static void
draw_emit_indirect(struct fd_batch *batch, struct fd_ringbuffer *ring,
enum pc_di_primtype primtype,
enum pc_di_vis_cull_mode vismode,
const struct pipe_draw_info *info,
unsigned index_offset)
{
struct fd_resource *ind = fd_resource(info->indirect->buffer);
if (info->index_size) {
struct pipe_resource *idx = info->index.resource;
unsigned max_indicies = (idx->width0 - info->indirect->offset) /
info->index_size;
OUT_PKT7(ring, CP_DRAW_INDX_INDIRECT, 6);
OUT_RINGP(ring, DRAW4(primtype, DI_SRC_SEL_DMA,
fd4_size2indextype(info->index_size), 0),
&batch->draw_patches);
OUT_RELOC(ring, fd_resource(idx)->bo,
index_offset, 0, 0);
// XXX: Check A5xx vs A6xx
OUT_RING(ring, A5XX_CP_DRAW_INDX_INDIRECT_3_MAX_INDICES(max_indicies));
OUT_RELOC(ring, ind->bo, info->indirect->offset, 0, 0);
} else {
OUT_PKT7(ring, CP_DRAW_INDIRECT, 3);
OUT_RINGP(ring, DRAW4(primtype, DI_SRC_SEL_AUTO_INDEX, 0, 0),
&batch->draw_patches);
OUT_RELOC(ring, ind->bo, info->indirect->offset, 0, 0);
}
}
static void
draw_emit(struct fd_batch *batch, struct fd_ringbuffer *ring,
enum pc_di_primtype primtype,
enum pc_di_vis_cull_mode vismode,
const struct pipe_draw_info *info,
unsigned index_offset)
{
if (info->index_size) {
assert(!info->has_user_indices);
struct pipe_resource *idx_buffer = info->index.resource;
uint32_t idx_size = info->index_size * info->count;
uint32_t idx_offset = index_offset + info->start * info->index_size;
/* leave vis mode blank for now, it will be patched up when
* we know if we are binning or not
*/
uint32_t draw = CP_DRAW_INDX_OFFSET_0_PRIM_TYPE(primtype) |
CP_DRAW_INDX_OFFSET_0_SOURCE_SELECT(DI_SRC_SEL_DMA) |
CP_DRAW_INDX_OFFSET_0_INDEX_SIZE(fd4_size2indextype(info->index_size)) |
0x2000;
OUT_PKT7(ring, CP_DRAW_INDX_OFFSET, 7);
if (vismode == USE_VISIBILITY) {
OUT_RINGP(ring, draw, &batch->draw_patches);
} else {
OUT_RING(ring, draw);
}
OUT_RING(ring, info->instance_count); /* NumInstances */
OUT_RING(ring, info->count); /* NumIndices */
OUT_RING(ring, 0x0); /* XXX */
OUT_RELOC(ring, fd_resource(idx_buffer)->bo, idx_offset, 0, 0);
OUT_RING (ring, idx_size);
} else {
/* leave vis mode blank for now, it will be patched up when
* we know if we are binning or not
*/
uint32_t draw = CP_DRAW_INDX_OFFSET_0_PRIM_TYPE(primtype) |
CP_DRAW_INDX_OFFSET_0_SOURCE_SELECT(DI_SRC_SEL_AUTO_INDEX) |
0x2000;
OUT_PKT7(ring, CP_DRAW_INDX_OFFSET, 3);
if (vismode == USE_VISIBILITY) {
OUT_RINGP(ring, draw, &batch->draw_patches);
} else {
OUT_RING(ring, draw);
}
OUT_RING(ring, info->instance_count); /* NumInstances */
OUT_RING(ring, info->count); /* NumIndices */
}
}
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 bool
fd6_clear(struct fd_context *ctx, unsigned buffers,
const union pipe_color_union *color, double depth, unsigned stencil)
{
struct pipe_framebuffer_state *pfb = &ctx->batch->framebuffer;
struct pipe_scissor_state *scissor = fd_context_get_scissor(ctx);
struct fd_ringbuffer *ring = ctx->batch->draw;
if ((buffers & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL)) &&
is_z32(pfb->zsbuf->format))
return false;
fd6_emit_render_cntl(ctx, true, false);
OUT_PKT4(ring, REG_A6XX_RB_BLIT_SCISSOR_TL, 2);
OUT_RING(ring, A6XX_RB_BLIT_SCISSOR_TL_X(scissor->minx) |
A6XX_RB_BLIT_SCISSOR_TL_Y(scissor->miny));
OUT_RING(ring, A6XX_RB_BLIT_SCISSOR_BR_X(scissor->maxx - 1) |
A6XX_RB_BLIT_SCISSOR_BR_Y(scissor->maxy - 1));
if (buffers & PIPE_CLEAR_COLOR) {
for (int i = 0; i < pfb->nr_cbufs; i++) {
union util_color uc = {0};
if (!pfb->cbufs[i])
continue;
if (!(buffers & (PIPE_CLEAR_COLOR0 << i)))
continue;
enum pipe_format pfmt = pfb->cbufs[i]->format;
// XXX I think RB_CLEAR_COLOR_DWn wants to take into account SWAP??
union pipe_color_union swapped;
switch (fd6_pipe2swap(pfmt)) {
case WZYX:
swapped.ui[0] = color->ui[0];
swapped.ui[1] = color->ui[1];
swapped.ui[2] = color->ui[2];
swapped.ui[3] = color->ui[3];
break;
case WXYZ:
swapped.ui[2] = color->ui[0];
swapped.ui[1] = color->ui[1];
swapped.ui[0] = color->ui[2];
swapped.ui[3] = color->ui[3];
break;
case ZYXW:
swapped.ui[3] = color->ui[0];
swapped.ui[0] = color->ui[1];
swapped.ui[1] = color->ui[2];
swapped.ui[2] = color->ui[3];
break;
case XYZW:
swapped.ui[3] = color->ui[0];
swapped.ui[2] = color->ui[1];
swapped.ui[1] = color->ui[2];
swapped.ui[0] = color->ui[3];
break;
}
if (util_format_is_pure_uint(pfmt)) {
util_format_write_4ui(pfmt, swapped.ui, 0, &uc, 0, 0, 0, 1, 1);
} else if (util_format_is_pure_sint(pfmt)) {
util_format_write_4i(pfmt, swapped.i, 0, &uc, 0, 0, 0, 1, 1);
} else {
util_pack_color(swapped.f, pfmt, &uc);
}
OUT_PKT4(ring, REG_A6XX_RB_BLIT_DST_INFO, 1);
OUT_RING(ring, A6XX_RB_BLIT_DST_INFO_TILE_MODE(TILE6_LINEAR) |
A6XX_RB_BLIT_DST_INFO_COLOR_FORMAT(fd6_pipe2color(pfmt)));
OUT_PKT4(ring, REG_A6XX_RB_BLIT_INFO, 1);
OUT_RING(ring, A6XX_RB_BLIT_INFO_GMEM |
A6XX_RB_BLIT_INFO_CLEAR_MASK(0xf));
OUT_PKT4(ring, REG_A6XX_RB_BLIT_BASE_GMEM, 1);
OUT_RINGP(ring, i, &ctx->batch->gmem_patches);
OUT_PKT4(ring, REG_A6XX_RB_UNKNOWN_88D0, 1);
OUT_RING(ring, 0);
OUT_PKT4(ring, REG_A6XX_RB_BLIT_CLEAR_COLOR_DW0, 4);
OUT_RING(ring, uc.ui[0]);
OUT_RING(ring, uc.ui[1]);
OUT_RING(ring, uc.ui[2]);
OUT_RING(ring, uc.ui[3]);
fd6_emit_blit(ctx->batch, ring);
}
}
if (pfb->zsbuf && (buffers & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL))) {
enum pipe_format pfmt = pfb->zsbuf->format;
uint32_t clear = util_pack_z_stencil(pfmt, depth, stencil);
uint32_t mask = 0;
if (buffers & PIPE_CLEAR_DEPTH)
mask |= 0x1;
if (buffers & PIPE_CLEAR_STENCIL)
mask |= 0x2;
OUT_PKT4(ring, REG_A6XX_RB_BLIT_DST_INFO, 1);
OUT_RING(ring, A6XX_RB_BLIT_DST_INFO_TILE_MODE(TILE6_LINEAR) |
A6XX_RB_BLIT_DST_INFO_COLOR_FORMAT(fd6_pipe2color(pfmt)));
OUT_PKT4(ring, REG_A6XX_RB_BLIT_INFO, 1);
OUT_RING(ring, A6XX_RB_BLIT_INFO_GMEM |
// XXX UNK0 for separate stencil ??
A6XX_RB_BLIT_INFO_DEPTH |
A6XX_RB_BLIT_INFO_CLEAR_MASK(mask));
OUT_PKT4(ring, REG_A6XX_RB_BLIT_BASE_GMEM, 1);
OUT_RINGP(ring, MAX_RENDER_TARGETS, &ctx->batch->gmem_patches);
OUT_PKT4(ring, REG_A6XX_RB_UNKNOWN_88D0, 1);
OUT_RING(ring, 0);
OUT_PKT4(ring, REG_A6XX_RB_BLIT_CLEAR_COLOR_DW0, 1);
OUT_RING(ring, clear);
fd6_emit_blit(ctx->batch, ring);
if (pfb->zsbuf && (buffers & PIPE_CLEAR_DEPTH)) {
struct fd_resource *zsbuf = fd_resource(pfb->zsbuf->texture);
if (zsbuf->lrz) {
zsbuf->lrz_valid = true;
fd6_clear_lrz(ctx->batch, zsbuf, depth);
}
}
}
return true;
}
void
fd4_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
struct fd4_emit *emit)
{
struct ir3_shader_variant *vp = fd4_emit_get_vp(emit);
struct ir3_shader_variant *fp = fd4_emit_get_fp(emit);
uint32_t dirty = emit->dirty;
emit_marker(ring, 5);
if ((dirty & (FD_DIRTY_ZSA | FD_DIRTY_PROG)) && !emit->key.binning_pass) {
uint32_t val = fd4_zsa_stateobj(ctx->zsa)->rb_render_control;
/* I suppose if we needed to (which I don't *think* we need
* to), we could emit this for binning pass too. But we
* would need to keep a different patch-list for binning
* vs render pass.
*/
OUT_PKT0(ring, REG_A4XX_RB_RENDER_CONTROL, 1);
OUT_RINGP(ring, val, &fd4_context(ctx)->rbrc_patches);
}
if (dirty & FD_DIRTY_ZSA) {
struct fd4_zsa_stateobj *zsa = fd4_zsa_stateobj(ctx->zsa);
OUT_PKT0(ring, REG_A4XX_RB_ALPHA_CONTROL, 1);
OUT_RING(ring, zsa->rb_alpha_control);
OUT_PKT0(ring, REG_A4XX_RB_STENCIL_CONTROL, 2);
OUT_RING(ring, zsa->rb_stencil_control);
OUT_RING(ring, zsa->rb_stencil_control2);
}
if (dirty & (FD_DIRTY_ZSA | FD_DIRTY_STENCIL_REF)) {
struct fd4_zsa_stateobj *zsa = fd4_zsa_stateobj(ctx->zsa);
struct pipe_stencil_ref *sr = &ctx->stencil_ref;
OUT_PKT0(ring, REG_A4XX_RB_STENCILREFMASK, 2);
OUT_RING(ring, zsa->rb_stencilrefmask |
A4XX_RB_STENCILREFMASK_STENCILREF(sr->ref_value[0]));
OUT_RING(ring, zsa->rb_stencilrefmask_bf |
A4XX_RB_STENCILREFMASK_BF_STENCILREF(sr->ref_value[1]));
}
if (dirty & (FD_DIRTY_ZSA | FD_DIRTY_PROG)) {
struct fd4_zsa_stateobj *zsa = fd4_zsa_stateobj(ctx->zsa);
bool fragz = fp->has_kill | fp->writes_pos;
OUT_PKT0(ring, REG_A4XX_RB_DEPTH_CONTROL, 1);
OUT_RING(ring, zsa->rb_depth_control |
COND(fragz, A4XX_RB_DEPTH_CONTROL_EARLY_Z_DISABLE));
/* maybe this register/bitfield needs a better name.. this
* appears to be just disabling early-z
*/
OUT_PKT0(ring, REG_A4XX_GRAS_ALPHA_CONTROL, 1);
OUT_RING(ring, zsa->gras_alpha_control |
COND(fragz, A4XX_GRAS_ALPHA_CONTROL_ALPHA_TEST_ENABLE));
}
if (dirty & FD_DIRTY_RASTERIZER) {
struct fd4_rasterizer_stateobj *rasterizer =
fd4_rasterizer_stateobj(ctx->rasterizer);
OUT_PKT0(ring, REG_A4XX_GRAS_SU_MODE_CONTROL, 1);
OUT_RING(ring, rasterizer->gras_su_mode_control |
A4XX_GRAS_SU_MODE_CONTROL_RENDERING_PASS);
OUT_PKT0(ring, REG_A4XX_GRAS_SU_POINT_MINMAX, 2);
OUT_RING(ring, rasterizer->gras_su_point_minmax);
OUT_RING(ring, rasterizer->gras_su_point_size);
OUT_PKT0(ring, REG_A4XX_GRAS_SU_POLY_OFFSET_SCALE, 2);
OUT_RING(ring, rasterizer->gras_su_poly_offset_scale);
OUT_RING(ring, rasterizer->gras_su_poly_offset_offset);
OUT_PKT0(ring, REG_A4XX_GRAS_CL_CLIP_CNTL, 1);
OUT_RING(ring, rasterizer->gras_cl_clip_cntl);
}
/* NOTE: since primitive_restart is not actually part of any
* state object, we need to make sure that we always emit
* PRIM_VTX_CNTL.. either that or be more clever and detect
* when it changes.
*/
if (emit->info) {
const struct pipe_draw_info *info = emit->info;
uint32_t val = fd4_rasterizer_stateobj(ctx->rasterizer)
->pc_prim_vtx_cntl;
if (info->indexed && info->primitive_restart)
val |= A4XX_PC_PRIM_VTX_CNTL_PRIMITIVE_RESTART;
val |= COND(vp->writes_psize, A4XX_PC_PRIM_VTX_CNTL_PSIZE);
if (fp->total_in > 0) {
uint32_t varout = align(fp->total_in, 16) / 16;
if (varout > 1)
varout = align(varout, 2);
val |= A4XX_PC_PRIM_VTX_CNTL_VAROUT(varout);
}
OUT_PKT0(ring, REG_A4XX_PC_PRIM_VTX_CNTL, 2);
OUT_RING(ring, val);
OUT_RING(ring, 0x12); /* XXX UNKNOWN_21C5 */
}
if (dirty & FD_DIRTY_SCISSOR) {
struct pipe_scissor_state *scissor = fd_context_get_scissor(ctx);
OUT_PKT0(ring, REG_A4XX_GRAS_SC_WINDOW_SCISSOR_BR, 2);
OUT_RING(ring, A4XX_GRAS_SC_WINDOW_SCISSOR_BR_X(scissor->maxx - 1) |
A4XX_GRAS_SC_WINDOW_SCISSOR_BR_Y(scissor->maxy - 1));
OUT_RING(ring, A4XX_GRAS_SC_WINDOW_SCISSOR_TL_X(scissor->minx) |
A4XX_GRAS_SC_WINDOW_SCISSOR_TL_Y(scissor->miny));
ctx->max_scissor.minx = MIN2(ctx->max_scissor.minx, scissor->minx);
ctx->max_scissor.miny = MIN2(ctx->max_scissor.miny, scissor->miny);
ctx->max_scissor.maxx = MAX2(ctx->max_scissor.maxx, scissor->maxx);
ctx->max_scissor.maxy = MAX2(ctx->max_scissor.maxy, scissor->maxy);
}
if (dirty & FD_DIRTY_VIEWPORT) {
fd_wfi(ctx, ring);
OUT_PKT0(ring, REG_A4XX_GRAS_CL_VPORT_XOFFSET_0, 6);
OUT_RING(ring, A4XX_GRAS_CL_VPORT_XOFFSET_0(ctx->viewport.translate[0]));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_XSCALE_0(ctx->viewport.scale[0]));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_YOFFSET_0(ctx->viewport.translate[1]));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_YSCALE_0(ctx->viewport.scale[1]));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_ZOFFSET_0(ctx->viewport.translate[2]));
OUT_RING(ring, A4XX_GRAS_CL_VPORT_ZSCALE_0(ctx->viewport.scale[2]));
}
if (dirty & FD_DIRTY_PROG)
fd4_program_emit(ring, emit);
if ((dirty & (FD_DIRTY_PROG | FD_DIRTY_CONSTBUF)) &&
/* evil hack to deal sanely with clear path: */
(emit->prog == &ctx->prog)) {
fd_wfi(ctx, ring);
emit_constants(ring, SB_VERT_SHADER,
&ctx->constbuf[PIPE_SHADER_VERTEX],
vp, emit->prog->dirty & FD_SHADER_DIRTY_VP);
if (!emit->key.binning_pass) {
emit_constants(ring, SB_FRAG_SHADER,
&ctx->constbuf[PIPE_SHADER_FRAGMENT],
fp, emit->prog->dirty & FD_SHADER_DIRTY_FP);
}
}
/* emit driver params every time */
if (emit->info && emit->prog == &ctx->prog) {
uint32_t vertex_params[4] = {
emit->info->indexed ? emit->info->index_bias : emit->info->start,
0,
0,
0
};
if (vp->constlen >= vp->first_driver_param + 4) {
fd4_emit_constant(ring, SB_VERT_SHADER,
(vp->first_driver_param + 4) * 4,
0, 4, vertex_params, NULL);
}
}
if ((dirty & FD_DIRTY_BLEND) && ctx->blend) {
struct fd4_blend_stateobj *blend = fd4_blend_stateobj(ctx->blend);
uint32_t i;
for (i = 0; i < 8; i++) {
OUT_PKT0(ring, REG_A4XX_RB_MRT_CONTROL(i), 1);
OUT_RING(ring, blend->rb_mrt[i].control);
OUT_PKT0(ring, REG_A4XX_RB_MRT_BLEND_CONTROL(i), 1);
OUT_RING(ring, blend->rb_mrt[i].blend_control);
}
OUT_PKT0(ring, REG_A4XX_RB_FS_OUTPUT, 1);
OUT_RING(ring, blend->rb_fs_output |
A4XX_RB_FS_OUTPUT_SAMPLE_MASK(0xffff));
}
if (dirty & FD_DIRTY_BLEND_COLOR) {
struct pipe_blend_color *bcolor = &ctx->blend_color;
OUT_PKT0(ring, REG_A4XX_RB_BLEND_RED, 4);
OUT_RING(ring, A4XX_RB_BLEND_RED_UINT(bcolor->color[0] * 255.0) |
A4XX_RB_BLEND_RED_FLOAT(bcolor->color[0]));
OUT_RING(ring, A4XX_RB_BLEND_GREEN_UINT(bcolor->color[1] * 255.0) |
A4XX_RB_BLEND_GREEN_FLOAT(bcolor->color[1]));
OUT_RING(ring, A4XX_RB_BLEND_BLUE_UINT(bcolor->color[2] * 255.0) |
A4XX_RB_BLEND_BLUE_FLOAT(bcolor->color[2]));
OUT_RING(ring, A4XX_RB_BLEND_ALPHA_UINT(bcolor->color[3] * 255.0) |
A4XX_RB_BLEND_ALPHA_FLOAT(bcolor->color[3]));
}
if (dirty & FD_DIRTY_VERTTEX) {
if (vp->has_samp)
emit_textures(ctx, ring, SB_VERT_TEX, &ctx->verttex);
else
dirty &= ~FD_DIRTY_VERTTEX;
}
if (dirty & FD_DIRTY_FRAGTEX) {
if (fp->has_samp)
emit_textures(ctx, ring, SB_FRAG_TEX, &ctx->fragtex);
else
dirty &= ~FD_DIRTY_FRAGTEX;
}
ctx->dirty &= ~dirty;
}
