#include "util/u_memory.h"
#include "util/u_helpers.h"
#include "util/u_format.h"
+#include "util/u_viewport.h"
#include "freedreno_resource.h"
+#include "freedreno_query_hw.h"
#include "fd3_emit.h"
#include "fd3_blend.h"
#include "fd3_program.h"
#include "fd3_rasterizer.h"
#include "fd3_texture.h"
-#include "fd3_util.h"
+#include "fd3_format.h"
#include "fd3_zsa.h"
+static const enum adreno_state_block sb[] = {
+ [SHADER_VERTEX] = SB_VERT_SHADER,
+ [SHADER_FRAGMENT] = SB_FRAG_SHADER,
+};
+
/* 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,
+fd3_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 = SS_INDIRECT;
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_STATE_BLOCK(sb[type]) |
CP_LOAD_STATE_0_NUM_UNIT(sizedwords/2));
if (prsc) {
struct fd_bo *bo = fd_resource(prsc)->bo;
}
static void
-emit_constants(struct fd_ringbuffer *ring,
- enum adreno_state_block sb,
- struct fd_constbuf_stateobj *constbuf,
- struct fd3_shader_variant *shader)
+fd3_emit_const_bo(struct fd_ringbuffer *ring, enum shader_t type, boolean write,
+ uint32_t regid, uint32_t num, struct pipe_resource **prscs, uint32_t *offsets)
{
- uint32_t enabled_mask = constbuf->enabled_mask;
- uint32_t first_immediate;
- uint32_t base = 0;
- 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;
-
- /* in particular, with binning shader and a unneeded consts no
- * longer referenced, 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
- */
- first_immediate = MIN2(shader->first_immediate, shader->constlen);
-
- /* 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));
-
- /* gallium could leave const buffers bound above what the
- * current shader uses.. don't let that confuse us.
- */
- if (base >= (4 * first_immediate))
- break;
-
- if (constbuf->dirty_mask & (1 << index)) {
- /* and even if the start of the const buffer is before
- * first_immediate, the end may not be:
- */
- size = MIN2(size, (4 * first_immediate) - base);
- fd3_emit_constant(ring, sb, base,
- cb->buffer_offset, size,
- cb->user_buffer, cb->buffer);
- constbuf->dirty_mask &= ~(1 << index);
- }
+ uint32_t i;
- base += size;
- enabled_mask &= ~(1 << index);
- }
+ debug_assert((regid % 4) == 0);
+ debug_assert((num % 4) == 0);
- /* emit shader immediates: */
- if (shader) {
- for (i = 0; i < shader->immediates_count; i++) {
- base = 4 * (shader->first_immediate + i);
- if (base >= (4 * shader->constlen))
- break;
- fd3_emit_constant(ring, sb, base,
- 0, 4, shader->immediates[i].val, NULL);
+ OUT_PKT3(ring, CP_LOAD_STATE, 2 + num);
+ OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(regid/2) |
+ CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
+ CP_LOAD_STATE_0_STATE_BLOCK(sb[type]) |
+ CP_LOAD_STATE_0_NUM_UNIT(num/2));
+ OUT_RING(ring, CP_LOAD_STATE_1_EXT_SRC_ADDR(0) |
+ CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS));
+
+ for (i = 0; i < num; i++) {
+ if (prscs[i]) {
+ if (write) {
+ OUT_RELOCW(ring, fd_resource(prscs[i])->bo, offsets[i], 0, 0);
+ } else {
+ OUT_RELOC(ring, fd_resource(prscs[i])->bo, offsets[i], 0, 0);
+ }
+ } else {
+ OUT_RING(ring, 0xbad00000 | (i << 16));
}
}
}
#define BASETABLE_SZ A3XX_MAX_MIP_LEVELS
static void
-emit_textures(struct fd_ringbuffer *ring,
- enum adreno_state_block sb,
- struct fd_texture_stateobj *tex)
+emit_textures(struct fd_context *ctx, struct fd_ringbuffer *ring,
+ enum adreno_state_block sb, struct fd_texture_stateobj *tex)
{
static const unsigned tex_off[] = {
[SB_VERT_TEX] = VERT_TEX_OFF,
[SB_VERT_TEX] = SB_VERT_MIPADDR,
[SB_FRAG_TEX] = SB_FRAG_MIPADDR,
};
+ static const uint32_t bcolor_reg[] = {
+ [SB_VERT_TEX] = REG_A3XX_TPL1_TP_VS_BORDER_COLOR_BASE_ADDR,
+ [SB_FRAG_TEX] = REG_A3XX_TPL1_TP_FS_BORDER_COLOR_BASE_ADDR,
+ };
+ struct fd3_context *fd3_ctx = fd3_context(ctx);
+ bool needs_border = false;
unsigned i, j;
if (tex->num_samplers > 0) {
const struct fd3_sampler_stateobj *sampler = tex->samplers[i] ?
fd3_sampler_stateobj(tex->samplers[i]) :
&dummy_sampler;
+
OUT_RING(ring, sampler->texsamp0);
OUT_RING(ring, sampler->texsamp1);
+
+ needs_border |= sampler->needs_border;
}
}
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 fd3_pipe_sampler_view dummy_view = {};
+ static const struct fd3_pipe_sampler_view dummy_view = {
+ .base.target = PIPE_TEXTURE_1D, /* anything !PIPE_BUFFER */
+ .base.u.tex.first_level = 1,
+ };
const struct fd3_pipe_sampler_view *view = tex->textures[i] ?
fd3_pipe_sampler_view(tex->textures[i]) :
&dummy_view;
- struct fd_resource *rsc = view->tex_resource;
-
- for (j = 0; j < view->mipaddrs; j++) {
- struct fd_resource_slice *slice = fd_resource_slice(rsc, j);
- OUT_RELOC(ring, rsc->bo, slice->offset, 0, 0);
+ struct fd_resource *rsc = fd_resource(view->base.texture);
+ if (rsc && rsc->base.b.target == PIPE_BUFFER) {
+ OUT_RELOC(ring, rsc->bo, view->base.u.buf.offset, 0, 0);
+ j = 1;
+ } else {
+ unsigned start = fd_sampler_first_level(&view->base);
+ unsigned end = fd_sampler_last_level(&view->base);
+
+ for (j = 0; j < (end - start + 1); j++) {
+ struct fd_resource_slice *slice =
+ fd_resource_slice(rsc, j + start);
+ OUT_RELOC(ring, rsc->bo, slice->offset, 0, 0);
+ }
}
/* pad the remaining entries w/ null: */
}
}
}
+
+ if (needs_border) {
+ unsigned off;
+ void *ptr;
+
+ u_upload_alloc(fd3_ctx->border_color_uploader,
+ 0, BORDER_COLOR_UPLOAD_SIZE,
+ BORDER_COLOR_UPLOAD_SIZE, &off,
+ &fd3_ctx->border_color_buf,
+ &ptr);
+
+ fd_setup_border_colors(tex, ptr, tex_off[sb]);
+
+ OUT_PKT0(ring, bcolor_reg[sb], 1);
+ OUT_RELOC(ring, fd_resource(fd3_ctx->border_color_buf)->bo, off, 0, 0);
+
+ u_upload_unmap(fd3_ctx->border_color_uploader);
+ }
}
/* emit texture state for mem->gmem restore operation.. eventually it would
* be good to get rid of this and use normal CSO/etc state for more of these
* special cases, but for now the compiler is not sufficient..
+ *
+ * Also, for using normal state, not quite sure how to handle the special
+ * case format (fd3_gmem_restore_format()) stuff for restoring depth/stencil.
*/
void
-fd3_emit_gmem_restore_tex(struct fd_ringbuffer *ring, struct pipe_surface *psurf)
+fd3_emit_gmem_restore_tex(struct fd_ringbuffer *ring,
+ struct pipe_surface **psurf,
+ int bufs)
{
- struct fd_resource *rsc = fd_resource(psurf->texture);
- enum pipe_format format = fd3_gmem_restore_format(psurf->format);
+ int i, j;
/* output sampler state: */
- OUT_PKT3(ring, CP_LOAD_STATE, 4);
+ OUT_PKT3(ring, CP_LOAD_STATE, 2 + 2 * bufs);
OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(FRAG_TEX_OFF) |
CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
CP_LOAD_STATE_0_STATE_BLOCK(SB_FRAG_TEX) |
- CP_LOAD_STATE_0_NUM_UNIT(1));
+ CP_LOAD_STATE_0_NUM_UNIT(bufs));
OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_SHADER) |
CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
- OUT_RING(ring, A3XX_TEX_SAMP_0_XY_MAG(A3XX_TEX_NEAREST) |
- A3XX_TEX_SAMP_0_XY_MIN(A3XX_TEX_NEAREST) |
- A3XX_TEX_SAMP_0_WRAP_S(A3XX_TEX_CLAMP_TO_EDGE) |
- A3XX_TEX_SAMP_0_WRAP_T(A3XX_TEX_CLAMP_TO_EDGE) |
- A3XX_TEX_SAMP_0_WRAP_R(A3XX_TEX_REPEAT));
- OUT_RING(ring, 0x00000000);
+ for (i = 0; i < bufs; i++) {
+ OUT_RING(ring, A3XX_TEX_SAMP_0_XY_MAG(A3XX_TEX_NEAREST) |
+ A3XX_TEX_SAMP_0_XY_MIN(A3XX_TEX_NEAREST) |
+ A3XX_TEX_SAMP_0_WRAP_S(A3XX_TEX_CLAMP_TO_EDGE) |
+ A3XX_TEX_SAMP_0_WRAP_T(A3XX_TEX_CLAMP_TO_EDGE) |
+ A3XX_TEX_SAMP_0_WRAP_R(A3XX_TEX_REPEAT));
+ OUT_RING(ring, 0x00000000);
+ }
/* emit texture state: */
- OUT_PKT3(ring, CP_LOAD_STATE, 6);
+ OUT_PKT3(ring, CP_LOAD_STATE, 2 + 4 * bufs);
OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(FRAG_TEX_OFF) |
CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
CP_LOAD_STATE_0_STATE_BLOCK(SB_FRAG_TEX) |
- CP_LOAD_STATE_0_NUM_UNIT(1));
+ CP_LOAD_STATE_0_NUM_UNIT(bufs));
OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS) |
CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
- OUT_RING(ring, A3XX_TEX_CONST_0_FMT(fd3_pipe2tex(psurf->format)) |
- A3XX_TEX_CONST_0_TYPE(A3XX_TEX_2D) |
- fd3_tex_swiz(format, PIPE_SWIZZLE_RED, PIPE_SWIZZLE_GREEN,
- PIPE_SWIZZLE_BLUE, PIPE_SWIZZLE_ALPHA));
- OUT_RING(ring, A3XX_TEX_CONST_1_FETCHSIZE(TFETCH_DISABLE) |
- A3XX_TEX_CONST_1_WIDTH(psurf->width) |
- A3XX_TEX_CONST_1_HEIGHT(psurf->height));
- OUT_RING(ring, A3XX_TEX_CONST_2_PITCH(rsc->slices[0].pitch * rsc->cpp) |
- A3XX_TEX_CONST_2_INDX(0));
- OUT_RING(ring, 0x00000000);
+ for (i = 0; i < bufs; i++) {
+ if (!psurf[i]) {
+ OUT_RING(ring, A3XX_TEX_CONST_0_TYPE(A3XX_TEX_2D) |
+ A3XX_TEX_CONST_0_SWIZ_X(A3XX_TEX_ONE) |
+ A3XX_TEX_CONST_0_SWIZ_Y(A3XX_TEX_ONE) |
+ A3XX_TEX_CONST_0_SWIZ_Z(A3XX_TEX_ONE) |
+ A3XX_TEX_CONST_0_SWIZ_W(A3XX_TEX_ONE));
+ OUT_RING(ring, 0x00000000);
+ OUT_RING(ring, A3XX_TEX_CONST_2_INDX(BASETABLE_SZ * i));
+ OUT_RING(ring, 0x00000000);
+ continue;
+ }
+
+ struct fd_resource *rsc = fd_resource(psurf[i]->texture);
+ enum pipe_format format = fd3_gmem_restore_format(psurf[i]->format);
+ /* The restore blit_zs shader expects stencil in sampler 0, and depth
+ * in sampler 1
+ */
+ if (rsc->stencil && i == 0) {
+ rsc = rsc->stencil;
+ format = fd3_gmem_restore_format(rsc->base.b.format);
+ }
+
+ /* note: PIPE_BUFFER disallowed for surfaces */
+ unsigned lvl = psurf[i]->u.tex.level;
+ struct fd_resource_slice *slice = fd_resource_slice(rsc, lvl);
+
+ debug_assert(psurf[i]->u.tex.first_layer == psurf[i]->u.tex.last_layer);
+
+ OUT_RING(ring, A3XX_TEX_CONST_0_FMT(fd3_pipe2tex(format)) |
+ A3XX_TEX_CONST_0_TYPE(A3XX_TEX_2D) |
+ fd3_tex_swiz(format, PIPE_SWIZZLE_X, PIPE_SWIZZLE_Y,
+ PIPE_SWIZZLE_Z, PIPE_SWIZZLE_W));
+ OUT_RING(ring, A3XX_TEX_CONST_1_FETCHSIZE(TFETCH_DISABLE) |
+ A3XX_TEX_CONST_1_WIDTH(psurf[i]->width) |
+ A3XX_TEX_CONST_1_HEIGHT(psurf[i]->height));
+ OUT_RING(ring, A3XX_TEX_CONST_2_PITCH(slice->pitch * rsc->cpp) |
+ A3XX_TEX_CONST_2_INDX(BASETABLE_SZ * i));
+ OUT_RING(ring, 0x00000000);
+ }
/* emit mipaddrs: */
- OUT_PKT3(ring, CP_LOAD_STATE, 3);
+ OUT_PKT3(ring, CP_LOAD_STATE, 2 + BASETABLE_SZ * bufs);
OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(BASETABLE_SZ * FRAG_TEX_OFF) |
CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
CP_LOAD_STATE_0_STATE_BLOCK(SB_FRAG_MIPADDR) |
- CP_LOAD_STATE_0_NUM_UNIT(1));
+ CP_LOAD_STATE_0_NUM_UNIT(BASETABLE_SZ * bufs));
OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS) |
CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
- OUT_RELOC(ring, rsc->bo, 0, 0, 0);
+ for (i = 0; i < bufs; i++) {
+ if (psurf[i]) {
+ struct fd_resource *rsc = fd_resource(psurf[i]->texture);
+ /* Matches above logic for blit_zs shader */
+ if (rsc->stencil && i == 0)
+ rsc = rsc->stencil;
+ unsigned lvl = psurf[i]->u.tex.level;
+ uint32_t offset = fd_resource_offset(rsc, lvl, psurf[i]->u.tex.first_layer);
+ OUT_RELOC(ring, rsc->bo, offset, 0, 0);
+ } else {
+ OUT_RING(ring, 0x00000000);
+ }
+
+ /* pad the remaining entries w/ null: */
+ for (j = 1; j < BASETABLE_SZ; j++) {
+ OUT_RING(ring, 0x00000000);
+ }
+ }
}
void
-fd3_emit_vertex_bufs(struct fd_ringbuffer *ring,
- struct fd3_shader_variant *vp,
- struct fd3_vertex_buf *vbufs, uint32_t n)
+fd3_emit_vertex_bufs(struct fd_ringbuffer *ring, struct fd3_emit *emit)
{
- uint32_t i, j, last = 0;
+ int32_t i, j, last = -1;
uint32_t total_in = 0;
-
- n = MIN2(n, vp->inputs_count);
-
- for (i = 0; i < n; i++)
- if (vp->inputs[i].compmask)
+ const struct fd_vertex_state *vtx = emit->vtx;
+ const struct ir3_shader_variant *vp = fd3_emit_get_vp(emit);
+ unsigned vertex_regid = regid(63, 0);
+ unsigned instance_regid = regid(63, 0);
+ unsigned vtxcnt_regid = regid(63, 0);
+
+ /* Note that sysvals come *after* normal inputs: */
+ for (i = 0; i < vp->inputs_count; i++) {
+ if (!vp->inputs[i].compmask)
+ continue;
+ if (vp->inputs[i].sysval) {
+ switch(vp->inputs[i].slot) {
+ case SYSTEM_VALUE_BASE_VERTEX:
+ /* handled elsewhere */
+ break;
+ case SYSTEM_VALUE_VERTEX_ID_ZERO_BASE:
+ vertex_regid = vp->inputs[i].regid;
+ break;
+ case SYSTEM_VALUE_INSTANCE_ID:
+ instance_regid = vp->inputs[i].regid;
+ break;
+ case SYSTEM_VALUE_VERTEX_CNT:
+ vtxcnt_regid = vp->inputs[i].regid;
+ break;
+ default:
+ unreachable("invalid system value");
+ break;
+ }
+ } else if (i < vtx->vtx->num_elements) {
last = i;
+ }
+ }
for (i = 0, j = 0; i <= last; i++) {
+ assert(!vp->inputs[i].sysval);
if (vp->inputs[i].compmask) {
- struct pipe_resource *prsc = vbufs[i].prsc;
- struct fd_resource *rsc = fd_resource(prsc);
- enum pipe_format pfmt = vbufs[i].format;
+ struct pipe_vertex_element *elem = &vtx->vtx->pipe[i];
+ const struct pipe_vertex_buffer *vb =
+ &vtx->vertexbuf.vb[elem->vertex_buffer_index];
+ struct fd_resource *rsc = fd_resource(vb->buffer);
+ enum pipe_format pfmt = elem->src_format;
enum a3xx_vtx_fmt fmt = fd3_pipe2vtx(pfmt);
- bool switchnext = (i != last);
+ bool switchnext = (i != last) ||
+ (vertex_regid != regid(63, 0)) ||
+ (instance_regid != regid(63, 0)) ||
+ (vtxcnt_regid != regid(63, 0));
+ bool isint = util_format_is_pure_integer(pfmt);
uint32_t fs = util_format_get_blocksize(pfmt);
debug_assert(fmt != ~0);
OUT_PKT0(ring, REG_A3XX_VFD_FETCH(j), 2);
OUT_RING(ring, A3XX_VFD_FETCH_INSTR_0_FETCHSIZE(fs - 1) |
- A3XX_VFD_FETCH_INSTR_0_BUFSTRIDE(vbufs[i].stride) |
+ A3XX_VFD_FETCH_INSTR_0_BUFSTRIDE(vb->stride) |
COND(switchnext, A3XX_VFD_FETCH_INSTR_0_SWITCHNEXT) |
A3XX_VFD_FETCH_INSTR_0_INDEXCODE(j) |
- A3XX_VFD_FETCH_INSTR_0_STEPRATE(1));
- OUT_RELOC(ring, rsc->bo, vbufs[i].offset, 0, 0);
+ COND(elem->instance_divisor, A3XX_VFD_FETCH_INSTR_0_INSTANCED) |
+ A3XX_VFD_FETCH_INSTR_0_STEPRATE(MAX2(1, elem->instance_divisor)));
+ OUT_RELOC(ring, rsc->bo, vb->buffer_offset + elem->src_offset, 0, 0);
OUT_PKT0(ring, REG_A3XX_VFD_DECODE_INSTR(j), 1);
OUT_RING(ring, A3XX_VFD_DECODE_INSTR_CONSTFILL |
A3XX_VFD_DECODE_INSTR_REGID(vp->inputs[i].regid) |
A3XX_VFD_DECODE_INSTR_SHIFTCNT(fs) |
A3XX_VFD_DECODE_INSTR_LASTCOMPVALID |
+ COND(isint, A3XX_VFD_DECODE_INSTR_INT) |
COND(switchnext, A3XX_VFD_DECODE_INSTR_SWITCHNEXT));
total_in += vp->inputs[i].ncomp;
}
}
+ /* hw doesn't like to be configured for zero vbo's, it seems: */
+ if (last < 0) {
+ /* just recycle the shader bo, we just need to point to *something*
+ * valid:
+ */
+ struct fd_bo *dummy_vbo = vp->bo;
+ bool switchnext = (vertex_regid != regid(63, 0)) ||
+ (instance_regid != regid(63, 0)) ||
+ (vtxcnt_regid != regid(63, 0));
+
+ OUT_PKT0(ring, REG_A3XX_VFD_FETCH(0), 2);
+ OUT_RING(ring, A3XX_VFD_FETCH_INSTR_0_FETCHSIZE(0) |
+ A3XX_VFD_FETCH_INSTR_0_BUFSTRIDE(0) |
+ COND(switchnext, A3XX_VFD_FETCH_INSTR_0_SWITCHNEXT) |
+ A3XX_VFD_FETCH_INSTR_0_INDEXCODE(0) |
+ A3XX_VFD_FETCH_INSTR_0_STEPRATE(1));
+ OUT_RELOC(ring, dummy_vbo, 0, 0, 0);
+
+ OUT_PKT0(ring, REG_A3XX_VFD_DECODE_INSTR(0), 1);
+ OUT_RING(ring, A3XX_VFD_DECODE_INSTR_CONSTFILL |
+ A3XX_VFD_DECODE_INSTR_WRITEMASK(0x1) |
+ A3XX_VFD_DECODE_INSTR_FORMAT(VFMT_8_UNORM) |
+ A3XX_VFD_DECODE_INSTR_SWAP(XYZW) |
+ A3XX_VFD_DECODE_INSTR_REGID(regid(0,0)) |
+ A3XX_VFD_DECODE_INSTR_SHIFTCNT(1) |
+ A3XX_VFD_DECODE_INSTR_LASTCOMPVALID |
+ COND(switchnext, A3XX_VFD_DECODE_INSTR_SWITCHNEXT));
+
+ total_in = 1;
+ j = 1;
+ }
+
OUT_PKT0(ring, REG_A3XX_VFD_CONTROL_0, 2);
OUT_RING(ring, A3XX_VFD_CONTROL_0_TOTALATTRTOVS(total_in) |
A3XX_VFD_CONTROL_0_PACKETSIZE(2) |
A3XX_VFD_CONTROL_0_STRMDECINSTRCNT(j) |
A3XX_VFD_CONTROL_0_STRMFETCHINSTRCNT(j));
OUT_RING(ring, A3XX_VFD_CONTROL_1_MAXSTORAGE(1) | // XXX
- A3XX_VFD_CONTROL_1_REGID4VTX(regid(63,0)) |
- A3XX_VFD_CONTROL_1_REGID4INST(regid(63,0)));
+ A3XX_VFD_CONTROL_1_REGID4VTX(vertex_regid) |
+ A3XX_VFD_CONTROL_1_REGID4INST(instance_regid));
+
+ OUT_PKT0(ring, REG_A3XX_VFD_VS_THREADING_THRESHOLD, 1);
+ OUT_RING(ring, A3XX_VFD_VS_THREADING_THRESHOLD_REGID_THRESHOLD(15) |
+ A3XX_VFD_VS_THREADING_THRESHOLD_REGID_VTXCNT(vtxcnt_regid));
}
void
fd3_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
- struct fd_program_stateobj *prog, uint32_t dirty,
- struct fd3_shader_key key)
+ struct fd3_emit *emit)
{
- struct fd3_shader_variant *vp;
- struct fd3_shader_variant *fp;
-
- fp = fd3_shader_variant(prog->fp, key);
- vp = fd3_shader_variant(prog->vp, key);
+ const struct ir3_shader_variant *vp = fd3_emit_get_vp(emit);
+ const struct ir3_shader_variant *fp = fd3_emit_get_fp(emit);
+ uint32_t dirty = emit->dirty;
emit_marker(ring, 5);
A3XX_RB_MSAA_CONTROL_SAMPLE_MASK(ctx->sample_mask));
}
- if ((dirty & (FD_DIRTY_ZSA | FD_DIRTY_PROG)) && !key.binning_pass) {
- uint32_t val = fd3_zsa_stateobj(ctx->zsa)->rb_render_control;
+ if ((dirty & (FD_DIRTY_ZSA | FD_DIRTY_PROG | FD_DIRTY_BLEND_DUAL)) &&
+ !emit->key.binning_pass) {
+ uint32_t val = fd3_zsa_stateobj(ctx->zsa)->rb_render_control |
+ fd3_blend_stateobj(ctx->blend)->rb_render_control;
val |= COND(fp->frag_face, A3XX_RB_RENDER_CONTROL_FACENESS);
val |= COND(fp->frag_coord, A3XX_RB_RENDER_CONTROL_XCOORD |
*/
OUT_PKT0(ring, REG_A3XX_RB_RENDER_CONTROL, 1);
- OUT_RINGP(ring, val, &fd3_context(ctx)->rbrc_patches);
+ OUT_RINGP(ring, val, &ctx->batch->rbrc_patches);
}
if (dirty & (FD_DIRTY_ZSA | FD_DIRTY_STENCIL_REF)) {
A3XX_RB_STENCILREFMASK_BF_STENCILREF(sr->ref_value[1]));
}
- if (dirty & (FD_DIRTY_ZSA | FD_DIRTY_PROG)) {
+ if (dirty & (FD_DIRTY_ZSA | FD_DIRTY_RASTERIZER | FD_DIRTY_PROG)) {
uint32_t val = fd3_zsa_stateobj(ctx->zsa)->rb_depth_control;
if (fp->writes_pos) {
val |= A3XX_RB_DEPTH_CONTROL_FRAG_WRITES_Z;
val |= A3XX_RB_DEPTH_CONTROL_EARLY_Z_DISABLE;
}
+ if (fp->has_kill) {
+ val |= A3XX_RB_DEPTH_CONTROL_EARLY_Z_DISABLE;
+ }
+ if (!ctx->rasterizer->depth_clip) {
+ val |= A3XX_RB_DEPTH_CONTROL_Z_CLAMP_ENABLE;
+ }
OUT_PKT0(ring, REG_A3XX_RB_DEPTH_CONTROL, 1);
OUT_RING(ring, val);
}
val |= COND(fp->writes_pos, A3XX_GRAS_CL_CLIP_CNTL_ZCLIP_DISABLE);
val |= COND(fp->frag_coord, A3XX_GRAS_CL_CLIP_CNTL_ZCOORD |
A3XX_GRAS_CL_CLIP_CNTL_WCOORD);
+ /* TODO only use if prog doesn't use clipvertex/clipdist */
+ val |= A3XX_GRAS_CL_CLIP_CNTL_NUM_USER_CLIP_PLANES(
+ MIN2(util_bitcount(ctx->rasterizer->clip_plane_enable), 6));
OUT_PKT0(ring, REG_A3XX_GRAS_CL_CLIP_CNTL, 1);
OUT_RING(ring, val);
}
- if (dirty & (FD_DIRTY_RASTERIZER | FD_DIRTY_PROG)) {
+ if (dirty & (FD_DIRTY_RASTERIZER | FD_DIRTY_UCP)) {
+ uint32_t planes = ctx->rasterizer->clip_plane_enable;
+ int count = 0;
+
+ while (planes && count < 6) {
+ int i = ffs(planes) - 1;
+
+ planes &= ~(1U << i);
+ fd_wfi(ctx->batch, ring);
+ OUT_PKT0(ring, REG_A3XX_GRAS_CL_USER_PLANE(count++), 4);
+ OUT_RING(ring, fui(ctx->ucp.ucp[i][0]));
+ OUT_RING(ring, fui(ctx->ucp.ucp[i][1]));
+ OUT_RING(ring, fui(ctx->ucp.ucp[i][2]));
+ OUT_RING(ring, fui(ctx->ucp.ucp[i][3]));
+ }
+ }
+
+ /* 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 = fd3_rasterizer_stateobj(ctx->rasterizer)
->pc_prim_vtx_cntl;
- if (!key.binning_pass) {
+ if (!emit->key.binning_pass) {
uint32_t stride_in_vpc = align(fp->total_in, 4) / 4;
if (stride_in_vpc > 0)
stride_in_vpc = MAX2(stride_in_vpc, 2);
val |= A3XX_PC_PRIM_VTX_CNTL_STRIDE_IN_VPC(stride_in_vpc);
}
+ if (info->indexed && info->primitive_restart) {
+ val |= A3XX_PC_PRIM_VTX_CNTL_PRIMITIVE_RESTART;
+ }
+
val |= COND(vp->writes_psize, A3XX_PC_PRIM_VTX_CNTL_PSIZE);
OUT_PKT0(ring, REG_A3XX_PC_PRIM_VTX_CNTL, 1);
OUT_RING(ring, A3XX_GRAS_SC_WINDOW_SCISSOR_BR_X(scissor->maxx - 1) |
A3XX_GRAS_SC_WINDOW_SCISSOR_BR_Y(scissor->maxy - 1));
- 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);
+ 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) {
- fd_wfi(ctx, ring);
+ fd_wfi(ctx->batch, ring);
OUT_PKT0(ring, REG_A3XX_GRAS_CL_VPORT_XOFFSET, 6);
OUT_RING(ring, A3XX_GRAS_CL_VPORT_XOFFSET(ctx->viewport.translate[0] - 0.5));
OUT_RING(ring, A3XX_GRAS_CL_VPORT_XSCALE(ctx->viewport.scale[0]));
OUT_RING(ring, A3XX_GRAS_CL_VPORT_ZSCALE(ctx->viewport.scale[2]));
}
- if (dirty & FD_DIRTY_PROG) {
- fd3_program_emit(ring, prog, key);
+ if (dirty & (FD_DIRTY_VIEWPORT | FD_DIRTY_RASTERIZER | FD_DIRTY_FRAMEBUFFER)) {
+ float zmin, zmax;
+ int depth = 24;
+ if (ctx->batch->framebuffer.zsbuf) {
+ depth = util_format_get_component_bits(
+ pipe_surface_format(ctx->batch->framebuffer.zsbuf),
+ UTIL_FORMAT_COLORSPACE_ZS, 0);
+ }
+ util_viewport_zmin_zmax(&ctx->viewport, ctx->rasterizer->clip_halfz,
+ &zmin, &zmax);
+
+ OUT_PKT0(ring, REG_A3XX_RB_Z_CLAMP_MIN, 2);
+ if (depth == 32) {
+ OUT_RING(ring, (uint32_t)(zmin * 0xffffffff));
+ OUT_RING(ring, (uint32_t)(zmax * 0xffffffff));
+ } else if (depth == 16) {
+ OUT_RING(ring, (uint32_t)(zmin * 0xffff));
+ OUT_RING(ring, (uint32_t)(zmax * 0xffff));
+ } else {
+ OUT_RING(ring, (uint32_t)(zmin * 0xffffff));
+ OUT_RING(ring, (uint32_t)(zmax * 0xffffff));
+ }
}
- if ((dirty & (FD_DIRTY_PROG | FD_DIRTY_CONSTBUF)) &&
- /* evil hack to deal sanely with clear path: */
- (prog == &ctx->prog)) {
- emit_constants(ring, SB_VERT_SHADER,
- &ctx->constbuf[PIPE_SHADER_VERTEX],
- (prog->dirty & FD_SHADER_DIRTY_VP) ? vp : NULL);
- if (!key.binning_pass) {
- emit_constants(ring, SB_FRAG_SHADER,
- &ctx->constbuf[PIPE_SHADER_FRAGMENT],
- (prog->dirty & FD_SHADER_DIRTY_FP) ? fp : NULL);
- }
+ if (dirty & (FD_DIRTY_PROG | FD_DIRTY_FRAMEBUFFER | FD_DIRTY_BLEND_DUAL)) {
+ struct pipe_framebuffer_state *pfb = &ctx->batch->framebuffer;
+ int nr_cbufs = pfb->nr_cbufs;
+ if (fd3_blend_stateobj(ctx->blend)->rb_render_control &
+ A3XX_RB_RENDER_CONTROL_DUAL_COLOR_IN_ENABLE)
+ nr_cbufs++;
+ fd3_program_emit(ring, emit, nr_cbufs, pfb->cbufs);
}
- if ((dirty & FD_DIRTY_BLEND) && ctx->blend) {
+ /* TODO we should not need this or fd_wfi() before emit_constants():
+ */
+ OUT_PKT3(ring, CP_EVENT_WRITE, 1);
+ OUT_RING(ring, HLSQ_FLUSH);
+
+ if (emit->prog == &ctx->prog) { /* evil hack to deal sanely with clear path */
+ ir3_emit_consts(vp, ring, ctx, emit->info, dirty);
+ if (!emit->key.binning_pass)
+ ir3_emit_consts(fp, ring, ctx, emit->info, dirty);
+ }
+
+ if (dirty & (FD_DIRTY_BLEND | FD_DIRTY_FRAMEBUFFER)) {
struct fd3_blend_stateobj *blend = fd3_blend_stateobj(ctx->blend);
uint32_t i;
for (i = 0; i < ARRAY_SIZE(blend->rb_mrt); i++) {
+ enum pipe_format format =
+ pipe_surface_format(ctx->batch->framebuffer.cbufs[i]);
+ const struct util_format_description *desc =
+ util_format_description(format);
+ bool is_float = util_format_is_float(format);
+ bool is_int = util_format_is_pure_integer(format);
+ bool has_alpha = util_format_has_alpha(format);
+ uint32_t control = blend->rb_mrt[i].control;
+ uint32_t blend_control = blend->rb_mrt[i].blend_control_alpha;
+
+ if (is_int) {
+ control &= (A3XX_RB_MRT_CONTROL_COMPONENT_ENABLE__MASK |
+ A3XX_RB_MRT_CONTROL_DITHER_MODE__MASK);
+ control |= A3XX_RB_MRT_CONTROL_ROP_CODE(ROP_COPY);
+ }
+
+ if (format == PIPE_FORMAT_NONE)
+ control &= ~A3XX_RB_MRT_CONTROL_COMPONENT_ENABLE__MASK;
+
+ if (has_alpha) {
+ blend_control |= blend->rb_mrt[i].blend_control_rgb;
+ } else {
+ blend_control |= blend->rb_mrt[i].blend_control_no_alpha_rgb;
+ control &= ~A3XX_RB_MRT_CONTROL_BLEND2;
+ }
+
+ if (format && util_format_get_component_bits(
+ format, UTIL_FORMAT_COLORSPACE_RGB, 0) < 8) {
+ const struct pipe_rt_blend_state *rt;
+ if (ctx->blend->independent_blend_enable)
+ rt = &ctx->blend->rt[i];
+ else
+ rt = &ctx->blend->rt[0];
+
+ if (!util_format_colormask_full(desc, rt->colormask))
+ control |= A3XX_RB_MRT_CONTROL_READ_DEST_ENABLE;
+ }
+
OUT_PKT0(ring, REG_A3XX_RB_MRT_CONTROL(i), 1);
- OUT_RING(ring, blend->rb_mrt[i].control);
+ OUT_RING(ring, control);
OUT_PKT0(ring, REG_A3XX_RB_MRT_BLEND_CONTROL(i), 1);
- OUT_RING(ring, blend->rb_mrt[i].blend_control);
+ OUT_RING(ring, blend_control |
+ COND(!is_float, A3XX_RB_MRT_BLEND_CONTROL_CLAMP_ENABLE));
}
}
A3XX_RB_BLEND_ALPHA_FLOAT(bcolor->color[3]));
}
+ if (dirty & (FD_DIRTY_VERTTEX | FD_DIRTY_FRAGTEX))
+ fd_wfi(ctx->batch, ring);
if (dirty & FD_DIRTY_VERTTEX) {
if (vp->has_samp)
- emit_textures(ring, SB_VERT_TEX, &ctx->verttex);
+ 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(ring, SB_FRAG_TEX, &ctx->fragtex);
+ emit_textures(ctx, ring, SB_FRAG_TEX, &ctx->fragtex);
else
dirty &= ~FD_DIRTY_FRAGTEX;
}
* state, there could have been a context switch between ioctls):
*/
void
-fd3_emit_restore(struct fd_context *ctx)
+fd3_emit_restore(struct fd_batch *batch, struct fd_ringbuffer *ring)
{
+ struct fd_context *ctx = batch->ctx;
struct fd3_context *fd3_ctx = fd3_context(ctx);
- struct fd_ringbuffer *ring = ctx->ring;
int i;
if (ctx->screen->gpu_id == 320) {
OUT_RING(ring, 0x00000000);
}
- fd_wfi(ctx, ring);
+ fd_wfi(batch, ring);
OUT_PKT3(ring, CP_INVALIDATE_STATE, 1);
OUT_RING(ring, 0x00007fff);
OUT_RING(ring, A3XX_HLSQ_CONST_FSPRESV_RANGE_REG_STARTENTRY(0) |
A3XX_HLSQ_CONST_FSPRESV_RANGE_REG_ENDENTRY(0));
- 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);
+ fd3_emit_cache_flush(batch, ring);
OUT_PKT0(ring, REG_A3XX_GRAS_CL_CLIP_CNTL, 1);
OUT_RING(ring, 0x00000000); /* GRAS_CL_CLIP_CNTL */
OUT_PKT0(ring, REG_A3XX_PC_VSTREAM_CONTROL, 1);
OUT_RING(ring, 0x00000000);
- fd_event_write(ctx, ring, CACHE_FLUSH);
+ fd_event_write(batch, ring, CACHE_FLUSH);
- /* probably only really needed on a320: */
- OUT_PKT3(ring, CP_DRAW_INDX, 3);
- OUT_RING(ring, 0x00000000);
- OUT_RING(ring, DRAW(1, DI_SRC_SEL_AUTO_INDEX,
- INDEX_SIZE_IGN, IGNORE_VISIBILITY));
- OUT_RING(ring, 0); /* NumIndices */
+ if (is_a3xx_p0(ctx->screen)) {
+ OUT_PKT3(ring, CP_DRAW_INDX, 3);
+ OUT_RING(ring, 0x00000000);
+ OUT_RING(ring, DRAW(1, DI_SRC_SEL_AUTO_INDEX,
+ INDEX_SIZE_IGN, IGNORE_VISIBILITY, 0));
+ OUT_RING(ring, 0); /* NumIndices */
+ }
OUT_PKT3(ring, CP_NOP, 4);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000000);
- fd_wfi(ctx, ring);
+ fd_wfi(batch, ring);
- ctx->needs_rb_fbd = true;
+ fd_hw_query_enable(batch, ring);
+}
+
+static void
+fd3_emit_ib(struct fd_ringbuffer *ring, struct fd_ringbuffer *target)
+{
+ __OUT_IB(ring, true, target);
+}
+
+void
+fd3_emit_init(struct pipe_context *pctx)
+{
+ struct fd_context *ctx = fd_context(pctx);
+ ctx->emit_const = fd3_emit_const;
+ ctx->emit_const_bo = fd3_emit_const_bo;
+ ctx->emit_ib = fd3_emit_ib;
}
projects / mesa.git / blobdiff