static void
gen6_emit_launch_grid_surface_const(struct ilo_render *r,
const struct ilo_state_vector *vec,
struct ilo_render_launch_grid_session *session)
{
const struct ilo_shader_state *cs = vec->cs;
uint32_t *surface_state = r->state.cs.SURFACE_STATE;
struct ilo_view_surface view;
int base, count;
ILO_DEV_ASSERT(r->dev, 7, 7.5);
base = ilo_shader_get_kernel_param(cs, ILO_KERNEL_SURFACE_CONST_BASE);
count = ilo_shader_get_kernel_param(cs, ILO_KERNEL_SURFACE_CONST_COUNT);
if (!count)
return;
ilo_gpe_init_view_surface_for_buffer(r->dev,
ilo_buffer(session->input->buffer),
session->input->buffer_offset,
session->input->buffer_size,
1, PIPE_FORMAT_NONE,
false, false, &view);
assert(count == 1 && session->input->buffer);
surface_state[base] = gen6_SURFACE_STATE(r->builder, &view, false);
}
static void
finalize_global_binding(struct ilo_state_vector *vec)
{
struct ilo_shader_state *cs = vec->cs;
int base, count, shift;
int i;
count = ilo_shader_get_kernel_param(cs,
ILO_KERNEL_CS_SURFACE_GLOBAL_COUNT);
if (!count)
return;
base = ilo_shader_get_kernel_param(cs, ILO_KERNEL_CS_SURFACE_GLOBAL_BASE);
shift = 32 - util_last_bit(base + count - 1);
if (count > vec->global_binding.count)
count = vec->global_binding.count;
for (i = 0; i < count; i++) {
struct ilo_global_binding_cso *cso =
util_dynarray_element(&vec->global_binding.bindings,
struct ilo_global_binding_cso, i);
const uint32_t offset = *cso->handle & ((1 << shift) - 1);
*cso->handle = ((base + i) << shift) | offset;
}
}
static void
gen6_emit_launch_grid_surface_view(struct ilo_render *r,
const struct ilo_state_vector *vec,
struct ilo_render_launch_grid_session *session)
{
const struct ilo_shader_state *cs = vec->cs;
const struct ilo_view_state *view = &vec->view[PIPE_SHADER_COMPUTE];
uint32_t *surface_state = r->state.cs.SURFACE_STATE;
int base, count, i;
ILO_DEV_ASSERT(r->dev, 7, 7.5);
base = ilo_shader_get_kernel_param(cs, ILO_KERNEL_SURFACE_TEX_BASE);
count = ilo_shader_get_kernel_param(cs, ILO_KERNEL_SURFACE_TEX_COUNT);
/* SURFACE_STATEs for sampler views */
surface_state += base;
for (i = 0; i < count; i++) {
if (i < view->count && view->states[i]) {
const struct ilo_view_cso *cso =
(const struct ilo_view_cso *) view->states[i];
surface_state[i] =
gen6_SURFACE_STATE(r->builder, &cso->surface, false);
} else {
surface_state[i] = 0;
}
}
}
static void
finalize_vertex_elements(struct ilo_context *ilo)
{
struct ilo_state_vector *vec = &ilo->state_vector;
if (!(vec->dirty & (ILO_DIRTY_VE | ILO_DIRTY_VS)))
return;
vec->dirty |= ILO_DIRTY_VE;
vec->ve->last_cso_edgeflag = false;
if (vec->ve->count && vec->vs &&
ilo_shader_get_kernel_param(vec->vs, ILO_KERNEL_VS_INPUT_EDGEFLAG)) {
vec->ve->edgeflag_cso = vec->ve->cso[vec->ve->count - 1];
ilo_gpe_set_ve_edgeflag(ilo->dev, &vec->ve->edgeflag_cso);
vec->ve->last_cso_edgeflag = true;
}
vec->ve->prepend_nosrc_cso = false;
if (vec->vs &&
(ilo_shader_get_kernel_param(vec->vs,
ILO_KERNEL_VS_INPUT_INSTANCEID) ||
ilo_shader_get_kernel_param(vec->vs,
ILO_KERNEL_VS_INPUT_VERTEXID))) {
ilo_gpe_init_ve_nosrc(ilo->dev,
GEN6_VFCOMP_STORE_VID,
GEN6_VFCOMP_STORE_IID,
GEN6_VFCOMP_NOSTORE,
GEN6_VFCOMP_NOSTORE,
&vec->ve->nosrc_cso);
vec->ve->prepend_nosrc_cso = true;
} else if (!vec->vs) {
/* generate VUE header */
ilo_gpe_init_ve_nosrc(ilo->dev,
GEN6_VFCOMP_STORE_0, /* Reserved */
GEN6_VFCOMP_STORE_0, /* Render Target Array Index */
GEN6_VFCOMP_STORE_0, /* Viewport Index */
GEN6_VFCOMP_STORE_0, /* Point Width */
&vec->ve->nosrc_cso);
vec->ve->prepend_nosrc_cso = true;
} else if (!vec->ve->count) {
/*
* From the Sandy Bridge PRM, volume 2 part 1, page 92:
*
* "SW must ensure that at least one vertex element is defined prior
* to issuing a 3DPRIMTIVE command, or operation is UNDEFINED."
*/
ilo_gpe_init_ve_nosrc(ilo->dev,
GEN6_VFCOMP_STORE_0,
GEN6_VFCOMP_STORE_0,
GEN6_VFCOMP_STORE_0,
GEN6_VFCOMP_STORE_1_FP,
&vec->ve->nosrc_cso);
vec->ve->prepend_nosrc_cso = true;
}
}
static void
gen6_emit_draw_surface_binding_tables(struct ilo_render *r,
const struct ilo_state_vector *vec,
int shader_type,
struct ilo_render_draw_session *session)
{
int count;
ILO_DEV_ASSERT(r->dev, 6, 7.5);
/* BINDING_TABLE_STATE */
switch (shader_type) {
case PIPE_SHADER_VERTEX:
if (!session->binding_table_vs_changed)
return;
if (!vec->vs)
return;
count = ilo_shader_get_kernel_param(vec->vs,
ILO_KERNEL_SURFACE_TOTAL_COUNT);
r->state.vs.BINDING_TABLE_STATE = gen6_BINDING_TABLE_STATE(r->builder,
r->state.vs.SURFACE_STATE, count);
break;
case PIPE_SHADER_GEOMETRY:
if (!session->binding_table_gs_changed)
return;
if (vec->gs) {
count = ilo_shader_get_kernel_param(vec->gs,
ILO_KERNEL_SURFACE_TOTAL_COUNT);
} else if (ilo_dev_gen(r->dev) == ILO_GEN(6) && vec->vs) {
count = ilo_shader_get_kernel_param(vec->vs,
ILO_KERNEL_VS_GEN6_SO_SURFACE_COUNT);
} else {
return;
}
r->state.gs.BINDING_TABLE_STATE = gen6_BINDING_TABLE_STATE(r->builder,
r->state.gs.SURFACE_STATE, count);
break;
case PIPE_SHADER_FRAGMENT:
if (!session->binding_table_fs_changed)
return;
if (!vec->fs)
return;
count = ilo_shader_get_kernel_param(vec->fs,
ILO_KERNEL_SURFACE_TOTAL_COUNT);
r->state.wm.BINDING_TABLE_STATE = gen6_BINDING_TABLE_STATE(r->builder,
r->state.wm.SURFACE_STATE, count);
break;
default:
break;
}
}
static void
gen6_emit_draw_surface_view(struct ilo_render *r,
const struct ilo_state_vector *vec,
int shader_type,
struct ilo_render_draw_session *session)
{
const struct ilo_view_state *view = &vec->view[shader_type];
const struct ilo_shader_state *sh;
uint32_t *surface_state;
int base, count, i;
ILO_DEV_ASSERT(r->dev, 6, 7.5);
switch (shader_type) {
case PIPE_SHADER_VERTEX:
if (!DIRTY(VS) && !DIRTY(VIEW_VS))
return;
if (!vec->vs)
return;
sh = vec->vs;
surface_state = r->state.vs.SURFACE_STATE;
session->binding_table_vs_changed = true;
break;
case PIPE_SHADER_FRAGMENT:
if (!DIRTY(FS) && !DIRTY(VIEW_FS))
return;
if (!vec->fs)
return;
sh = vec->fs;
surface_state = r->state.wm.SURFACE_STATE;
session->binding_table_fs_changed = true;
break;
default:
return;
break;
}
base = ilo_shader_get_kernel_param(sh, ILO_KERNEL_SURFACE_TEX_BASE);
count = ilo_shader_get_kernel_param(sh, ILO_KERNEL_SURFACE_TEX_COUNT);
/* SURFACE_STATEs for sampler views */
surface_state += base;
for (i = 0; i < count; i++) {
if (i < view->count && view->states[i]) {
const struct ilo_view_cso *cso =
(const struct ilo_view_cso *) view->states[i];
surface_state[i] =
gen6_SURFACE_STATE(r->builder, &cso->surface, false);
} else {
surface_state[i] = 0;
}
}
}
static void
gen6_emit_draw_surface_const(struct ilo_render *r,
const struct ilo_state_vector *vec,
int shader_type,
struct ilo_render_draw_session *session)
{
const struct ilo_cbuf_state *cbuf = &vec->cbuf[shader_type];
const struct ilo_shader_state *sh;
uint32_t *surface_state;
int base, count, i;
ILO_DEV_ASSERT(r->dev, 6, 7.5);
switch (shader_type) {
case PIPE_SHADER_VERTEX:
if (!DIRTY(VS) && !DIRTY(CBUF))
return;
if (!vec->vs)
return;
sh = vec->vs;
surface_state = r->state.vs.SURFACE_STATE;
session->binding_table_vs_changed = true;
break;
case PIPE_SHADER_FRAGMENT:
if (!DIRTY(FS) && !DIRTY(CBUF))
return;
if (!vec->fs)
return;
sh = vec->fs;
surface_state = r->state.wm.SURFACE_STATE;
session->binding_table_fs_changed = true;
break;
default:
return;
break;
}
base = ilo_shader_get_kernel_param(sh, ILO_KERNEL_SURFACE_CONST_BASE);
count = ilo_shader_get_kernel_param(sh, ILO_KERNEL_SURFACE_CONST_COUNT);
/* SURFACE_STATEs for constant buffers */
surface_state += base;
for (i = 0; i < count; i++) {
const struct ilo_cbuf_cso *cso = &cbuf->cso[i];
if (cso->resource) {
surface_state[i] = gen6_SURFACE_STATE(r->builder,
&cso->surface, false);
} else {
surface_state[i] = 0;
}
}
}
int
ilo_render_get_draw_surface_states_len(const struct ilo_render *render,
const struct ilo_state_vector *vec)
{
int sh_type, len;
ILO_DEV_ASSERT(render->dev, 6, 7.5);
len = 0;
for (sh_type = 0; sh_type < PIPE_SHADER_TYPES; sh_type++) {
const int alignment = 32 / 4;
int num_surfaces = 0;
switch (sh_type) {
case PIPE_SHADER_VERTEX:
if (vec->vs) {
num_surfaces = ilo_shader_get_kernel_param(vec->vs,
ILO_KERNEL_SURFACE_TOTAL_COUNT);
if (ilo_dev_gen(render->dev) == ILO_GEN(6)) {
num_surfaces += ilo_shader_get_kernel_param(vec->vs,
ILO_KERNEL_VS_GEN6_SO_SURFACE_COUNT);
}
}
break;
case PIPE_SHADER_GEOMETRY:
if (vec->gs) {
num_surfaces = ilo_shader_get_kernel_param(vec->gs,
ILO_KERNEL_SURFACE_TOTAL_COUNT);
}
break;
case PIPE_SHADER_FRAGMENT:
if (vec->fs) {
num_surfaces = ilo_shader_get_kernel_param(vec->fs,
ILO_KERNEL_SURFACE_TOTAL_COUNT);
}
break;
default:
break;
}
/* BINDING_TABLE_STATE and SURFACE_STATEs */
if (num_surfaces) {
len += align(num_surfaces, alignment) +
align(GEN6_SURFACE_STATE__SIZE, alignment) * num_surfaces;
}
}
return len;
}
static void
gen6_emit_draw_surface_so(struct ilo_render *r,
const struct ilo_state_vector *vec,
struct ilo_render_draw_session *session)
{
const struct ilo_shader_state *vs = vec->vs;
const struct ilo_shader_state *gs = vec->gs;
const struct ilo_so_state *so = &vec->so;
const struct pipe_stream_output_info *so_info;
uint32_t *surface_state;
int base, count, i;
ILO_DEV_ASSERT(r->dev, 6, 6);
if (!DIRTY(VS) && !DIRTY(GS) && !DIRTY(SO))
return;
if (gs) {
so_info = ilo_shader_get_kernel_so_info(gs);
base = ilo_shader_get_kernel_param(gs,
ILO_KERNEL_GS_GEN6_SURFACE_SO_BASE);
count = ilo_shader_get_kernel_param(gs,
ILO_KERNEL_GS_GEN6_SURFACE_SO_COUNT);
} else if (vs) {
so_info = ilo_shader_get_kernel_so_info(vs);
base = 0;
count = ilo_shader_get_kernel_param(vs,
ILO_KERNEL_VS_GEN6_SO_SURFACE_COUNT);
} else {
return;
}
session->binding_table_gs_changed = true;
/* SURFACE_STATEs for stream output targets */
surface_state = &r->state.gs.SURFACE_STATE[base];
for (i = 0; i < count; i++) {
if (so_info && i < so_info->num_outputs &&
so_info->output[i].output_buffer < so->count &&
so->states[so_info->output[i].output_buffer]) {
const struct pipe_stream_output_target *so_target =
so->states[so_info->output[i].output_buffer];
surface_state[i] = gen6_so_SURFACE_STATE(r->builder,
so_target, so_info, i);
} else {
surface_state[i] = 0;
}
}
}
int
ilo_render_get_launch_grid_dynamic_states_len(const struct ilo_render *render,
const struct ilo_state_vector *vec)
{
const int alignment = 32 / 4;
int num_samplers;
int len = 0;
ILO_DEV_ASSERT(render->dev, 7, 7.5);
num_samplers = ilo_shader_get_kernel_param(vec->cs,
ILO_KERNEL_SAMPLER_COUNT);
/* SAMPLER_STATE array and SAMPLER_BORDER_COLORs */
if (num_samplers) {
/* prefetches are done in multiples of 4 */
num_samplers = align(num_samplers, 4);
len += align(GEN6_SAMPLER_STATE__SIZE * num_samplers, alignment) +
align(GEN6_SAMPLER_BORDER_COLOR__SIZE, alignment) * num_samplers;
}
len += GEN6_INTERFACE_DESCRIPTOR_DATA__SIZE;
return len;
}
static void
gen6_emit_launch_grid_dynamic_samplers(struct ilo_render *r,
const struct ilo_state_vector *vec,
struct ilo_render_launch_grid_session *session)
{
const unsigned shader_type = PIPE_SHADER_COMPUTE;
const struct ilo_shader_state *cs = vec->cs;
const struct ilo_sampler_cso * const *samplers =
vec->sampler[shader_type].cso;
const struct pipe_sampler_view * const *views =
(const struct pipe_sampler_view **) vec->view[shader_type].states;
int sampler_count, i;
ILO_DEV_ASSERT(r->dev, 7, 7.5);
sampler_count = ilo_shader_get_kernel_param(cs, ILO_KERNEL_SAMPLER_COUNT);
assert(sampler_count <= Elements(vec->view[shader_type].states) &&
sampler_count <= Elements(vec->sampler[shader_type].cso));
for (i = 0; i < sampler_count; i++) {
r->state.cs.SAMPLER_BORDER_COLOR_STATE[i] = (samplers[i]) ?
gen6_SAMPLER_BORDER_COLOR_STATE(r->builder, samplers[i]) : 0;
}
r->state.cs.SAMPLER_STATE = gen6_SAMPLER_STATE(r->builder, samplers, views,
r->state.cs.SAMPLER_BORDER_COLOR_STATE, sampler_count);
}
void
ilo_gpe_init_gs_cso_gen7(const struct ilo_dev_info *dev,
const struct ilo_shader_state *gs,
struct ilo_shader_cso *cso)
{
int start_grf, vue_read_len, max_threads;
uint32_t dw2, dw4, dw5;
ILO_GPE_VALID_GEN(dev, 7, 7.5);
start_grf = ilo_shader_get_kernel_param(gs, ILO_KERNEL_URB_DATA_START_REG);
vue_read_len = ilo_shader_get_kernel_param(gs, ILO_KERNEL_INPUT_COUNT);
/* in pairs */
vue_read_len = (vue_read_len + 1) / 2;
switch (dev->gen) {
case ILO_GEN(7.5):
max_threads = (dev->gt >= 2) ? 256 : 70;
break;
case ILO_GEN(7):
max_threads = (dev->gt == 2) ? 128 : 36;
break;
default:
max_threads = 1;
break;
}
dw2 = (true) ? 0 : GEN6_THREADDISP_FP_MODE_ALT;
dw4 = vue_read_len << GEN7_GS_DW4_URB_READ_LEN__SHIFT |
GEN7_GS_DW4_INCLUDE_VERTEX_HANDLES |
0 << GEN7_GS_DW4_URB_READ_OFFSET__SHIFT |
start_grf << GEN7_GS_DW4_URB_GRF_START__SHIFT;
dw5 = (max_threads - 1) << GEN7_GS_DW5_MAX_THREADS__SHIFT |
GEN7_GS_DW5_STATISTICS |
GEN7_GS_DW5_GS_ENABLE;
STATIC_ASSERT(Elements(cso->payload) >= 3);
cso->payload[0] = dw2;
cso->payload[1] = dw4;
cso->payload[2] = dw5;
}
static void
gen6_emit_launch_grid_surface_global(struct ilo_render *r,
const struct ilo_state_vector *vec,
struct ilo_render_launch_grid_session *session)
{
const struct ilo_shader_state *cs = vec->cs;
const struct ilo_global_binding_cso *bindings =
util_dynarray_begin(&vec->global_binding.bindings);
uint32_t *surface_state = r->state.cs.SURFACE_STATE;
int base, count, i;
ILO_DEV_ASSERT(r->dev, 7, 7.5);
base = ilo_shader_get_kernel_param(cs, ILO_KERNEL_CS_SURFACE_GLOBAL_BASE);
count = ilo_shader_get_kernel_param(cs, ILO_KERNEL_CS_SURFACE_GLOBAL_COUNT);
if (!count)
return;
if (base + count > Elements(r->state.cs.SURFACE_STATE)) {
ilo_warn("too many global bindings\n");
count = Elements(r->state.cs.SURFACE_STATE) - base;
}
/* SURFACE_STATEs for global bindings */
surface_state += base;
for (i = 0; i < count; i++) {
if (i < vec->global_binding.count && bindings[i].resource) {
const struct ilo_buffer *buf = ilo_buffer(bindings[i].resource);
struct ilo_view_surface view;
assert(bindings[i].resource->target == PIPE_BUFFER);
ilo_gpe_init_view_surface_for_buffer(r->dev, buf, 0, buf->bo_size,
1, PIPE_FORMAT_NONE, true, true, &view);
surface_state[i] =
gen6_SURFACE_STATE(r->builder, &view, true);
} else {
surface_state[i] = 0;
}
}
}
static void
gen6_emit_draw_surface_rt(struct ilo_render *r,
const struct ilo_state_vector *vec,
struct ilo_render_draw_session *session)
{
const struct ilo_shader_state *fs = vec->fs;
const struct ilo_fb_state *fb = &vec->fb;
uint32_t *surface_state;
int base, count, i;
ILO_DEV_ASSERT(r->dev, 6, 7.5);
if (!DIRTY(FS) && !DIRTY(FB))
return;
if (!fs)
return;
session->binding_table_fs_changed = true;
base = ilo_shader_get_kernel_param(fs, ILO_KERNEL_FS_SURFACE_RT_BASE);
count = ilo_shader_get_kernel_param(fs, ILO_KERNEL_FS_SURFACE_RT_COUNT);
/* SURFACE_STATEs for render targets */
surface_state = &r->state.wm.SURFACE_STATE[base];
for (i = 0; i < count; i++) {
if (i < fb->state.nr_cbufs && fb->state.cbufs[i]) {
const struct ilo_surface_cso *surface =
(const struct ilo_surface_cso *) fb->state.cbufs[i];
assert(surface->is_rt);
surface_state[i] =
gen6_SURFACE_STATE(r->builder, &surface->u.rt, true);
} else {
surface_state[i] =
gen6_SURFACE_STATE(r->builder, &fb->null_rt, true);
}
}
}
static void
gen6_emit_launch_grid_surface_binding_table(struct ilo_render *r,
const struct ilo_state_vector *vec,
struct ilo_render_launch_grid_session *session)
{
const struct ilo_shader_state *cs = vec->cs;
int count;
ILO_DEV_ASSERT(r->dev, 7, 7.5);
count = ilo_shader_get_kernel_param(cs, ILO_KERNEL_SURFACE_TOTAL_COUNT);
if (count) {
r->state.cs.BINDING_TABLE_STATE = gen6_BINDING_TABLE_STATE(r->builder,
r->state.cs.SURFACE_STATE, count);
}
}
int
ilo_render_get_launch_grid_surface_states_len(const struct ilo_render *render,
const struct ilo_state_vector *vec)
{
const int alignment = 32 / 4;
int num_surfaces;
int len = 0;
ILO_DEV_ASSERT(render->dev, 7, 7.5);
num_surfaces = ilo_shader_get_kernel_param(vec->cs,
ILO_KERNEL_SURFACE_TOTAL_COUNT);
/* BINDING_TABLE_STATE and SURFACE_STATEs */
if (num_surfaces) {
len += align(num_surfaces, alignment) +
align(GEN6_SURFACE_STATE__SIZE, alignment) * num_surfaces;
}
return len;
}
static void
gen6_draw_gs(struct ilo_render *r,
const struct ilo_state_vector *vec,
struct ilo_render_draw_session *session)
{
/* 3DSTATE_CONSTANT_GS */
if (session->pcb_gs_changed)
gen6_3DSTATE_CONSTANT_GS(r->builder, NULL, NULL, 0);
/* 3DSTATE_GS */
if (DIRTY(GS) || DIRTY(VS) ||
session->prim_changed || r->instruction_bo_changed) {
if (vec->gs) {
gen6_3DSTATE_GS(r->builder, vec->gs);
} else if (vec->vs &&
ilo_shader_get_kernel_param(vec->vs, ILO_KERNEL_VS_GEN6_SO)) {
const int verts_per_prim = u_vertices_per_prim(session->reduced_prim);
gen6_so_3DSTATE_GS(r->builder, vec->vs, verts_per_prim);
} else {
gen6_disable_3DSTATE_GS(r->builder);
}
}
}
void
ilo_gpe_init_fs_cso_gen7(const struct ilo_dev_info *dev,
const struct ilo_shader_state *fs,
struct ilo_shader_cso *cso)
{
int start_grf, max_threads;
uint32_t dw2, dw4, dw5;
uint32_t wm_interps, wm_dw1;
ILO_GPE_VALID_GEN(dev, 7, 7.5);
start_grf = ilo_shader_get_kernel_param(fs, ILO_KERNEL_URB_DATA_START_REG);
dw2 = (true) ? 0 : GEN6_THREADDISP_FP_MODE_ALT;
dw4 = GEN7_PS_DW4_POSOFFSET_NONE;
/* see brwCreateContext() */
switch (dev->gen) {
case ILO_GEN(7.5):
max_threads = (dev->gt == 3) ? 408 : (dev->gt == 2) ? 204 : 102;
dw4 |= (max_threads - 1) << GEN75_PS_DW4_MAX_THREADS__SHIFT;
dw4 |= 1 << GEN75_PS_DW4_SAMPLE_MASK__SHIFT;
break;
case ILO_GEN(7):
default:
max_threads = (dev->gt == 2) ? 172 : 48;
dw4 |= (max_threads - 1) << GEN7_PS_DW4_MAX_THREADS__SHIFT;
break;
}
if (ilo_shader_get_kernel_param(fs, ILO_KERNEL_PCB_CBUF0_SIZE))
dw4 |= GEN7_PS_DW4_PUSH_CONSTANT_ENABLE;
if (ilo_shader_get_kernel_param(fs, ILO_KERNEL_INPUT_COUNT))
dw4 |= GEN7_PS_DW4_ATTR_ENABLE;
assert(!ilo_shader_get_kernel_param(fs, ILO_KERNEL_FS_DISPATCH_16_OFFSET));
dw4 |= GEN7_PS_DW4_8_PIXEL_DISPATCH;
dw5 = start_grf << GEN7_PS_DW5_URB_GRF_START0__SHIFT |
0 << GEN7_PS_DW5_URB_GRF_START1__SHIFT |
0 << GEN7_PS_DW5_URB_GRF_START2__SHIFT;
/* FS affects 3DSTATE_WM too */
wm_dw1 = 0;
/*
* TODO set this bit only when
*
* a) fs writes colors and color is not masked, or
* b) fs writes depth, or
* c) fs or cc kills
*/
wm_dw1 |= GEN7_WM_DW1_PS_ENABLE;
/*
* From the Ivy Bridge PRM, volume 2 part 1, page 278:
*
* "This bit (Pixel Shader Kill Pixel), if ENABLED, indicates that
* the PS kernel or color calculator has the ability to kill
* (discard) pixels or samples, other than due to depth or stencil
* testing. This bit is required to be ENABLED in the following
* situations:
*
* - The API pixel shader program contains "killpix" or "discard"
* instructions, or other code in the pixel shader kernel that
* can cause the final pixel mask to differ from the pixel mask
* received on dispatch.
*
* - A sampler with chroma key enabled with kill pixel mode is used
* by the pixel shader.
*
* - Any render target has Alpha Test Enable or AlphaToCoverage
* Enable enabled.
*
* - The pixel shader kernel generates and outputs oMask.
*
* Note: As ClipDistance clipping is fully supported in hardware
* and therefore not via PS instructions, there should be no need
* to ENABLE this bit due to ClipDistance clipping."
*/
if (ilo_shader_get_kernel_param(fs, ILO_KERNEL_FS_USE_KILL))
wm_dw1 |= GEN7_WM_DW1_PS_KILL;
if (ilo_shader_get_kernel_param(fs, ILO_KERNEL_FS_OUTPUT_Z))
wm_dw1 |= GEN7_WM_DW1_PSCDEPTH_ON;
if (ilo_shader_get_kernel_param(fs, ILO_KERNEL_FS_INPUT_Z))
wm_dw1 |= GEN7_WM_DW1_PS_USE_DEPTH;
if (ilo_shader_get_kernel_param(fs, ILO_KERNEL_FS_INPUT_W))
wm_dw1 |= GEN7_WM_DW1_PS_USE_W;
wm_interps = ilo_shader_get_kernel_param(fs,
ILO_KERNEL_FS_BARYCENTRIC_INTERPOLATIONS);
wm_dw1 |= wm_interps << GEN7_WM_DW1_BARYCENTRIC_INTERP__SHIFT;
STATIC_ASSERT(Elements(cso->payload) >= 4);
cso->payload[0] = dw2;
cso->payload[1] = dw4;
cso->payload[2] = dw5;
cso->payload[3] = wm_dw1;
}
static void
gen6_draw_common_urb(struct ilo_render *r,
const struct ilo_state_vector *vec,
struct ilo_render_draw_session *session)
{
/* 3DSTATE_URB */
if (DIRTY(VE) || DIRTY(VS) || DIRTY(GS)) {
const bool gs_active = (vec->gs || (vec->vs &&
ilo_shader_get_kernel_param(vec->vs, ILO_KERNEL_VS_GEN6_SO)));
int vs_entry_size, gs_entry_size;
int vs_total_size, gs_total_size;
vs_entry_size = (vec->vs) ?
ilo_shader_get_kernel_param(vec->vs, ILO_KERNEL_OUTPUT_COUNT) : 0;
/*
* As indicated by 2e712e41db0c0676e9f30fc73172c0e8de8d84d4, VF and VS
* share VUE handles. The VUE allocation size must be large enough to
* store either VF outputs (number of VERTEX_ELEMENTs) and VS outputs.
*
* I am not sure if the PRM explicitly states that VF and VS share VUE
* handles. But here is a citation that implies so:
*
* From the Sandy Bridge PRM, volume 2 part 1, page 44:
*
* "Once a FF stage that spawn threads has sufficient input to
* initiate a thread, it must guarantee that it is safe to request
* the thread initiation. For all these FF stages, this check is
* based on :
*
* - The availability of output URB entries:
* - VS: As the input URB entries are overwritten with the
* VS-generated output data, output URB availability isn't a
* factor."
*/
if (vs_entry_size < vec->ve->count + vec->ve->prepend_nosrc_cso)
vs_entry_size = vec->ve->count + vec->ve->prepend_nosrc_cso;
gs_entry_size = (vec->gs) ?
ilo_shader_get_kernel_param(vec->gs, ILO_KERNEL_OUTPUT_COUNT) :
(gs_active) ? vs_entry_size : 0;
/* in bytes */
vs_entry_size *= sizeof(float) * 4;
gs_entry_size *= sizeof(float) * 4;
vs_total_size = r->dev->urb_size;
if (gs_active) {
vs_total_size /= 2;
gs_total_size = vs_total_size;
}
else {
gs_total_size = 0;
}
gen6_3DSTATE_URB(r->builder, vs_total_size, gs_total_size,
vs_entry_size, gs_entry_size);
/*
* From the Sandy Bridge PRM, volume 2 part 1, page 27:
*
* "Because of a urb corruption caused by allocating a previous
* gsunit's urb entry to vsunit software is required to send a
* "GS NULL Fence" (Send URB fence with VS URB size == 1 and GS URB
* size == 0) plus a dummy DRAW call before any case where VS will
* be taking over GS URB space."
*/
if (r->state.gs.active && !gs_active)
ilo_render_emit_flush(r);
r->state.gs.active = gs_active;
}
}
int
ilo_render_get_draw_dynamic_states_len(const struct ilo_render *render,
const struct ilo_state_vector *vec)
{
static int static_len;
int sh_type, len;
ILO_DEV_ASSERT(render->dev, 6, 7.5);
if (!static_len) {
/* 64 bytes, or 16 dwords */
const int alignment = 64 / 4;
/* pad first */
len = alignment - 1;
/* CC states */
len += align(GEN6_BLEND_STATE__SIZE * ILO_MAX_DRAW_BUFFERS, alignment);
len += align(GEN6_DEPTH_STENCIL_STATE__SIZE, alignment);
len += align(GEN6_COLOR_CALC_STATE__SIZE, alignment);
/* viewport arrays */
if (ilo_dev_gen(render->dev) >= ILO_GEN(7)) {
len += 15 + /* pad first */
align(GEN7_SF_CLIP_VIEWPORT__SIZE * ILO_MAX_VIEWPORTS, 16) +
align(GEN6_CC_VIEWPORT__SIZE * ILO_MAX_VIEWPORTS, 8) +
align(GEN6_SCISSOR_RECT__SIZE * ILO_MAX_VIEWPORTS, 8);
} else {
len += 7 + /* pad first */
align(GEN6_SF_VIEWPORT__SIZE * ILO_MAX_VIEWPORTS, 8) +
align(GEN6_CLIP_VIEWPORT__SIZE * ILO_MAX_VIEWPORTS, 8) +
align(GEN6_CC_VIEWPORT__SIZE * ILO_MAX_VIEWPORTS, 8) +
align(GEN6_SCISSOR_RECT__SIZE * ILO_MAX_VIEWPORTS, 8);
}
static_len = len;
}
len = static_len;
for (sh_type = 0; sh_type < PIPE_SHADER_TYPES; sh_type++) {
const int alignment = 32 / 4;
int num_samplers = 0, pcb_len = 0;
switch (sh_type) {
case PIPE_SHADER_VERTEX:
if (vec->vs) {
num_samplers = ilo_shader_get_kernel_param(vec->vs,
ILO_KERNEL_SAMPLER_COUNT);
pcb_len = ilo_shader_get_kernel_param(vec->vs,
ILO_KERNEL_PCB_CBUF0_SIZE);
pcb_len += ilo_shader_get_kernel_param(vec->vs,
ILO_KERNEL_VS_PCB_UCP_SIZE);
}
break;
case PIPE_SHADER_GEOMETRY:
break;
case PIPE_SHADER_FRAGMENT:
if (vec->fs) {
num_samplers = ilo_shader_get_kernel_param(vec->fs,
ILO_KERNEL_SAMPLER_COUNT);
pcb_len = ilo_shader_get_kernel_param(vec->fs,
ILO_KERNEL_PCB_CBUF0_SIZE);
}
break;
default:
break;
}
/* SAMPLER_STATE array and SAMPLER_BORDER_COLORs */
if (num_samplers) {
/* prefetches are done in multiples of 4 */
num_samplers = align(num_samplers, 4);
len += align(GEN6_SAMPLER_STATE__SIZE * num_samplers, alignment) +
align(GEN6_SAMPLER_BORDER_COLOR__SIZE, alignment) * num_samplers;
}
/* PCB */
if (pcb_len)
len += align(pcb_len, alignment);
}
return len;
}
static void
gen6_emit_draw_dynamic_pcb(struct ilo_render *r,
const struct ilo_state_vector *vec,
struct ilo_render_draw_session *session)
{
ILO_DEV_ASSERT(r->dev, 6, 7.5);
/* push constant buffer for VS */
if (DIRTY(VS) || DIRTY(CBUF) || DIRTY(CLIP)) {
const int cbuf0_size = (vec->vs) ?
ilo_shader_get_kernel_param(vec->vs,
ILO_KERNEL_PCB_CBUF0_SIZE) : 0;
const int clip_state_size = (vec->vs) ?
ilo_shader_get_kernel_param(vec->vs,
ILO_KERNEL_VS_PCB_UCP_SIZE) : 0;
const int total_size = cbuf0_size + clip_state_size;
if (total_size) {
void *pcb;
r->state.vs.PUSH_CONSTANT_BUFFER =
gen6_push_constant_buffer(r->builder, total_size, &pcb);
r->state.vs.PUSH_CONSTANT_BUFFER_size = total_size;
if (cbuf0_size) {
const struct ilo_cbuf_state *cbuf =
&vec->cbuf[PIPE_SHADER_VERTEX];
if (cbuf0_size <= cbuf->cso[0].user_buffer_size) {
memcpy(pcb, cbuf->cso[0].user_buffer, cbuf0_size);
} else {
memcpy(pcb, cbuf->cso[0].user_buffer,
cbuf->cso[0].user_buffer_size);
memset(pcb + cbuf->cso[0].user_buffer_size, 0,
cbuf0_size - cbuf->cso[0].user_buffer_size);
}
pcb += cbuf0_size;
}
if (clip_state_size)
memcpy(pcb, &vec->clip, clip_state_size);
session->pcb_vs_changed = true;
} else if (r->state.vs.PUSH_CONSTANT_BUFFER_size) {
r->state.vs.PUSH_CONSTANT_BUFFER = 0;
r->state.vs.PUSH_CONSTANT_BUFFER_size = 0;
session->pcb_vs_changed = true;
}
}
/* push constant buffer for FS */
if (DIRTY(FS) || DIRTY(CBUF)) {
const int cbuf0_size = (vec->fs) ?
ilo_shader_get_kernel_param(vec->fs, ILO_KERNEL_PCB_CBUF0_SIZE) : 0;
if (cbuf0_size) {
const struct ilo_cbuf_state *cbuf = &vec->cbuf[PIPE_SHADER_FRAGMENT];
void *pcb;
r->state.wm.PUSH_CONSTANT_BUFFER =
gen6_push_constant_buffer(r->builder, cbuf0_size, &pcb);
r->state.wm.PUSH_CONSTANT_BUFFER_size = cbuf0_size;
if (cbuf0_size <= cbuf->cso[0].user_buffer_size) {
memcpy(pcb, cbuf->cso[0].user_buffer, cbuf0_size);
} else {
memcpy(pcb, cbuf->cso[0].user_buffer,
cbuf->cso[0].user_buffer_size);
memset(pcb + cbuf->cso[0].user_buffer_size, 0,
cbuf0_size - cbuf->cso[0].user_buffer_size);
}
session->pcb_fs_changed = true;
} else if (r->state.wm.PUSH_CONSTANT_BUFFER_size) {
r->state.wm.PUSH_CONSTANT_BUFFER = 0;
r->state.wm.PUSH_CONSTANT_BUFFER_size = 0;
session->pcb_fs_changed = true;
}
}
}
static void
gen6_emit_draw_dynamic_samplers(struct ilo_render *r,
const struct ilo_state_vector *vec,
int shader_type,
struct ilo_render_draw_session *session)
{
const struct ilo_sampler_cso * const *samplers =
vec->sampler[shader_type].cso;
const struct pipe_sampler_view * const *views =
(const struct pipe_sampler_view **) vec->view[shader_type].states;
uint32_t *sampler_state, *border_color_state;
int sampler_count;
bool emit_border_color = false;
bool skip = false;
ILO_DEV_ASSERT(r->dev, 6, 7.5);
/* SAMPLER_BORDER_COLOR_STATE and SAMPLER_STATE */
switch (shader_type) {
case PIPE_SHADER_VERTEX:
if (DIRTY(VS) || DIRTY(SAMPLER_VS) || DIRTY(VIEW_VS)) {
sampler_state = &r->state.vs.SAMPLER_STATE;
border_color_state = r->state.vs.SAMPLER_BORDER_COLOR_STATE;
if (DIRTY(VS) || DIRTY(SAMPLER_VS))
emit_border_color = true;
sampler_count = (vec->vs) ? ilo_shader_get_kernel_param(vec->vs,
ILO_KERNEL_SAMPLER_COUNT) : 0;
session->sampler_vs_changed = true;
} else {
skip = true;
}
break;
case PIPE_SHADER_FRAGMENT:
if (DIRTY(FS) || DIRTY(SAMPLER_FS) || DIRTY(VIEW_FS)) {
sampler_state = &r->state.wm.SAMPLER_STATE;
border_color_state = r->state.wm.SAMPLER_BORDER_COLOR_STATE;
if (DIRTY(VS) || DIRTY(SAMPLER_FS))
emit_border_color = true;
sampler_count = (vec->fs) ? ilo_shader_get_kernel_param(vec->fs,
ILO_KERNEL_SAMPLER_COUNT) : 0;
session->sampler_fs_changed = true;
} else {
skip = true;
}
break;
default:
skip = true;
break;
}
if (skip)
return;
assert(sampler_count <= Elements(vec->view[shader_type].states) &&
sampler_count <= Elements(vec->sampler[shader_type].cso));
if (emit_border_color) {
int i;
for (i = 0; i < sampler_count; i++) {
border_color_state[i] = (samplers[i]) ?
gen6_SAMPLER_BORDER_COLOR_STATE(r->builder, samplers[i]) : 0;
}
}
*sampler_state = gen6_SAMPLER_STATE(r->builder,
samplers, views, border_color_state, sampler_count);
}
