s) {
case PIPE_TEX_WRAP_REPEAT:
state->unit[i].wrap_mode = RC_WRAP_REPEAT;
break;
case PIPE_TEX_WRAP_MIRROR_REPEAT:
state->unit[i].wrap_mode = RC_WRAP_MIRRORED_REPEAT;
break;
case PIPE_TEX_WRAP_MIRROR_CLAMP:
case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
state->unit[i].wrap_mode = RC_WRAP_MIRRORED_CLAMP;
break;
default:
state->unit[i].wrap_mode = RC_WRAP_NONE;
}
if (t->b.b.b.target == PIPE_TEXTURE_3D)
state->unit[i].clamp_and_scale_before_fetch = TRUE;
}
}
}