#include "util/u_string.h"
#include "util/u_memory.h"
#include "util/u_inlines.h"
-#include "util/u_format.h"
+#include "util/format/u_format.h"
#include "util/bitset.h"
#include "freedreno_program.h"
#include "fd5_texture.h"
#include "fd5_format.h"
-static void
-delete_shader_stateobj(struct fd5_shader_stateobj *so)
-{
- ir3_shader_destroy(so->shader);
- free(so);
-}
-
-static struct fd5_shader_stateobj *
-create_shader_stateobj(struct pipe_context *pctx, const struct pipe_shader_state *cso,
- enum shader_t type)
-{
- struct fd_context *ctx = fd_context(pctx);
- struct ir3_compiler *compiler = ctx->screen->compiler;
- struct fd5_shader_stateobj *so = CALLOC_STRUCT(fd5_shader_stateobj);
- so->shader = ir3_shader_create(compiler, cso, type, &ctx->debug);
- return so;
-}
-
-static void *
-fd5_fp_state_create(struct pipe_context *pctx,
- const struct pipe_shader_state *cso)
-{
- return create_shader_stateobj(pctx, cso, SHADER_FRAGMENT);
-}
-
-static void
-fd5_fp_state_delete(struct pipe_context *pctx, void *hwcso)
-{
- struct fd5_shader_stateobj *so = hwcso;
- delete_shader_stateobj(so);
-}
+#include "ir3_cache.h"
-static void *
-fd5_vp_state_create(struct pipe_context *pctx,
- const struct pipe_shader_state *cso)
-{
- return create_shader_stateobj(pctx, cso, SHADER_VERTEX);
-}
-
-static void
-fd5_vp_state_delete(struct pipe_context *pctx, void *hwcso)
-{
- struct fd5_shader_stateobj *so = hwcso;
- delete_shader_stateobj(so);
-}
-
-static void
-emit_shader(struct fd_ringbuffer *ring, const struct ir3_shader_variant *so)
+void
+fd5_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;
+ enum a4xx_state_block sb = fd4_stage2shadersb(so->type);
+ enum a4xx_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;
+ src = SS4_DIRECT;
bin = fd_bo_map(so->bo);
} else {
sz = 0;
- src = 2; // enums different on a5xx..
+ src = SS4_INDIRECT;
bin = NULL;
}
- OUT_PKT7(ring, CP_LOAD_STATE, 3 + 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));
+ OUT_PKT7(ring, CP_LOAD_STATE4, 3 + sz);
+ OUT_RING(ring, CP_LOAD_STATE4_0_DST_OFF(0) |
+ CP_LOAD_STATE4_0_STATE_SRC(src) |
+ CP_LOAD_STATE4_0_STATE_BLOCK(sb) |
+ CP_LOAD_STATE4_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));
- OUT_RING(ring, CP_LOAD_STATE_2_EXT_SRC_ADDR_HI(0));
+ OUT_RING(ring, CP_LOAD_STATE4_1_EXT_SRC_ADDR(0) |
+ CP_LOAD_STATE4_1_STATE_TYPE(ST4_SHADER));
+ OUT_RING(ring, CP_LOAD_STATE4_2_EXT_SRC_ADDR_HI(0));
} else {
OUT_RELOC(ring, so->bo, 0,
- CP_LOAD_STATE_1_STATE_TYPE(ST_SHADER), 0);
+ CP_LOAD_STATE4_1_STATE_TYPE(ST4_SHADER), 0);
}
/* for how clever coverity is, it is sometimes rather dull, and
}
}
+/* Add any missing varyings needed for stream-out. Otherwise varyings not
+ * used by fragment shader will be stripped out.
+ */
+static void
+link_stream_out(struct ir3_shader_linkage *l, const struct ir3_shader_variant *v)
+{
+ const struct ir3_stream_output_info *strmout = &v->shader->stream_output;
+
+ /*
+ * First, any stream-out varyings not already in linkage map (ie. also
+ * consumed by frag shader) need to be added:
+ */
+ for (unsigned i = 0; i < strmout->num_outputs; i++) {
+ const struct ir3_stream_output *out = &strmout->output[i];
+ unsigned k = out->register_index;
+ unsigned compmask =
+ (1 << (out->num_components + out->start_component)) - 1;
+ unsigned idx, nextloc = 0;
+
+ /* psize/pos need to be the last entries in linkage map, and will
+ * get added link_stream_out, so skip over them:
+ */
+ if ((v->outputs[k].slot == VARYING_SLOT_PSIZ) ||
+ (v->outputs[k].slot == VARYING_SLOT_POS))
+ continue;
+
+ for (idx = 0; idx < l->cnt; idx++) {
+ if (l->var[idx].regid == v->outputs[k].regid)
+ break;
+ nextloc = MAX2(nextloc, l->var[idx].loc + 4);
+ }
+
+ /* add if not already in linkage map: */
+ if (idx == l->cnt)
+ ir3_link_add(l, v->outputs[k].regid, compmask, nextloc);
+
+ /* expand component-mask if needed, ie streaming out all components
+ * but frag shader doesn't consume all components:
+ */
+ if (compmask & ~l->var[idx].compmask) {
+ l->var[idx].compmask |= compmask;
+ l->max_loc = MAX2(l->max_loc,
+ l->var[idx].loc + util_last_bit(l->var[idx].compmask));
+ }
+ }
+}
+
+/* TODO maybe some of this we could pre-compute once rather than having
+ * so much draw-time logic?
+ */
+static void
+emit_stream_out(struct fd_ringbuffer *ring, const struct ir3_shader_variant *v,
+ struct ir3_shader_linkage *l)
+{
+ const struct ir3_stream_output_info *strmout = &v->shader->stream_output;
+ unsigned ncomp[PIPE_MAX_SO_BUFFERS] = {0};
+ unsigned prog[align(l->max_loc, 2) / 2];
+
+ memset(prog, 0, sizeof(prog));
+
+ for (unsigned i = 0; i < strmout->num_outputs; i++) {
+ const struct ir3_stream_output *out = &strmout->output[i];
+ unsigned k = out->register_index;
+ unsigned idx;
+
+ ncomp[out->output_buffer] += out->num_components;
+
+ /* linkage map sorted by order frag shader wants things, so
+ * a bit less ideal here..
+ */
+ for (idx = 0; idx < l->cnt; idx++)
+ if (l->var[idx].regid == v->outputs[k].regid)
+ break;
+
+ debug_assert(idx < l->cnt);
+
+ for (unsigned j = 0; j < out->num_components; j++) {
+ unsigned c = j + out->start_component;
+ unsigned loc = l->var[idx].loc + c;
+ unsigned off = j + out->dst_offset; /* in dwords */
+
+ if (loc & 1) {
+ prog[loc/2] |= A5XX_VPC_SO_PROG_B_EN |
+ A5XX_VPC_SO_PROG_B_BUF(out->output_buffer) |
+ A5XX_VPC_SO_PROG_B_OFF(off * 4);
+ } else {
+ prog[loc/2] |= A5XX_VPC_SO_PROG_A_EN |
+ A5XX_VPC_SO_PROG_A_BUF(out->output_buffer) |
+ A5XX_VPC_SO_PROG_A_OFF(off * 4);
+ }
+ }
+ }
+
+ OUT_PKT7(ring, CP_CONTEXT_REG_BUNCH, 12 + (2 * ARRAY_SIZE(prog)));
+ OUT_RING(ring, REG_A5XX_VPC_SO_BUF_CNTL);
+ OUT_RING(ring, A5XX_VPC_SO_BUF_CNTL_ENABLE |
+ COND(ncomp[0] > 0, A5XX_VPC_SO_BUF_CNTL_BUF0) |
+ COND(ncomp[1] > 0, A5XX_VPC_SO_BUF_CNTL_BUF1) |
+ COND(ncomp[2] > 0, A5XX_VPC_SO_BUF_CNTL_BUF2) |
+ COND(ncomp[3] > 0, A5XX_VPC_SO_BUF_CNTL_BUF3));
+ OUT_RING(ring, REG_A5XX_VPC_SO_NCOMP(0));
+ OUT_RING(ring, ncomp[0]);
+ OUT_RING(ring, REG_A5XX_VPC_SO_NCOMP(1));
+ OUT_RING(ring, ncomp[1]);
+ OUT_RING(ring, REG_A5XX_VPC_SO_NCOMP(2));
+ OUT_RING(ring, ncomp[2]);
+ OUT_RING(ring, REG_A5XX_VPC_SO_NCOMP(3));
+ OUT_RING(ring, ncomp[3]);
+ OUT_RING(ring, REG_A5XX_VPC_SO_CNTL);
+ OUT_RING(ring, A5XX_VPC_SO_CNTL_ENABLE);
+ for (unsigned i = 0; i < ARRAY_SIZE(prog); i++) {
+ OUT_RING(ring, REG_A5XX_VPC_SO_PROG);
+ OUT_RING(ring, prog[i]);
+ }
+}
+
struct stage {
const struct ir3_shader_variant *v;
const struct ir3_info *i;
if (s[i].v) {
s[i].i = &s[i].v->info;
/* constlen is in units of 4 * vec4: */
- s[i].constlen = align(s[i].v->constlen, 4) / 4;
+ assert(s[i].v->constlen % 4 == 0);
+ s[i].constlen = s[i].v->constlen / 4;
/* instrlen is already in units of 16 instr.. although
* probably we should ditch that and not make the compiler
* care about instruction group size of a3xx vs a5xx
}
void
-fd5_program_emit(struct fd_ringbuffer *ring, struct fd5_emit *emit,
- int nr, struct pipe_surface **bufs)
+fd5_program_emit(struct fd_context *ctx, struct fd_ringbuffer *ring,
+ struct fd5_emit *emit)
{
struct stage s[MAX_STAGES];
- uint32_t pos_regid, posz_regid, psize_regid, color_regid[8];
- uint32_t face_regid, coord_regid, zwcoord_regid;
- uint32_t vcoord_regid, vertex_regid, instance_regid;
+ uint32_t pos_regid, psize_regid, color_regid[8];
+ uint32_t face_regid, coord_regid, zwcoord_regid, samp_id_regid, samp_mask_regid;
+ uint32_t ij_regid[IJ_COUNT], vertex_regid, instance_regid;
+ enum a3xx_threadsize fssz;
+ uint8_t psize_loc = ~0;
int i, j;
- debug_assert(nr <= ARRAY_SIZE(color_regid));
-
- if (emit->key.binning_pass)
- nr = 0;
-
setup_stages(emit, s);
+ fssz = (s[FS].i->max_reg >= 24) ? TWO_QUADS : FOUR_QUADS;
+
pos_regid = ir3_find_output_regid(s[VS].v, VARYING_SLOT_POS);
- posz_regid = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DEPTH);
psize_regid = ir3_find_output_regid(s[VS].v, VARYING_SLOT_PSIZ);
- vertex_regid = ir3_find_output_regid(s[VS].v, SYSTEM_VALUE_VERTEX_ID_ZERO_BASE);
- instance_regid = ir3_find_output_regid(s[VS].v, SYSTEM_VALUE_INSTANCE_ID);
+ vertex_regid = ir3_find_sysval_regid(s[VS].v, SYSTEM_VALUE_VERTEX_ID_ZERO_BASE);
+ instance_regid = ir3_find_sysval_regid(s[VS].v, SYSTEM_VALUE_INSTANCE_ID);
if (s[FS].v->color0_mrt) {
color_regid[0] = color_regid[1] = color_regid[2] = color_regid[3] =
color_regid[7] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA7);
}
- /* TODO get these dynamically: */
- face_regid = s[FS].v->frag_face ? regid(0,0) : regid(63,0);
- coord_regid = s[FS].v->frag_coord ? regid(0,0) : regid(63,0);
- zwcoord_regid = s[FS].v->frag_coord ? regid(0,2) : regid(63,0);
- vcoord_regid = (s[FS].v->total_in > 0) ? regid(0,0) : regid(63,0);
+ samp_id_regid = ir3_find_sysval_regid(s[FS].v, SYSTEM_VALUE_SAMPLE_ID);
+ samp_mask_regid = ir3_find_sysval_regid(s[FS].v, SYSTEM_VALUE_SAMPLE_MASK_IN);
+ face_regid = ir3_find_sysval_regid(s[FS].v, SYSTEM_VALUE_FRONT_FACE);
+ coord_regid = ir3_find_sysval_regid(s[FS].v, SYSTEM_VALUE_FRAG_COORD);
+ zwcoord_regid = (coord_regid == regid(63,0)) ? regid(63,0) : (coord_regid + 2);
+ for (unsigned i = 0; i < ARRAY_SIZE(ij_regid); i++)
+ ij_regid[i] = ir3_find_sysval_regid(s[FS].v, SYSTEM_VALUE_BARYCENTRIC_PERSP_PIXEL + i);
/* we could probably divide this up into things that need to be
* emitted if frag-prog is dirty vs if vert-prog is dirty..
*/
- OUT_PKT4(ring, REG_A5XX_HLSQ_VS_CONTROL_REG, 5);
- OUT_RING(ring, A5XX_HLSQ_VS_CONTROL_REG_CONSTOBJECTOFFSET(s[VS].constoff) |
- A5XX_HLSQ_VS_CONTROL_REG_SHADEROBJOFFSET(s[VS].instroff) |
- COND(s[VS].v, A5XX_HLSQ_VS_CONTROL_REG_ENABLED));
- OUT_RING(ring, A5XX_HLSQ_FS_CONTROL_REG_CONSTOBJECTOFFSET(s[FS].constoff) |
- A5XX_HLSQ_FS_CONTROL_REG_SHADEROBJOFFSET(s[FS].instroff) |
- COND(s[FS].v, A5XX_HLSQ_FS_CONTROL_REG_ENABLED));
- OUT_RING(ring, A5XX_HLSQ_HS_CONTROL_REG_CONSTOBJECTOFFSET(s[HS].constoff) |
- A5XX_HLSQ_HS_CONTROL_REG_SHADEROBJOFFSET(s[HS].instroff) |
- COND(s[HS].v, A5XX_HLSQ_HS_CONTROL_REG_ENABLED));
- OUT_RING(ring, A5XX_HLSQ_DS_CONTROL_REG_CONSTOBJECTOFFSET(s[DS].constoff) |
- A5XX_HLSQ_DS_CONTROL_REG_SHADEROBJOFFSET(s[DS].instroff) |
- COND(s[DS].v, A5XX_HLSQ_DS_CONTROL_REG_ENABLED));
- OUT_RING(ring, A5XX_HLSQ_GS_CONTROL_REG_CONSTOBJECTOFFSET(s[GS].constoff) |
- A5XX_HLSQ_GS_CONTROL_REG_SHADEROBJOFFSET(s[GS].instroff) |
- COND(s[GS].v, A5XX_HLSQ_GS_CONTROL_REG_ENABLED));
+ OUT_PKT4(ring, REG_A5XX_HLSQ_VS_CONFIG, 5);
+ OUT_RING(ring, A5XX_HLSQ_VS_CONFIG_CONSTOBJECTOFFSET(s[VS].constoff) |
+ A5XX_HLSQ_VS_CONFIG_SHADEROBJOFFSET(s[VS].instroff) |
+ COND(s[VS].v, A5XX_HLSQ_VS_CONFIG_ENABLED));
+ OUT_RING(ring, A5XX_HLSQ_FS_CONFIG_CONSTOBJECTOFFSET(s[FS].constoff) |
+ A5XX_HLSQ_FS_CONFIG_SHADEROBJOFFSET(s[FS].instroff) |
+ COND(s[FS].v, A5XX_HLSQ_FS_CONFIG_ENABLED));
+ OUT_RING(ring, A5XX_HLSQ_HS_CONFIG_CONSTOBJECTOFFSET(s[HS].constoff) |
+ A5XX_HLSQ_HS_CONFIG_SHADEROBJOFFSET(s[HS].instroff) |
+ COND(s[HS].v, A5XX_HLSQ_HS_CONFIG_ENABLED));
+ OUT_RING(ring, A5XX_HLSQ_DS_CONFIG_CONSTOBJECTOFFSET(s[DS].constoff) |
+ A5XX_HLSQ_DS_CONFIG_SHADEROBJOFFSET(s[DS].instroff) |
+ COND(s[DS].v, A5XX_HLSQ_DS_CONFIG_ENABLED));
+ OUT_RING(ring, A5XX_HLSQ_GS_CONFIG_CONSTOBJECTOFFSET(s[GS].constoff) |
+ A5XX_HLSQ_GS_CONFIG_SHADEROBJOFFSET(s[GS].instroff) |
+ COND(s[GS].v, A5XX_HLSQ_GS_CONFIG_ENABLED));
OUT_PKT4(ring, REG_A5XX_HLSQ_CS_CONFIG, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT4(ring, REG_A5XX_HLSQ_VS_CNTL, 5);
- OUT_RING(ring, A5XX_HLSQ_VS_CNTL_INSTRLEN(s[VS].instrlen));
- OUT_RING(ring, A5XX_HLSQ_FS_CNTL_INSTRLEN(s[FS].instrlen));
- OUT_RING(ring, A5XX_HLSQ_HS_CNTL_INSTRLEN(s[HS].instrlen));
- OUT_RING(ring, A5XX_HLSQ_DS_CNTL_INSTRLEN(s[DS].instrlen));
- OUT_RING(ring, A5XX_HLSQ_GS_CNTL_INSTRLEN(s[GS].instrlen));
-
- OUT_PKT4(ring, REG_A5XX_SP_VS_CONTROL_REG, 5);
- OUT_RING(ring, A5XX_SP_VS_CONTROL_REG_CONSTOBJECTOFFSET(s[VS].constoff) |
- A5XX_SP_VS_CONTROL_REG_SHADEROBJOFFSET(s[VS].instroff) |
- COND(s[VS].v, A5XX_SP_VS_CONTROL_REG_ENABLED));
- OUT_RING(ring, A5XX_SP_FS_CONTROL_REG_CONSTOBJECTOFFSET(s[FS].constoff) |
- A5XX_SP_FS_CONTROL_REG_SHADEROBJOFFSET(s[FS].instroff) |
- COND(s[FS].v, A5XX_SP_FS_CONTROL_REG_ENABLED));
- OUT_RING(ring, A5XX_SP_HS_CONTROL_REG_CONSTOBJECTOFFSET(s[HS].constoff) |
- A5XX_SP_HS_CONTROL_REG_SHADEROBJOFFSET(s[HS].instroff) |
- COND(s[HS].v, A5XX_SP_HS_CONTROL_REG_ENABLED));
- OUT_RING(ring, A5XX_SP_DS_CONTROL_REG_CONSTOBJECTOFFSET(s[DS].constoff) |
- A5XX_SP_DS_CONTROL_REG_SHADEROBJOFFSET(s[DS].instroff) |
- COND(s[DS].v, A5XX_SP_DS_CONTROL_REG_ENABLED));
- OUT_RING(ring, A5XX_SP_GS_CONTROL_REG_CONSTOBJECTOFFSET(s[GS].constoff) |
- A5XX_SP_GS_CONTROL_REG_SHADEROBJOFFSET(s[GS].instroff) |
- COND(s[GS].v, A5XX_SP_GS_CONTROL_REG_ENABLED));
+ OUT_RING(ring, A5XX_HLSQ_VS_CNTL_INSTRLEN(s[VS].instrlen) |
+ COND(s[VS].v && s[VS].v->has_ssbo, A5XX_HLSQ_VS_CNTL_SSBO_ENABLE));
+ OUT_RING(ring, A5XX_HLSQ_FS_CNTL_INSTRLEN(s[FS].instrlen) |
+ COND(s[FS].v && s[FS].v->has_ssbo, A5XX_HLSQ_FS_CNTL_SSBO_ENABLE));
+ OUT_RING(ring, A5XX_HLSQ_HS_CNTL_INSTRLEN(s[HS].instrlen) |
+ COND(s[HS].v && s[HS].v->has_ssbo, A5XX_HLSQ_HS_CNTL_SSBO_ENABLE));
+ OUT_RING(ring, A5XX_HLSQ_DS_CNTL_INSTRLEN(s[DS].instrlen) |
+ COND(s[DS].v && s[DS].v->has_ssbo, A5XX_HLSQ_DS_CNTL_SSBO_ENABLE));
+ OUT_RING(ring, A5XX_HLSQ_GS_CNTL_INSTRLEN(s[GS].instrlen) |
+ COND(s[GS].v && s[GS].v->has_ssbo, A5XX_HLSQ_GS_CNTL_SSBO_ENABLE));
+
+ OUT_PKT4(ring, REG_A5XX_SP_VS_CONFIG, 5);
+ OUT_RING(ring, A5XX_SP_VS_CONFIG_CONSTOBJECTOFFSET(s[VS].constoff) |
+ A5XX_SP_VS_CONFIG_SHADEROBJOFFSET(s[VS].instroff) |
+ COND(s[VS].v, A5XX_SP_VS_CONFIG_ENABLED));
+ OUT_RING(ring, A5XX_SP_FS_CONFIG_CONSTOBJECTOFFSET(s[FS].constoff) |
+ A5XX_SP_FS_CONFIG_SHADEROBJOFFSET(s[FS].instroff) |
+ COND(s[FS].v, A5XX_SP_FS_CONFIG_ENABLED));
+ OUT_RING(ring, A5XX_SP_HS_CONFIG_CONSTOBJECTOFFSET(s[HS].constoff) |
+ A5XX_SP_HS_CONFIG_SHADEROBJOFFSET(s[HS].instroff) |
+ COND(s[HS].v, A5XX_SP_HS_CONFIG_ENABLED));
+ OUT_RING(ring, A5XX_SP_DS_CONFIG_CONSTOBJECTOFFSET(s[DS].constoff) |
+ A5XX_SP_DS_CONFIG_SHADEROBJOFFSET(s[DS].instroff) |
+ COND(s[DS].v, A5XX_SP_DS_CONFIG_ENABLED));
+ OUT_RING(ring, A5XX_SP_GS_CONFIG_CONSTOBJECTOFFSET(s[GS].constoff) |
+ A5XX_SP_GS_CONFIG_SHADEROBJOFFSET(s[GS].instroff) |
+ COND(s[GS].v, A5XX_SP_GS_CONFIG_ENABLED));
OUT_PKT4(ring, REG_A5XX_SP_CS_CONFIG, 1);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, s[GS].constlen); /* HLSQ_GS_CONSTLEN */
OUT_RING(ring, s[GS].instrlen); /* HLSQ_GS_INSTRLEN */
- OUT_PKT4(ring, REG_A5XX_HLSQ_CONTEXT_SWITCH_CS_SW_3, 2);
- OUT_RING(ring, 0x00000000); /* HLSQ_CONTEXT_SWITCH_CS_SW_3 */
- OUT_RING(ring, 0x00000000); /* HLSQ_CONTEXT_SWITCH_CS_SW_4 */
+ OUT_PKT4(ring, REG_A5XX_HLSQ_CS_CONSTLEN, 2);
+ OUT_RING(ring, 0x00000000); /* HLSQ_CS_CONSTLEN */
+ OUT_RING(ring, 0x00000000); /* HLSQ_CS_INSTRLEN */
OUT_PKT4(ring, REG_A5XX_SP_VS_CTRL_REG0, 1);
OUT_RING(ring, A5XX_SP_VS_CTRL_REG0_HALFREGFOOTPRINT(s[VS].i->max_half_reg + 1) |
A5XX_SP_VS_CTRL_REG0_FULLREGFOOTPRINT(s[VS].i->max_reg + 1) |
0x6 | /* XXX seems to be always set? */
- A5XX_SP_VS_CTRL_REG0_BRANCHSTACK(0x3) | // XXX need to figure this out somehow..
- COND(s[VS].v->has_samp, A5XX_SP_VS_CTRL_REG0_PIXLODENABLE));
+ A5XX_SP_VS_CTRL_REG0_BRANCHSTACK(s[VS].v->branchstack) |
+ COND(s[VS].v->need_pixlod, A5XX_SP_VS_CTRL_REG0_PIXLODENABLE));
struct ir3_shader_linkage l = {0};
- ir3_link_shaders(&l, s[VS].v, s[FS].v);
-
- BITSET_DECLARE(varbs, 128) = {0};
- uint32_t *varmask = (uint32_t *)varbs;
+ ir3_link_shaders(&l, s[VS].v, s[FS].v, true);
- for (i = 0; i < l.cnt; i++)
- for (j = 0; j < util_last_bit(l.var[i].compmask); j++)
- BITSET_SET(varbs, l.var[i].loc + j);
+ if ((s[VS].v->shader->stream_output.num_outputs > 0) &&
+ !emit->binning_pass)
+ link_stream_out(&l, s[VS].v);
OUT_PKT4(ring, REG_A5XX_VPC_VAR_DISABLE(0), 4);
- OUT_RING(ring, ~varmask[0]); /* VPC_VAR[0].DISABLE */
- OUT_RING(ring, ~varmask[1]); /* VPC_VAR[1].DISABLE */
- OUT_RING(ring, ~varmask[2]); /* VPC_VAR[2].DISABLE */
- OUT_RING(ring, ~varmask[3]); /* VPC_VAR[3].DISABLE */
+ OUT_RING(ring, ~l.varmask[0]); /* VPC_VAR[0].DISABLE */
+ OUT_RING(ring, ~l.varmask[1]); /* VPC_VAR[1].DISABLE */
+ OUT_RING(ring, ~l.varmask[2]); /* VPC_VAR[2].DISABLE */
+ OUT_RING(ring, ~l.varmask[3]); /* VPC_VAR[3].DISABLE */
/* a5xx appends pos/psize to end of the linkage map: */
if (pos_regid != regid(63,0))
ir3_link_add(&l, pos_regid, 0xf, l.max_loc);
- if (psize_regid != regid(63,0))
+ if (psize_regid != regid(63,0)) {
+ psize_loc = l.max_loc;
ir3_link_add(&l, psize_regid, 0x1, l.max_loc);
+ }
+
+ if ((s[VS].v->shader->stream_output.num_outputs > 0) &&
+ !emit->binning_pass) {
+ emit_stream_out(ring, s[VS].v, &l);
+
+ OUT_PKT4(ring, REG_A5XX_VPC_SO_OVERRIDE, 1);
+ OUT_RING(ring, 0x00000000);
+ } else {
+ OUT_PKT4(ring, REG_A5XX_VPC_SO_OVERRIDE, 1);
+ OUT_RING(ring, A5XX_VPC_SO_OVERRIDE_SO_DISABLE);
+ }
for (i = 0, j = 0; (i < 16) && (j < l.cnt); i++) {
uint32_t reg = 0;
OUT_RELOC(ring, s[VS].v->bo, 0, 0, 0); /* SP_VS_OBJ_START_LO/HI */
if (s[VS].instrlen)
- emit_shader(ring, s[VS].v);
+ fd5_emit_shader(ring, s[VS].v);
// TODO depending on other bits in this reg (if any) set somewhere else?
OUT_PKT4(ring, REG_A5XX_PC_PRIM_VTX_CNTL, 1);
OUT_RING(ring, COND(s[VS].v->writes_psize, A5XX_PC_PRIM_VTX_CNTL_PSIZE));
- if (emit->key.binning_pass) {
+ OUT_PKT4(ring, REG_A5XX_SP_PRIMITIVE_CNTL, 1);
+ OUT_RING(ring, A5XX_SP_PRIMITIVE_CNTL_VSOUT(l.cnt));
+
+ OUT_PKT4(ring, REG_A5XX_VPC_CNTL_0, 1);
+ OUT_RING(ring, A5XX_VPC_CNTL_0_STRIDE_IN_VPC(l.max_loc) |
+ COND(s[FS].v->total_in > 0, A5XX_VPC_CNTL_0_VARYING) |
+ COND(s[FS].v->fragcoord_compmask != 0, A5XX_VPC_CNTL_0_VARYING) |
+ 0x10000); // XXX
+
+ fd5_context(ctx)->max_loc = l.max_loc;
+
+ if (emit->binning_pass) {
OUT_PKT4(ring, REG_A5XX_SP_FS_OBJ_START_LO, 2);
OUT_RING(ring, 0x00000000); /* SP_FS_OBJ_START_LO */
OUT_RING(ring, 0x00000000); /* SP_FS_OBJ_START_HI */
} else {
- uint32_t stride_in_vpc = align(s[FS].v->total_in, 4) + 4;
-
- if (s[VS].v->writes_psize)
- stride_in_vpc++;
-
- // TODO if some of these other bits depend on something other than
- // program state we should probably move these next three regs:
-
- OUT_PKT4(ring, REG_A5XX_SP_PRIMITIVE_CNTL, 1);
- OUT_RING(ring, A5XX_SP_PRIMITIVE_CNTL_VSOUT(l.cnt));
-
- OUT_PKT4(ring, REG_A5XX_VPC_CNTL_0, 1);
- OUT_RING(ring, A5XX_VPC_CNTL_0_STRIDE_IN_VPC(stride_in_vpc) |
- COND(s[FS].v->total_in > 0, A5XX_VPC_CNTL_0_VARYING) |
- 0x10000); // XXX
-
- OUT_PKT4(ring, REG_A5XX_PC_PRIMITIVE_CNTL, 1);
- OUT_RING(ring, A5XX_PC_PRIMITIVE_CNTL_STRIDE_IN_VPC(stride_in_vpc) |
- 0x400); // XXX
-
OUT_PKT4(ring, REG_A5XX_SP_FS_OBJ_START_LO, 2);
OUT_RELOC(ring, s[FS].v->bo, 0, 0, 0); /* SP_FS_OBJ_START_LO/HI */
}
OUT_PKT4(ring, REG_A5XX_HLSQ_CONTROL_0_REG, 5);
- OUT_RING(ring, 0x00000881); /* XXX HLSQ_CONTROL_0 */
+ OUT_RING(ring, A5XX_HLSQ_CONTROL_0_REG_FSTHREADSIZE(fssz) |
+ A5XX_HLSQ_CONTROL_0_REG_CSTHREADSIZE(TWO_QUADS) |
+ 0x00000880); /* XXX HLSQ_CONTROL_0 */
OUT_RING(ring, A5XX_HLSQ_CONTROL_1_REG_PRIMALLOCTHRESHOLD(63));
OUT_RING(ring, A5XX_HLSQ_CONTROL_2_REG_FACEREGID(face_regid) |
- 0xfcfcfc00); /* XXX */
- OUT_RING(ring, A5XX_HLSQ_CONTROL_3_REG_FRAGCOORDXYREGID(vcoord_regid) |
- 0xfcfcfc00); /* XXX */
+ A5XX_HLSQ_CONTROL_2_REG_SAMPLEID(samp_id_regid) |
+ A5XX_HLSQ_CONTROL_2_REG_SAMPLEMASK(samp_mask_regid) |
+ A5XX_HLSQ_CONTROL_2_REG_SIZE(ij_regid[IJ_PERSP_SIZE]));
+ OUT_RING(ring,
+ A5XX_HLSQ_CONTROL_3_REG_IJ_PERSP_PIXEL(ij_regid[IJ_PERSP_PIXEL]) |
+ A5XX_HLSQ_CONTROL_3_REG_IJ_LINEAR_PIXEL(ij_regid[IJ_LINEAR_PIXEL]) |
+ A5XX_HLSQ_CONTROL_3_REG_IJ_PERSP_CENTROID(ij_regid[IJ_PERSP_CENTROID]) |
+ A5XX_HLSQ_CONTROL_3_REG_IJ_PERSP_CENTROID(ij_regid[IJ_LINEAR_CENTROID]));
OUT_RING(ring, A5XX_HLSQ_CONTROL_4_REG_XYCOORDREGID(coord_regid) |
A5XX_HLSQ_CONTROL_4_REG_ZWCOORDREGID(zwcoord_regid) |
- 0x0000fcfc); /* XXX */
+ A5XX_HLSQ_CONTROL_4_REG_IJ_PERSP_SAMPLE(ij_regid[IJ_PERSP_SAMPLE]) |
+ A5XX_HLSQ_CONTROL_4_REG_IJ_LINEAR_SAMPLE(ij_regid[IJ_LINEAR_SAMPLE]));
OUT_PKT4(ring, REG_A5XX_SP_FS_CTRL_REG0, 1);
OUT_RING(ring, COND(s[FS].v->total_in > 0, A5XX_SP_FS_CTRL_REG0_VARYING) |
- 0x4000e | /* XXX set pretty much everywhere */
+ COND(s[FS].v->fragcoord_compmask != 0, A5XX_SP_FS_CTRL_REG0_VARYING) |
+ 0x40006 | /* XXX set pretty much everywhere */
+ A5XX_SP_FS_CTRL_REG0_THREADSIZE(fssz) |
A5XX_SP_FS_CTRL_REG0_HALFREGFOOTPRINT(s[FS].i->max_half_reg + 1) |
A5XX_SP_FS_CTRL_REG0_FULLREGFOOTPRINT(s[FS].i->max_reg + 1) |
- A5XX_SP_FS_CTRL_REG0_BRANCHSTACK(0x3) | // XXX need to figure this out somehow..
- COND(s[FS].v->has_samp, A5XX_SP_FS_CTRL_REG0_PIXLODENABLE));
+ A5XX_SP_FS_CTRL_REG0_BRANCHSTACK(s[FS].v->branchstack) |
+ COND(s[FS].v->need_pixlod, A5XX_SP_FS_CTRL_REG0_PIXLODENABLE));
OUT_PKT4(ring, REG_A5XX_HLSQ_UPDATE_CNTL, 1);
OUT_RING(ring, 0x020fffff); /* XXX */
OUT_PKT4(ring, REG_A5XX_SP_SP_CNTL, 1);
OUT_RING(ring, 0x00000010); /* XXX */
+ /* XXX: missing enable bits for per-sample bary linear centroid and IJ_PERSP_SIZE
+ * (should be identical to a6xx)
+ */
+
OUT_PKT4(ring, REG_A5XX_GRAS_CNTL, 1);
- OUT_RING(ring, COND(s[FS].v->total_in > 0, A5XX_GRAS_CNTL_VARYING) |
- COND(s[FS].v->frag_coord, A5XX_GRAS_CNTL_XCOORD |
- A5XX_GRAS_CNTL_YCOORD |
- A5XX_GRAS_CNTL_ZCOORD |
- A5XX_GRAS_CNTL_WCOORD |
- A5XX_GRAS_CNTL_UNK3) |
- COND(s[FS].v->frag_face, A5XX_GRAS_CNTL_UNK3));
-
- OUT_PKT4(ring, REG_A5XX_RB_RENDER_CONTROL0, 3);
OUT_RING(ring,
- COND(s[FS].v->total_in > 0, A5XX_RB_RENDER_CONTROL0_VARYING) |
- COND(s[FS].v->frag_coord, A5XX_RB_RENDER_CONTROL0_XCOORD |
- A5XX_RB_RENDER_CONTROL0_YCOORD |
- A5XX_RB_RENDER_CONTROL0_ZCOORD |
- A5XX_RB_RENDER_CONTROL0_WCOORD |
- A5XX_RB_RENDER_CONTROL0_UNK3) |
- COND(s[FS].v->frag_face, A5XX_RB_RENDER_CONTROL0_UNK3));
-
+ CONDREG(ij_regid[IJ_PERSP_PIXEL], A5XX_GRAS_CNTL_IJ_PERSP_PIXEL) |
+ CONDREG(ij_regid[IJ_PERSP_CENTROID], A5XX_GRAS_CNTL_IJ_PERSP_CENTROID) |
+ COND(s[FS].v->fragcoord_compmask != 0,
+ A5XX_GRAS_CNTL_COORD_MASK(s[FS].v->fragcoord_compmask) |
+ A5XX_GRAS_CNTL_SIZE) |
+ COND(s[FS].v->frag_face, A5XX_GRAS_CNTL_SIZE) |
+ CONDREG(ij_regid[IJ_LINEAR_PIXEL], A5XX_GRAS_CNTL_SIZE));
+
+ OUT_PKT4(ring, REG_A5XX_RB_RENDER_CONTROL0, 2);
OUT_RING(ring,
- COND(s[FS].v->frag_face, A5XX_RB_RENDER_CONTROL1_FACENESS));
- OUT_RING(ring, A5XX_RB_FS_OUTPUT_CNTL_MRT(nr) |
- COND(s[FS].v->writes_pos, A5XX_RB_FS_OUTPUT_CNTL_FRAG_WRITES_Z));
-
- OUT_PKT4(ring, REG_A5XX_SP_FS_OUTPUT_CNTL, 9);
- OUT_RING(ring, A5XX_SP_FS_OUTPUT_CNTL_MRT(nr) |
- A5XX_SP_FS_OUTPUT_CNTL_DEPTH_REGID(posz_regid) |
- A5XX_SP_FS_OUTPUT_CNTL_SAMPLEMASK_REGID(regid(63, 0)));
+ CONDREG(ij_regid[IJ_PERSP_PIXEL], A5XX_RB_RENDER_CONTROL0_IJ_PERSP_PIXEL) |
+ CONDREG(ij_regid[IJ_PERSP_CENTROID], A5XX_RB_RENDER_CONTROL0_IJ_PERSP_CENTROID) |
+ COND(s[FS].v->fragcoord_compmask != 0,
+ A5XX_RB_RENDER_CONTROL0_COORD_MASK(s[FS].v->fragcoord_compmask) |
+ A5XX_RB_RENDER_CONTROL0_SIZE) |
+ COND(s[FS].v->frag_face, A5XX_RB_RENDER_CONTROL0_SIZE) |
+ CONDREG(ij_regid[IJ_LINEAR_PIXEL], A5XX_RB_RENDER_CONTROL0_SIZE));
+ OUT_RING(ring,
+ COND(samp_mask_regid != regid(63, 0),
+ A5XX_RB_RENDER_CONTROL1_SAMPLEMASK) |
+ COND(s[FS].v->frag_face, A5XX_RB_RENDER_CONTROL1_FACENESS) |
+ COND(samp_id_regid != regid(63, 0),
+ A5XX_RB_RENDER_CONTROL1_SAMPLEID));
+
+ OUT_PKT4(ring, REG_A5XX_SP_FS_OUTPUT_REG(0), 8);
for (i = 0; i < 8; i++) {
OUT_RING(ring, A5XX_SP_FS_OUTPUT_REG_REGID(color_regid[i]) |
- COND(emit->key.half_precision,
- A5XX_SP_FS_OUTPUT_REG_HALF_PRECISION));
+ COND(color_regid[i] & HALF_REG_ID, A5XX_SP_FS_OUTPUT_REG_HALF_PRECISION));
}
- if (emit->key.binning_pass) {
- OUT_PKT4(ring, REG_A5XX_VPC_PACK, 1);
- OUT_RING(ring, A5XX_VPC_PACK_NUMNONPOSVAR(0));
- } else {
+
+ OUT_PKT4(ring, REG_A5XX_VPC_PACK, 1);
+ OUT_RING(ring, A5XX_VPC_PACK_NUMNONPOSVAR(s[FS].v->total_in) |
+ A5XX_VPC_PACK_PSIZELOC(psize_loc));
+
+ if (!emit->binning_pass) {
uint32_t vinterp[8], vpsrepl[8];
memset(vinterp, 0, sizeof(vinterp));
}
}
- gl_varying_slot slot = s[FS].v->inputs[j].slot;
-
- /* since we don't enable PIPE_CAP_TGSI_TEXCOORD: */
- if (slot >= VARYING_SLOT_VAR0) {
- unsigned texmask = 1 << (slot - VARYING_SLOT_VAR0);
- /* Replace the .xy coordinates with S/T from the point sprite. Set
- * interpolation bits for .zw such that they become .01
+ bool coord_mode = emit->sprite_coord_mode;
+ if (ir3_point_sprite(s[FS].v, j, emit->sprite_coord_enable, &coord_mode)) {
+ /* mask is two 2-bit fields, where:
+ * '01' -> S
+ * '10' -> T
+ * '11' -> 1 - T (flip mode)
*/
- if (emit->sprite_coord_enable & texmask) {
- /* mask is two 2-bit fields, where:
- * '01' -> S
- * '10' -> T
- * '11' -> 1 - T (flip mode)
- */
- unsigned mask = emit->sprite_coord_mode ? 0b1101 : 0b1001;
- uint32_t loc = inloc;
- if (compmask & 0x1) {
- vpsrepl[loc / 16] |= ((mask >> 0) & 0x3) << ((loc % 16) * 2);
- loc++;
- }
- if (compmask & 0x2) {
- vpsrepl[loc / 16] |= ((mask >> 2) & 0x3) << ((loc % 16) * 2);
- loc++;
- }
- if (compmask & 0x4) {
- /* .z <- 0.0f */
- vinterp[loc / 16] |= 0b10 << ((loc % 16) * 2);
- loc++;
- }
- if (compmask & 0x8) {
- /* .w <- 1.0f */
- vinterp[loc / 16] |= 0b11 << ((loc % 16) * 2);
- loc++;
- }
+ unsigned mask = coord_mode ? 0b1101 : 0b1001;
+ uint32_t loc = inloc;
+ if (compmask & 0x1) {
+ vpsrepl[loc / 16] |= ((mask >> 0) & 0x3) << ((loc % 16) * 2);
+ loc++;
+ }
+ if (compmask & 0x2) {
+ vpsrepl[loc / 16] |= ((mask >> 2) & 0x3) << ((loc % 16) * 2);
+ loc++;
+ }
+ if (compmask & 0x4) {
+ /* .z <- 0.0f */
+ vinterp[loc / 16] |= 0b10 << ((loc % 16) * 2);
+ loc++;
+ }
+ if (compmask & 0x8) {
+ /* .w <- 1.0f */
+ vinterp[loc / 16] |= 0b11 << ((loc % 16) * 2);
+ loc++;
}
}
}
- OUT_PKT4(ring, REG_A5XX_VPC_PACK, 1);
- OUT_RING(ring, A5XX_VPC_PACK_NUMNONPOSVAR(s[FS].v->total_in) |
- (s[VS].v->writes_psize ? 0x0c00 : 0xff00)); // XXX
-
OUT_PKT4(ring, REG_A5XX_VPC_VARYING_INTERP_MODE(0), 8);
for (i = 0; i < 8; i++)
OUT_RING(ring, vinterp[i]); /* VPC_VARYING_INTERP[i].MODE */
OUT_RING(ring, vpsrepl[i]); /* VPC_VARYING_PS_REPL[i] */
}
- if (!emit->key.binning_pass)
+ if (!emit->binning_pass)
if (s[FS].instrlen)
- emit_shader(ring, s[FS].v);
+ fd5_emit_shader(ring, s[FS].v);
OUT_PKT4(ring, REG_A5XX_VFD_CONTROL_1, 5);
OUT_RING(ring, A5XX_VFD_CONTROL_1_REGID4VTX(vertex_regid) |
A5XX_VFD_CONTROL_1_REGID4INST(instance_regid) |
- 0xfc);
+ 0xfc0000);
OUT_RING(ring, 0x0000fcfc); /* VFD_CONTROL_2 */
OUT_RING(ring, 0x0000fcfc); /* VFD_CONTROL_3 */
OUT_RING(ring, 0x000000fc); /* VFD_CONTROL_4 */
void
fd5_prog_init(struct pipe_context *pctx)
{
- pctx->create_fs_state = fd5_fp_state_create;
- pctx->delete_fs_state = fd5_fp_state_delete;
-
- pctx->create_vs_state = fd5_vp_state_create;
- pctx->delete_vs_state = fd5_vp_state_delete;
-
+ ir3_prog_init(pctx);
fd_prog_init(pctx);
}