nv50: handle all and more system values
authorChristoph Bumiller <e0425955@student.tuwien.ac.at>
Sat, 16 Jan 2010 14:39:32 +0000 (15:39 +0100)
committerChristoph Bumiller <e0425955@student.tuwien.ac.at>
Sat, 16 Jan 2010 17:07:31 +0000 (18:07 +0100)
src/gallium/drivers/nv50/nv50_program.c
src/gallium/drivers/nv50/nv50_program.h

index 6647f759b5abb428f7da2d417004077d6cc568e4..acd695b39bcde27e7c33969d1d4e3c13d45f0d20 100644 (file)
@@ -138,6 +138,8 @@ struct nv50_pc {
        int immd_nr;
        struct nv50_reg **addr;
        int addr_nr;
+       struct nv50_reg *sysval;
+       int sysval_nr;
 
        struct nv50_reg *temp_temp[16];
        struct nv50_program_exec *temp_temp_exec[16];
@@ -2281,6 +2283,10 @@ tgsi_dst(struct nv50_pc *pc, int c, const struct tgsi_full_dst_register *dst)
        }
        case TGSI_FILE_NULL:
                return NULL;
+       case TGSI_FILE_SYSTEM_VALUE:
+               assert(pc->sysval[dst->Register.Index].type == P_RESULT);
+               assert(c == 0);
+               return &pc->sysval[dst->Register.Index];
        default:
                break;
        }
@@ -2351,6 +2357,10 @@ tgsi_src(struct nv50_pc *pc, int chan, const struct tgsi_full_src_register *src,
                        r = pc->addr[src->Register.Index * 4 + c];
                        assert(r);
                        break;
+               case TGSI_FILE_SYSTEM_VALUE:
+                       assert(c == 0);
+                       r = &pc->sysval[src->Register.Index];
+                       break;
                default:
                        assert(0);
                        break;
@@ -3454,17 +3464,22 @@ load_interpolant(struct nv50_pc *pc, struct nv50_reg *reg)
  * value of 0 for back-facing, and 0xffffffff for front-facing.
  */
 static void
-load_frontfacing(struct nv50_pc *pc, struct nv50_reg *a)
+load_frontfacing(struct nv50_pc *pc, struct nv50_reg *sv)
 {
-       struct nv50_reg *one = alloc_immd(pc, 1.0f);
+       struct nv50_reg *temp = alloc_temp(pc, NULL);
+       int r_pred = 0;
 
-       assert(a->rhw == -1);
-       alloc_reg(pc, a); /* do this before rhw is set */
-       a->rhw = 255;
-       load_interpolant(pc, a);
-       emit_bitop2(pc, a, a, one, TGSI_OPCODE_AND);
+       temp->rhw = 255;
+       emit_interp(pc, temp, NULL, INTERP_FLAT);
 
-       FREE(one);
+       emit_cvt(pc, sv, temp, r_pred, CVT_ABS | CVT_F32_S32);
+
+       emit_not(pc, temp, temp);
+       set_pred(pc, 0x2, r_pred, pc->p->exec_tail);
+       emit_cvt(pc, sv, temp, -1, CVT_F32_S32);
+       set_pred(pc, 0x2, r_pred, pc->p->exec_tail);
+
+       free_temp(pc, temp);
 }
 
 static void
@@ -3491,7 +3506,7 @@ nv50_program_tx_prep(struct nv50_pc *pc)
        struct tgsi_parse_context tp;
        struct nv50_program *p = pc->p;
        boolean ret = FALSE;
-       unsigned i, c, flat_nr = 0;
+       unsigned i, c, instance_id, vertex_id, flat_nr = 0;
 
        tgsi_parse_init(&tp, pc->p->pipe.tokens);
        while (!tgsi_parse_end_of_tokens(&tp)) {
@@ -3577,6 +3592,37 @@ nv50_program_tx_prep(struct nv50_pc *pc)
                                        pc->interp_mode[i] = mode;
                        }
                                break;
+                       case TGSI_FILE_SYSTEM_VALUE:
+                               assert(d->Declaration.Semantic);
+                               switch (d->Semantic.Name) {
+                               case TGSI_SEMANTIC_FACE:
+                                       assert(p->type == PIPE_SHADER_FRAGMENT);
+                                       load_frontfacing(pc,
+                                                        &pc->sysval[first]);
+                                       break;
+                               case TGSI_SEMANTIC_INSTANCEID:
+                                       assert(p->type == PIPE_SHADER_VERTEX);
+                                       instance_id = first;
+                                       p->cfg.regs[0] |= (1 << 4);
+                                       break;
+                               case TGSI_SEMANTIC_PRIMID:
+                                       assert(p->type != PIPE_SHADER_VERTEX);
+                                       p->cfg.prim_id = first;
+                                       break;
+                                       /*
+                               case TGSI_SEMANTIC_PRIMIDIN:
+                                       assert(p->type == PIPE_SHADER_GEOMETRY);
+                                       pc->sysval[first].hw = 6;
+                                       p->cfg.regs[0] |= (1 << 8);
+                                       break;
+                               case TGSI_SEMANTIC_VERTEXID:
+                                       assert(p->type == PIPE_SHADER_VERTEX);
+                                       vertex_id = first;
+                                       p->cfg.regs[0] |= (1 << 12) | (1 << 0);
+                                       break;
+                                       */
+                               }
+                               break;
                        case TGSI_FILE_ADDRESS:
                        case TGSI_FILE_CONSTANT:
                        case TGSI_FILE_SAMPLER:
@@ -3620,6 +3666,10 @@ nv50_program_tx_prep(struct nv50_pc *pc)
                                        p->cfg.attr[i / 32] |= 1 << (i % 32);
                                }
                        }
+                       if (p->cfg.regs[0] & (1 << 0))
+                               pc->sysval[vertex_id].hw = rid++;
+                       if (p->cfg.regs[0] & (1 << 4))
+                               pc->sysval[instance_id].hw = rid++;
                }
 
                for (i = 0, rid = 0; i < pc->result_nr; ++i) {
@@ -3634,6 +3684,12 @@ nv50_program_tx_prep(struct nv50_pc *pc)
                                p->cfg.out[i].mask |= 1 << c;
                        }
                }
+               if (p->cfg.prim_id < 0x40) {
+                       /* GP has to write to PrimitiveID */
+                       ctor_reg(&pc->sysval[p->cfg.prim_id],
+                                P_RESULT, p->cfg.prim_id, rid);
+                       p->cfg.prim_id = rid++;
+               }
 
                for (c = 0; c < 2; ++c)
                        if (p->cfg.two_side[c].hw < 0x40)
@@ -3725,6 +3781,14 @@ nv50_program_tx_prep(struct nv50_pc *pc)
                if (n < 0x80)
                        p->cfg.regs[0] += n;
 
+               if (p->cfg.prim_id < 0x40) {
+                       pc->sysval[p->cfg.prim_id].rhw = rid++;
+                       emit_interp(pc, &pc->sysval[p->cfg.prim_id], NULL,
+                                   INTERP_FLAT);
+                       /* increase FP_INTERPOLANT_CTRL_COUNT */
+                       p->cfg.regs[1] += 1;
+               }
+
                /* Initialize FP results:
                 * FragDepth is always first TGSI and last hw output
                 */
@@ -3778,6 +3842,8 @@ free_nv50_pc(struct nv50_pc *pc)
                FREE(pc->attr);
        if (pc->temp)
                FREE(pc->temp);
+       if (pc->sysval)
+               FREE(pc->sysval);
        if (pc->insn_pos)
                FREE(pc->insn_pos);
 
@@ -3814,11 +3880,13 @@ ctor_nv50_pc(struct nv50_pc *pc, struct nv50_program *p)
        pc->param_nr = p->info.file_max[TGSI_FILE_CONSTANT] + 1;
        pc->addr_nr = p->info.file_max[TGSI_FILE_ADDRESS] + 1;
        assert(pc->addr_nr <= 2);
+       pc->sysval_nr = p->info.file_max[TGSI_FILE_SYSTEM_VALUE] + 1;
 
        p->cfg.high_temp = 4;
 
        p->cfg.two_side[0].hw = 0x40;
        p->cfg.two_side[1].hw = 0x40;
+       p->cfg.prim_id = 0x40;
 
        p->cfg.edgeflag_in = pc->edgeflag_out = 0xff;
 
@@ -3849,6 +3917,7 @@ ctor_nv50_pc(struct nv50_pc *pc, struct nv50_program *p)
 
                p->cfg.psiz = 0x80;
                p->cfg.clpd = 0x80;
+               p->cfg.prim_id = 0x80;
                p->cfg.out_nr = pc->result_nr;
                p->cfg.in_nr = pc->attr_nr;
 
@@ -3919,6 +3988,15 @@ ctor_nv50_pc(struct nv50_pc *pc, struct nv50_program *p)
        for (i = 0; i < NV50_SU_MAX_ADDR; ++i)
                ctor_reg(&pc->r_addr[i], P_ADDR, -1, i + 1);
 
+       if (pc->sysval_nr) {
+               pc->sysval = CALLOC(pc->sysval_nr, sizeof(struct nv50_reg *));
+               if (!pc->sysval)
+                       return FALSE;
+               /* will only ever use SYSTEM_VALUE[i].x (hopefully) */
+               for (i = 0; i < pc->sysval_nr; ++i)
+                       ctor_reg(&pc->sysval[i], rtype[0], i, -1);
+       }
+
        return TRUE;
 }
 
@@ -4362,7 +4440,7 @@ nv50_fp_linkage_validate(struct nv50_context *nv50)
        struct nouveau_stateobj *so;
        struct nv50_sreg4 dummy;
        int i, n, c, m = 0;
-       uint32_t map[16], lin[4], reg[5], pcrd[8];
+       uint32_t map[16], lin[4], reg[6], pcrd[8];
        uint8_t zval = 0x40;
 
        if (nv50->geomprog) {
@@ -4375,6 +4453,7 @@ nv50_fp_linkage_validate(struct nv50_context *nv50)
        reg[1] = 0x00000004; /* low and high clip distance map ids */
        reg[2] = 0x00000000; /* layer index map id (disabled, GP only) */
        reg[3] = 0x00000000; /* point size map id & enable */
+       reg[5] = 0x00000000; /* primitive ID map slot */
        reg[0] = fp->cfg.regs[0]; /* colour semantic reg */
        reg[4] = fp->cfg.regs[1]; /* interpolant info */
 
@@ -4421,6 +4500,13 @@ nv50_fp_linkage_validate(struct nv50_context *nv50)
                                  (n < vp->cfg.out_nr) ?
                                  &vp->cfg.out[n] : &dummy);
        }
+       /* PrimitiveID either is replaced by the system value, or
+        * written by the geometry shader into an output register
+        */
+       if (fp->cfg.prim_id < 0x40) {
+               map[m / 4] |= vp->cfg.prim_id << ((m % 4) * 8);
+               reg[5] = m++;
+       }
 
        if (nv50->rasterizer->pipe.point_size_per_vertex) {
                map[m / 4] |= vp->cfg.psiz << ((m % 4) * 8);
@@ -4428,7 +4514,7 @@ nv50_fp_linkage_validate(struct nv50_context *nv50)
        }
 
        /* now fill the stateobj (at most 28 so_data)  */
-       so = so_new(8, 56, 0);
+       so = so_new(10, 54, 0);
 
        n = (m + 3) / 4;
        assert(m <= 32);
@@ -4438,6 +4524,12 @@ nv50_fp_linkage_validate(struct nv50_context *nv50)
                so_method(so, tesla, NV50TCL_GP_RESULT_MAP(0), n);
                so_datap (so, map, n);
        } else {
+               so_method(so, tesla, NV50TCL_VP_GP_BUILTIN_ATTR_EN, 1);
+               so_data  (so, vp->cfg.regs[0]);
+
+               so_method(so, tesla, NV50TCL_MAP_SEMANTIC_4, 1);
+               so_data  (so, reg[5]);
+
                so_method(so, tesla, NV50TCL_VP_RESULT_MAP_SIZE, 1);
                so_data  (so, m);
                so_method(so, tesla, NV50TCL_VP_RESULT_MAP(0), n);
@@ -4518,7 +4610,10 @@ nv50_gp_linkage_validate(struct nv50_context *nv50)
 
        m = construct_vp_gp_mapping(map, m, vp, gp);
 
-       so = so_new(2, 14, 0);
+       so = so_new(3, 24 - 3, 0);
+
+       so_method(so, tesla, NV50TCL_VP_GP_BUILTIN_ATTR_EN, 1);
+       so_data  (so, vp->cfg.regs[0] | gp->cfg.regs[0]);
 
        assert(m <= 32);
        so_method(so, tesla, NV50TCL_VP_RESULT_MAP_SIZE, 1);
index e35900aa8d0844ebe77db1d32fd0ff15843cb4f0..1e3ad6bff058d0baab1bb215401faae4a5dc3537 100644 (file)
@@ -66,6 +66,9 @@ struct nv50_program {
                uint8_t clpd, clpd_nr;
                uint8_t psiz;
                uint8_t edgeflag_in;
+
+               /* FP & GP only */
+               uint8_t prim_id;
        } cfg;
 };