static void r600_blitter_begin(struct pipe_context *ctx, enum r600_blitter_op op)
{
struct r600_context *rctx = (struct r600_context *)ctx;
r600_suspend_nontimer_queries(rctx);
util_blitter_save_vertex_buffers(rctx->blitter,
util_last_bit(rctx->vertex_buffer_state.enabled_mask),
rctx->vertex_buffer_state.vb);
util_blitter_save_vertex_elements(rctx->blitter, rctx->vertex_elements);
util_blitter_save_vertex_shader(rctx->blitter, rctx->vs_shader);
util_blitter_save_so_targets(rctx->blitter, rctx->num_so_targets,
(struct pipe_stream_output_target**)rctx->so_targets);
util_blitter_save_rasterizer(rctx->blitter, rctx->states[R600_PIPE_STATE_RASTERIZER]);
if (op & R600_SAVE_FRAGMENT_STATE) {
if (rctx->states[R600_PIPE_STATE_VIEWPORT]) {
util_blitter_save_viewport(rctx->blitter, &rctx->viewport);
}
util_blitter_save_fragment_shader(rctx->blitter, rctx->ps_shader);
util_blitter_save_blend(rctx->blitter, rctx->states[R600_PIPE_STATE_BLEND]);
util_blitter_save_depth_stencil_alpha(rctx->blitter, rctx->states[R600_PIPE_STATE_DSA]);
if (rctx->states[R600_PIPE_STATE_STENCIL_REF]) {
util_blitter_save_stencil_ref(rctx->blitter, &rctx->stencil_ref);
}
util_blitter_save_sample_mask(rctx->blitter, rctx->sample_mask.sample_mask);
}
if (op & R600_SAVE_FRAMEBUFFER)
util_blitter_save_framebuffer(rctx->blitter, &rctx->framebuffer);
if (op & R600_SAVE_TEXTURES) {
util_blitter_save_fragment_sampler_states(
rctx->blitter, rctx->ps_samplers.n_samplers,
(void**)rctx->ps_samplers.samplers);
util_blitter_save_fragment_sampler_views(
rctx->blitter, util_last_bit(rctx->ps_samplers.views.enabled_mask),
(struct pipe_sampler_view**)rctx->ps_samplers.views.views);
}
if ((op & R600_DISABLE_RENDER_COND) && rctx->current_render_cond) {
rctx->saved_render_cond = rctx->current_render_cond;
rctx->saved_render_cond_mode = rctx->current_render_cond_mode;
rctx->context.render_condition(&rctx->context, NULL, 0);
}
}
static void
fd_set_vertex_buffers(struct pipe_context *pctx,
unsigned start_slot, unsigned count,
const struct pipe_vertex_buffer *vb)
{
struct fd_context *ctx = fd_context(pctx);
struct fd_vertexbuf_stateobj *so = &ctx->vertexbuf;
int i;
/* on a2xx, pitch is encoded in the vtx fetch instruction, so
* we need to mark VTXSTATE as dirty as well to trigger patching
* and re-emitting the vtx shader:
*/
for (i = 0; i < count; i++) {
bool new_enabled = vb && (vb[i].buffer || vb[i].user_buffer);
bool old_enabled = so->vb[i].buffer || so->vb[i].user_buffer;
uint32_t new_stride = vb ? vb[i].stride : 0;
uint32_t old_stride = so->vb[i].stride;
if ((new_enabled != old_enabled) || (new_stride != old_stride)) {
ctx->dirty |= FD_DIRTY_VTXSTATE;
break;
}
}
util_set_vertex_buffers_mask(so->vb, &so->enabled_mask, vb, start_slot, count);
so->count = util_last_bit(so->enabled_mask);
ctx->dirty |= FD_DIRTY_VTXBUF;
}
void
nv50_fp_linkage_validate(struct nv50_context *nv50)
{
struct nouveau_pushbuf *push = nv50->base.pushbuf;
struct nv50_program *vp = nv50->gmtyprog ? nv50->gmtyprog : nv50->vertprog;
struct nv50_program *fp = nv50->fragprog;
struct nv50_varying dummy;
int i, n, c, m;
uint32_t primid = 0;
uint32_t layerid = 0;
uint32_t viewportid = 0;
uint32_t psiz = 0x000;
uint32_t interp = fp->fp.interp;
uint32_t colors = fp->fp.colors;
uint32_t clpd_nr = util_last_bit(vp->vp.clip_enable | vp->vp.cull_enable);
uint32_t lin[4];
uint8_t map[64];
uint8_t so_map[64];
if (!(nv50->dirty_3d & (NV50_NEW_3D_VERTPROG |
NV50_NEW_3D_FRAGPROG |
NV50_NEW_3D_GMTYPROG))) {
uint8_t bfc, ffc;
ffc = (nv50->state.semantic_color & NV50_3D_SEMANTIC_COLOR_FFC0_ID__MASK);
bfc = (nv50->state.semantic_color & NV50_3D_SEMANTIC_COLOR_BFC0_ID__MASK)
>> 8;
if (nv50->rast->pipe.light_twoside == ((ffc == bfc) ? 0 : 1))
return;
}
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
etna_set_vertex_buffers(struct pipe_context *pctx, unsigned start_slot,
unsigned num_buffers, const struct pipe_vertex_buffer *vb)
{
struct etna_context *ctx = etna_context(pctx);
struct etna_vertexbuf_state *so = &ctx->vertex_buffer;
util_set_vertex_buffers_mask(so->vb, &so->enabled_mask, vb, start_slot, num_buffers);
so->count = util_last_bit(so->enabled_mask);
for (unsigned idx = start_slot; idx < start_slot + num_buffers; ++idx) {
struct compiled_set_vertex_buffer *cs = &so->cvb[idx];
struct pipe_vertex_buffer *vbi = &so->vb[idx];
assert(!vbi->is_user_buffer); /* XXX support user_buffer using
etna_usermem_map */
if (vbi->buffer.resource) { /* GPU buffer */
cs->FE_VERTEX_STREAM_BASE_ADDR.bo = etna_resource(vbi->buffer.resource)->bo;
cs->FE_VERTEX_STREAM_BASE_ADDR.offset = vbi->buffer_offset;
cs->FE_VERTEX_STREAM_BASE_ADDR.flags = ETNA_RELOC_READ;
cs->FE_VERTEX_STREAM_CONTROL =
FE_VERTEX_STREAM_CONTROL_VERTEX_STRIDE(vbi->stride);
} else {
cs->FE_VERTEX_STREAM_BASE_ADDR.bo = NULL;
cs->FE_VERTEX_STREAM_CONTROL = 0;
}
}
ctx->dirty |= ETNA_DIRTY_VERTEX_BUFFERS;
}
/* for vertex shader, the inputs are loaded into registers before the shader
* is executed, so max_regs from the shader instructions might not properly
* reflect the # of registers actually used, especially in case passthrough
* varyings.
*
* Likewise, for fragment shader, we can have some regs which are passed
* input values but never touched by the resulting shader (ie. as result
* of dead code elimination or simply because we don't know how to turn
* the reg off.
*/
static void
fixup_regfootprint(struct ir3_shader_variant *v, uint32_t gpu_id)
{
unsigned i;
for (i = 0; i < v->inputs_count; i++) {
/* skip frag inputs fetch via bary.f since their reg's are
* not written by gpu before shader starts (and in fact the
* regid's might not even be valid)
*/
if (v->inputs[i].bary)
continue;
/* ignore high regs that are global to all threads in a warp
* (they exist by default) (a5xx+)
*/
if (v->inputs[i].regid >= regid(48,0))
continue;
if (v->inputs[i].compmask) {
unsigned n = util_last_bit(v->inputs[i].compmask) - 1;
int32_t regid = v->inputs[i].regid + n;
if (v->inputs[i].half) {
if (gpu_id < 500) {
v->info.max_half_reg = MAX2(v->info.max_half_reg, regid >> 2);
} else {
v->info.max_reg = MAX2(v->info.max_reg, regid >> 3);
}
} else {
void
brw_upload_cs_prog(struct brw_context *brw)
{
struct gl_context *ctx = &brw->ctx;
struct brw_cs_prog_key key;
struct brw_program *cp = (struct brw_program *) brw->compute_program;
if (!cp)
return;
if (!brw_state_dirty(brw, _NEW_TEXTURE, BRW_NEW_COMPUTE_PROGRAM))
return;
brw->cs.base.sampler_count =
util_last_bit(ctx->ComputeProgram._Current->SamplersUsed);
brw_cs_populate_key(brw, &key);
if (!brw_search_cache(&brw->cache, BRW_CACHE_CS_PROG,
&key, sizeof(key),
&brw->cs.base.prog_offset,
&brw->cs.base.prog_data)) {
bool success =
brw_codegen_cs_prog(brw,
ctx->Shader.CurrentProgram[MESA_SHADER_COMPUTE],
cp, &key);
(void) success;
assert(success);
}
}
static void r600_blitter_begin(struct pipe_context *ctx, enum r600_blitter_op op)
{
struct r600_context *rctx = (struct r600_context *)ctx;
r600_suspend_nontimer_queries(&rctx->b);
util_blitter_save_vertex_buffer_slot(rctx->blitter, rctx->vertex_buffer_state.vb);
util_blitter_save_vertex_elements(rctx->blitter, rctx->vertex_fetch_shader.cso);
util_blitter_save_vertex_shader(rctx->blitter, rctx->vs_shader);
util_blitter_save_geometry_shader(rctx->blitter, rctx->gs_shader);
util_blitter_save_so_targets(rctx->blitter, rctx->b.streamout.num_targets,
(struct pipe_stream_output_target**)rctx->b.streamout.targets);
util_blitter_save_rasterizer(rctx->blitter, rctx->rasterizer_state.cso);
if (op & R600_SAVE_FRAGMENT_STATE) {
util_blitter_save_viewport(rctx->blitter, &rctx->viewport[0].state);
util_blitter_save_scissor(rctx->blitter, &rctx->scissor[0].scissor);
util_blitter_save_fragment_shader(rctx->blitter, rctx->ps_shader);
util_blitter_save_blend(rctx->blitter, rctx->blend_state.cso);
util_blitter_save_depth_stencil_alpha(rctx->blitter, rctx->dsa_state.cso);
util_blitter_save_stencil_ref(rctx->blitter, &rctx->stencil_ref.pipe_state);
util_blitter_save_sample_mask(rctx->blitter, rctx->sample_mask.sample_mask);
}
if (op & R600_SAVE_FRAMEBUFFER)
util_blitter_save_framebuffer(rctx->blitter, &rctx->framebuffer.state);
if (op & R600_SAVE_TEXTURES) {
util_blitter_save_fragment_sampler_states(
rctx->blitter, util_last_bit(rctx->samplers[PIPE_SHADER_FRAGMENT].states.enabled_mask),
(void**)rctx->samplers[PIPE_SHADER_FRAGMENT].states.states);
util_blitter_save_fragment_sampler_views(
rctx->blitter, util_last_bit(rctx->samplers[PIPE_SHADER_FRAGMENT].views.enabled_mask),
(struct pipe_sampler_view**)rctx->samplers[PIPE_SHADER_FRAGMENT].views.views);
}
if ((op & R600_DISABLE_RENDER_COND) && rctx->b.current_render_cond) {
util_blitter_save_render_condition(rctx->blitter,
rctx->b.current_render_cond,
rctx->b.current_render_cond_cond,
rctx->b.current_render_cond_mode);
}
}
static void
vc4_set_vertex_buffers(struct pipe_context *pctx,
unsigned start_slot, unsigned count,
const struct pipe_vertex_buffer *vb)
{
struct vc4_context *vc4 = vc4_context(pctx);
struct vc4_vertexbuf_stateobj *so = &vc4->vertexbuf;
util_set_vertex_buffers_mask(so->vb, &so->enabled_mask, vb,
start_slot, count);
so->count = util_last_bit(so->enabled_mask);
vc4->dirty |= VC4_DIRTY_VTXBUF;
}
static void bind_sampler_states(struct fd_texture_stateobj *tex,
unsigned start, unsigned nr, void **hwcso)
{
unsigned i;
for (i = 0; i < nr; i++) {
unsigned p = i + start;
tex->samplers[p] = hwcso[i];
if (tex->samplers[p])
tex->valid_samplers |= (1 << p);
else
tex->valid_samplers &= ~(1 << p);
}
tex->num_samplers = util_last_bit(tex->valid_samplers);
}
static void set_sampler_views(struct fd_texture_stateobj *tex,
unsigned start, unsigned nr, struct pipe_sampler_view **views)
{
unsigned i;
for (i = 0; i < nr; i++) {
struct pipe_sampler_view *view = views ? views[i] : NULL;
unsigned p = i + start;
pipe_sampler_view_reference(&tex->textures[p], view);
if (tex->textures[p])
tex->valid_textures |= (1 << p);
else
tex->valid_textures &= ~(1 << p);
}
tex->num_textures = util_last_bit(tex->valid_textures);
}
/**
* Same as util_set_vertex_buffers_mask, but it only returns the number
* of bound buffers.
*/
void util_set_vertex_buffers_count(struct pipe_vertex_buffer *dst,
unsigned *dst_count,
const struct pipe_vertex_buffer *src,
unsigned start_slot, unsigned count)
{
unsigned i;
uint32_t enabled_buffers = 0;
for (i = 0; i < *dst_count; i++) {
if (dst[i].buffer || dst[i].user_buffer)
enabled_buffers |= (1ull << i);
}
util_set_vertex_buffers_mask(dst, &enabled_buffers, src, start_slot,
count);
*dst_count = util_last_bit(enabled_buffers);
}
/* Add any missing varyings needed for stream-out. Otherwise varyings not
* used by fragment shader will be stripped out.
*/
static void
link_stream_out(struct ir3_shader_linkage *l, const struct ir3_shader_variant *v)
{
const struct pipe_stream_output_info *strmout = &v->shader->stream_output;
/*
* First, any stream-out varyings not already in linkage map (ie. also
* consumed by frag shader) need to be added:
*/
for (unsigned i = 0; i < strmout->num_outputs; i++) {
const struct pipe_stream_output *out = &strmout->output[i];
unsigned k = out->register_index;
unsigned compmask =
(1 << (out->num_components + out->start_component)) - 1;
unsigned idx, nextloc = 0;
/* psize/pos need to be the last entries in linkage map, and will
* get added link_stream_out, so skip over them:
*/
if ((v->outputs[k].slot == VARYING_SLOT_PSIZ) ||
(v->outputs[k].slot == VARYING_SLOT_POS))
continue;
for (idx = 0; idx < l->cnt; idx++) {
if (l->var[idx].regid == v->outputs[k].regid)
break;
nextloc = MAX2(nextloc, l->var[idx].loc + 4);
}
/* add if not already in linkage map: */
if (idx == l->cnt)
ir3_link_add(l, v->outputs[k].regid, compmask, nextloc);
/* expand component-mask if needed, ie streaming out all components
* but frag shader doesn't consume all components:
*/
if (compmask & ~l->var[idx].compmask) {
l->var[idx].compmask |= compmask;
l->max_loc = MAX2(l->max_loc,
l->var[idx].loc + util_last_bit(l->var[idx].compmask));
}
}
}
static void si_blitter_begin(struct pipe_context *ctx, enum si_blitter_op op)
{
struct si_context *sctx = (struct si_context *)ctx;
r600_suspend_nontimer_queries(&sctx->b);
util_blitter_save_blend(sctx->blitter, sctx->queued.named.blend);
util_blitter_save_depth_stencil_alpha(sctx->blitter, sctx->queued.named.dsa);
util_blitter_save_stencil_ref(sctx->blitter, &sctx->stencil_ref);
util_blitter_save_rasterizer(sctx->blitter, sctx->queued.named.rasterizer);
util_blitter_save_fragment_shader(sctx->blitter, sctx->ps_shader);
util_blitter_save_geometry_shader(sctx->blitter, sctx->gs_shader);
util_blitter_save_vertex_shader(sctx->blitter, sctx->vs_shader);
util_blitter_save_vertex_elements(sctx->blitter, sctx->vertex_elements);
if (sctx->queued.named.viewport) {
util_blitter_save_viewport(sctx->blitter, &sctx->queued.named.viewport->viewport);
}
util_blitter_save_vertex_buffer_slot(sctx->blitter, sctx->vertex_buffer);
util_blitter_save_so_targets(sctx->blitter, sctx->b.streamout.num_targets,
(struct pipe_stream_output_target**)sctx->b.streamout.targets);
if (op & SI_SAVE_FRAMEBUFFER)
util_blitter_save_framebuffer(sctx->blitter, &sctx->framebuffer.state);
if (op & SI_SAVE_TEXTURES) {
util_blitter_save_fragment_sampler_states(
sctx->blitter, sctx->samplers[PIPE_SHADER_FRAGMENT].n_samplers,
(void**)sctx->samplers[PIPE_SHADER_FRAGMENT].samplers);
util_blitter_save_fragment_sampler_views(sctx->blitter,
util_last_bit(sctx->samplers[PIPE_SHADER_FRAGMENT].views.desc.enabled_mask &
((1 << NUM_TEX_UNITS) - 1)),
sctx->samplers[PIPE_SHADER_FRAGMENT].views.views);
}
if ((op & SI_DISABLE_RENDER_COND) && sctx->b.current_render_cond) {
util_blitter_save_render_condition(sctx->blitter,
sctx->b.current_render_cond,
sctx->b.current_render_cond_cond,
sctx->b.current_render_cond_mode);
}
}
static void r600_blitter_begin(struct pipe_context *ctx, enum r600_blitter_op op)
{
struct r600_context *rctx = (struct r600_context *)ctx;
r600_context_queries_suspend(rctx);
util_blitter_save_blend(rctx->blitter, rctx->queued.named.blend);
util_blitter_save_depth_stencil_alpha(rctx->blitter, rctx->queued.named.dsa);
util_blitter_save_stencil_ref(rctx->blitter, &rctx->stencil_ref);
util_blitter_save_rasterizer(rctx->blitter, rctx->queued.named.rasterizer);
util_blitter_save_fragment_shader(rctx->blitter, rctx->ps_shader);
util_blitter_save_vertex_shader(rctx->blitter, rctx->vs_shader);
util_blitter_save_vertex_elements(rctx->blitter, rctx->vertex_elements);
if (rctx->queued.named.viewport) {
util_blitter_save_viewport(rctx->blitter, &rctx->queued.named.viewport->viewport);
}
util_blitter_save_vertex_buffer_slot(rctx->blitter, rctx->vertex_buffer);
util_blitter_save_so_targets(rctx->blitter, rctx->b.streamout.num_targets,
(struct pipe_stream_output_target**)rctx->b.streamout.targets);
if (op & R600_SAVE_FRAMEBUFFER)
util_blitter_save_framebuffer(rctx->blitter, &rctx->framebuffer);
if (op & R600_SAVE_TEXTURES) {
util_blitter_save_fragment_sampler_states(
rctx->blitter, rctx->samplers[PIPE_SHADER_FRAGMENT].n_samplers,
(void**)rctx->samplers[PIPE_SHADER_FRAGMENT].samplers);
util_blitter_save_fragment_sampler_views(rctx->blitter,
util_last_bit(rctx->samplers[PIPE_SHADER_FRAGMENT].views.desc.enabled_mask &
((1 << NUM_TEX_UNITS) - 1)),
rctx->samplers[PIPE_SHADER_FRAGMENT].views.views);
}
if ((op & R600_DISABLE_RENDER_COND) && rctx->current_render_cond) {
rctx->saved_render_cond = rctx->current_render_cond;
rctx->saved_render_cond_cond = rctx->current_render_cond_cond;
rctx->saved_render_cond_mode = rctx->current_render_cond_mode;
rctx->b.b.render_condition(&rctx->b.b, NULL, FALSE, 0);
}
}
static float
etna_screen_get_paramf(struct pipe_screen *pscreen, enum pipe_capf param)
{
struct etna_screen *screen = etna_screen(pscreen);
switch (param) {
case PIPE_CAPF_MAX_LINE_WIDTH:
case PIPE_CAPF_MAX_LINE_WIDTH_AA:
case PIPE_CAPF_MAX_POINT_WIDTH:
case PIPE_CAPF_MAX_POINT_WIDTH_AA:
return 8192.0f;
case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
return 16.0f;
case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
return util_last_bit(screen->specs.max_texture_size);
}
debug_printf("unknown paramf %d", param);
return 0;
}
/**
* Write out a batch of 32 control data bits from the control_data_bits
* register to the URB.
*
* The current value of the vertex_count register determines which DWORD in
* the URB receives the control data bits. The control_data_bits register is
* assumed to contain the correct data for the vertex that was most recently
* output, and all previous vertices that share the same DWORD.
*
* This function takes care of ensuring that if no vertices have been output
* yet, no control bits are emitted.
*/
void
vec4_gs_visitor::emit_control_data_bits()
{
assert(c->control_data_bits_per_vertex != 0);
/* Since the URB_WRITE_OWORD message operates with 128-bit (vec4 sized)
* granularity, we need to use two tricks to ensure that the batch of 32
* control data bits is written to the appropriate DWORD in the URB. To
* select which vec4 we are writing to, we use the "slot {0,1} offset"
* fields of the message header. To select which DWORD in the vec4 we are
* writing to, we use the channel mask fields of the message header. To
* avoid penalizing geometry shaders that emit a small number of vertices
* with extra bookkeeping, we only do each of these tricks when
* c->prog_data.control_data_header_size_bits is large enough to make it
* necessary.
*
* Note: this means that if we're outputting just a single DWORD of control
* data bits, we'll actually replicate it four times since we won't do any
* channel masking. But that's not a problem since in this case the
* hardware only pays attention to the first DWORD.
*/
enum brw_urb_write_flags urb_write_flags = BRW_URB_WRITE_OWORD;
if (c->control_data_header_size_bits > 32)
urb_write_flags = urb_write_flags | BRW_URB_WRITE_USE_CHANNEL_MASKS;
if (c->control_data_header_size_bits > 128)
urb_write_flags = urb_write_flags | BRW_URB_WRITE_PER_SLOT_OFFSET;
/* If we are using either channel masks or a per-slot offset, then we
* need to figure out which DWORD we are trying to write to, using the
* formula:
*
* dword_index = (vertex_count - 1) * bits_per_vertex / 32
*
* Since bits_per_vertex is a power of two, and is known at compile
* time, this can be optimized to:
*
* dword_index = (vertex_count - 1) >> (6 - log2(bits_per_vertex))
*/
src_reg dword_index(this, glsl_type::uint_type);
if (urb_write_flags) {
src_reg prev_count(this, glsl_type::uint_type);
emit(ADD(dst_reg(prev_count), this->vertex_count,
brw_imm_ud(0xffffffffu)));
unsigned log2_bits_per_vertex =
util_last_bit(c->control_data_bits_per_vertex);
emit(SHR(dst_reg(dword_index), prev_count,
brw_imm_ud(6 - log2_bits_per_vertex)));
}
/* Start building the URB write message. The first MRF gets a copy of
* R0.
*/
int base_mrf = 1;
dst_reg mrf_reg(MRF, base_mrf);
src_reg r0(retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UD));
vec4_instruction *inst = emit(MOV(mrf_reg, r0));
inst->force_writemask_all = true;
if (urb_write_flags & BRW_URB_WRITE_PER_SLOT_OFFSET) {
/* Set the per-slot offset to dword_index / 4, to that we'll write to
* the appropriate OWORD within the control data header.
*/
src_reg per_slot_offset(this, glsl_type::uint_type);
emit(SHR(dst_reg(per_slot_offset), dword_index, brw_imm_ud(2u)));
emit(GS_OPCODE_SET_WRITE_OFFSET, mrf_reg, per_slot_offset,
brw_imm_ud(1u));
}
if (urb_write_flags & BRW_URB_WRITE_USE_CHANNEL_MASKS) {
/* Set the channel masks to 1 << (dword_index % 4), so that we'll
* write to the appropriate DWORD within the OWORD. We need to do
* this computation with force_writemask_all, otherwise garbage data
* from invocation 0 might clobber the mask for invocation 1 when
* GS_OPCODE_PREPARE_CHANNEL_MASKS tries to OR the two masks
* together.
*/
src_reg channel(this, glsl_type::uint_type);
inst = emit(AND(dst_reg(channel), dword_index, brw_imm_ud(3u)));
inst->force_writemask_all = true;
src_reg one(this, glsl_type::uint_type);
inst = emit(MOV(dst_reg(one), brw_imm_ud(1u)));
inst->force_writemask_all = true;
src_reg channel_mask(this, glsl_type::uint_type);
inst = emit(SHL(dst_reg(channel_mask), one, channel));
inst->force_writemask_all = true;
emit(GS_OPCODE_PREPARE_CHANNEL_MASKS, dst_reg(channel_mask),
channel_mask);
emit(GS_OPCODE_SET_CHANNEL_MASKS, mrf_reg, channel_mask);
}
/* Store the control data bits in the message payload and send it. */
dst_reg mrf_reg2(MRF, base_mrf + 1);
inst = emit(MOV(mrf_reg2, this->control_data_bits));
inst->force_writemask_all = true;
inst = emit(GS_OPCODE_URB_WRITE);
inst->urb_write_flags = urb_write_flags;
/* We need to increment Global Offset by 256-bits to make room for
* Broadwell's extra "Vertex Count" payload at the beginning of the
* URB entry. Since this is an OWord message, Global Offset is counted
* in 128-bit units, so we must set it to 2.
*/
if (devinfo->gen >= 8 && gs_prog_data->static_vertex_count == -1)
inst->offset = 2;
inst->base_mrf = base_mrf;
inst->mlen = 2;
}
void
fd5_program_emit(struct fd_context *ctx, struct fd_ringbuffer *ring,
struct fd5_emit *emit)
{
struct stage s[MAX_STAGES];
uint32_t pos_regid, psize_regid, color_regid[8];
uint32_t face_regid, coord_regid, zwcoord_regid;
uint32_t vcoord_regid, vertex_regid, instance_regid;
enum a3xx_threadsize fssz;
uint8_t psize_loc = ~0;
int i, j;
setup_stages(emit, s);
fssz = (s[FS].i->max_reg >= 24) ? TWO_QUADS : FOUR_QUADS;
pos_regid = ir3_find_output_regid(s[VS].v, VARYING_SLOT_POS);
psize_regid = ir3_find_output_regid(s[VS].v, VARYING_SLOT_PSIZ);
vertex_regid = ir3_find_sysval_regid(s[VS].v, SYSTEM_VALUE_VERTEX_ID_ZERO_BASE);
instance_regid = ir3_find_sysval_regid(s[VS].v, SYSTEM_VALUE_INSTANCE_ID);
if (s[FS].v->color0_mrt) {
color_regid[0] = color_regid[1] = color_regid[2] = color_regid[3] =
color_regid[4] = color_regid[5] = color_regid[6] = color_regid[7] =
ir3_find_output_regid(s[FS].v, FRAG_RESULT_COLOR);
} else {
color_regid[0] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA0);
color_regid[1] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA1);
color_regid[2] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA2);
color_regid[3] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA3);
color_regid[4] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA4);
color_regid[5] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA5);
color_regid[6] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA6);
color_regid[7] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA7);
}
/* TODO get these dynamically: */
face_regid = s[FS].v->frag_face ? regid(0,0) : regid(63,0);
coord_regid = s[FS].v->frag_coord ? regid(0,0) : regid(63,0);
zwcoord_regid = s[FS].v->frag_coord ? regid(0,2) : regid(63,0);
vcoord_regid = (s[FS].v->total_in > 0) ? s[FS].v->pos_regid : regid(63,0);
/* we could probably divide this up into things that need to be
* emitted if frag-prog is dirty vs if vert-prog is dirty..
*/
OUT_PKT4(ring, REG_A5XX_HLSQ_VS_CONFIG, 5);
OUT_RING(ring, A5XX_HLSQ_VS_CONFIG_CONSTOBJECTOFFSET(s[VS].constoff) |
A5XX_HLSQ_VS_CONFIG_SHADEROBJOFFSET(s[VS].instroff) |
COND(s[VS].v, A5XX_HLSQ_VS_CONFIG_ENABLED));
OUT_RING(ring, A5XX_HLSQ_FS_CONFIG_CONSTOBJECTOFFSET(s[FS].constoff) |
A5XX_HLSQ_FS_CONFIG_SHADEROBJOFFSET(s[FS].instroff) |
COND(s[FS].v, A5XX_HLSQ_FS_CONFIG_ENABLED));
OUT_RING(ring, A5XX_HLSQ_HS_CONFIG_CONSTOBJECTOFFSET(s[HS].constoff) |
A5XX_HLSQ_HS_CONFIG_SHADEROBJOFFSET(s[HS].instroff) |
COND(s[HS].v, A5XX_HLSQ_HS_CONFIG_ENABLED));
OUT_RING(ring, A5XX_HLSQ_DS_CONFIG_CONSTOBJECTOFFSET(s[DS].constoff) |
A5XX_HLSQ_DS_CONFIG_SHADEROBJOFFSET(s[DS].instroff) |
COND(s[DS].v, A5XX_HLSQ_DS_CONFIG_ENABLED));
OUT_RING(ring, A5XX_HLSQ_GS_CONFIG_CONSTOBJECTOFFSET(s[GS].constoff) |
A5XX_HLSQ_GS_CONFIG_SHADEROBJOFFSET(s[GS].instroff) |
COND(s[GS].v, A5XX_HLSQ_GS_CONFIG_ENABLED));
OUT_PKT4(ring, REG_A5XX_HLSQ_CS_CONFIG, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT4(ring, REG_A5XX_HLSQ_VS_CNTL, 5);
OUT_RING(ring, A5XX_HLSQ_VS_CNTL_INSTRLEN(s[VS].instrlen) |
COND(s[VS].v && s[VS].v->has_ssbo, A5XX_HLSQ_VS_CNTL_SSBO_ENABLE));
OUT_RING(ring, A5XX_HLSQ_FS_CNTL_INSTRLEN(s[FS].instrlen) |
COND(s[FS].v && s[FS].v->has_ssbo, A5XX_HLSQ_FS_CNTL_SSBO_ENABLE));
OUT_RING(ring, A5XX_HLSQ_HS_CNTL_INSTRLEN(s[HS].instrlen) |
COND(s[HS].v && s[HS].v->has_ssbo, A5XX_HLSQ_HS_CNTL_SSBO_ENABLE));
OUT_RING(ring, A5XX_HLSQ_DS_CNTL_INSTRLEN(s[DS].instrlen) |
COND(s[DS].v && s[DS].v->has_ssbo, A5XX_HLSQ_DS_CNTL_SSBO_ENABLE));
OUT_RING(ring, A5XX_HLSQ_GS_CNTL_INSTRLEN(s[GS].instrlen) |
COND(s[GS].v && s[GS].v->has_ssbo, A5XX_HLSQ_GS_CNTL_SSBO_ENABLE));
OUT_PKT4(ring, REG_A5XX_SP_VS_CONFIG, 5);
OUT_RING(ring, A5XX_SP_VS_CONFIG_CONSTOBJECTOFFSET(s[VS].constoff) |
A5XX_SP_VS_CONFIG_SHADEROBJOFFSET(s[VS].instroff) |
COND(s[VS].v, A5XX_SP_VS_CONFIG_ENABLED));
OUT_RING(ring, A5XX_SP_FS_CONFIG_CONSTOBJECTOFFSET(s[FS].constoff) |
A5XX_SP_FS_CONFIG_SHADEROBJOFFSET(s[FS].instroff) |
COND(s[FS].v, A5XX_SP_FS_CONFIG_ENABLED));
OUT_RING(ring, A5XX_SP_HS_CONFIG_CONSTOBJECTOFFSET(s[HS].constoff) |
A5XX_SP_HS_CONFIG_SHADEROBJOFFSET(s[HS].instroff) |
COND(s[HS].v, A5XX_SP_HS_CONFIG_ENABLED));
OUT_RING(ring, A5XX_SP_DS_CONFIG_CONSTOBJECTOFFSET(s[DS].constoff) |
A5XX_SP_DS_CONFIG_SHADEROBJOFFSET(s[DS].instroff) |
COND(s[DS].v, A5XX_SP_DS_CONFIG_ENABLED));
OUT_RING(ring, A5XX_SP_GS_CONFIG_CONSTOBJECTOFFSET(s[GS].constoff) |
A5XX_SP_GS_CONFIG_SHADEROBJOFFSET(s[GS].instroff) |
COND(s[GS].v, A5XX_SP_GS_CONFIG_ENABLED));
OUT_PKT4(ring, REG_A5XX_SP_CS_CONFIG, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT4(ring, REG_A5XX_HLSQ_VS_CONSTLEN, 2);
OUT_RING(ring, s[VS].constlen); /* HLSQ_VS_CONSTLEN */
OUT_RING(ring, s[VS].instrlen); /* HLSQ_VS_INSTRLEN */
OUT_PKT4(ring, REG_A5XX_HLSQ_FS_CONSTLEN, 2);
OUT_RING(ring, s[FS].constlen); /* HLSQ_FS_CONSTLEN */
OUT_RING(ring, s[FS].instrlen); /* HLSQ_FS_INSTRLEN */
OUT_PKT4(ring, REG_A5XX_HLSQ_HS_CONSTLEN, 2);
OUT_RING(ring, s[HS].constlen); /* HLSQ_HS_CONSTLEN */
OUT_RING(ring, s[HS].instrlen); /* HLSQ_HS_INSTRLEN */
OUT_PKT4(ring, REG_A5XX_HLSQ_DS_CONSTLEN, 2);
OUT_RING(ring, s[DS].constlen); /* HLSQ_DS_CONSTLEN */
OUT_RING(ring, s[DS].instrlen); /* HLSQ_DS_INSTRLEN */
OUT_PKT4(ring, REG_A5XX_HLSQ_GS_CONSTLEN, 2);
OUT_RING(ring, s[GS].constlen); /* HLSQ_GS_CONSTLEN */
OUT_RING(ring, s[GS].instrlen); /* HLSQ_GS_INSTRLEN */
OUT_PKT4(ring, REG_A5XX_HLSQ_CS_CONSTLEN, 2);
OUT_RING(ring, 0x00000000); /* HLSQ_CS_CONSTLEN */
OUT_RING(ring, 0x00000000); /* HLSQ_CS_INSTRLEN */
OUT_PKT4(ring, REG_A5XX_SP_VS_CTRL_REG0, 1);
OUT_RING(ring, A5XX_SP_VS_CTRL_REG0_HALFREGFOOTPRINT(s[VS].i->max_half_reg + 1) |
A5XX_SP_VS_CTRL_REG0_FULLREGFOOTPRINT(s[VS].i->max_reg + 1) |
0x6 | /* XXX seems to be always set? */
A5XX_SP_VS_CTRL_REG0_BRANCHSTACK(0x3) | // XXX need to figure this out somehow..
COND(s[VS].v->has_samp, A5XX_SP_VS_CTRL_REG0_PIXLODENABLE));
struct ir3_shader_linkage l = {0};
ir3_link_shaders(&l, s[VS].v, s[FS].v);
if ((s[VS].v->shader->stream_output.num_outputs > 0) &&
!emit->key.binning_pass)
link_stream_out(&l, s[VS].v);
BITSET_DECLARE(varbs, 128) = {0};
uint32_t *varmask = (uint32_t *)varbs;
for (i = 0; i < l.cnt; i++)
for (j = 0; j < util_last_bit(l.var[i].compmask); j++)
BITSET_SET(varbs, l.var[i].loc + j);
OUT_PKT4(ring, REG_A5XX_VPC_VAR_DISABLE(0), 4);
OUT_RING(ring, ~varmask[0]); /* VPC_VAR[0].DISABLE */
OUT_RING(ring, ~varmask[1]); /* VPC_VAR[1].DISABLE */
OUT_RING(ring, ~varmask[2]); /* VPC_VAR[2].DISABLE */
OUT_RING(ring, ~varmask[3]); /* VPC_VAR[3].DISABLE */
/* a5xx appends pos/psize to end of the linkage map: */
if (pos_regid != regid(63,0))
ir3_link_add(&l, pos_regid, 0xf, l.max_loc);
if (psize_regid != regid(63,0)) {
psize_loc = l.max_loc;
ir3_link_add(&l, psize_regid, 0x1, l.max_loc);
}
if ((s[VS].v->shader->stream_output.num_outputs > 0) &&
!emit->key.binning_pass) {
emit_stream_out(ring, s[VS].v, &l);
OUT_PKT4(ring, REG_A5XX_VPC_SO_OVERRIDE, 1);
OUT_RING(ring, 0x00000000);
} else {
OUT_PKT4(ring, REG_A5XX_VPC_SO_OVERRIDE, 1);
OUT_RING(ring, A5XX_VPC_SO_OVERRIDE_SO_DISABLE);
}
for (i = 0, j = 0; (i < 16) && (j < l.cnt); i++) {
uint32_t reg = 0;
OUT_PKT4(ring, REG_A5XX_SP_VS_OUT_REG(i), 1);
reg |= A5XX_SP_VS_OUT_REG_A_REGID(l.var[j].regid);
reg |= A5XX_SP_VS_OUT_REG_A_COMPMASK(l.var[j].compmask);
j++;
reg |= A5XX_SP_VS_OUT_REG_B_REGID(l.var[j].regid);
reg |= A5XX_SP_VS_OUT_REG_B_COMPMASK(l.var[j].compmask);
j++;
OUT_RING(ring, reg);
}
for (i = 0, j = 0; (i < 8) && (j < l.cnt); i++) {
uint32_t reg = 0;
OUT_PKT4(ring, REG_A5XX_SP_VS_VPC_DST_REG(i), 1);
reg |= A5XX_SP_VS_VPC_DST_REG_OUTLOC0(l.var[j++].loc);
reg |= A5XX_SP_VS_VPC_DST_REG_OUTLOC1(l.var[j++].loc);
reg |= A5XX_SP_VS_VPC_DST_REG_OUTLOC2(l.var[j++].loc);
reg |= A5XX_SP_VS_VPC_DST_REG_OUTLOC3(l.var[j++].loc);
OUT_RING(ring, reg);
}
OUT_PKT4(ring, REG_A5XX_SP_VS_OBJ_START_LO, 2);
OUT_RELOC(ring, s[VS].v->bo, 0, 0, 0); /* SP_VS_OBJ_START_LO/HI */
if (s[VS].instrlen)
fd5_emit_shader(ring, s[VS].v);
// TODO depending on other bits in this reg (if any) set somewhere else?
OUT_PKT4(ring, REG_A5XX_PC_PRIM_VTX_CNTL, 1);
OUT_RING(ring, COND(s[VS].v->writes_psize, A5XX_PC_PRIM_VTX_CNTL_PSIZE));
OUT_PKT4(ring, REG_A5XX_SP_PRIMITIVE_CNTL, 1);
OUT_RING(ring, A5XX_SP_PRIMITIVE_CNTL_VSOUT(l.cnt));
OUT_PKT4(ring, REG_A5XX_VPC_CNTL_0, 1);
OUT_RING(ring, A5XX_VPC_CNTL_0_STRIDE_IN_VPC(l.max_loc) |
COND(s[FS].v->total_in > 0, A5XX_VPC_CNTL_0_VARYING) |
COND(s[FS].v->frag_coord, A5XX_VPC_CNTL_0_VARYING) |
0x10000); // XXX
fd5_context(ctx)->max_loc = l.max_loc;
if (emit->key.binning_pass) {
OUT_PKT4(ring, REG_A5XX_SP_FS_OBJ_START_LO, 2);
OUT_RING(ring, 0x00000000); /* SP_FS_OBJ_START_LO */
OUT_RING(ring, 0x00000000); /* SP_FS_OBJ_START_HI */
} else {
OUT_PKT4(ring, REG_A5XX_SP_FS_OBJ_START_LO, 2);
OUT_RELOC(ring, s[FS].v->bo, 0, 0, 0); /* SP_FS_OBJ_START_LO/HI */
}
OUT_PKT4(ring, REG_A5XX_HLSQ_CONTROL_0_REG, 5);
OUT_RING(ring, A5XX_HLSQ_CONTROL_0_REG_FSTHREADSIZE(fssz) |
A5XX_HLSQ_CONTROL_0_REG_CSTHREADSIZE(TWO_QUADS) |
0x00000880); /* XXX HLSQ_CONTROL_0 */
OUT_RING(ring, A5XX_HLSQ_CONTROL_1_REG_PRIMALLOCTHRESHOLD(63));
OUT_RING(ring, A5XX_HLSQ_CONTROL_2_REG_FACEREGID(face_regid) |
0xfcfcfc00); /* XXX */
OUT_RING(ring, A5XX_HLSQ_CONTROL_3_REG_FRAGCOORDXYREGID(vcoord_regid) |
0xfcfcfc00); /* XXX */
OUT_RING(ring, A5XX_HLSQ_CONTROL_4_REG_XYCOORDREGID(coord_regid) |
A5XX_HLSQ_CONTROL_4_REG_ZWCOORDREGID(zwcoord_regid) |
0x0000fcfc); /* XXX */
OUT_PKT4(ring, REG_A5XX_SP_FS_CTRL_REG0, 1);
OUT_RING(ring, COND(s[FS].v->total_in > 0, A5XX_SP_FS_CTRL_REG0_VARYING) |
COND(s[FS].v->frag_coord, A5XX_SP_FS_CTRL_REG0_VARYING) |
0x40006 | /* XXX set pretty much everywhere */
A5XX_SP_FS_CTRL_REG0_THREADSIZE(fssz) |
A5XX_SP_FS_CTRL_REG0_HALFREGFOOTPRINT(s[FS].i->max_half_reg + 1) |
A5XX_SP_FS_CTRL_REG0_FULLREGFOOTPRINT(s[FS].i->max_reg + 1) |
A5XX_SP_FS_CTRL_REG0_BRANCHSTACK(0x3) | // XXX need to figure this out somehow..
COND(s[FS].v->has_samp, A5XX_SP_FS_CTRL_REG0_PIXLODENABLE));
OUT_PKT4(ring, REG_A5XX_HLSQ_UPDATE_CNTL, 1);
OUT_RING(ring, 0x020fffff); /* XXX */
OUT_PKT4(ring, REG_A5XX_VPC_GS_SIV_CNTL, 1);
OUT_RING(ring, 0x0000ffff); /* XXX */
OUT_PKT4(ring, REG_A5XX_SP_SP_CNTL, 1);
OUT_RING(ring, 0x00000010); /* XXX */
OUT_PKT4(ring, REG_A5XX_GRAS_CNTL, 1);
OUT_RING(ring, COND(s[FS].v->total_in > 0, A5XX_GRAS_CNTL_VARYING) |
COND(s[FS].v->frag_coord, A5XX_GRAS_CNTL_XCOORD |
A5XX_GRAS_CNTL_YCOORD |
A5XX_GRAS_CNTL_ZCOORD |
A5XX_GRAS_CNTL_WCOORD |
A5XX_GRAS_CNTL_UNK3) |
COND(s[FS].v->frag_face, A5XX_GRAS_CNTL_UNK3));
OUT_PKT4(ring, REG_A5XX_RB_RENDER_CONTROL0, 2);
OUT_RING(ring, COND(s[FS].v->total_in > 0, A5XX_RB_RENDER_CONTROL0_VARYING) |
COND(s[FS].v->frag_coord, A5XX_RB_RENDER_CONTROL0_XCOORD |
A5XX_RB_RENDER_CONTROL0_YCOORD |
A5XX_RB_RENDER_CONTROL0_ZCOORD |
A5XX_RB_RENDER_CONTROL0_WCOORD |
A5XX_RB_RENDER_CONTROL0_UNK3) |
COND(s[FS].v->frag_face, A5XX_RB_RENDER_CONTROL0_UNK3));
OUT_RING(ring, COND(s[FS].v->frag_face, A5XX_RB_RENDER_CONTROL1_FACENESS));
OUT_PKT4(ring, REG_A5XX_SP_FS_OUTPUT_REG(0), 8);
for (i = 0; i < 8; i++) {
OUT_RING(ring, A5XX_SP_FS_OUTPUT_REG_REGID(color_regid[i]) |
COND(emit->key.half_precision,
A5XX_SP_FS_OUTPUT_REG_HALF_PRECISION));
}
OUT_PKT4(ring, REG_A5XX_VPC_PACK, 1);
OUT_RING(ring, A5XX_VPC_PACK_NUMNONPOSVAR(s[FS].v->total_in) |
A5XX_VPC_PACK_PSIZELOC(psize_loc));
if (!emit->key.binning_pass) {
uint32_t vinterp[8], vpsrepl[8];
memset(vinterp, 0, sizeof(vinterp));
memset(vpsrepl, 0, sizeof(vpsrepl));
/* looks like we need to do int varyings in the frag
* shader on a5xx (no flatshad reg? or a420.0 bug?):
*
* (sy)(ss)nop
* (sy)ldlv.u32 r0.x,l[r0.x], 1
* ldlv.u32 r0.y,l[r0.x+1], 1
* (ss)bary.f (ei)r63.x, 0, r0.x
* (ss)(rpt1)cov.s32f16 hr0.x, (r)r0.x
* (rpt5)nop
* sam (f16)(xyzw)hr0.x, hr0.x, s#0, t#0
*
* Possibly on later a5xx variants we'll be able to use
* something like the code below instead of workaround
* in the shader:
*/
/* figure out VARYING_INTERP / VARYING_PS_REPL register values: */
for (j = -1; (j = ir3_next_varying(s[FS].v, j)) < (int)s[FS].v->inputs_count; ) {
/* NOTE: varyings are packed, so if compmask is 0xb
* then first, third, and fourth component occupy
* three consecutive varying slots:
*/
unsigned compmask = s[FS].v->inputs[j].compmask;
uint32_t inloc = s[FS].v->inputs[j].inloc;
if ((s[FS].v->inputs[j].interpolate == INTERP_MODE_FLAT) ||
(s[FS].v->inputs[j].rasterflat && emit->rasterflat)) {
uint32_t loc = inloc;
for (i = 0; i < 4; i++) {
if (compmask & (1 << i)) {
vinterp[loc / 16] |= 1 << ((loc % 16) * 2);
//flatshade[loc / 32] |= 1 << (loc % 32);
loc++;
}
}
}
gl_varying_slot slot = s[FS].v->inputs[j].slot;
/* since we don't enable PIPE_CAP_TGSI_TEXCOORD: */
if (slot >= VARYING_SLOT_VAR0) {
unsigned texmask = 1 << (slot - VARYING_SLOT_VAR0);
/* Replace the .xy coordinates with S/T from the point sprite. Set
* interpolation bits for .zw such that they become .01
*/
if (emit->sprite_coord_enable & texmask) {
/* mask is two 2-bit fields, where:
* '01' -> S
* '10' -> T
* '11' -> 1 - T (flip mode)
*/
unsigned mask = emit->sprite_coord_mode ? 0b1101 : 0b1001;
uint32_t loc = inloc;
if (compmask & 0x1) {
vpsrepl[loc / 16] |= ((mask >> 0) & 0x3) << ((loc % 16) * 2);
loc++;
}
if (compmask & 0x2) {
vpsrepl[loc / 16] |= ((mask >> 2) & 0x3) << ((loc % 16) * 2);
loc++;
}
static bool
ps_get_gen6_ff_kernels(const struct ilo_dev *dev,
const struct ilo_state_ps_info *info,
struct pixel_ff *ff)
{
const struct ilo_state_shader_kernel_info *kernel_8 = &info->kernel_8;
const struct ilo_state_shader_kernel_info *kernel_16 = &info->kernel_16;
const struct ilo_state_shader_kernel_info *kernel_32 = &info->kernel_32;
uint32_t scratch_size;
ILO_DEV_ASSERT(dev, 6, 8);
ff->dispatch_modes = ps_get_gen6_dispatch_modes(dev, info);
/* initialize kernel offsets and GRF starts */
if (util_is_power_of_two(ff->dispatch_modes)) {
if (ff->dispatch_modes & GEN6_PS_DISPATCH_8) {
ff->kernel_offsets[0] = kernel_8->offset;
ff->grf_starts[0] = kernel_8->grf_start;
} else if (ff->dispatch_modes & GEN6_PS_DISPATCH_16) {
ff->kernel_offsets[0] = kernel_16->offset;
ff->grf_starts[0] = kernel_16->grf_start;
} else if (ff->dispatch_modes & GEN6_PS_DISPATCH_32) {
ff->kernel_offsets[0] = kernel_32->offset;
ff->grf_starts[0] = kernel_32->grf_start;
}
} else {
ff->kernel_offsets[0] = kernel_8->offset;
ff->kernel_offsets[1] = kernel_32->offset;
ff->kernel_offsets[2] = kernel_16->offset;
ff->grf_starts[0] = kernel_8->grf_start;
ff->grf_starts[1] = kernel_32->grf_start;
ff->grf_starts[2] = kernel_16->grf_start;
}
/* we do not want to save it */
assert(ff->kernel_offsets[0] == 0);
ff->pcb_enable = (((ff->dispatch_modes & GEN6_PS_DISPATCH_8) &&
kernel_8->pcb_attr_count) ||
((ff->dispatch_modes & GEN6_PS_DISPATCH_16) &&
kernel_16->pcb_attr_count) ||
((ff->dispatch_modes & GEN6_PS_DISPATCH_32) &&
kernel_32->pcb_attr_count));
scratch_size = 0;
if ((ff->dispatch_modes & GEN6_PS_DISPATCH_8) &&
scratch_size < kernel_8->scratch_size)
scratch_size = kernel_8->scratch_size;
if ((ff->dispatch_modes & GEN6_PS_DISPATCH_16) &&
scratch_size < kernel_16->scratch_size)
scratch_size = kernel_16->scratch_size;
if ((ff->dispatch_modes & GEN6_PS_DISPATCH_32) &&
scratch_size < kernel_32->scratch_size)
scratch_size = kernel_32->scratch_size;
/* next power of two, starting from 1KB */
ff->scratch_space = (scratch_size > 1024) ?
(util_last_bit(scratch_size - 1) - 10): 0;
/* GPU hangs on Haswell if none of the dispatch mode bits is set */
if (ilo_dev_gen(dev) == ILO_GEN(7.5) && !ff->dispatch_modes)
ff->dispatch_modes |= GEN6_PS_DISPATCH_8;
return true;
}
static int
etna_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
{
struct etna_screen *screen = etna_screen(pscreen);
switch (param) {
/* Supported features (boolean caps). */
case PIPE_CAP_TWO_SIDED_STENCIL:
case PIPE_CAP_ANISOTROPIC_FILTER:
case PIPE_CAP_POINT_SPRITE:
case PIPE_CAP_TEXTURE_SHADOW_MAP:
case PIPE_CAP_BLEND_EQUATION_SEPARATE:
case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
case PIPE_CAP_SM3:
case PIPE_CAP_TEXTURE_BARRIER:
case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
case PIPE_CAP_USER_CONSTANT_BUFFERS:
case PIPE_CAP_TGSI_TEXCOORD:
case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
return 1;
/* Memory */
case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
return 256;
case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
return 4; /* XXX could easily be supported */
case PIPE_CAP_GLSL_FEATURE_LEVEL:
return 120;
case PIPE_CAP_NPOT_TEXTURES:
return true; /* VIV_FEATURE(priv->dev, chipMinorFeatures1,
NON_POWER_OF_TWO); */
case PIPE_CAP_PRIMITIVE_RESTART:
return VIV_FEATURE(screen, chipMinorFeatures1, HALTI0);
case PIPE_CAP_ENDIANNESS:
return PIPE_ENDIAN_LITTLE; /* on most Viv hw this is configurable (feature
ENDIANNESS_CONFIG) */
/* Unsupported features. */
case PIPE_CAP_SEAMLESS_CUBE_MAP:
case PIPE_CAP_TEXTURE_SWIZZLE: /* XXX supported on gc2000 */
case PIPE_CAP_COMPUTE: /* XXX supported on gc2000 */
case PIPE_CAP_MIXED_COLORBUFFER_FORMATS: /* only one colorbuffer supported, so mixing makes no sense */
case PIPE_CAP_CONDITIONAL_RENDER: /* no occlusion queries */
case PIPE_CAP_TGSI_INSTANCEID: /* no idea, really */
case PIPE_CAP_START_INSTANCE: /* instancing not supported AFAIK */
case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR: /* instancing not supported AFAIK */
case PIPE_CAP_SHADER_STENCIL_EXPORT: /* Fragment shader cannot export stencil value */
case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS: /* no dual-source supported */
case PIPE_CAP_TEXTURE_MULTISAMPLE: /* no texture multisample */
case PIPE_CAP_TEXTURE_MIRROR_CLAMP: /* only mirrored repeat */
case PIPE_CAP_INDEP_BLEND_ENABLE:
case PIPE_CAP_INDEP_BLEND_FUNC:
case PIPE_CAP_DEPTH_CLIP_DISABLE:
case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS: /* Don't skip strict max uniform limit check */
case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
case PIPE_CAP_VERTEX_COLOR_CLAMPED:
case PIPE_CAP_USER_VERTEX_BUFFERS:
case PIPE_CAP_USER_INDEX_BUFFERS:
case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES: /* TODO: test me out with piglit */
case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
case PIPE_CAP_TEXTURE_GATHER_SM5:
case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
case PIPE_CAP_FAKE_SW_MSAA:
case PIPE_CAP_TEXTURE_QUERY_LOD:
case PIPE_CAP_SAMPLE_SHADING:
case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
case PIPE_CAP_DRAW_INDIRECT:
case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
case PIPE_CAP_SAMPLER_VIEW_TARGET:
case PIPE_CAP_CLIP_HALFZ:
case PIPE_CAP_VERTEXID_NOBASE:
case PIPE_CAP_POLYGON_OFFSET_CLAMP:
case PIPE_CAP_MULTISAMPLE_Z_RESOLVE:
case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
case PIPE_CAP_DEPTH_BOUNDS_TEST:
case PIPE_CAP_TGSI_TXQS:
case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
case PIPE_CAP_SHAREABLE_SHADERS:
case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
case PIPE_CAP_CLEAR_TEXTURE:
case PIPE_CAP_DRAW_PARAMETERS:
case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
case PIPE_CAP_MULTI_DRAW_INDIRECT:
case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL:
case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL:
case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
case PIPE_CAP_INVALIDATE_BUFFER:
case PIPE_CAP_GENERATE_MIPMAP:
case PIPE_CAP_STRING_MARKER:
case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS:
case PIPE_CAP_QUERY_BUFFER_OBJECT:
case PIPE_CAP_QUERY_MEMORY_INFO:
case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT:
case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR:
case PIPE_CAP_CULL_DISTANCE:
case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES:
case PIPE_CAP_TGSI_VOTE:
case PIPE_CAP_MAX_WINDOW_RECTANGLES:
case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED:
case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS:
case PIPE_CAP_MIXED_COLOR_DEPTH_BITS:
case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS:
case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
case PIPE_CAP_NATIVE_FENCE_FD:
case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
case PIPE_CAP_TGSI_FS_FBFETCH:
case PIPE_CAP_TGSI_MUL_ZERO_WINS:
case PIPE_CAP_DOUBLES:
case PIPE_CAP_INT64:
case PIPE_CAP_INT64_DIVMOD:
return 0;
/* Stream output. */
case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
return 0;
/* Geometry shader output, unsupported. */
case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
case PIPE_CAP_MAX_VERTEX_STREAMS:
return 0;
case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
return 128;
/* Texturing. */
case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
{
int log2_max_tex_size = util_last_bit(screen->specs.max_texture_size);
assert(log2_max_tex_size > 0);
return log2_max_tex_size;
}
case PIPE_CAP_MAX_TEXTURE_3D_LEVELS: /* 3D textures not supported - fake it */
return 5;
case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
return 0;
case PIPE_CAP_CUBE_MAP_ARRAY:
return 0;
case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
case PIPE_CAP_MIN_TEXEL_OFFSET:
return -8;
case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
case PIPE_CAP_MAX_TEXEL_OFFSET:
return 7;
case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
return 0;
case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
return 65536;
/* Render targets. */
case PIPE_CAP_MAX_RENDER_TARGETS:
return 1;
/* Viewports and scissors. */
case PIPE_CAP_MAX_VIEWPORTS:
return 1;
/* Timer queries. */
case PIPE_CAP_QUERY_TIME_ELAPSED:
case PIPE_CAP_OCCLUSION_QUERY:
return 0;
case PIPE_CAP_QUERY_TIMESTAMP:
return 1;
case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
return 0;
/* Preferences */
case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
return 0;
case PIPE_CAP_PCI_GROUP:
case PIPE_CAP_PCI_BUS:
case PIPE_CAP_PCI_DEVICE:
case PIPE_CAP_PCI_FUNCTION:
return 0;
case PIPE_CAP_VENDOR_ID:
case PIPE_CAP_DEVICE_ID:
return 0xFFFFFFFF;
case PIPE_CAP_ACCELERATED:
return 1;
case PIPE_CAP_VIDEO_MEMORY:
return 0;
case PIPE_CAP_UMA:
return 1;
}
debug_printf("unknown param %d", param);
return 0;
}