diff options
Diffstat (limited to 'src/gallium/drivers/nv50')
-rw-r--r-- | src/gallium/drivers/nv50/nv50_miptree.c | 21 | ||||
-rw-r--r-- | src/gallium/drivers/nv50/nv50_program.c | 111 | ||||
-rw-r--r-- | src/gallium/drivers/nv50/nv50_program.h | 3 | ||||
-rw-r--r-- | src/gallium/drivers/nv50/nv50_state.c | 4 | ||||
-rw-r--r-- | src/gallium/drivers/nv50/nv50_state_validate.c | 8 | ||||
-rw-r--r-- | src/gallium/drivers/nv50/nv50_tex.c | 4 | ||||
-rw-r--r-- | src/gallium/drivers/nv50/nv50_transfer.c | 11 |
7 files changed, 85 insertions, 77 deletions
diff --git a/src/gallium/drivers/nv50/nv50_miptree.c b/src/gallium/drivers/nv50/nv50_miptree.c index 9c20c5cc28..3d58746793 100644 --- a/src/gallium/drivers/nv50/nv50_miptree.c +++ b/src/gallium/drivers/nv50/nv50_miptree.c @@ -61,8 +61,8 @@ nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *tmp) struct nouveau_device *dev = nouveau_screen(pscreen)->device; struct nv50_miptree *mt = CALLOC_STRUCT(nv50_miptree); struct pipe_texture *pt = &mt->base.base; - unsigned width = tmp->width[0], height = tmp->height[0]; - unsigned depth = tmp->depth[0], image_alignment; + unsigned width = tmp->width0, height = tmp->height0; + unsigned depth = tmp->depth0, image_alignment; uint32_t tile_flags; int ret, i, l; @@ -92,9 +92,6 @@ nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *tmp) for (l = 0; l <= pt->last_level; l++) { struct nv50_miptree_level *lvl = &mt->level[l]; - pt->width[l] = width; - pt->height[l] = height; - pt->depth[l] = depth; pt->nblocksx[l] = pf_get_nblocksx(&pt->block, width); pt->nblocksy[l] = pf_get_nblocksy(&pt->block, height); @@ -102,9 +99,9 @@ nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *tmp) lvl->pitch = align(pt->nblocksx[l] * pt->block.size, 64); lvl->tile_mode = get_tile_mode(pt->nblocksy[l], depth); - width = MAX2(1, width >> 1); - height = MAX2(1, height >> 1); - depth = MAX2(1, depth >> 1); + width = u_minify(width, 1); + height = u_minify(height, 1); + depth = u_minify(depth, 1); } image_alignment = get_tile_height(mt->level[0].tile_mode) * 64; @@ -122,7 +119,7 @@ nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *tmp) size = lvl->pitch; size *= align(pt->nblocksy[l], tile_h); - size *= align(pt->depth[l], tile_d); + size *= align(u_minify(pt->depth0, l), tile_d); lvl->image_offset[i] = mt->total_size; @@ -151,7 +148,7 @@ nv50_miptree_blanket(struct pipe_screen *pscreen, const struct pipe_texture *pt, /* Only supports 2D, non-mipmapped textures for the moment */ if (pt->target != PIPE_TEXTURE_2D || pt->last_level != 0 || - pt->depth[0] != 1) + pt->depth0 != 1) return NULL; mt = CALLOC_STRUCT(nv50_miptree); @@ -202,8 +199,8 @@ nv50_miptree_surface_new(struct pipe_screen *pscreen, struct pipe_texture *pt, return NULL; pipe_texture_reference(&ps->texture, pt); ps->format = pt->format; - ps->width = pt->width[level]; - ps->height = pt->height[level]; + ps->width = u_minify(pt->width0, level); + ps->height = u_minify(pt->height0, level); ps->usage = flags; pipe_reference_init(&ps->reference, 1); ps->face = face; diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c index 1509cecaac..f0fe7e6168 100644 --- a/src/gallium/drivers/nv50/nv50_program.c +++ b/src/gallium/drivers/nv50/nv50_program.c @@ -1440,19 +1440,25 @@ emit_ddx(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src) static void emit_ddy(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src) { + struct nv50_reg *r = src; struct nv50_program_exec *e = exec(pc); assert(src->type == P_TEMP); - if (!(src->mod & NV50_MOD_NEG)) /* ! double negation */ - emit_neg(pc, src, src); + if (!(src->mod & NV50_MOD_NEG)) { /* ! double negation */ + r = alloc_temp(pc, NULL); + emit_neg(pc, r, src); + } e->inst[0] = 0xc0150000; e->inst[1] = 0x8a400000; set_long(pc, e); set_dst(pc, dst, e); - set_src_0(pc, src, e); - set_src_2(pc, src, e); + set_src_0(pc, r, e); + set_src_2(pc, r, e); + + if (r != src) + free_temp(pc, r); emit(pc, e); } @@ -2637,7 +2643,7 @@ nv50_program_tx_prep(struct nv50_pc *pc) for (i = 0, rid = 0; i < pc->result_nr; ++i) { p->cfg.io[i].hw = rid; - p->cfg.io[i].id_vp = i; + p->cfg.io[i].id = i; for (c = 0; c < 4; ++c) { int n = i * 4 + c; @@ -2669,14 +2675,12 @@ nv50_program_tx_prep(struct nv50_pc *pc) * the lower hardware IDs, so sort them: */ for (i = 0; i < pc->attr_nr; i++) { - if (pc->interp_mode[i] == INTERP_FLAT) { - p->cfg.io[m].id_vp = i + base; - p->cfg.io[m++].id_fp = i; - } else { + if (pc->interp_mode[i] == INTERP_FLAT) + p->cfg.io[m++].id = i; + else { if (!(pc->interp_mode[i] & INTERP_PERSPECTIVE)) p->cfg.io[n].linear = TRUE; - p->cfg.io[n].id_vp = i + base; - p->cfg.io[n++].id_fp = i; + p->cfg.io[n++].id = i; } } @@ -2688,7 +2692,7 @@ nv50_program_tx_prep(struct nv50_pc *pc) for (n = 0; n < pc->attr_nr; ++n) { p->cfg.io[n].hw = rid = aid; - i = p->cfg.io[n].id_fp; + i = p->cfg.io[n].id; if (p->info.input_semantic_name[n] == TGSI_SEMANTIC_FACE) { @@ -2728,8 +2732,8 @@ nv50_program_tx_prep(struct nv50_pc *pc) for (i = 0; i < pc->attr_nr; i++) { ubyte si, sn; - sn = p->info.input_semantic_name[p->cfg.io[i].id_fp]; - si = p->info.input_semantic_index[p->cfg.io[i].id_fp]; + sn = p->info.input_semantic_name[p->cfg.io[i].id]; + si = p->info.input_semantic_index[p->cfg.io[i].id]; if (sn == TGSI_SEMANTIC_COLOR) { p->cfg.two_side[si] = p->cfg.io[i]; @@ -2914,7 +2918,7 @@ nv50_fp_move_results(struct nv50_pc *pc) static void nv50_program_fixup_insns(struct nv50_pc *pc) { - struct nv50_program_exec *e, *prev = NULL, **bra_list; + struct nv50_program_exec *e, **bra_list; unsigned i, n, pos; bra_list = CALLOC(pc->p->exec_size, sizeof(struct nv50_program_exec *)); @@ -2926,6 +2930,16 @@ nv50_program_fixup_insns(struct nv50_pc *pc) if (e->param.index >= 0 && !e->param.mask) bra_list[n++] = e; + /* last instruction must be long so it can have the exit bit set */ + if (!is_long(pc->p->exec_tail)) + convert_to_long(pc, pc->p->exec_tail); + /* set exit bit */ + pc->p->exec_tail->inst[1] |= 1; + + /* !immd on exit insn simultaneously means !join */ + assert(!is_immd(pc->p->exec_head)); + assert(!is_immd(pc->p->exec_tail)); + /* Make sure we don't have any single 32 bit instructions. */ for (e = pc->p->exec_head, pos = 0; e; e = e->next) { pos += is_long(e) ? 2 : 1; @@ -2937,22 +2951,7 @@ nv50_program_fixup_insns(struct nv50_pc *pc) convert_to_long(pc, e); ++pos; } - if (e->next) - prev = e; - } - - assert(!is_immd(pc->p->exec_head)); - assert(!is_immd(pc->p->exec_tail)); - - /* last instruction must be long so it can have the end bit set */ - if (!is_long(pc->p->exec_tail)) { - convert_to_long(pc, pc->p->exec_tail); - if (prev) - convert_to_long(pc, prev); } - assert(!(pc->p->exec_tail->inst[1] & 2)); - /* set the end-bit */ - pc->p->exec_tail->inst[1] |= 1; FREE(bra_list); } @@ -3236,15 +3235,15 @@ nv50_pntc_replace(struct nv50_context *nv50, uint32_t pntc[8], unsigned base) struct nv50_program *vp = nv50->vertprog; unsigned i, c, m = base; - /* XXX: This can't work correctly in all cases yet, we either - * have to create TGSI_SEMANTIC_PNTC or sprite_coord_mode has - * to be per FP input instead of per VP output + /* XXX: this might not work correctly in all cases yet - we'll + * just assume that an FP generic input that is not written in + * the VP is PointCoord. */ memset(pntc, 0, 8 * sizeof(uint32_t)); for (i = 0; i < fp->cfg.io_nr; i++) { uint8_t sn, si; - uint8_t j = fp->cfg.io[i].id_vp, k = fp->cfg.io[i].id_fp; + uint8_t j, k = fp->cfg.io[i].id; unsigned n = popcnt4(fp->cfg.io[i].mask); if (fp->info.input_semantic_name[k] != TGSI_SEMANTIC_GENERIC) { @@ -3252,10 +3251,16 @@ nv50_pntc_replace(struct nv50_context *nv50, uint32_t pntc[8], unsigned base) continue; } - sn = vp->info.input_semantic_name[j]; - si = vp->info.input_semantic_index[j]; + for (j = 0; j < vp->info.num_outputs; ++j) { + sn = vp->info.output_semantic_name[j]; + si = vp->info.output_semantic_index[j]; - if (j < fp->cfg.io_nr && sn == TGSI_SEMANTIC_GENERIC) { + if (sn == fp->info.input_semantic_name[k] && + si == fp->info.input_semantic_index[k]) + break; + } + + if (j < vp->info.num_outputs) { ubyte mode = nv50->rasterizer->pipe.sprite_coord_mode[si]; @@ -3343,20 +3348,24 @@ nv50_linkage_validate(struct nv50_context *nv50) reg[0] += m - 4; /* adjust FFC0 id */ reg[4] |= m << 8; /* set mid where 'normal' FP inputs start */ - i = 0; - if (fp->info.input_semantic_name[0] == TGSI_SEMANTIC_POSITION) - i = 1; - for (; i < fp->cfg.io_nr; i++) { - ubyte sn = fp->info.input_semantic_name[fp->cfg.io[i].id_fp]; - ubyte si = fp->info.input_semantic_index[fp->cfg.io[i].id_fp]; - - n = fp->cfg.io[i].id_vp; - if (n >= vp->cfg.io_nr || - vp->info.output_semantic_name[n] != sn || - vp->info.output_semantic_index[n] != si) - vpo = &dummy; - else - vpo = &vp->cfg.io[n]; + for (i = 0; i < fp->cfg.io_nr; i++) { + ubyte sn = fp->info.input_semantic_name[fp->cfg.io[i].id]; + ubyte si = fp->info.input_semantic_index[fp->cfg.io[i].id]; + + /* position must be mapped first */ + assert(i == 0 || sn != TGSI_SEMANTIC_POSITION); + + /* maybe even remove these from cfg.io */ + if (sn == TGSI_SEMANTIC_POSITION || sn == TGSI_SEMANTIC_FACE) + continue; + + /* VP outputs and vp->cfg.io are in the same order */ + for (n = 0; n < vp->info.num_outputs; ++n) { + if (vp->info.output_semantic_name[n] == sn && + vp->info.output_semantic_index[n] == si) + break; + } + vpo = (n < vp->info.num_outputs) ? &vp->cfg.io[n] : &dummy; m = nv50_sreg4_map(map, m, lin, &fp->cfg.io[i], vpo); } diff --git a/src/gallium/drivers/nv50/nv50_program.h b/src/gallium/drivers/nv50/nv50_program.h index d78dee083f..255c7c737e 100644 --- a/src/gallium/drivers/nv50/nv50_program.h +++ b/src/gallium/drivers/nv50/nv50_program.h @@ -17,8 +17,7 @@ struct nv50_program_exec { struct nv50_sreg4 { uint8_t hw; - uint8_t id_vp; - uint8_t id_fp; + uint8_t id; /* tgsi index, nv50 needs them sorted: flat ones last */ uint8_t mask; boolean linear; diff --git a/src/gallium/drivers/nv50/nv50_state.c b/src/gallium/drivers/nv50/nv50_state.c index ffaa5e29d1..07318f2394 100644 --- a/src/gallium/drivers/nv50/nv50_state.c +++ b/src/gallium/drivers/nv50/nv50_state.c @@ -648,9 +648,9 @@ nv50_init_state_functions(struct nv50_context *nv50) nv50->pipe.delete_blend_state = nv50_blend_state_delete; nv50->pipe.create_sampler_state = nv50_sampler_state_create; - nv50->pipe.bind_sampler_states = nv50_sampler_state_bind; + nv50->pipe.bind_fragment_sampler_states = nv50_sampler_state_bind; nv50->pipe.delete_sampler_state = nv50_sampler_state_delete; - nv50->pipe.set_sampler_textures = nv50_set_sampler_texture; + nv50->pipe.set_fragment_sampler_textures = nv50_set_sampler_texture; nv50->pipe.create_rasterizer_state = nv50_rasterizer_state_create; nv50->pipe.bind_rasterizer_state = nv50_rasterizer_state_bind; diff --git a/src/gallium/drivers/nv50/nv50_state_validate.c b/src/gallium/drivers/nv50/nv50_state_validate.c index 799d2758fe..c871acaab8 100644 --- a/src/gallium/drivers/nv50/nv50_state_validate.c +++ b/src/gallium/drivers/nv50/nv50_state_validate.c @@ -201,7 +201,8 @@ nv50_state_emit(struct nv50_context *nv50) so_emit(chan, nv50->state.vertprog); if (nv50->state.dirty & NV50_NEW_FRAGPROG) so_emit(chan, nv50->state.fragprog); - if (nv50->state.dirty & (NV50_NEW_FRAGPROG | NV50_NEW_VERTPROG)) + if (nv50->state.dirty & (NV50_NEW_FRAGPROG | NV50_NEW_VERTPROG | + NV50_NEW_RASTERIZER)) so_emit(chan, nv50->state.programs); if (nv50->state.dirty & NV50_NEW_RASTERIZER) so_emit(chan, nv50->state.rast); @@ -264,7 +265,8 @@ nv50_state_validate(struct nv50_context *nv50) if (nv50->dirty & (NV50_NEW_FRAGPROG | NV50_NEW_FRAGPROG_CB)) nv50_fragprog_validate(nv50); - if (nv50->dirty & (NV50_NEW_FRAGPROG | NV50_NEW_VERTPROG)) + if (nv50->dirty & (NV50_NEW_FRAGPROG | NV50_NEW_VERTPROG | + NV50_NEW_RASTERIZER)) nv50_linkage_validate(nv50); if (nv50->dirty & NV50_NEW_RASTERIZER) @@ -285,7 +287,7 @@ nv50_state_validate(struct nv50_context *nv50) so = so_new(33, 0); so_method(so, tesla, NV50TCL_POLYGON_STIPPLE_PATTERN(0), 32); for (i = 0; i < 32; i++) - so_data(so, nv50->stipple.stipple[i]); + so_data(so, util_bswap32(nv50->stipple.stipple[i])); so_ref(so, &nv50->state.stipple); so_ref(NULL, &so); } diff --git a/src/gallium/drivers/nv50/nv50_tex.c b/src/gallium/drivers/nv50/nv50_tex.c index 2813f54477..417d367942 100644 --- a/src/gallium/drivers/nv50/nv50_tex.c +++ b/src/gallium/drivers/nv50/nv50_tex.c @@ -131,9 +131,9 @@ nv50_tex_construct(struct nv50_context *nv50, struct nouveau_stateobj *so, NOUVEAU_BO_RD, 0, 0); so_data (so, mode); so_data (so, 0x00300000); - so_data (so, mt->base.base.width[0] | (1 << 31)); + so_data (so, mt->base.base.width0 | (1 << 31)); so_data (so, (mt->base.base.last_level << 28) | - (mt->base.base.depth[0] << 16) | mt->base.base.height[0]); + (mt->base.base.depth0 << 16) | mt->base.base.height0); so_data (so, 0x03000000); so_data (so, mt->base.base.last_level << 4); diff --git a/src/gallium/drivers/nv50/nv50_transfer.c b/src/gallium/drivers/nv50/nv50_transfer.c index ea61357aaa..39d65279fc 100644 --- a/src/gallium/drivers/nv50/nv50_transfer.c +++ b/src/gallium/drivers/nv50/nv50_transfer.c @@ -1,6 +1,7 @@ #include "pipe/p_context.h" #include "pipe/p_inlines.h" +#include "util/u_math.h" #include "nv50_context.h" @@ -156,9 +157,9 @@ nv50_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt, tx->base.block = pt->block; if (!pt->nblocksx[level]) { tx->base.nblocksx = pf_get_nblocksx(&pt->block, - pt->width[level]); + u_minify(pt->width0, level)); tx->base.nblocksy = pf_get_nblocksy(&pt->block, - pt->height[level]); + u_minify(pt->height0, level)); } else { tx->base.nblocksx = pt->nblocksx[level]; tx->base.nblocksy = pt->nblocksy[level]; @@ -167,9 +168,9 @@ nv50_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt, tx->base.usage = usage; tx->level_pitch = lvl->pitch; - tx->level_width = mt->base.base.width[level]; - tx->level_height = mt->base.base.height[level]; - tx->level_depth = mt->base.base.depth[level]; + tx->level_width = u_minify(mt->base.base.width0, level); + tx->level_height = u_minify(mt->base.base.height0, level); + tx->level_depth = u_minify(mt->base.base.depth0, level); tx->level_offset = lvl->image_offset[image]; tx->level_tiling = lvl->tile_mode; tx->level_x = pf_get_nblocksx(&tx->base.block, x); |