void
fd2_emit_vertex_bufs(struct fd_ringbuffer *ring, uint32_t val,
struct fd2_vertex_buf *vbufs, uint32_t n)
{
unsigned i;
OUT_PKT3(ring, CP_SET_CONSTANT, 1 + (2 * n));
OUT_RING(ring, (0x1 << 16) | (val & 0xffff));
for (i = 0; i < n; i++) {
struct fd_resource *rsc = fd_resource(vbufs[i].prsc);
OUT_RELOC(ring, rsc->bo, vbufs[i].offset, 3, 0);
OUT_RING (ring, vbufs[i].size);
}
}
static struct fd_hw_sample *
occlusion_get_sample(struct fd_context *ctx, struct fd_ringbuffer *ring)
{
struct fd_hw_sample *samp =
fd_hw_sample_init(ctx, sizeof(struct fd_rb_samp_ctrs));
/* Set RB_SAMPLE_COUNT_ADDR to samp->offset plus value of
* HW_QUERY_BASE_REG register:
*/
OUT_PKT3(ring, CP_SET_CONSTANT, 3);
OUT_RING(ring, CP_REG(REG_A3XX_RB_SAMPLE_COUNT_ADDR) | 0x80000000);
OUT_RING(ring, HW_QUERY_BASE_REG);
OUT_RING(ring, samp->offset);
OUT_PKT0(ring, REG_A3XX_RB_SAMPLE_COUNT_CONTROL, 1);
OUT_RING(ring, A3XX_RB_SAMPLE_COUNT_CONTROL_COPY);
OUT_PKT3(ring, CP_DRAW_INDX, 3);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, DRAW(DI_PT_POINTLIST_A2XX, DI_SRC_SEL_AUTO_INDEX,
INDEX_SIZE_IGN, USE_VISIBILITY));
OUT_RING(ring, 0); /* NumIndices */
fd_event_write(ctx, ring, ZPASS_DONE);
OUT_PKT0(ring, REG_A3XX_RBBM_PERFCTR_CTL, 1);
OUT_RING(ring, A3XX_RBBM_PERFCTR_CTL_ENABLE);
OUT_PKT0(ring, REG_A3XX_VBIF_PERF_CNT_EN, 1);
OUT_RING(ring, A3XX_VBIF_PERF_CNT_EN_CNT0 |
A3XX_VBIF_PERF_CNT_EN_CNT1 |
A3XX_VBIF_PERF_CNT_EN_PWRCNT0 |
A3XX_VBIF_PERF_CNT_EN_PWRCNT1 |
A3XX_VBIF_PERF_CNT_EN_PWRCNT2);
return samp;
}
static void
fd2_draw(struct fd_context *ctx, const struct pipe_draw_info *info)
{
struct fd_ringbuffer *ring = ctx->ring;
if (ctx->dirty & FD_DIRTY_VTXBUF)
emit_vertexbufs(ctx);
fd2_emit_state(ctx, ctx->dirty);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_VGT_INDX_OFFSET));
OUT_RING(ring, info->start);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_VGT_VERTEX_REUSE_BLOCK_CNTL));
OUT_RING(ring, 0x0000003b);
OUT_PKT0(ring, REG_A2XX_TC_CNTL_STATUS, 1);
OUT_RING(ring, A2XX_TC_CNTL_STATUS_L2_INVALIDATE);
OUT_WFI (ring);
OUT_PKT3(ring, CP_SET_CONSTANT, 3);
OUT_RING(ring, CP_REG(REG_A2XX_VGT_MAX_VTX_INDX));
OUT_RING(ring, info->max_index); /* VGT_MAX_VTX_INDX */
OUT_RING(ring, info->min_index); /* VGT_MIN_VTX_INDX */
fd_draw_emit(ctx, ring, IGNORE_VISIBILITY, info);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_UNKNOWN_2010));
OUT_RING(ring, 0x00000000);
emit_cacheflush(ring);
}
static void
emit_shader(struct fd_ringbuffer *ring, struct fd_shader *shader,
enum adreno_state_block state_block)
{
uint32_t i;
OUT_PKT3(ring, CP_LOAD_STATE, 2 + shader->sizedwords);
OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(0) |
CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
CP_LOAD_STATE_0_STATE_BLOCK(state_block) |
CP_LOAD_STATE_0_NUM_UNIT(instrlen(shader)));
OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_SHADER) |
CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
for (i = 0; i < shader->sizedwords; i++)
OUT_RING(ring, shader->bin[i]);
}
static void emit_uniconst(struct fd_ringbuffer *ring,
struct fd_shader *shader, struct fd_parameters *uniforms,
enum adreno_state_block state_block)
{
static uint32_t buf[512]; /* cheesy, but test code isn't multithreaded */
uint32_t i, j, k, sz = 0;
memset(buf, 0, sizeof(buf));
for (i = 0; i < shader->ir->consts_count; i++) {
struct ir3_const *c = shader->ir->consts[i];
uint32_t off = c->cstart->num;
memcpy(&buf[off], c->val, sizeof(c->val));
sz = max(sz, off + (sizeof(c->val) / sizeof(buf[0])));
}
for (i = 0; i < shader->ir->uniforms_count; i++) {
struct ir3_uniform *u = shader->ir->uniforms[i];
struct fd_param *p = find_param(uniforms, u->name);
const uint32_t *dwords = p->data;
uint32_t off = u->cstart->num;
for (j = 0; j < p->count; j++) {
for (k = 0; k < p->size; k++)
buf[off++] = *(dwords++);
/* zero pad if needed: */
for (; k < ALIGN(p->size, 4); k++)
buf[off++] = 0;
}
sz = max(sz, off);
}
/* if no constants, don't emit the CP_LOAD_STATE */
if (sz == 0)
return;
OUT_PKT3(ring, CP_LOAD_STATE, 2 + sz);
OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(0) |
CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
CP_LOAD_STATE_0_STATE_BLOCK(state_block) |
CP_LOAD_STATE_0_NUM_UNIT(sz/2));
OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS) |
CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
for (i = 0; i < sz; i++)
OUT_RING(ring, buf[i]);
}
static void
emit_shader(struct fd_ringbuffer *ring, const struct ir3_shader_variant *so)
{
const struct ir3_info *si = &so->info;
enum adreno_state_block sb;
enum adreno_state_src src;
uint32_t i, sz, *bin;
if (so->type == SHADER_VERTEX) {
sb = SB_VERT_SHADER;
} else {
sb = SB_FRAG_SHADER;
}
if (fd_mesa_debug & FD_DBG_DIRECT) {
sz = si->sizedwords;
src = SS_DIRECT;
bin = fd_bo_map(so->bo);
} else {
sz = 0;
src = 2; // enums different on a4xx..
bin = NULL;
}
OUT_PKT3(ring, CP_LOAD_STATE, 2 + sz);
OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(0) |
CP_LOAD_STATE_0_STATE_SRC(src) |
CP_LOAD_STATE_0_STATE_BLOCK(sb) |
CP_LOAD_STATE_0_NUM_UNIT(so->instrlen));
if (bin) {
OUT_RING(ring, CP_LOAD_STATE_1_EXT_SRC_ADDR(0) |
CP_LOAD_STATE_1_STATE_TYPE(ST_SHADER));
} else {
OUT_RELOC(ring, so->bo, 0,
CP_LOAD_STATE_1_STATE_TYPE(ST_SHADER), 0);
}
/* for how clever coverity is, it is sometimes rather dull, and
* doesn't realize that the only case where bin==NULL, sz==0:
*/
assume(bin || (sz == 0));
for (i = 0; i < sz; i++) {
OUT_RING(ring, bin[i]);
}
}
/* regid: base const register
* prsc or dwords: buffer containing constant values
* sizedwords: size of const value buffer
*/
void
fd3_emit_constant(struct fd_ringbuffer *ring,
enum adreno_state_block sb,
uint32_t regid, uint32_t offset, uint32_t sizedwords,
const uint32_t *dwords, struct pipe_resource *prsc)
{
uint32_t i, sz;
enum adreno_state_src src;
if (prsc) {
sz = 0;
src = SS_INDIRECT;
} else {
sz = sizedwords;
src = SS_DIRECT;
}
/* we have this sometimes, not others.. perhaps we could be clever
* and figure out actually when we need to invalidate cache:
*/
OUT_PKT0(ring, REG_A3XX_UCHE_CACHE_INVALIDATE0_REG, 2);
OUT_RING(ring, A3XX_UCHE_CACHE_INVALIDATE0_REG_ADDR(0));
OUT_RING(ring, A3XX_UCHE_CACHE_INVALIDATE1_REG_ADDR(0) |
A3XX_UCHE_CACHE_INVALIDATE1_REG_OPCODE(INVALIDATE) |
A3XX_UCHE_CACHE_INVALIDATE1_REG_ENTIRE_CACHE);
OUT_PKT3(ring, CP_LOAD_STATE, 2 + sz);
OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(regid/2) |
CP_LOAD_STATE_0_STATE_SRC(src) |
CP_LOAD_STATE_0_STATE_BLOCK(sb) |
CP_LOAD_STATE_0_NUM_UNIT(sizedwords/2));
if (prsc) {
struct fd_bo *bo = fd_resource(prsc)->bo;
OUT_RELOC(ring, bo, offset,
CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS), 0);
} else {
OUT_RING(ring, CP_LOAD_STATE_1_EXT_SRC_ADDR(0) |
CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS));
dwords = (uint32_t *)&((uint8_t *)dwords)[offset];
}
for (i = 0; i < sz; i++) {
OUT_RING(ring, dwords[i]);
}
}
static texmask
emit_texture(struct fd_ringbuffer *ring, struct fd_context *ctx,
struct fd_texture_stateobj *tex, unsigned samp_id, texmask emitted)
{
unsigned const_idx = fd2_get_const_idx(ctx, tex, samp_id);
static const struct fd2_sampler_stateobj dummy_sampler = {};
static const struct fd2_pipe_sampler_view dummy_view = {};
const struct fd2_sampler_stateobj *sampler;
const struct fd2_pipe_sampler_view *view;
struct fd_resource *rsc;
if (emitted & (1 << const_idx))
return 0;
sampler = tex->samplers[samp_id] ?
fd2_sampler_stateobj(tex->samplers[samp_id]) :
&dummy_sampler;
view = tex->textures[samp_id] ?
fd2_pipe_sampler_view(tex->textures[samp_id]) :
&dummy_view;
rsc = view->base.texture ? fd_resource(view->base.texture) : NULL;
OUT_PKT3(ring, CP_SET_CONSTANT, 7);
OUT_RING(ring, 0x00010000 + (0x6 * const_idx));
OUT_RING(ring, sampler->tex0 | view->tex0);
if (rsc)
OUT_RELOC(ring, rsc->bo, 0, view->tex1, 0);
else
OUT_RING(ring, 0);
OUT_RING(ring, view->tex2);
OUT_RING(ring, sampler->tex3 | view->tex3);
OUT_RING(ring, sampler->tex4 | view->tex4);
if (rsc && rsc->base.last_level)
OUT_RELOC(ring, rsc->bo, fd_resource_offset(rsc, 1, 0), view->tex5, 0);
else
OUT_RING(ring, view->tex5);
return (1 << const_idx);
}
static void
emit_shader(struct fd_ringbuffer *ring, const struct ir3_shader_variant *so)
{
const struct ir3_info *si = &so->info;
enum adreno_state_block sb;
enum adreno_state_src src;
uint32_t i, sz, *bin;
if (so->type == SHADER_VERTEX) {
sb = SB_VERT_SHADER;
} else {
sb = SB_FRAG_SHADER;
}
if (fd_mesa_debug & FD_DBG_DIRECT) {
sz = si->sizedwords;
src = SS_DIRECT;
bin = fd_bo_map(so->bo);
} else {
sz = 0;
src = SS_INDIRECT;
bin = NULL;
}
OUT_PKT3(ring, CP_LOAD_STATE, 2 + sz);
OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(0) |
CP_LOAD_STATE_0_STATE_SRC(src) |
CP_LOAD_STATE_0_STATE_BLOCK(sb) |
CP_LOAD_STATE_0_NUM_UNIT(so->instrlen));
if (bin) {
OUT_RING(ring, CP_LOAD_STATE_1_EXT_SRC_ADDR(0) |
CP_LOAD_STATE_1_STATE_TYPE(ST_SHADER));
} else {
OUT_RELOC(ring, so->bo, 0,
CP_LOAD_STATE_1_STATE_TYPE(ST_SHADER), 0);
}
for (i = 0; i < sz; i++) {
OUT_RING(ring, bin[i]);
}
}
/* regid: base const register
* prsc or dwords: buffer containing constant values
* sizedwords: size of const value buffer
*/
void
fd4_emit_const(struct fd_ringbuffer *ring, enum shader_t type,
uint32_t regid, uint32_t offset, uint32_t sizedwords,
const uint32_t *dwords, struct pipe_resource *prsc)
{
uint32_t i, sz;
enum adreno_state_src src;
debug_assert((regid % 4) == 0);
debug_assert((sizedwords % 4) == 0);
if (prsc) {
sz = 0;
src = 0x2; // TODO ??
} else {
sz = sizedwords;
src = SS_DIRECT;
}
OUT_PKT3(ring, CP_LOAD_STATE, 2 + sz);
OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(regid/4) |
CP_LOAD_STATE_0_STATE_SRC(src) |
CP_LOAD_STATE_0_STATE_BLOCK(sb[type]) |
CP_LOAD_STATE_0_NUM_UNIT(sizedwords/4));
if (prsc) {
struct fd_bo *bo = fd_resource(prsc)->bo;
OUT_RELOC(ring, bo, offset,
CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS), 0);
} else {
OUT_RING(ring, CP_LOAD_STATE_1_EXT_SRC_ADDR(0) |
CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS));
dwords = (uint32_t *)&((uint8_t *)dwords)[offset];
}
for (i = 0; i < sz; i++) {
OUT_RING(ring, dwords[i]);
}
}
/* this is same for a2xx/a3xx, so split into helper: */
void
fd_draw_emit(struct fd_context *ctx, const struct pipe_draw_info *info)
{
struct fd_ringbuffer *ring = ctx->ring;
struct pipe_index_buffer *idx = &ctx->indexbuf;
struct fd_bo *idx_bo = NULL;
enum pc_di_index_size idx_type = INDEX_SIZE_IGN;
enum pc_di_src_sel src_sel;
uint32_t idx_size, idx_offset;
if (info->indexed) {
assert(!idx->user_buffer);
idx_bo = fd_resource(idx->buffer)->bo;
idx_type = size2indextype(idx->index_size);
idx_size = idx->index_size * info->count;
idx_offset = idx->offset;
src_sel = DI_SRC_SEL_DMA;
} else {
idx_bo = NULL;
idx_type = INDEX_SIZE_IGN;
idx_size = 0;
idx_offset = 0;
src_sel = DI_SRC_SEL_AUTO_INDEX;
}
OUT_PKT3(ring, CP_DRAW_INDX, info->indexed ? 5 : 3);
OUT_RING(ring, 0x00000000); /* viz query info. */
OUT_RING(ring, DRAW(mode2primtype(info->mode),
src_sel, idx_type, IGNORE_VISIBILITY));
OUT_RING(ring, info->count); /* NumIndices */
if (info->indexed) {
OUT_RELOC(ring, idx_bo, idx_offset, 0);
OUT_RING (ring, idx_size);
}
}
void
fd_state_emit(struct pipe_context *pctx, uint32_t dirty)
{
struct fd_context *ctx = fd_context(pctx);
struct fd_ringbuffer *ring = ctx->ring;
/* NOTE: we probably want to eventually refactor this so each state
* object handles emitting it's own state.. although the mapping of
* state to registers is not always orthogonal, sometimes a single
* register contains bitfields coming from multiple state objects,
* so not sure the best way to deal with that yet.
*/
if (dirty & FD_DIRTY_SAMPLE_MASK) {
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_AA_MASK));
OUT_RING(ring, ctx->sample_mask);
}
if (dirty & FD_DIRTY_ZSA) {
struct pipe_stencil_ref *sr = &ctx->stencil_ref;
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_RB_DEPTHCONTROL));
OUT_RING(ring, ctx->zsa->rb_depthcontrol);
OUT_PKT3(ring, CP_SET_CONSTANT, 4);
OUT_RING(ring, CP_REG(REG_A2XX_RB_STENCILREFMASK_BF));
OUT_RING(ring, ctx->zsa->rb_stencilrefmask_bf |
A2XX_RB_STENCILREFMASK_STENCILREF(sr->ref_value[1]));
OUT_RING(ring, ctx->zsa->rb_stencilrefmask |
A2XX_RB_STENCILREFMASK_STENCILREF(sr->ref_value[0]));
OUT_RING(ring, ctx->zsa->rb_alpha_ref);
}
if (dirty & (FD_DIRTY_RASTERIZER | FD_DIRTY_FRAMEBUFFER)) {
OUT_PKT3(ring, CP_SET_CONSTANT, 3);
OUT_RING(ring, CP_REG(REG_A2XX_PA_CL_CLIP_CNTL));
OUT_RING(ring, ctx->rasterizer->pa_cl_clip_cntl);
OUT_RING(ring, ctx->rasterizer->pa_su_sc_mode_cntl |
A2XX_PA_SU_SC_MODE_CNTL_VTX_WINDOW_OFFSET_ENABLE);
OUT_PKT3(ring, CP_SET_CONSTANT, 5);
OUT_RING(ring, CP_REG(REG_A2XX_PA_SU_POINT_SIZE));
OUT_RING(ring, ctx->rasterizer->pa_su_point_size);
OUT_RING(ring, ctx->rasterizer->pa_su_point_minmax);
OUT_RING(ring, ctx->rasterizer->pa_su_line_cntl);
OUT_RING(ring, ctx->rasterizer->pa_sc_line_stipple);
OUT_PKT3(ring, CP_SET_CONSTANT, 6);
OUT_RING(ring, CP_REG(REG_A2XX_PA_SU_VTX_CNTL));
OUT_RING(ring, ctx->rasterizer->pa_su_vtx_cntl);
OUT_RING(ring, fui(1.0)); /* PA_CL_GB_VERT_CLIP_ADJ */
OUT_RING(ring, fui(1.0)); /* PA_CL_GB_VERT_DISC_ADJ */
OUT_RING(ring, fui(1.0)); /* PA_CL_GB_HORZ_CLIP_ADJ */
OUT_RING(ring, fui(1.0)); /* PA_CL_GB_HORZ_DISC_ADJ */
}
if (dirty & FD_DIRTY_SCISSOR) {
OUT_PKT3(ring, CP_SET_CONSTANT, 3);
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_WINDOW_SCISSOR_TL));
OUT_RING(ring, xy2d(ctx->scissor.minx, /* PA_SC_WINDOW_SCISSOR_TL */
ctx->scissor.miny));
OUT_RING(ring, xy2d(ctx->scissor.maxx, /* PA_SC_WINDOW_SCISSOR_BR */
ctx->scissor.maxy));
ctx->max_scissor.minx = MIN2(ctx->max_scissor.minx, ctx->scissor.minx);
ctx->max_scissor.miny = MIN2(ctx->max_scissor.miny, ctx->scissor.miny);
ctx->max_scissor.maxx = MAX2(ctx->max_scissor.maxx, ctx->scissor.maxx);
ctx->max_scissor.maxy = MAX2(ctx->max_scissor.maxy, ctx->scissor.maxy);
}
if (dirty & FD_DIRTY_VIEWPORT) {
OUT_PKT3(ring, CP_SET_CONSTANT, 7);
OUT_RING(ring, CP_REG(REG_A2XX_PA_CL_VPORT_XSCALE));
OUT_RING(ring, fui(ctx->viewport.scale[0])); /* PA_CL_VPORT_XSCALE */
OUT_RING(ring, fui(ctx->viewport.translate[0])); /* PA_CL_VPORT_XOFFSET */
OUT_RING(ring, fui(ctx->viewport.scale[1])); /* PA_CL_VPORT_YSCALE */
OUT_RING(ring, fui(ctx->viewport.translate[1])); /* PA_CL_VPORT_YOFFSET */
OUT_RING(ring, fui(ctx->viewport.scale[2])); /* PA_CL_VPORT_ZSCALE */
OUT_RING(ring, fui(ctx->viewport.translate[2])); /* PA_CL_VPORT_ZOFFSET */
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_PA_CL_VTE_CNTL));
OUT_RING(ring, A2XX_PA_CL_VTE_CNTL_VTX_W0_FMT |
A2XX_PA_CL_VTE_CNTL_VPORT_X_SCALE_ENA |
A2XX_PA_CL_VTE_CNTL_VPORT_X_OFFSET_ENA |
A2XX_PA_CL_VTE_CNTL_VPORT_Y_SCALE_ENA |
A2XX_PA_CL_VTE_CNTL_VPORT_Y_OFFSET_ENA |
A2XX_PA_CL_VTE_CNTL_VPORT_Z_SCALE_ENA |
A2XX_PA_CL_VTE_CNTL_VPORT_Z_OFFSET_ENA);
}
if (dirty & (FD_DIRTY_PROG | FD_DIRTY_VTXSTATE | FD_DIRTY_TEXSTATE)) {
fd_program_validate(ctx);
fd_program_emit(ring, &ctx->prog);
}
if (dirty & (FD_DIRTY_PROG | FD_DIRTY_CONSTBUF)) {
emit_constants(ring, VS_CONST_BASE * 4,
&ctx->constbuf[PIPE_SHADER_VERTEX],
(dirty & FD_DIRTY_PROG) ? ctx->prog.vp : NULL);
emit_constants(ring, PS_CONST_BASE * 4,
&ctx->constbuf[PIPE_SHADER_FRAGMENT],
(dirty & FD_DIRTY_PROG) ? ctx->prog.fp : NULL);
}
if (dirty & (FD_DIRTY_BLEND | FD_DIRTY_ZSA)) {
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_RB_COLORCONTROL));
OUT_RING(ring, ctx->zsa->rb_colorcontrol | ctx->blend->rb_colorcontrol);
}
if (dirty & FD_DIRTY_BLEND) {
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_RB_BLEND_CONTROL));
OUT_RING(ring, ctx->blend->rb_blendcontrol);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_RB_COLOR_MASK));
OUT_RING(ring, ctx->blend->rb_colormask);
}
if (dirty & (FD_DIRTY_VERTTEX | FD_DIRTY_FRAGTEX | FD_DIRTY_PROG))
emit_textures(ring, ctx);
ctx->dirty &= ~dirty;
}
static void
emit_constants(struct fd_ringbuffer *ring, uint32_t base,
struct fd_constbuf_stateobj *constbuf,
struct fd_shader_stateobj *shader)
{
uint32_t enabled_mask = constbuf->enabled_mask;
uint32_t start_base = base;
unsigned i;
// XXX TODO only emit dirty consts.. but we need to keep track if
// they are clobbered by a clear, gmem2mem, or mem2gmem..
constbuf->dirty_mask = enabled_mask;
/* emit user constants: */
while (enabled_mask) {
unsigned index = ffs(enabled_mask) - 1;
struct pipe_constant_buffer *cb = &constbuf->cb[index];
unsigned size = align(cb->buffer_size, 4) / 4; /* size in dwords */
// I expect that size should be a multiple of vec4's:
assert(size == align(size, 4));
/* hmm, sometimes we still seem to end up with consts bound,
* even if shader isn't using them, which ends up overwriting
* const reg's used for immediates.. this is a hack to work
* around that:
*/
if (shader && ((base - start_base) >= (shader->first_immediate * 4)))
break;
if (constbuf->dirty_mask & (1 << index)) {
const uint32_t *dwords;
if (cb->user_buffer) {
dwords = cb->user_buffer;
} else {
struct fd_resource *rsc = fd_resource(cb->buffer);
dwords = fd_bo_map(rsc->bo);
}
dwords = (uint32_t *)(((uint8_t *)dwords) + cb->buffer_offset);
OUT_PKT3(ring, CP_SET_CONSTANT, size + 1);
OUT_RING(ring, base);
for (i = 0; i < size; i++)
OUT_RING(ring, *(dwords++));
constbuf->dirty_mask &= ~(1 << index);
}
base += size;
enabled_mask &= ~(1 << index);
}
/* emit shader immediates: */
if (shader) {
for (i = 0; i < shader->num_immediates; i++) {
OUT_PKT3(ring, CP_SET_CONSTANT, 5);
OUT_RING(ring, start_base + (4 * (shader->first_immediate + i)));
OUT_RING(ring, shader->immediates[i].val[0]);
OUT_RING(ring, shader->immediates[i].val[1]);
OUT_RING(ring, shader->immediates[i].val[2]);
OUT_RING(ring, shader->immediates[i].val[3]);
base += 4;
}
}
}
static void
emit_constants(struct fd_ringbuffer *ring,
enum adreno_state_block sb,
struct fd_constbuf_stateobj *constbuf,
struct ir3_shader_variant *shader,
bool emit_immediates)
{
uint32_t enabled_mask = constbuf->enabled_mask;
uint32_t max_const;
int i;
// XXX TODO only emit dirty consts.. but we need to keep track if
// they are clobbered by a clear, gmem2mem, or mem2gmem..
constbuf->dirty_mask = enabled_mask;
/* in particular, with binning shader we may end up with unused
* consts, ie. we could end up w/ constlen that is smaller
* than first_immediate. In that case truncate the user consts
* early to avoid HLSQ lockup caused by writing too many consts
*/
max_const = MIN2(shader->first_driver_param, shader->constlen);
/* emit user constants: */
if (enabled_mask & 1) {
const unsigned index = 0;
struct pipe_constant_buffer *cb = &constbuf->cb[index];
unsigned size = align(cb->buffer_size, 4) / 4; /* size in dwords */
// I expect that size should be a multiple of vec4's:
assert(size == align(size, 4));
/* and even if the start of the const buffer is before
* first_immediate, the end may not be:
*/
size = MIN2(size, 4 * max_const);
if (size && (constbuf->dirty_mask & (1 << index))) {
fd4_emit_constant(ring, sb, 0,
cb->buffer_offset, size,
cb->user_buffer, cb->buffer);
constbuf->dirty_mask &= ~(1 << index);
}
enabled_mask &= ~(1 << index);
}
/* emit ubos: */
if (shader->constlen > shader->first_driver_param) {
uint32_t params = MIN2(4, shader->constlen - shader->first_driver_param);
OUT_PKT3(ring, CP_LOAD_STATE, 2 + params * 4);
OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(shader->first_driver_param) |
CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
CP_LOAD_STATE_0_STATE_BLOCK(sb) |
CP_LOAD_STATE_0_NUM_UNIT(params));
OUT_RING(ring, CP_LOAD_STATE_1_EXT_SRC_ADDR(0) |
CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS));
for (i = 1; i <= params * 4; i++) {
struct pipe_constant_buffer *cb = &constbuf->cb[i];
assert(!cb->user_buffer);
if ((enabled_mask & (1 << i)) && cb->buffer)
OUT_RELOC(ring, fd_resource(cb->buffer)->bo, cb->buffer_offset, 0, 0);
else
OUT_RING(ring, 0xbad00000 | ((i - 1) << 16));
}
}
/* emit shader immediates: */
if (shader && emit_immediates) {
int size = shader->immediates_count;
uint32_t base = shader->first_immediate;
/* truncate size to avoid writing constants that shader
* does not use:
*/
size = MIN2(size + base, shader->constlen) - base;
/* convert out of vec4: */
base *= 4;
size *= 4;
if (size > 0) {
fd4_emit_constant(ring, sb, base,
0, size, shader->immediates[0].val, NULL);
}
}
}
/* emit setup at begin of new cmdstream buffer (don't rely on previous
* state, there could have been a context switch between ioctls):
*/
void
fd4_emit_restore(struct fd_context *ctx)
{
struct fd4_context *fd4_ctx = fd4_context(ctx);
struct fd_ringbuffer *ring = ctx->ring;
OUT_PKT0(ring, REG_A4XX_RBBM_PERFCTR_CTL, 1);
OUT_RING(ring, 0x00000001);
OUT_PKT0(ring, REG_A4XX_GRAS_DEBUG_ECO_CONTROL, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_0EC3, 1);
OUT_RING(ring, 0x00000006);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_0F03, 1);
OUT_RING(ring, 0x0000003a);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_0D01, 1);
OUT_RING(ring, 0x00000001);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_0E42, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A4XX_UCHE_CACHE_WAYS_VFD, 1);
OUT_RING(ring, 0x00000007);
OUT_PKT0(ring, REG_A4XX_UCHE_CACHE_MODE_CONTROL, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A4XX_UCHE_INVALIDATE0, 2);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000012);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_0E05, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_0CC5, 1);
OUT_RING(ring, 0x00000006);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_0CC6, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_0EC2, 1);
OUT_RING(ring, 0x00040000);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_2001, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT3(ring, CP_INVALIDATE_STATE, 1);
OUT_RING(ring, 0x00001000);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_20EF, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_20F0, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_20F1, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_20F2, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A4XX_RB_BLEND_RED, 4);
OUT_RING(ring, A4XX_RB_BLEND_RED_UINT(0) |
A4XX_RB_BLEND_RED_FLOAT(0.0));
OUT_RING(ring, A4XX_RB_BLEND_GREEN_UINT(0) |
A4XX_RB_BLEND_GREEN_FLOAT(0.0));
OUT_RING(ring, A4XX_RB_BLEND_BLUE_UINT(0) |
A4XX_RB_BLEND_BLUE_FLOAT(0.0));
OUT_RING(ring, A4XX_RB_BLEND_ALPHA_UINT(0x7fff) |
A4XX_RB_BLEND_ALPHA_FLOAT(1.0));
OUT_PKT0(ring, REG_A4XX_UNKNOWN_20F7, 1);
OUT_RING(ring, 0x3f800000);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_2152, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_2153, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_2154, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_2155, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_2156, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_2157, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_21C3, 1);
OUT_RING(ring, 0x0000001d);
OUT_PKT0(ring, REG_A4XX_PC_GS_PARAM, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_21E6, 1);
OUT_RING(ring, 0x00000001);
OUT_PKT0(ring, REG_A4XX_PC_HS_PARAM, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A4XX_UNKNOWN_22D7, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A4XX_TPL1_TP_TEX_OFFSET, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A4XX_TPL1_TP_TEX_COUNT, 1);
OUT_RING(ring, A4XX_TPL1_TP_TEX_COUNT_VS(16) |
A4XX_TPL1_TP_TEX_COUNT_HS(0) |
A4XX_TPL1_TP_TEX_COUNT_DS(0) |
A4XX_TPL1_TP_TEX_COUNT_GS(0));
OUT_PKT0(ring, REG_A4XX_TPL1_TP_FS_TEX_COUNT, 1);
OUT_RING(ring, 16);
/* we don't use this yet.. probably best to disable.. */
OUT_PKT3(ring, CP_SET_DRAW_STATE, 2);
OUT_RING(ring, CP_SET_DRAW_STATE_0_COUNT(0) |
CP_SET_DRAW_STATE_0_DISABLE_ALL_GROUPS |
CP_SET_DRAW_STATE_0_GROUP_ID(0));
OUT_RING(ring, CP_SET_DRAW_STATE_1_ADDR(0));
OUT_PKT0(ring, REG_A4XX_SP_VS_PVT_MEM_PARAM, 2);
OUT_RING(ring, 0x08000001); /* SP_VS_PVT_MEM_PARAM */
OUT_RELOC(ring, fd4_ctx->vs_pvt_mem, 0,0,0); /* SP_VS_PVT_MEM_ADDR */
OUT_PKT0(ring, REG_A4XX_SP_FS_PVT_MEM_PARAM, 2);
OUT_RING(ring, 0x08000001); /* SP_FS_PVT_MEM_PARAM */
OUT_RELOC(ring, fd4_ctx->fs_pvt_mem, 0,0,0); /* SP_FS_PVT_MEM_ADDR */
OUT_PKT0(ring, REG_A4XX_GRAS_SC_CONTROL, 1);
OUT_RING(ring, A4XX_GRAS_SC_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
A4XX_GRAS_SC_CONTROL_MSAA_DISABLE |
A4XX_GRAS_SC_CONTROL_MSAA_SAMPLES(MSAA_ONE) |
A4XX_GRAS_SC_CONTROL_RASTER_MODE(0));
OUT_PKT0(ring, REG_A4XX_RB_MSAA_CONTROL, 1);
OUT_RING(ring, A4XX_RB_MSAA_CONTROL_DISABLE |
A4XX_RB_MSAA_CONTROL_SAMPLES(MSAA_ONE));
OUT_PKT0(ring, REG_A4XX_GRAS_CL_GB_CLIP_ADJ, 1);
OUT_RING(ring, A4XX_GRAS_CL_GB_CLIP_ADJ_HORZ(0) |
A4XX_GRAS_CL_GB_CLIP_ADJ_VERT(0));
OUT_PKT0(ring, REG_A4XX_RB_ALPHA_CONTROL, 1);
OUT_RING(ring, A4XX_RB_ALPHA_CONTROL_ALPHA_TEST_FUNC(FUNC_ALWAYS));
OUT_PKT0(ring, REG_A4XX_RB_FS_OUTPUT, 1);
OUT_RING(ring, A4XX_RB_FS_OUTPUT_SAMPLE_MASK(0xffff));
OUT_PKT0(ring, REG_A4XX_RB_RENDER_COMPONENTS, 1);
OUT_RING(ring, A4XX_RB_RENDER_COMPONENTS_RT0(0xf));
OUT_PKT0(ring, REG_A4XX_GRAS_CLEAR_CNTL, 1);
OUT_RING(ring, A4XX_GRAS_CLEAR_CNTL_NOT_FASTCLEAR);
OUT_PKT0(ring, REG_A4XX_GRAS_ALPHA_CONTROL, 1);
OUT_RING(ring, 0x0);
ctx->needs_rb_fbd = true;
}
static void
emit_textures(struct fd_context *ctx, struct fd_ringbuffer *ring,
enum adreno_state_block sb, struct fd_texture_stateobj *tex)
{
unsigned i;
if (tex->num_samplers > 0) {
int num_samplers;
/* not sure if this is an a420.0 workaround, but we seem
* to need to emit these in pairs.. emit a final dummy
* entry if odd # of samplers:
*/
num_samplers = align(tex->num_samplers, 2);
/* output sampler state: */
OUT_PKT3(ring, CP_LOAD_STATE, 2 + (2 * num_samplers));
OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(0) |
CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
CP_LOAD_STATE_0_STATE_BLOCK(sb) |
CP_LOAD_STATE_0_NUM_UNIT(num_samplers));
OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_SHADER) |
CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
for (i = 0; i < tex->num_samplers; i++) {
static const struct fd4_sampler_stateobj dummy_sampler = {};
const struct fd4_sampler_stateobj *sampler = tex->samplers[i] ?
fd4_sampler_stateobj(tex->samplers[i]) :
&dummy_sampler;
OUT_RING(ring, sampler->texsamp0);
OUT_RING(ring, sampler->texsamp1);
}
for (; i < num_samplers; i++) {
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000000);
}
}
if (tex->num_textures > 0) {
/* emit texture state: */
OUT_PKT3(ring, CP_LOAD_STATE, 2 + (8 * tex->num_textures));
OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(0) |
CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
CP_LOAD_STATE_0_STATE_BLOCK(sb) |
CP_LOAD_STATE_0_NUM_UNIT(tex->num_textures));
OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS) |
CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
for (i = 0; i < tex->num_textures; i++) {
static const struct fd4_pipe_sampler_view dummy_view = {};
const struct fd4_pipe_sampler_view *view = tex->textures[i] ?
fd4_pipe_sampler_view(tex->textures[i]) :
&dummy_view;
unsigned start = view->base.u.tex.first_level;
OUT_RING(ring, view->texconst0);
OUT_RING(ring, view->texconst1);
OUT_RING(ring, view->texconst2);
OUT_RING(ring, view->texconst3);
if (view->base.texture) {
struct fd_resource *rsc = fd_resource(view->base.texture);
uint32_t offset = fd_resource_offset(rsc, start, 0);
OUT_RELOC(ring, rsc->bo, offset, view->textconst4, 0);
} else {
OUT_RING(ring, 0x00000000);
}
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000000);
}
}
}
/* emit per-context initialization:
*/
void
fd_state_emit_setup(struct pipe_context *pctx)
{
struct fd_context *ctx = fd_context(pctx);
struct fd_ringbuffer *ring = ctx->ring;
OUT_PKT0(ring, REG_A2XX_TP0_CHICKEN, 1);
OUT_RING(ring, 0x00000002);
OUT_PKT3(ring, CP_INVALIDATE_STATE, 1);
OUT_RING(ring, 0x00007fff);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_SQ_VS_CONST));
OUT_RING(ring, A2XX_SQ_VS_CONST_BASE(VS_CONST_BASE) |
A2XX_SQ_VS_CONST_SIZE(0x100));
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_SQ_PS_CONST));
OUT_RING(ring, A2XX_SQ_PS_CONST_BASE(PS_CONST_BASE) |
A2XX_SQ_PS_CONST_SIZE(0xe0));
OUT_PKT3(ring, CP_SET_CONSTANT, 3);
OUT_RING(ring, CP_REG(REG_A2XX_VGT_MAX_VTX_INDX));
OUT_RING(ring, 0xffffffff); /* VGT_MAX_VTX_INDX */
OUT_RING(ring, 0x00000000); /* VGT_MIN_VTX_INDX */
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_VGT_INDX_OFFSET));
OUT_RING(ring, 0x00000000);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_VGT_VERTEX_REUSE_BLOCK_CNTL));
OUT_RING(ring, 0x0000003b);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_SQ_CONTEXT_MISC));
OUT_RING(ring, A2XX_SQ_CONTEXT_MISC_SC_SAMPLE_CNTL(CENTERS_ONLY));
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_SQ_INTERPOLATOR_CNTL));
OUT_RING(ring, 0xffffffff);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_AA_CONFIG));
OUT_RING(ring, 0x00000000);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_LINE_CNTL));
OUT_RING(ring, 0x00000000);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_WINDOW_OFFSET));
OUT_RING(ring, 0x00000000);
// XXX we change this dynamically for draw/clear.. vs gmem<->mem..
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_RB_MODECONTROL));
OUT_RING(ring, A2XX_RB_MODECONTROL_EDRAM_MODE(COLOR_DEPTH));
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_RB_SAMPLE_POS));
OUT_RING(ring, 0x88888888);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_RB_COLOR_DEST_MASK));
OUT_RING(ring, 0xffffffff);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_RB_COPY_DEST_INFO));
OUT_RING(ring, A2XX_RB_COPY_DEST_INFO_FORMAT(COLORX_4_4_4_4) |
A2XX_RB_COPY_DEST_INFO_WRITE_RED |
A2XX_RB_COPY_DEST_INFO_WRITE_GREEN |
A2XX_RB_COPY_DEST_INFO_WRITE_BLUE |
A2XX_RB_COPY_DEST_INFO_WRITE_ALPHA);
OUT_PKT3(ring, CP_SET_CONSTANT, 3);
OUT_RING(ring, CP_REG(REG_A2XX_SQ_WRAPPING_0));
OUT_RING(ring, 0x00000000); /* SQ_WRAPPING_0 */
OUT_RING(ring, 0x00000000); /* SQ_WRAPPING_1 */
OUT_PKT3(ring, CP_SET_DRAW_INIT_FLAGS, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT3(ring, CP_WAIT_REG_EQ, 4);
OUT_RING(ring, 0x000005d0);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x5f601000);
OUT_RING(ring, 0x00000001);
OUT_PKT0(ring, REG_A2XX_SQ_INST_STORE_MANAGMENT, 1);
OUT_RING(ring, 0x00000180);
OUT_PKT3(ring, CP_INVALIDATE_STATE, 1);
OUT_RING(ring, 0x00000300);
OUT_PKT3(ring, CP_SET_SHADER_BASES, 1);
OUT_RING(ring, 0x80000180);
/* not sure what this form of CP_SET_CONSTANT is.. */
OUT_PKT3(ring, CP_SET_CONSTANT, 13);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x469c4000);
OUT_RING(ring, 0x3f800000);
OUT_RING(ring, 0x3f000000);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x40000000);
OUT_RING(ring, 0x3f400000);
OUT_RING(ring, 0x3ec00000);
OUT_RING(ring, 0x3e800000);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_RB_COLOR_MASK));
OUT_RING(ring, A2XX_RB_COLOR_MASK_WRITE_RED |
A2XX_RB_COLOR_MASK_WRITE_GREEN |
A2XX_RB_COLOR_MASK_WRITE_BLUE |
A2XX_RB_COLOR_MASK_WRITE_ALPHA);
OUT_PKT3(ring, CP_SET_CONSTANT, 5);
OUT_RING(ring, CP_REG(REG_A2XX_RB_BLEND_RED));
OUT_RING(ring, 0x00000000); /* RB_BLEND_RED */
OUT_RING(ring, 0x00000000); /* RB_BLEND_GREEN */
OUT_RING(ring, 0x00000000); /* RB_BLEND_BLUE */
OUT_RING(ring, 0x000000ff); /* RB_BLEND_ALPHA */
fd_ringbuffer_flush(ring);
fd_ringmarker_mark(ctx->draw_start);
}
/* emit per-context initialization:
*/
void
fd2_emit_restore(struct fd_context *ctx, struct fd_ringbuffer *ring)
{
if (is_a20x(ctx->screen)) {
OUT_PKT0(ring, REG_A2XX_RB_BC_CONTROL, 1);
OUT_RING(ring,
A2XX_RB_BC_CONTROL_ACCUM_TIMEOUT_SELECT(3) |
A2XX_RB_BC_CONTROL_DISABLE_LZ_NULL_ZCMD_DROP |
A2XX_RB_BC_CONTROL_ENABLE_CRC_UPDATE |
A2XX_RB_BC_CONTROL_ACCUM_DATA_FIFO_LIMIT(8) |
A2XX_RB_BC_CONTROL_MEM_EXPORT_TIMEOUT_SELECT(3));
/* not sure why this is required */
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_VIZ_QUERY));
OUT_RING(ring, A2XX_PA_SC_VIZ_QUERY_VIZ_QUERY_ID(16));
}
OUT_PKT0(ring, REG_A2XX_TP0_CHICKEN, 1);
OUT_RING(ring, 0x00000002);
OUT_PKT3(ring, CP_INVALIDATE_STATE, 1);
OUT_RING(ring, 0x00007fff);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_SQ_VS_CONST));
OUT_RING(ring, A2XX_SQ_VS_CONST_BASE(VS_CONST_BASE) |
A2XX_SQ_VS_CONST_SIZE(0x100));
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_SQ_PS_CONST));
OUT_RING(ring, A2XX_SQ_PS_CONST_BASE(PS_CONST_BASE) |
A2XX_SQ_PS_CONST_SIZE(0xe0));
OUT_PKT3(ring, CP_SET_CONSTANT, 3);
OUT_RING(ring, CP_REG(REG_A2XX_VGT_MAX_VTX_INDX));
OUT_RING(ring, 0xffffffff); /* VGT_MAX_VTX_INDX */
OUT_RING(ring, 0x00000000); /* VGT_MIN_VTX_INDX */
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_VGT_INDX_OFFSET));
OUT_RING(ring, 0x00000000);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_VGT_VERTEX_REUSE_BLOCK_CNTL));
OUT_RING(ring, 0x0000003b);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_SQ_CONTEXT_MISC));
OUT_RING(ring, A2XX_SQ_CONTEXT_MISC_SC_SAMPLE_CNTL(CENTERS_ONLY));
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_SQ_INTERPOLATOR_CNTL));
OUT_RING(ring, 0xffffffff);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_AA_CONFIG));
OUT_RING(ring, 0x00000000);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_LINE_CNTL));
OUT_RING(ring, 0x00000000);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_WINDOW_OFFSET));
OUT_RING(ring, 0x00000000);
// XXX we change this dynamically for draw/clear.. vs gmem<->mem..
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_RB_MODECONTROL));
OUT_RING(ring, A2XX_RB_MODECONTROL_EDRAM_MODE(COLOR_DEPTH));
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_RB_SAMPLE_POS));
OUT_RING(ring, 0x88888888);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_RB_COLOR_DEST_MASK));
OUT_RING(ring, 0xffffffff);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_RB_COPY_DEST_INFO));
OUT_RING(ring, A2XX_RB_COPY_DEST_INFO_FORMAT(COLORX_4_4_4_4) |
A2XX_RB_COPY_DEST_INFO_WRITE_RED |
A2XX_RB_COPY_DEST_INFO_WRITE_GREEN |
A2XX_RB_COPY_DEST_INFO_WRITE_BLUE |
A2XX_RB_COPY_DEST_INFO_WRITE_ALPHA);
OUT_PKT3(ring, CP_SET_CONSTANT, 3);
OUT_RING(ring, CP_REG(REG_A2XX_SQ_WRAPPING_0));
OUT_RING(ring, 0x00000000); /* SQ_WRAPPING_0 */
OUT_RING(ring, 0x00000000); /* SQ_WRAPPING_1 */
OUT_PKT3(ring, CP_SET_DRAW_INIT_FLAGS, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT3(ring, CP_WAIT_REG_EQ, 4);
OUT_RING(ring, 0x000005d0);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x5f601000);
OUT_RING(ring, 0x00000001);
OUT_PKT0(ring, REG_A2XX_SQ_INST_STORE_MANAGMENT, 1);
OUT_RING(ring, 0x00000180);
OUT_PKT3(ring, CP_INVALIDATE_STATE, 1);
OUT_RING(ring, 0x00000300);
OUT_PKT3(ring, CP_SET_SHADER_BASES, 1);
OUT_RING(ring, 0x80000180);
/* not sure what this form of CP_SET_CONSTANT is.. */
OUT_PKT3(ring, CP_SET_CONSTANT, 13);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x469c4000);
OUT_RING(ring, 0x3f800000);
OUT_RING(ring, 0x3f000000);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x40000000);
OUT_RING(ring, 0x3f400000);
OUT_RING(ring, 0x3ec00000);
OUT_RING(ring, 0x3e800000);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_RB_COLOR_MASK));
OUT_RING(ring, A2XX_RB_COLOR_MASK_WRITE_RED |
A2XX_RB_COLOR_MASK_WRITE_GREEN |
A2XX_RB_COLOR_MASK_WRITE_BLUE |
A2XX_RB_COLOR_MASK_WRITE_ALPHA);
OUT_PKT3(ring, CP_SET_CONSTANT, 5);
OUT_RING(ring, CP_REG(REG_A2XX_RB_BLEND_RED));
OUT_RING(ring, 0x00000000); /* RB_BLEND_RED */
OUT_RING(ring, 0x00000000); /* RB_BLEND_GREEN */
OUT_RING(ring, 0x00000000); /* RB_BLEND_BLUE */
OUT_RING(ring, 0x000000ff); /* RB_BLEND_ALPHA */
}
void
fd2_emit_state(struct fd_context *ctx, const enum fd_dirty_3d_state dirty)
{
struct fd2_blend_stateobj *blend = fd2_blend_stateobj(ctx->blend);
struct fd2_zsa_stateobj *zsa = fd2_zsa_stateobj(ctx->zsa);
struct fd_ringbuffer *ring = ctx->batch->draw;
/* NOTE: we probably want to eventually refactor this so each state
* object handles emitting it's own state.. although the mapping of
* state to registers is not always orthogonal, sometimes a single
* register contains bitfields coming from multiple state objects,
* so not sure the best way to deal with that yet.
*/
if (dirty & FD_DIRTY_SAMPLE_MASK) {
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_AA_MASK));
OUT_RING(ring, ctx->sample_mask);
}
if (dirty & (FD_DIRTY_ZSA | FD_DIRTY_STENCIL_REF)) {
struct pipe_stencil_ref *sr = &ctx->stencil_ref;
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_RB_DEPTHCONTROL));
OUT_RING(ring, zsa->rb_depthcontrol);
OUT_PKT3(ring, CP_SET_CONSTANT, 4);
OUT_RING(ring, CP_REG(REG_A2XX_RB_STENCILREFMASK_BF));
OUT_RING(ring, zsa->rb_stencilrefmask_bf |
A2XX_RB_STENCILREFMASK_STENCILREF(sr->ref_value[1]));
OUT_RING(ring, zsa->rb_stencilrefmask |
A2XX_RB_STENCILREFMASK_STENCILREF(sr->ref_value[0]));
OUT_RING(ring, zsa->rb_alpha_ref);
}
if (ctx->rasterizer && dirty & FD_DIRTY_RASTERIZER) {
struct fd2_rasterizer_stateobj *rasterizer =
fd2_rasterizer_stateobj(ctx->rasterizer);
OUT_PKT3(ring, CP_SET_CONSTANT, 3);
OUT_RING(ring, CP_REG(REG_A2XX_PA_CL_CLIP_CNTL));
OUT_RING(ring, rasterizer->pa_cl_clip_cntl);
OUT_RING(ring, rasterizer->pa_su_sc_mode_cntl |
A2XX_PA_SU_SC_MODE_CNTL_VTX_WINDOW_OFFSET_ENABLE);
OUT_PKT3(ring, CP_SET_CONSTANT, 5);
OUT_RING(ring, CP_REG(REG_A2XX_PA_SU_POINT_SIZE));
OUT_RING(ring, rasterizer->pa_su_point_size);
OUT_RING(ring, rasterizer->pa_su_point_minmax);
OUT_RING(ring, rasterizer->pa_su_line_cntl);
OUT_RING(ring, rasterizer->pa_sc_line_stipple);
OUT_PKT3(ring, CP_SET_CONSTANT, 6);
OUT_RING(ring, CP_REG(REG_A2XX_PA_SU_VTX_CNTL));
OUT_RING(ring, rasterizer->pa_su_vtx_cntl);
OUT_RING(ring, fui(1.0)); /* PA_CL_GB_VERT_CLIP_ADJ */
OUT_RING(ring, fui(1.0)); /* PA_CL_GB_VERT_DISC_ADJ */
OUT_RING(ring, fui(1.0)); /* PA_CL_GB_HORZ_CLIP_ADJ */
OUT_RING(ring, fui(1.0)); /* PA_CL_GB_HORZ_DISC_ADJ */
}
/* NOTE: scissor enabled bit is part of rasterizer state: */
if (dirty & (FD_DIRTY_SCISSOR | FD_DIRTY_RASTERIZER)) {
struct pipe_scissor_state *scissor = fd_context_get_scissor(ctx);
OUT_PKT3(ring, CP_SET_CONSTANT, 3);
OUT_RING(ring, CP_REG(REG_A2XX_PA_SC_WINDOW_SCISSOR_TL));
OUT_RING(ring, xy2d(scissor->minx, /* PA_SC_WINDOW_SCISSOR_TL */
scissor->miny));
OUT_RING(ring, xy2d(scissor->maxx, /* PA_SC_WINDOW_SCISSOR_BR */
scissor->maxy));
ctx->batch->max_scissor.minx = MIN2(ctx->batch->max_scissor.minx, scissor->minx);
ctx->batch->max_scissor.miny = MIN2(ctx->batch->max_scissor.miny, scissor->miny);
ctx->batch->max_scissor.maxx = MAX2(ctx->batch->max_scissor.maxx, scissor->maxx);
ctx->batch->max_scissor.maxy = MAX2(ctx->batch->max_scissor.maxy, scissor->maxy);
}
if (dirty & FD_DIRTY_VIEWPORT) {
OUT_PKT3(ring, CP_SET_CONSTANT, 7);
OUT_RING(ring, CP_REG(REG_A2XX_PA_CL_VPORT_XSCALE));
OUT_RING(ring, fui(ctx->viewport.scale[0])); /* PA_CL_VPORT_XSCALE */
OUT_RING(ring, fui(ctx->viewport.translate[0])); /* PA_CL_VPORT_XOFFSET */
OUT_RING(ring, fui(ctx->viewport.scale[1])); /* PA_CL_VPORT_YSCALE */
OUT_RING(ring, fui(ctx->viewport.translate[1])); /* PA_CL_VPORT_YOFFSET */
OUT_RING(ring, fui(ctx->viewport.scale[2])); /* PA_CL_VPORT_ZSCALE */
OUT_RING(ring, fui(ctx->viewport.translate[2])); /* PA_CL_VPORT_ZOFFSET */
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_PA_CL_VTE_CNTL));
OUT_RING(ring, A2XX_PA_CL_VTE_CNTL_VTX_W0_FMT |
A2XX_PA_CL_VTE_CNTL_VPORT_X_SCALE_ENA |
A2XX_PA_CL_VTE_CNTL_VPORT_X_OFFSET_ENA |
A2XX_PA_CL_VTE_CNTL_VPORT_Y_SCALE_ENA |
A2XX_PA_CL_VTE_CNTL_VPORT_Y_OFFSET_ENA |
A2XX_PA_CL_VTE_CNTL_VPORT_Z_SCALE_ENA |
A2XX_PA_CL_VTE_CNTL_VPORT_Z_OFFSET_ENA);
}
if (dirty & (FD_DIRTY_PROG | FD_DIRTY_VTXSTATE | FD_DIRTY_TEXSTATE)) {
fd2_program_validate(ctx);
fd2_program_emit(ring, &ctx->prog);
}
if (dirty & (FD_DIRTY_PROG | FD_DIRTY_CONST)) {
emit_constants(ring, VS_CONST_BASE * 4,
&ctx->constbuf[PIPE_SHADER_VERTEX],
(dirty & FD_DIRTY_PROG) ? ctx->prog.vp : NULL);
emit_constants(ring, PS_CONST_BASE * 4,
&ctx->constbuf[PIPE_SHADER_FRAGMENT],
(dirty & FD_DIRTY_PROG) ? ctx->prog.fp : NULL);
}
if (dirty & (FD_DIRTY_BLEND | FD_DIRTY_ZSA)) {
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_RB_COLORCONTROL));
OUT_RING(ring, blend ? zsa->rb_colorcontrol | blend->rb_colorcontrol : 0);
}
if (dirty & (FD_DIRTY_BLEND | FD_DIRTY_FRAMEBUFFER)) {
enum pipe_format format =
pipe_surface_format(ctx->batch->framebuffer.cbufs[0]);
bool has_alpha = util_format_has_alpha(format);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_RB_BLEND_CONTROL));
OUT_RING(ring, blend ? blend->rb_blendcontrol_alpha |
COND(has_alpha, blend->rb_blendcontrol_rgb) |
COND(!has_alpha, blend->rb_blendcontrol_no_alpha_rgb) : 0);
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_RB_COLOR_MASK));
OUT_RING(ring, blend ? blend->rb_colormask : 0xf);
}
if (dirty & FD_DIRTY_BLEND_COLOR) {
OUT_PKT3(ring, CP_SET_CONSTANT, 5);
OUT_RING(ring, CP_REG(REG_A2XX_RB_BLEND_RED));
OUT_RING(ring, float_to_ubyte(ctx->blend_color.color[0]));
OUT_RING(ring, float_to_ubyte(ctx->blend_color.color[1]));
OUT_RING(ring, float_to_ubyte(ctx->blend_color.color[2]));
OUT_RING(ring, float_to_ubyte(ctx->blend_color.color[3]));
}
if (dirty & (FD_DIRTY_TEX | FD_DIRTY_PROG))
emit_textures(ring, ctx);
}
static void
fd4_clear(struct fd_context *ctx, unsigned buffers,
const union pipe_color_union *color, double depth, unsigned stencil)
{
struct fd4_context *fd4_ctx = fd4_context(ctx);
struct fd_ringbuffer *ring = ctx->ring;
struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
unsigned dirty = ctx->dirty;
unsigned ce, i;
struct fd4_emit emit = {
.vtx = &fd4_ctx->solid_vbuf_state,
.prog = &ctx->solid_prog,
.key = {
.half_precision = true,
},
};
uint32_t colr = 0;
if ((buffers & PIPE_CLEAR_COLOR) && pfb->nr_cbufs)
colr = pack_rgba(pfb->cbufs[0]->format, color->f);
dirty &= FD_DIRTY_FRAMEBUFFER | FD_DIRTY_SCISSOR;
dirty |= FD_DIRTY_PROG;
emit.dirty = dirty;
OUT_PKT0(ring, REG_A4XX_PC_PRIM_VTX_CNTL, 1);
OUT_RING(ring, A4XX_PC_PRIM_VTX_CNTL_PROVOKING_VTX_LAST);
/* emit generic state now: */
fd4_emit_state(ctx, ring, &emit);
reset_viewport(ring, pfb);
if (buffers & PIPE_CLEAR_DEPTH) {
OUT_PKT0(ring, REG_A4XX_RB_DEPTH_CONTROL, 1);
OUT_RING(ring, A4XX_RB_DEPTH_CONTROL_Z_WRITE_ENABLE |
A4XX_RB_DEPTH_CONTROL_Z_ENABLE |
A4XX_RB_DEPTH_CONTROL_ZFUNC(FUNC_ALWAYS));
fd_wfi(ctx, ring);
OUT_PKT0(ring, REG_A4XX_GRAS_CL_VPORT_ZOFFSET_0, 2);
OUT_RING(ring, A4XX_GRAS_CL_VPORT_ZOFFSET_0(0.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_ZSCALE_0(depth));
ctx->dirty |= FD_DIRTY_VIEWPORT;
} else {
OUT_PKT0(ring, REG_A4XX_RB_DEPTH_CONTROL, 1);
OUT_RING(ring, A4XX_RB_DEPTH_CONTROL_ZFUNC(FUNC_NEVER));
}
if (buffers & PIPE_CLEAR_STENCIL) {
OUT_PKT0(ring, REG_A4XX_RB_STENCILREFMASK, 2);
OUT_RING(ring, A4XX_RB_STENCILREFMASK_STENCILREF(stencil) |
A4XX_RB_STENCILREFMASK_STENCILMASK(stencil) |
A4XX_RB_STENCILREFMASK_STENCILWRITEMASK(0xff));
OUT_RING(ring, A4XX_RB_STENCILREFMASK_STENCILREF(0) |
A4XX_RB_STENCILREFMASK_STENCILMASK(0) |
0xff000000 | // XXX ???
A4XX_RB_STENCILREFMASK_STENCILWRITEMASK(0xff));
OUT_PKT0(ring, REG_A4XX_RB_STENCIL_CONTROL, 2);
OUT_RING(ring, A4XX_RB_STENCIL_CONTROL_STENCIL_ENABLE |
A4XX_RB_STENCIL_CONTROL_FUNC(FUNC_ALWAYS) |
A4XX_RB_STENCIL_CONTROL_FAIL(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZPASS(STENCIL_REPLACE) |
A4XX_RB_STENCIL_CONTROL_ZFAIL(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_FUNC_BF(FUNC_NEVER) |
A4XX_RB_STENCIL_CONTROL_FAIL_BF(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZPASS_BF(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZFAIL_BF(STENCIL_KEEP));
OUT_RING(ring, 0x00000000); /* RB_STENCIL_CONTROL2 */
} else {
OUT_PKT0(ring, REG_A4XX_RB_STENCILREFMASK, 2);
OUT_RING(ring, A4XX_RB_STENCILREFMASK_STENCILREF(0) |
A4XX_RB_STENCILREFMASK_STENCILMASK(0) |
A4XX_RB_STENCILREFMASK_STENCILWRITEMASK(0));
OUT_RING(ring, A4XX_RB_STENCILREFMASK_BF_STENCILREF(0) |
A4XX_RB_STENCILREFMASK_BF_STENCILMASK(0) |
A4XX_RB_STENCILREFMASK_BF_STENCILWRITEMASK(0));
OUT_PKT0(ring, REG_A4XX_RB_STENCIL_CONTROL, 2);
OUT_RING(ring, A4XX_RB_STENCIL_CONTROL_FUNC(FUNC_NEVER) |
A4XX_RB_STENCIL_CONTROL_FAIL(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZPASS(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZFAIL(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_FUNC_BF(FUNC_NEVER) |
A4XX_RB_STENCIL_CONTROL_FAIL_BF(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZPASS_BF(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZFAIL_BF(STENCIL_KEEP));
OUT_RING(ring, 0x00000000); /* RB_STENCIL_CONTROL2 */
}
if (buffers & PIPE_CLEAR_COLOR) {
OUT_PKT0(ring, REG_A4XX_RB_ALPHA_CONTROL, 1);
OUT_RING(ring, A4XX_RB_ALPHA_CONTROL_ALPHA_TEST_FUNC(FUNC_NEVER));
ce = 0xf;
} else {
ce = 0x0;
}
for (i = 0; i < 8; i++) {
OUT_PKT0(ring, REG_A4XX_RB_MRT_CONTROL(i), 1);
OUT_RING(ring, A4XX_RB_MRT_CONTROL_FASTCLEAR |
A4XX_RB_MRT_CONTROL_B11 |
A4XX_RB_MRT_CONTROL_COMPONENT_ENABLE(ce));
OUT_PKT0(ring, REG_A4XX_RB_MRT_BLEND_CONTROL(i), 1);
OUT_RING(ring, A4XX_RB_MRT_BLEND_CONTROL_RGB_SRC_FACTOR(FACTOR_ONE) |
A4XX_RB_MRT_BLEND_CONTROL_RGB_BLEND_OPCODE(BLEND_DST_PLUS_SRC) |
A4XX_RB_MRT_BLEND_CONTROL_RGB_DEST_FACTOR(FACTOR_ZERO) |
A4XX_RB_MRT_BLEND_CONTROL_ALPHA_SRC_FACTOR(FACTOR_ONE) |
A4XX_RB_MRT_BLEND_CONTROL_ALPHA_BLEND_OPCODE(BLEND_DST_PLUS_SRC) |
A4XX_RB_MRT_BLEND_CONTROL_ALPHA_DEST_FACTOR(FACTOR_ZERO));
}
fd4_emit_vertex_bufs(ring, &emit);
OUT_PKT0(ring, REG_A4XX_GRAS_ALPHA_CONTROL, 1);
OUT_RING(ring, 0x0); /* XXX GRAS_ALPHA_CONTROL */
OUT_PKT0(ring, REG_A4XX_GRAS_CLEAR_CNTL, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A4XX_RB_CLEAR_COLOR_DW0, 4);
OUT_RING(ring, colr); /* RB_CLEAR_COLOR_DW0 */
OUT_RING(ring, colr); /* RB_CLEAR_COLOR_DW1 */
OUT_RING(ring, colr); /* RB_CLEAR_COLOR_DW2 */
OUT_RING(ring, colr); /* RB_CLEAR_COLOR_DW3 */
/* until fastclear works: */
fd4_emit_constant(ring, SB_FRAG_SHADER, 0, 0, 4, color->ui, NULL);
OUT_PKT0(ring, REG_A4XX_VFD_INDEX_OFFSET, 2);
OUT_RING(ring, 0); /* VFD_INDEX_OFFSET */
OUT_RING(ring, 0); /* ??? UNKNOWN_2209 */
OUT_PKT0(ring, REG_A4XX_PC_RESTART_INDEX, 1);
OUT_RING(ring, 0xffffffff); /* PC_RESTART_INDEX */
OUT_PKT3(ring, CP_UNKNOWN_1A, 1);
OUT_RING(ring, 0x00000001);
fd4_draw(ctx, ring, DI_PT_RECTLIST, USE_VISIBILITY,
DI_SRC_SEL_AUTO_INDEX, 2, INDEX_SIZE_IGN, 0, 0, NULL);
OUT_PKT3(ring, CP_UNKNOWN_1A, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT0(ring, REG_A4XX_GRAS_CLEAR_CNTL, 1);
OUT_RING(ring, A4XX_GRAS_CLEAR_CNTL_NOT_FASTCLEAR);
OUT_PKT0(ring, REG_A4XX_GRAS_SC_CONTROL, 1);
OUT_RING(ring, A4XX_GRAS_SC_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
A4XX_GRAS_SC_CONTROL_MSAA_DISABLE |
A4XX_GRAS_SC_CONTROL_MSAA_SAMPLES(MSAA_ONE) |
A4XX_GRAS_SC_CONTROL_RASTER_MODE(0));
}
static bool
fd4_draw_vbo(struct fd_context *ctx, const struct pipe_draw_info *info)
{
struct fd4_context *fd4_ctx = fd4_context(ctx);
struct fd4_emit emit = {
.debug = &ctx->debug,
.vtx = &ctx->vtx,
.prog = &ctx->prog,
.info = info,
.key = {
.color_two_side = ctx->rasterizer->light_twoside,
.vclamp_color = ctx->rasterizer->clamp_vertex_color,
.fclamp_color = ctx->rasterizer->clamp_fragment_color,
.rasterflat = ctx->rasterizer->flatshade,
// TODO set .half_precision based on render target format,
// ie. float16 and smaller use half, float32 use full..
.half_precision = !!(fd_mesa_debug & FD_DBG_FRAGHALF),
.ucp_enables = ctx->rasterizer->clip_plane_enable,
.has_per_samp = (fd4_ctx->fsaturate || fd4_ctx->vsaturate ||
fd4_ctx->fastc_srgb || fd4_ctx->vastc_srgb),
.vsaturate_s = fd4_ctx->vsaturate_s,
.vsaturate_t = fd4_ctx->vsaturate_t,
.vsaturate_r = fd4_ctx->vsaturate_r,
.fsaturate_s = fd4_ctx->fsaturate_s,
.fsaturate_t = fd4_ctx->fsaturate_t,
.fsaturate_r = fd4_ctx->fsaturate_r,
.vastc_srgb = fd4_ctx->vastc_srgb,
.fastc_srgb = fd4_ctx->fastc_srgb,
},
.rasterflat = ctx->rasterizer->flatshade,
.sprite_coord_enable = ctx->rasterizer->sprite_coord_enable,
.sprite_coord_mode = ctx->rasterizer->sprite_coord_mode,
};
fixup_shader_state(ctx, &emit.key);
unsigned dirty = ctx->dirty;
/* do regular pass first, since that is more likely to fail compiling: */
if (!(fd4_emit_get_vp(&emit) && fd4_emit_get_fp(&emit)))
return false;
emit.key.binning_pass = false;
emit.dirty = dirty;
struct fd_ringbuffer *ring = ctx->batch->draw;
if (ctx->rasterizer->rasterizer_discard) {
fd_wfi(ctx->batch, ring);
OUT_PKT3(ring, CP_REG_RMW, 3);
OUT_RING(ring, REG_A4XX_RB_RENDER_CONTROL);
OUT_RING(ring, ~A4XX_RB_RENDER_CONTROL_DISABLE_COLOR_PIPE);
OUT_RING(ring, A4XX_RB_RENDER_CONTROL_DISABLE_COLOR_PIPE);
}
draw_impl(ctx, ctx->batch->draw, &emit);
if (ctx->rasterizer->rasterizer_discard) {
fd_wfi(ctx->batch, ring);
OUT_PKT3(ring, CP_REG_RMW, 3);
OUT_RING(ring, REG_A4XX_RB_RENDER_CONTROL);
OUT_RING(ring, ~A4XX_RB_RENDER_CONTROL_DISABLE_COLOR_PIPE);
OUT_RING(ring, 0);
}
/* and now binning pass: */
emit.key.binning_pass = true;
emit.dirty = dirty & ~(FD_DIRTY_BLEND);
emit.vp = NULL; /* we changed key so need to refetch vp */
emit.fp = NULL;
draw_impl(ctx, ctx->batch->binning, &emit);
return true;
}
static void
fd4_emit_tile_mem2gmem(struct fd_context *ctx, struct fd_tile *tile)
{
struct fd4_context *fd4_ctx = fd4_context(ctx);
struct fd_gmem_stateobj *gmem = &ctx->gmem;
struct fd_ringbuffer *ring = ctx->ring;
struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
struct fd4_emit emit = {
.vtx = &fd4_ctx->blit_vbuf_state,
.sprite_coord_enable = 1,
/* NOTE: They all use the same VP, this is for vtx bufs. */
.prog = &ctx->blit_prog[0],
.key = {
.half_precision = fd_half_precision(pfb),
},
.no_decode_srgb = true,
};
unsigned char mrt_comp[A4XX_MAX_RENDER_TARGETS] = {0};
float x0, y0, x1, y1;
unsigned bin_w = tile->bin_w;
unsigned bin_h = tile->bin_h;
unsigned i;
/* write texture coordinates to vertexbuf: */
x0 = ((float)tile->xoff) / ((float)pfb->width);
x1 = ((float)tile->xoff + bin_w) / ((float)pfb->width);
y0 = ((float)tile->yoff) / ((float)pfb->height);
y1 = ((float)tile->yoff + bin_h) / ((float)pfb->height);
OUT_PKT3(ring, CP_MEM_WRITE, 5);
OUT_RELOCW(ring, fd_resource(fd4_ctx->blit_texcoord_vbuf)->bo, 0, 0, 0);
OUT_RING(ring, fui(x0));
OUT_RING(ring, fui(y0));
OUT_RING(ring, fui(x1));
OUT_RING(ring, fui(y1));
for (i = 0; i < A4XX_MAX_RENDER_TARGETS; i++) {
mrt_comp[i] = ((i < pfb->nr_cbufs) && pfb->cbufs[i]) ? 0xf : 0;
OUT_PKT0(ring, REG_A4XX_RB_MRT_CONTROL(i), 1);
OUT_RING(ring, A4XX_RB_MRT_CONTROL_FASTCLEAR |
A4XX_RB_MRT_CONTROL_B11 |
A4XX_RB_MRT_CONTROL_COMPONENT_ENABLE(0xf));
OUT_PKT0(ring, REG_A4XX_RB_MRT_BLEND_CONTROL(i), 1);
OUT_RING(ring, A4XX_RB_MRT_BLEND_CONTROL_RGB_SRC_FACTOR(FACTOR_ONE) |
A4XX_RB_MRT_BLEND_CONTROL_RGB_BLEND_OPCODE(BLEND_DST_PLUS_SRC) |
A4XX_RB_MRT_BLEND_CONTROL_RGB_DEST_FACTOR(FACTOR_ZERO) |
A4XX_RB_MRT_BLEND_CONTROL_ALPHA_SRC_FACTOR(FACTOR_ONE) |
A4XX_RB_MRT_BLEND_CONTROL_ALPHA_BLEND_OPCODE(BLEND_DST_PLUS_SRC) |
A4XX_RB_MRT_BLEND_CONTROL_ALPHA_DEST_FACTOR(FACTOR_ZERO));
}
OUT_PKT0(ring, REG_A4XX_RB_RENDER_COMPONENTS, 1);
OUT_RING(ring, A4XX_RB_RENDER_COMPONENTS_RT0(mrt_comp[0]) |
A4XX_RB_RENDER_COMPONENTS_RT1(mrt_comp[1]) |
A4XX_RB_RENDER_COMPONENTS_RT2(mrt_comp[2]) |
A4XX_RB_RENDER_COMPONENTS_RT3(mrt_comp[3]) |
A4XX_RB_RENDER_COMPONENTS_RT4(mrt_comp[4]) |
A4XX_RB_RENDER_COMPONENTS_RT5(mrt_comp[5]) |
A4XX_RB_RENDER_COMPONENTS_RT6(mrt_comp[6]) |
A4XX_RB_RENDER_COMPONENTS_RT7(mrt_comp[7]));
OUT_PKT0(ring, REG_A4XX_RB_RENDER_CONTROL, 1);
OUT_RING(ring, 0x8); /* XXX RB_RENDER_CONTROL */
OUT_PKT0(ring, REG_A4XX_RB_DEPTH_CONTROL, 1);
OUT_RING(ring, A4XX_RB_DEPTH_CONTROL_ZFUNC(FUNC_LESS));
OUT_PKT0(ring, REG_A4XX_GRAS_CL_CLIP_CNTL, 1);
OUT_RING(ring, 0x280000); /* XXX GRAS_CL_CLIP_CNTL */
OUT_PKT0(ring, REG_A4XX_GRAS_SU_MODE_CONTROL, 1);
OUT_RING(ring, A4XX_GRAS_SU_MODE_CONTROL_LINEHALFWIDTH(0) |
A4XX_GRAS_SU_MODE_CONTROL_RENDERING_PASS);
OUT_PKT0(ring, REG_A4XX_GRAS_CL_VPORT_XOFFSET_0, 6);
OUT_RING(ring, A4XX_GRAS_CL_VPORT_XOFFSET_0((float)bin_w/2.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_XSCALE_0((float)bin_w/2.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_YOFFSET_0((float)bin_h/2.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_YSCALE_0(-(float)bin_h/2.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_ZOFFSET_0(0.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_ZSCALE_0(1.0));
OUT_PKT0(ring, REG_A4XX_GRAS_SC_WINDOW_SCISSOR_BR, 2);
OUT_RING(ring, A4XX_GRAS_SC_WINDOW_SCISSOR_BR_X(bin_w - 1) |
A4XX_GRAS_SC_WINDOW_SCISSOR_BR_Y(bin_h - 1));
OUT_RING(ring, A4XX_GRAS_SC_WINDOW_SCISSOR_TL_X(0) |
A4XX_GRAS_SC_WINDOW_SCISSOR_TL_Y(0));
OUT_PKT0(ring, REG_A4XX_GRAS_SC_SCREEN_SCISSOR_TL, 2);
OUT_RING(ring, A4XX_GRAS_SC_SCREEN_SCISSOR_TL_X(0) |
A4XX_GRAS_SC_SCREEN_SCISSOR_TL_Y(0));
OUT_RING(ring, A4XX_GRAS_SC_SCREEN_SCISSOR_BR_X(bin_w - 1) |
A4XX_GRAS_SC_SCREEN_SCISSOR_BR_Y(bin_h - 1));
OUT_PKT0(ring, REG_A4XX_RB_MODE_CONTROL, 1);
OUT_RING(ring, A4XX_RB_MODE_CONTROL_WIDTH(gmem->bin_w) |
A4XX_RB_MODE_CONTROL_HEIGHT(gmem->bin_h));
OUT_PKT0(ring, REG_A4XX_RB_STENCIL_CONTROL, 2);
OUT_RING(ring, A4XX_RB_STENCIL_CONTROL_FUNC(FUNC_ALWAYS) |
A4XX_RB_STENCIL_CONTROL_FAIL(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZPASS(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZFAIL(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_FUNC_BF(FUNC_ALWAYS) |
A4XX_RB_STENCIL_CONTROL_FAIL_BF(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZPASS_BF(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZFAIL_BF(STENCIL_KEEP));
OUT_RING(ring, 0x00000000); /* RB_STENCIL_CONTROL2 */
OUT_PKT0(ring, REG_A4XX_GRAS_SC_CONTROL, 1);
OUT_RING(ring, A4XX_GRAS_SC_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
A4XX_GRAS_SC_CONTROL_MSAA_DISABLE |
A4XX_GRAS_SC_CONTROL_MSAA_SAMPLES(MSAA_ONE) |
A4XX_GRAS_SC_CONTROL_RASTER_MODE(1));
OUT_PKT0(ring, REG_A4XX_PC_PRIM_VTX_CNTL, 1);
OUT_RING(ring, A4XX_PC_PRIM_VTX_CNTL_PROVOKING_VTX_LAST |
A4XX_PC_PRIM_VTX_CNTL_VAROUT(1));
OUT_PKT0(ring, REG_A4XX_VFD_INDEX_OFFSET, 2);
OUT_RING(ring, 0); /* VFD_INDEX_OFFSET */
OUT_RING(ring, 0); /* ??? UNKNOWN_2209 */
fd4_emit_vertex_bufs(ring, &emit);
/* for gmem pitch/base calculations, we need to use the non-
* truncated tile sizes:
*/
bin_w = gmem->bin_w;
bin_h = gmem->bin_h;
if (fd_gmem_needs_restore(ctx, tile, FD_BUFFER_COLOR)) {
emit.prog = &ctx->blit_prog[pfb->nr_cbufs - 1];
emit.fp = NULL; /* frag shader changed so clear cache */
fd4_program_emit(ring, &emit, pfb->nr_cbufs, pfb->cbufs);
emit_mem2gmem_surf(ctx, gmem->cbuf_base, pfb->cbufs, pfb->nr_cbufs, bin_w);
}
if (fd_gmem_needs_restore(ctx, tile, FD_BUFFER_DEPTH | FD_BUFFER_STENCIL)) {
switch (pfb->zsbuf->format) {
case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
case PIPE_FORMAT_Z32_FLOAT:
emit.prog = (pfb->zsbuf->format == PIPE_FORMAT_Z32_FLOAT) ?
&ctx->blit_z : &ctx->blit_zs;
emit.key.half_precision = false;
OUT_PKT0(ring, REG_A4XX_RB_DEPTH_CONTROL, 1);
OUT_RING(ring, A4XX_RB_DEPTH_CONTROL_Z_ENABLE |
A4XX_RB_DEPTH_CONTROL_Z_WRITE_ENABLE |
A4XX_RB_DEPTH_CONTROL_ZFUNC(FUNC_ALWAYS) |
A4XX_RB_DEPTH_CONTROL_EARLY_Z_DISABLE);
OUT_PKT0(ring, REG_A4XX_GRAS_ALPHA_CONTROL, 1);
OUT_RING(ring, A4XX_GRAS_ALPHA_CONTROL_ALPHA_TEST_ENABLE);
OUT_PKT0(ring, REG_A4XX_GRAS_CL_CLIP_CNTL, 1);
OUT_RING(ring, 0x80000); /* GRAS_CL_CLIP_CNTL */
break;
default:
/* Non-float can use a regular color write. It's split over 8-bit
* components, so half precision is always sufficient.
*/
emit.prog = &ctx->blit_prog[0];
emit.key.half_precision = true;
break;
}
emit.fp = NULL; /* frag shader changed so clear cache */
fd4_program_emit(ring, &emit, 1, &pfb->zsbuf);
emit_mem2gmem_surf(ctx, gmem->zsbuf_base, &pfb->zsbuf, 1, bin_w);
}
OUT_PKT0(ring, REG_A4XX_GRAS_SC_CONTROL, 1);
OUT_RING(ring, A4XX_GRAS_SC_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
A4XX_GRAS_SC_CONTROL_MSAA_SAMPLES(MSAA_ONE) |
A4XX_GRAS_SC_CONTROL_RASTER_MODE(0));
OUT_PKT0(ring, REG_A4XX_RB_MODE_CONTROL, 1);
OUT_RING(ring, A4XX_RB_MODE_CONTROL_WIDTH(gmem->bin_w) |
A4XX_RB_MODE_CONTROL_HEIGHT(gmem->bin_h) |
0x00010000); /* XXX */
}
static void
fd3_draw(struct fd_context *ctx, const struct pipe_draw_info *info)
{
unsigned dirty = ctx->dirty;
struct fd3_shader_key key = {
/* do binning pass first: */
.binning_pass = true,
.color_two_side = ctx->rasterizer ? ctx->rasterizer->light_twoside : false,
// TODO set .half_precision based on render target format,
// ie. float16 and smaller use half, float32 use full..
.half_precision = !!(fd_mesa_debug & FD_DBG_FRAGHALF),
};
draw_impl(ctx, info, ctx->binning_ring,
dirty & ~(FD_DIRTY_BLEND), key);
/* and now regular (non-binning) pass: */
key.binning_pass = false;
draw_impl(ctx, info, ctx->ring, dirty, key);
}
/* binning pass cmds for a clear:
* NOTE: newer blob drivers don't use binning for clear, which is probably
* preferable since it is low vtx count. However that doesn't seem to
* actually work for me. Not sure if it is depending on support for
* clear pass (rather than using solid-fill shader), or something else
* that newer blob is doing differently. Once that is figured out, we
* can remove fd3_clear_binning().
*/
static void
fd3_clear_binning(struct fd_context *ctx, unsigned dirty)
{
struct fd3_context *fd3_ctx = fd3_context(ctx);
struct fd_ringbuffer *ring = ctx->binning_ring;
struct fd3_shader_key key = {
.binning_pass = true,
.half_precision = true,
};
fd3_emit_state(ctx, ring, &ctx->solid_prog, dirty, key);
fd3_emit_vertex_bufs(ring, fd3_shader_variant(ctx->solid_prog.vp, key),
(struct fd3_vertex_buf[]) {{
.prsc = fd3_ctx->solid_vbuf,
.stride = 12,
.format = PIPE_FORMAT_R32G32B32_FLOAT,
}}, 1);
OUT_PKT0(ring, REG_A3XX_PC_PRIM_VTX_CNTL, 1);
OUT_RING(ring, A3XX_PC_PRIM_VTX_CNTL_STRIDE_IN_VPC(0) |
A3XX_PC_PRIM_VTX_CNTL_POLYMODE_FRONT_PTYPE(PC_DRAW_TRIANGLES) |
A3XX_PC_PRIM_VTX_CNTL_POLYMODE_BACK_PTYPE(PC_DRAW_TRIANGLES) |
A3XX_PC_PRIM_VTX_CNTL_PROVOKING_VTX_LAST);
OUT_PKT0(ring, REG_A3XX_VFD_INDEX_MIN, 4);
OUT_RING(ring, 0); /* VFD_INDEX_MIN */
OUT_RING(ring, 2); /* VFD_INDEX_MAX */
OUT_RING(ring, 0); /* VFD_INSTANCEID_OFFSET */
OUT_RING(ring, 0); /* VFD_INDEX_OFFSET */
OUT_PKT0(ring, REG_A3XX_PC_RESTART_INDEX, 1);
OUT_RING(ring, 0xffffffff); /* PC_RESTART_INDEX */
OUT_PKT3(ring, CP_EVENT_WRITE, 1);
OUT_RING(ring, PERFCOUNTER_STOP);
fd_draw(ctx, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
DI_SRC_SEL_AUTO_INDEX, 2, INDEX_SIZE_IGN, 0, 0, NULL);
}
static void
fd3_clear(struct fd_context *ctx, unsigned buffers,
const union pipe_color_union *color, double depth, unsigned stencil)
{
struct fd3_context *fd3_ctx = fd3_context(ctx);
struct fd_ringbuffer *ring = ctx->ring;
unsigned dirty = ctx->dirty;
unsigned ce, i;
struct fd3_shader_key key = {
.half_precision = true,
};
dirty &= FD_DIRTY_VIEWPORT | FD_DIRTY_FRAMEBUFFER | FD_DIRTY_SCISSOR;
dirty |= FD_DIRTY_PROG;
fd3_clear_binning(ctx, dirty);
/* emit generic state now: */
fd3_emit_state(ctx, ring, &ctx->solid_prog, dirty, key);
OUT_PKT0(ring, REG_A3XX_RB_BLEND_ALPHA, 1);
OUT_RING(ring, A3XX_RB_BLEND_ALPHA_UINT(0xff) |
A3XX_RB_BLEND_ALPHA_FLOAT(1.0));
OUT_PKT0(ring, REG_A3XX_RB_RENDER_CONTROL, 1);
OUT_RINGP(ring, A3XX_RB_RENDER_CONTROL_ALPHA_TEST_FUNC(FUNC_NEVER),
&fd3_ctx->rbrc_patches);
if (buffers & PIPE_CLEAR_DEPTH) {
OUT_PKT0(ring, REG_A3XX_RB_DEPTH_CONTROL, 1);
OUT_RING(ring, A3XX_RB_DEPTH_CONTROL_Z_WRITE_ENABLE |
A3XX_RB_DEPTH_CONTROL_Z_ENABLE |
A3XX_RB_DEPTH_CONTROL_ZFUNC(FUNC_ALWAYS));
OUT_PKT0(ring, REG_A3XX_GRAS_CL_VPORT_ZOFFSET, 2);
OUT_RING(ring, A3XX_GRAS_CL_VPORT_ZOFFSET(0.0));
OUT_RING(ring, A3XX_GRAS_CL_VPORT_ZSCALE(depth));
ctx->dirty |= FD_DIRTY_VIEWPORT;
} else {
OUT_PKT0(ring, REG_A3XX_RB_DEPTH_CONTROL, 1);
OUT_RING(ring, A3XX_RB_DEPTH_CONTROL_ZFUNC(FUNC_NEVER));
}
if (buffers & PIPE_CLEAR_STENCIL) {
OUT_PKT0(ring, REG_A3XX_RB_STENCILREFMASK, 2);
OUT_RING(ring, A3XX_RB_STENCILREFMASK_STENCILREF(stencil) |
A3XX_RB_STENCILREFMASK_STENCILMASK(stencil) |
A3XX_RB_STENCILREFMASK_STENCILWRITEMASK(0xff));
OUT_RING(ring, A3XX_RB_STENCILREFMASK_STENCILREF(0) |
A3XX_RB_STENCILREFMASK_STENCILMASK(0) |
0xff000000 | // XXX ???
A3XX_RB_STENCILREFMASK_STENCILWRITEMASK(0xff));
OUT_PKT0(ring, REG_A3XX_RB_STENCIL_CONTROL, 1);
OUT_RING(ring, A3XX_RB_STENCIL_CONTROL_STENCIL_ENABLE |
A3XX_RB_STENCIL_CONTROL_FUNC(FUNC_ALWAYS) |
A3XX_RB_STENCIL_CONTROL_FAIL(STENCIL_KEEP) |
A3XX_RB_STENCIL_CONTROL_ZPASS(STENCIL_REPLACE) |
A3XX_RB_STENCIL_CONTROL_ZFAIL(STENCIL_KEEP) |
A3XX_RB_STENCIL_CONTROL_FUNC_BF(FUNC_NEVER) |
A3XX_RB_STENCIL_CONTROL_FAIL_BF(STENCIL_KEEP) |
A3XX_RB_STENCIL_CONTROL_ZPASS_BF(STENCIL_KEEP) |
A3XX_RB_STENCIL_CONTROL_ZFAIL_BF(STENCIL_KEEP));
} else {
OUT_PKT0(ring, REG_A3XX_RB_STENCILREFMASK, 2);
OUT_RING(ring, A3XX_RB_STENCILREFMASK_STENCILREF(0) |
A3XX_RB_STENCILREFMASK_STENCILMASK(0) |
A3XX_RB_STENCILREFMASK_STENCILWRITEMASK(0));
OUT_RING(ring, A3XX_RB_STENCILREFMASK_BF_STENCILREF(0) |
A3XX_RB_STENCILREFMASK_BF_STENCILMASK(0) |
A3XX_RB_STENCILREFMASK_BF_STENCILWRITEMASK(0));
OUT_PKT0(ring, REG_A3XX_RB_STENCIL_CONTROL, 1);
OUT_RING(ring, A3XX_RB_STENCIL_CONTROL_FUNC(FUNC_NEVER) |
A3XX_RB_STENCIL_CONTROL_FAIL(STENCIL_KEEP) |
A3XX_RB_STENCIL_CONTROL_ZPASS(STENCIL_KEEP) |
A3XX_RB_STENCIL_CONTROL_ZFAIL(STENCIL_KEEP) |
A3XX_RB_STENCIL_CONTROL_FUNC_BF(FUNC_NEVER) |
A3XX_RB_STENCIL_CONTROL_FAIL_BF(STENCIL_KEEP) |
A3XX_RB_STENCIL_CONTROL_ZPASS_BF(STENCIL_KEEP) |
A3XX_RB_STENCIL_CONTROL_ZFAIL_BF(STENCIL_KEEP));
}
if (buffers & PIPE_CLEAR_COLOR) {
ce = 0xf;
} else {
ce = 0x0;
}
for (i = 0; i < 4; i++) {
OUT_PKT0(ring, REG_A3XX_RB_MRT_CONTROL(i), 1);
OUT_RING(ring, A3XX_RB_MRT_CONTROL_ROP_CODE(ROP_COPY) |
A3XX_RB_MRT_CONTROL_DITHER_MODE(DITHER_ALWAYS) |
A3XX_RB_MRT_CONTROL_COMPONENT_ENABLE(ce));
OUT_PKT0(ring, REG_A3XX_RB_MRT_BLEND_CONTROL(i), 1);
OUT_RING(ring, A3XX_RB_MRT_BLEND_CONTROL_RGB_SRC_FACTOR(FACTOR_ONE) |
A3XX_RB_MRT_BLEND_CONTROL_RGB_BLEND_OPCODE(BLEND_DST_PLUS_SRC) |
A3XX_RB_MRT_BLEND_CONTROL_RGB_DEST_FACTOR(FACTOR_ZERO) |
A3XX_RB_MRT_BLEND_CONTROL_ALPHA_SRC_FACTOR(FACTOR_ONE) |
A3XX_RB_MRT_BLEND_CONTROL_ALPHA_BLEND_OPCODE(BLEND_DST_PLUS_SRC) |
A3XX_RB_MRT_BLEND_CONTROL_ALPHA_DEST_FACTOR(FACTOR_ZERO) |
A3XX_RB_MRT_BLEND_CONTROL_CLAMP_ENABLE);
}
OUT_PKT0(ring, REG_A3XX_GRAS_SU_MODE_CONTROL, 1);
OUT_RING(ring, A3XX_GRAS_SU_MODE_CONTROL_LINEHALFWIDTH(0));
fd3_emit_vertex_bufs(ring, fd3_shader_variant(ctx->solid_prog.vp, key),
(struct fd3_vertex_buf[]) {{
.prsc = fd3_ctx->solid_vbuf,
.stride = 12,
.format = PIPE_FORMAT_R32G32B32_FLOAT,
}}, 1);
fd_wfi(ctx, ring);
fd3_emit_constant(ring, SB_FRAG_SHADER, 0, 0, 4, color->ui, NULL);
OUT_PKT0(ring, REG_A3XX_PC_PRIM_VTX_CNTL, 1);
OUT_RING(ring, A3XX_PC_PRIM_VTX_CNTL_STRIDE_IN_VPC(0) |
A3XX_PC_PRIM_VTX_CNTL_POLYMODE_FRONT_PTYPE(PC_DRAW_TRIANGLES) |
A3XX_PC_PRIM_VTX_CNTL_POLYMODE_BACK_PTYPE(PC_DRAW_TRIANGLES) |
A3XX_PC_PRIM_VTX_CNTL_PROVOKING_VTX_LAST);
OUT_PKT0(ring, REG_A3XX_VFD_INDEX_MIN, 4);
OUT_RING(ring, 0); /* VFD_INDEX_MIN */
OUT_RING(ring, 2); /* VFD_INDEX_MAX */
OUT_RING(ring, 0); /* VFD_INSTANCEID_OFFSET */
OUT_RING(ring, 0); /* VFD_INDEX_OFFSET */
OUT_PKT0(ring, REG_A3XX_PC_RESTART_INDEX, 1);
OUT_RING(ring, 0xffffffff); /* PC_RESTART_INDEX */
OUT_PKT3(ring, CP_EVENT_WRITE, 1);
OUT_RING(ring, PERFCOUNTER_STOP);
fd_draw(ctx, ring, DI_PT_RECTLIST, USE_VISIBILITY,
DI_SRC_SEL_AUTO_INDEX, 2, INDEX_SIZE_IGN, 0, 0, NULL);
}
void
fd3_draw_init(struct pipe_context *pctx)
{
struct fd_context *ctx = fd_context(pctx);
ctx->draw = fd3_draw;
ctx->clear = fd3_clear;
}
static void
fd4_emit_tile_gmem2mem(struct fd_context *ctx, struct fd_tile *tile)
{
struct fd4_context *fd4_ctx = fd4_context(ctx);
struct fd_ringbuffer *ring = ctx->ring;
struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
struct fd4_emit emit = {
.vtx = &fd4_ctx->solid_vbuf_state,
.prog = &ctx->solid_prog,
.key = key,
.format = fd4_emit_format(pfb->cbufs[0]),
};
OUT_PKT0(ring, REG_A4XX_RB_DEPTH_CONTROL, 1);
OUT_RING(ring, A4XX_RB_DEPTH_CONTROL_ZFUNC(FUNC_NEVER));
OUT_PKT0(ring, REG_A4XX_RB_STENCIL_CONTROL, 2);
OUT_RING(ring, A4XX_RB_STENCIL_CONTROL_FUNC(FUNC_NEVER) |
A4XX_RB_STENCIL_CONTROL_FAIL(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZPASS(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZFAIL(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_FUNC_BF(FUNC_NEVER) |
A4XX_RB_STENCIL_CONTROL_FAIL_BF(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZPASS_BF(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZFAIL_BF(STENCIL_KEEP));
OUT_RING(ring, 0x00000000); /* RB_STENCIL_CONTROL2 */
OUT_PKT0(ring, REG_A4XX_RB_STENCILREFMASK, 2);
OUT_RING(ring, 0xff000000 |
A4XX_RB_STENCILREFMASK_STENCILREF(0) |
A4XX_RB_STENCILREFMASK_STENCILMASK(0) |
A4XX_RB_STENCILREFMASK_STENCILWRITEMASK(0xff));
OUT_RING(ring, 0xff000000 |
A4XX_RB_STENCILREFMASK_BF_STENCILREF(0) |
A4XX_RB_STENCILREFMASK_BF_STENCILMASK(0) |
A4XX_RB_STENCILREFMASK_BF_STENCILWRITEMASK(0xff));
OUT_PKT0(ring, REG_A4XX_GRAS_SU_MODE_CONTROL, 1);
OUT_RING(ring, A4XX_GRAS_SU_MODE_CONTROL_LINEHALFWIDTH(0));
fd_wfi(ctx, ring);
OUT_PKT0(ring, REG_A4XX_GRAS_CL_CLIP_CNTL, 1);
OUT_RING(ring, 0x80000); /* GRAS_CL_CLIP_CNTL */
OUT_PKT0(ring, REG_A4XX_GRAS_CL_VPORT_XOFFSET_0, 6);
OUT_RING(ring, A4XX_GRAS_CL_VPORT_XOFFSET_0((float)pfb->width/2.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_XSCALE_0((float)pfb->width/2.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_YOFFSET_0((float)pfb->height/2.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_YSCALE_0(-(float)pfb->height/2.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_ZOFFSET_0(0.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_ZSCALE_0(1.0));
OUT_PKT0(ring, REG_A4XX_RB_RENDER_CONTROL, 1);
OUT_RING(ring, A4XX_RB_RENDER_CONTROL_DISABLE_COLOR_PIPE |
0xa); /* XXX */
OUT_PKT0(ring, REG_A4XX_GRAS_SC_CONTROL, 1);
OUT_RING(ring, A4XX_GRAS_SC_CONTROL_RENDER_MODE(RB_RESOLVE_PASS) |
A4XX_GRAS_SC_CONTROL_MSAA_DISABLE |
A4XX_GRAS_SC_CONTROL_MSAA_SAMPLES(MSAA_ONE) |
A4XX_GRAS_SC_CONTROL_RASTER_MODE(1));
OUT_PKT0(ring, REG_A4XX_PC_PRIM_VTX_CNTL, 1);
OUT_RING(ring, A4XX_PC_PRIM_VTX_CNTL_PROVOKING_VTX_LAST);
OUT_PKT0(ring, REG_A4XX_GRAS_ALPHA_CONTROL, 1);
OUT_RING(ring, 0x00000002);
OUT_PKT0(ring, REG_A4XX_GRAS_SC_WINDOW_SCISSOR_BR, 2);
OUT_RING(ring, A4XX_GRAS_SC_WINDOW_SCISSOR_BR_X(pfb->width - 1) |
A4XX_GRAS_SC_WINDOW_SCISSOR_BR_Y(pfb->height - 1));
OUT_RING(ring, A4XX_GRAS_SC_WINDOW_SCISSOR_TL_X(0) |
A4XX_GRAS_SC_WINDOW_SCISSOR_TL_Y(0));
OUT_PKT0(ring, REG_A4XX_VFD_INDEX_OFFSET, 2);
OUT_RING(ring, 0); /* VFD_INDEX_OFFSET */
OUT_RING(ring, 0); /* ??? UNKNOWN_2209 */
fd4_program_emit(ring, &emit);
fd4_emit_vertex_bufs(ring, &emit);
if (ctx->resolve & (FD_BUFFER_DEPTH | FD_BUFFER_STENCIL)) {
uint32_t base = depth_base(ctx);
emit_gmem2mem_surf(ctx, base, pfb->zsbuf);
}
if (ctx->resolve & FD_BUFFER_COLOR) {
emit_gmem2mem_surf(ctx, 0, pfb->cbufs[0]);
}
OUT_PKT0(ring, REG_A4XX_GRAS_SC_CONTROL, 1);
OUT_RING(ring, A4XX_GRAS_SC_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
A4XX_GRAS_SC_CONTROL_MSAA_DISABLE |
A4XX_GRAS_SC_CONTROL_MSAA_SAMPLES(MSAA_ONE) |
A4XX_GRAS_SC_CONTROL_RASTER_MODE(0));
}
/* transfer from system memory to gmem */
static void
emit_mem2gmem_surf(struct fd_context *ctx, uint32_t base,
struct pipe_surface *psurf, uint32_t bin_w)
{
struct fd_ringbuffer *ring = ctx->ring;
emit_mrt(ring, 1, &psurf, &base, bin_w);
fd4_emit_gmem_restore_tex(ring, psurf);
fd4_draw(ctx, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
DI_SRC_SEL_AUTO_INDEX, 2, 1, INDEX_SIZE_IGN, 0, 0, NULL);
}
static void
fd4_emit_tile_mem2gmem(struct fd_context *ctx, struct fd_tile *tile)
{
struct fd4_context *fd4_ctx = fd4_context(ctx);
struct fd_gmem_stateobj *gmem = &ctx->gmem;
struct fd_ringbuffer *ring = ctx->ring;
struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
struct fd4_emit emit = {
.vtx = &fd4_ctx->blit_vbuf_state,
.prog = &ctx->blit_prog[0],
.key = key,
.format = fd4_emit_format(pfb->cbufs[0]),
};
float x0, y0, x1, y1;
unsigned bin_w = tile->bin_w;
unsigned bin_h = tile->bin_h;
unsigned i;
/* write texture coordinates to vertexbuf: */
x0 = ((float)tile->xoff) / ((float)pfb->width);
x1 = ((float)tile->xoff + bin_w) / ((float)pfb->width);
y0 = ((float)tile->yoff) / ((float)pfb->height);
y1 = ((float)tile->yoff + bin_h) / ((float)pfb->height);
OUT_PKT3(ring, CP_MEM_WRITE, 5);
OUT_RELOCW(ring, fd_resource(fd4_ctx->blit_texcoord_vbuf)->bo, 0, 0, 0);
OUT_RING(ring, fui(x0));
OUT_RING(ring, fui(y0));
OUT_RING(ring, fui(x1));
OUT_RING(ring, fui(y1));
for (i = 0; i < 8; i++) {
OUT_PKT0(ring, REG_A4XX_RB_MRT_CONTROL(i), 1);
OUT_RING(ring, A4XX_RB_MRT_CONTROL_FASTCLEAR |
A4XX_RB_MRT_CONTROL_B11 |
A4XX_RB_MRT_CONTROL_COMPONENT_ENABLE(0xf));
OUT_PKT0(ring, REG_A4XX_RB_MRT_BLEND_CONTROL(i), 1);
OUT_RING(ring, A4XX_RB_MRT_BLEND_CONTROL_RGB_SRC_FACTOR(FACTOR_ONE) |
A4XX_RB_MRT_BLEND_CONTROL_RGB_BLEND_OPCODE(BLEND_DST_PLUS_SRC) |
A4XX_RB_MRT_BLEND_CONTROL_RGB_DEST_FACTOR(FACTOR_ZERO) |
A4XX_RB_MRT_BLEND_CONTROL_ALPHA_SRC_FACTOR(FACTOR_ONE) |
A4XX_RB_MRT_BLEND_CONTROL_ALPHA_BLEND_OPCODE(BLEND_DST_PLUS_SRC) |
A4XX_RB_MRT_BLEND_CONTROL_ALPHA_DEST_FACTOR(FACTOR_ZERO));
}
OUT_PKT0(ring, REG_A4XX_RB_RENDER_CONTROL, 1);
OUT_RING(ring, 0x8); /* XXX RB_RENDER_CONTROL */
OUT_PKT0(ring, REG_A4XX_RB_DEPTH_CONTROL, 1);
OUT_RING(ring, A4XX_RB_DEPTH_CONTROL_ZFUNC(FUNC_LESS));
OUT_PKT0(ring, REG_A4XX_GRAS_CL_CLIP_CNTL, 1);
OUT_RING(ring, 0x280000); /* XXX GRAS_CL_CLIP_CNTL */
OUT_PKT0(ring, REG_A4XX_GRAS_SU_MODE_CONTROL, 1);
OUT_RING(ring, A4XX_GRAS_SU_MODE_CONTROL_LINEHALFWIDTH(0) |
A4XX_GRAS_SU_MODE_CONTROL_RENDERING_PASS);
OUT_PKT0(ring, REG_A4XX_GRAS_CL_VPORT_XOFFSET_0, 6);
OUT_RING(ring, A4XX_GRAS_CL_VPORT_XOFFSET_0((float)bin_w/2.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_XSCALE_0((float)bin_w/2.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_YOFFSET_0((float)bin_h/2.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_YSCALE_0(-(float)bin_h/2.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_ZOFFSET_0(0.0));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_ZSCALE_0(1.0));
OUT_PKT0(ring, REG_A4XX_GRAS_SC_WINDOW_SCISSOR_BR, 2);
OUT_RING(ring, A4XX_GRAS_SC_WINDOW_SCISSOR_BR_X(bin_w - 1) |
A4XX_GRAS_SC_WINDOW_SCISSOR_BR_Y(bin_h - 1));
OUT_RING(ring, A4XX_GRAS_SC_WINDOW_SCISSOR_TL_X(0) |
A4XX_GRAS_SC_WINDOW_SCISSOR_TL_Y(0));
OUT_PKT0(ring, REG_A4XX_GRAS_SC_SCREEN_SCISSOR_TL, 2);
OUT_RING(ring, A4XX_GRAS_SC_SCREEN_SCISSOR_TL_X(0) |
A4XX_GRAS_SC_SCREEN_SCISSOR_TL_Y(0));
OUT_RING(ring, A4XX_GRAS_SC_SCREEN_SCISSOR_BR_X(bin_w - 1) |
A4XX_GRAS_SC_SCREEN_SCISSOR_BR_Y(bin_h - 1));
OUT_PKT0(ring, REG_A4XX_RB_MODE_CONTROL, 1);
OUT_RING(ring, A4XX_RB_MODE_CONTROL_WIDTH(gmem->bin_w) |
A4XX_RB_MODE_CONTROL_HEIGHT(gmem->bin_h));
OUT_PKT0(ring, REG_A4XX_RB_STENCIL_CONTROL, 2);
OUT_RING(ring, A4XX_RB_STENCIL_CONTROL_FUNC(FUNC_ALWAYS) |
A4XX_RB_STENCIL_CONTROL_FAIL(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZPASS(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZFAIL(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_FUNC_BF(FUNC_ALWAYS) |
A4XX_RB_STENCIL_CONTROL_FAIL_BF(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZPASS_BF(STENCIL_KEEP) |
A4XX_RB_STENCIL_CONTROL_ZFAIL_BF(STENCIL_KEEP));
OUT_RING(ring, 0x00000000); /* RB_STENCIL_CONTROL2 */
OUT_PKT0(ring, REG_A4XX_GRAS_SC_CONTROL, 1);
OUT_RING(ring, A4XX_GRAS_SC_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
A4XX_GRAS_SC_CONTROL_MSAA_DISABLE |
A4XX_GRAS_SC_CONTROL_MSAA_SAMPLES(MSAA_ONE) |
A4XX_GRAS_SC_CONTROL_RASTER_MODE(1));
OUT_PKT0(ring, REG_A4XX_PC_PRIM_VTX_CNTL, 1);
OUT_RING(ring, A4XX_PC_PRIM_VTX_CNTL_PROVOKING_VTX_LAST |
A4XX_PC_PRIM_VTX_CNTL_VAROUT(1));
OUT_PKT0(ring, REG_A4XX_VFD_INDEX_OFFSET, 2);
OUT_RING(ring, 0); /* VFD_INDEX_OFFSET */
OUT_RING(ring, 0); /* ??? UNKNOWN_2209 */
fd4_program_emit(ring, &emit);
fd4_emit_vertex_bufs(ring, &emit);
/* for gmem pitch/base calculations, we need to use the non-
* truncated tile sizes:
*/
bin_w = gmem->bin_w;
bin_h = gmem->bin_h;
if (fd_gmem_needs_restore(ctx, tile, FD_BUFFER_DEPTH | FD_BUFFER_STENCIL))
emit_mem2gmem_surf(ctx, depth_base(ctx), pfb->zsbuf, bin_w);
if (fd_gmem_needs_restore(ctx, tile, FD_BUFFER_COLOR))
emit_mem2gmem_surf(ctx, 0, pfb->cbufs[0], bin_w);
OUT_PKT0(ring, REG_A4XX_GRAS_SC_CONTROL, 1);
OUT_RING(ring, A4XX_GRAS_SC_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
A4XX_GRAS_SC_CONTROL_MSAA_SAMPLES(MSAA_ONE) |
A4XX_GRAS_SC_CONTROL_RASTER_MODE(0));
OUT_PKT0(ring, REG_A4XX_RB_MODE_CONTROL, 1);
OUT_RING(ring, A4XX_RB_MODE_CONTROL_WIDTH(gmem->bin_w) |
A4XX_RB_MODE_CONTROL_HEIGHT(gmem->bin_h) |
0x00010000); /* XXX */
}
static void
patch_draws(struct fd_context *ctx, enum pc_di_vis_cull_mode vismode)
{
unsigned i;
for (i = 0; i < fd_patch_num_elements(&ctx->draw_patches); i++) {
struct fd_cs_patch *patch = fd_patch_element(&ctx->draw_patches, i);
*patch->cs = patch->val | DRAW4(0, 0, 0, vismode);
}
util_dynarray_resize(&ctx->draw_patches, 0);
}
static void
patch_rbrc(struct fd_context *ctx, uint32_t val)
{
struct fd4_context *fd4_ctx = fd4_context(ctx);
unsigned i;
for (i = 0; i < fd_patch_num_elements(&fd4_ctx->rbrc_patches); i++) {
struct fd_cs_patch *patch = fd_patch_element(&fd4_ctx->rbrc_patches, i);
*patch->cs = patch->val | val;
}
util_dynarray_resize(&fd4_ctx->rbrc_patches, 0);
}
/* for rendering directly to system memory: */
static void
fd4_emit_sysmem_prep(struct fd_context *ctx)
{
struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
struct fd_ringbuffer *ring = ctx->ring;
fd4_emit_restore(ctx);
OUT_PKT0(ring, REG_A4XX_RB_FRAME_BUFFER_DIMENSION, 1);
OUT_RING(ring, A4XX_RB_FRAME_BUFFER_DIMENSION_WIDTH(pfb->width) |
A4XX_RB_FRAME_BUFFER_DIMENSION_HEIGHT(pfb->height));
emit_mrt(ring, pfb->nr_cbufs, pfb->cbufs, NULL, 0);
/* setup scissor/offset for current tile: */
OUT_PKT0(ring, REG_A4XX_RB_BIN_OFFSET, 1);
OUT_RING(ring, A4XX_RB_BIN_OFFSET_X(0) |
A4XX_RB_BIN_OFFSET_Y(0));
OUT_PKT0(ring, REG_A4XX_GRAS_SC_SCREEN_SCISSOR_TL, 2);
OUT_RING(ring, A4XX_GRAS_SC_SCREEN_SCISSOR_TL_X(0) |
A4XX_GRAS_SC_SCREEN_SCISSOR_TL_Y(0));
OUT_RING(ring, A4XX_GRAS_SC_SCREEN_SCISSOR_BR_X(pfb->width - 1) |
A4XX_GRAS_SC_SCREEN_SCISSOR_BR_Y(pfb->height - 1));
OUT_PKT0(ring, REG_A4XX_RB_MODE_CONTROL, 1);
OUT_RING(ring, A4XX_RB_MODE_CONTROL_WIDTH(0) |
A4XX_RB_MODE_CONTROL_HEIGHT(0) |
0x00c00000); /* XXX */
patch_draws(ctx, IGNORE_VISIBILITY);
patch_rbrc(ctx, 0); // XXX
}
static void
update_vsc_pipe(struct fd_context *ctx)
{
struct fd4_context *fd4_ctx = fd4_context(ctx);
struct fd_ringbuffer *ring = ctx->ring;
int i;
OUT_PKT0(ring, REG_A4XX_VSC_SIZE_ADDRESS, 1);
OUT_RELOCW(ring, fd4_ctx->vsc_size_mem, 0, 0, 0); /* VSC_SIZE_ADDRESS */
OUT_PKT0(ring, REG_A4XX_VSC_PIPE_CONFIG_REG(0), 8);
for (i = 0; i < 8; i++) {
struct fd_vsc_pipe *pipe = &ctx->pipe[i];
OUT_RING(ring, A4XX_VSC_PIPE_CONFIG_REG_X(pipe->x) |
A4XX_VSC_PIPE_CONFIG_REG_Y(pipe->y) |
A4XX_VSC_PIPE_CONFIG_REG_W(pipe->w) |
A4XX_VSC_PIPE_CONFIG_REG_H(pipe->h));
}
OUT_PKT0(ring, REG_A4XX_VSC_PIPE_DATA_ADDRESS_REG(0), 8);
for (i = 0; i < 8; i++) {
struct fd_vsc_pipe *pipe = &ctx->pipe[i];
if (!pipe->bo) {
pipe->bo = fd_bo_new(ctx->dev, 0x40000,
DRM_FREEDRENO_GEM_TYPE_KMEM);
}
OUT_RELOCW(ring, pipe->bo, 0, 0, 0); /* VSC_PIPE_DATA_ADDRESS[i] */
}
OUT_PKT0(ring, REG_A4XX_VSC_PIPE_DATA_LENGTH_REG(0), 8);
for (i = 0; i < 8; i++) {
struct fd_vsc_pipe *pipe = &ctx->pipe[i];
OUT_RING(ring, fd_bo_size(pipe->bo) - 32); /* VSC_PIPE_DATA_LENGTH[i] */
}
}
/* before first tile */
static void
fd4_emit_tile_init(struct fd_context *ctx)
{
struct fd_ringbuffer *ring = ctx->ring;
struct fd_gmem_stateobj *gmem = &ctx->gmem;
uint32_t rb_render_control;
fd4_emit_restore(ctx);
OUT_PKT0(ring, REG_A4XX_VSC_BIN_SIZE, 1);
OUT_RING(ring, A4XX_VSC_BIN_SIZE_WIDTH(gmem->bin_w) |
A4XX_VSC_BIN_SIZE_HEIGHT(gmem->bin_h));
OUT_PKT0(ring, REG_A4XX_RB_MODE_CONTROL, 1);
OUT_RING(ring, A4XX_RB_MODE_CONTROL_WIDTH(gmem->bin_w) |
A4XX_RB_MODE_CONTROL_HEIGHT(gmem->bin_h) |
0x00010000); /* XXX */
update_vsc_pipe(ctx);
patch_draws(ctx, IGNORE_VISIBILITY);
rb_render_control = 0; // XXX or BINNING_PASS.. but maybe we can emit only from gmem
patch_rbrc(ctx, rb_render_control);
}
/* before mem2gmem */
static void
fd4_emit_tile_prep(struct fd_context *ctx, struct fd_tile *tile)
{
struct fd_ringbuffer *ring = ctx->ring;
struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
struct fd_gmem_stateobj *gmem = &ctx->gmem;
uint32_t reg;
OUT_PKT0(ring, REG_A4XX_RB_DEPTH_INFO, 3);
reg = A4XX_RB_DEPTH_INFO_DEPTH_BASE(depth_base(ctx));
if (pfb->zsbuf) {
reg |= A4XX_RB_DEPTH_INFO_DEPTH_FORMAT(fd4_pipe2depth(pfb->zsbuf->format));
}
OUT_RING(ring, reg);
if (pfb->zsbuf) {
uint32_t cpp = util_format_get_blocksize(pfb->zsbuf->format);
OUT_RING(ring, A4XX_RB_DEPTH_PITCH(cpp * gmem->bin_w));
OUT_RING(ring, A4XX_RB_DEPTH_PITCH2(cpp * gmem->bin_w));
} else {
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000000);
}
OUT_PKT0(ring, REG_A4XX_GRAS_DEPTH_CONTROL, 1);
if (pfb->zsbuf) {
OUT_RING(ring, A4XX_GRAS_DEPTH_CONTROL_FORMAT(
fd4_pipe2depth(pfb->zsbuf->format)));
} else {
OUT_RING(ring, A4XX_GRAS_DEPTH_CONTROL_FORMAT(DEPTH4_NONE));
}
if (ctx->needs_rb_fbd) {
fd_wfi(ctx, ring);
OUT_PKT0(ring, REG_A4XX_RB_FRAME_BUFFER_DIMENSION, 1);
OUT_RING(ring, A4XX_RB_FRAME_BUFFER_DIMENSION_WIDTH(pfb->width) |
A4XX_RB_FRAME_BUFFER_DIMENSION_HEIGHT(pfb->height));
ctx->needs_rb_fbd = false;
}
}
/* before IB to rendering cmds: */
static void
fd4_emit_tile_renderprep(struct fd_context *ctx, struct fd_tile *tile)
{
struct fd_ringbuffer *ring = ctx->ring;
struct fd_gmem_stateobj *gmem = &ctx->gmem;
struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
uint32_t x1 = tile->xoff;
uint32_t y1 = tile->yoff;
uint32_t x2 = tile->xoff + tile->bin_w - 1;
uint32_t y2 = tile->yoff + tile->bin_h - 1;
OUT_PKT3(ring, CP_SET_BIN, 3);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, CP_SET_BIN_1_X1(x1) | CP_SET_BIN_1_Y1(y1));
OUT_RING(ring, CP_SET_BIN_2_X2(x2) | CP_SET_BIN_2_Y2(y2));
emit_mrt(ring, pfb->nr_cbufs, pfb->cbufs, NULL, gmem->bin_w);
/* setup scissor/offset for current tile: */
OUT_PKT0(ring, REG_A4XX_RB_BIN_OFFSET, 1);
OUT_RING(ring, A4XX_RB_BIN_OFFSET_X(tile->xoff) |
A4XX_RB_BIN_OFFSET_Y(tile->yoff));
OUT_PKT0(ring, REG_A4XX_GRAS_SC_SCREEN_SCISSOR_TL, 2);
OUT_RING(ring, A4XX_GRAS_SC_SCREEN_SCISSOR_TL_X(x1) |
A4XX_GRAS_SC_SCREEN_SCISSOR_TL_Y(y1));
OUT_RING(ring, A4XX_GRAS_SC_SCREEN_SCISSOR_BR_X(x2) |
A4XX_GRAS_SC_SCREEN_SCISSOR_BR_Y(y2));
}
void
fd4_gmem_init(struct pipe_context *pctx)
{
struct fd_context *ctx = fd_context(pctx);
ctx->emit_sysmem_prep = fd4_emit_sysmem_prep;
ctx->emit_tile_init = fd4_emit_tile_init;
ctx->emit_tile_prep = fd4_emit_tile_prep;
ctx->emit_tile_mem2gmem = fd4_emit_tile_mem2gmem;
ctx->emit_tile_renderprep = fd4_emit_tile_renderprep;
ctx->emit_tile_gmem2mem = fd4_emit_tile_gmem2mem;
}
void adreno_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,
struct msm_file_private *ctx)
{
struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
struct msm_drm_private *priv = gpu->dev->dev_private;
struct msm_ringbuffer *ring = gpu->rb;
unsigned i;
for (i = 0; i < submit->nr_cmds; i++) {
switch (submit->cmd[i].type) {
case MSM_SUBMIT_CMD_IB_TARGET_BUF:
/* ignore IB-targets */
break;
case MSM_SUBMIT_CMD_CTX_RESTORE_BUF:
/* ignore if there has not been a ctx switch: */
if (priv->lastctx == ctx)
break;
case MSM_SUBMIT_CMD_BUF:
OUT_PKT3(ring, adreno_is_a430(adreno_gpu) ?
CP_INDIRECT_BUFFER_PFE : CP_INDIRECT_BUFFER_PFD, 2);
OUT_RING(ring, submit->cmd[i].iova);
OUT_RING(ring, submit->cmd[i].size);
OUT_PKT2(ring);
break;
}
}
OUT_PKT0(ring, REG_AXXX_CP_SCRATCH_REG2, 1);
OUT_RING(ring, submit->fence->seqno);
if (adreno_is_a3xx(adreno_gpu) || adreno_is_a4xx(adreno_gpu)) {
/* Flush HLSQ lazy updates to make sure there is nothing
* pending for indirect loads after the timestamp has
* passed:
*/
OUT_PKT3(ring, CP_EVENT_WRITE, 1);
OUT_RING(ring, HLSQ_FLUSH);
OUT_PKT3(ring, CP_WAIT_FOR_IDLE, 1);
OUT_RING(ring, 0x00000000);
}
OUT_PKT3(ring, CP_EVENT_WRITE, 3);
OUT_RING(ring, CACHE_FLUSH_TS);
OUT_RING(ring, rbmemptr(adreno_gpu, fence));
OUT_RING(ring, submit->fence->seqno);
/* we could maybe be clever and only CP_COND_EXEC the interrupt: */
OUT_PKT3(ring, CP_INTERRUPT, 1);
OUT_RING(ring, 0x80000000);
/* Workaround for missing irq issue on 8x16/a306. Unsure if the
* root cause is a platform issue or some a306 quirk, but this
* keeps things humming along:
*/
if (adreno_is_a306(adreno_gpu)) {
OUT_PKT3(ring, CP_WAIT_FOR_IDLE, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT3(ring, CP_INTERRUPT, 1);
OUT_RING(ring, 0x80000000);
}
#if 0
if (adreno_is_a3xx(adreno_gpu)) {
/* Dummy set-constant to trigger context rollover */
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A3XX_HLSQ_CL_KERNEL_GROUP_X_REG));
OUT_RING(ring, 0x00000000);
}
#endif
gpu->funcs->flush(gpu);
}
int main(int argc, char *argv[])
{
struct fd_device *dev;
struct fd_pipe *pipe;
struct fd_ringbuffer *ring;
struct fd_bo *bo;
uint32_t i = 0;
uint32_t *ptr;
int fd, ret;
fd = drmOpen("msm", NULL);
if (fd < 0) {
printf("failed to initialize DRM\n");
return fd;
}
dev = fd_device_new(fd);
if (!dev) {
printf("failed to initialize freedreno device\n");
return -1;
}
pipe = fd_pipe_new(dev, FD_PIPE_3D);
if (!pipe) {
printf("failed to initialize freedreno pipe\n");
return -1;
}
ring = fd_ringbuffer_new(pipe, 4096);
if (!ring) {
printf("failed to initialize freedreno ring\n");
return -1;
}
bo = fd_bo_new(dev, 4096, 0);
#define BASE REG_A3XX_GRAS_CL_VPORT_XOFFSET
#define SIZE 6
OUT_PKT0(ring, REG_AXXX_CP_SCRATCH_REG4, 1);
OUT_RING(ring, 0x123);
OUT_PKT0(ring, BASE, SIZE);
for (i = 0; i < SIZE; i++)
OUT_RING(ring, i);
/* this adds the value of CP_SCRATCH_REG4 to 0x111 and writes
* to REG_BASE+2
*/
OUT_PKT3(ring, CP_SET_CONSTANT, 3);
OUT_RING(ring, 0x80000000 | CP_REG(BASE + 2));
OUT_RING(ring, REG_AXXX_CP_SCRATCH_REG4);
OUT_RING(ring, 0x111);
/* read back all the regs: */
for (i = 0; i < SIZE; i++) {
OUT_PKT3(ring, CP_REG_TO_MEM, 2);
OUT_RING(ring, BASE + i);
OUT_RELOCW(ring, bo, i * 4, 0, 0);
}
fd_ringbuffer_flush(ring);
/* and read back the values: */
fd_bo_cpu_prep(bo, pipe, DRM_FREEDRENO_PREP_READ);
ptr = fd_bo_map(bo);
for (i = 0; i < SIZE; i++) {
printf("%02x: %08x\n", i, ptr[i]);
}
fd_bo_cpu_fini(bo);
return 0;
}
