diff options
Diffstat (limited to 'src/gallium/drivers/nv50')
| -rw-r--r-- | src/gallium/drivers/nv50/nv50_clear.c | 76 | ||||
| -rw-r--r-- | src/gallium/drivers/nv50/nv50_context.c | 28 | ||||
| -rw-r--r-- | src/gallium/drivers/nv50/nv50_context.h | 18 | ||||
| -rw-r--r-- | src/gallium/drivers/nv50/nv50_miptree.c | 31 | ||||
| -rw-r--r-- | src/gallium/drivers/nv50/nv50_program.c | 1223 | ||||
| -rw-r--r-- | src/gallium/drivers/nv50/nv50_program.h | 11 | ||||
| -rw-r--r-- | src/gallium/drivers/nv50/nv50_query.c | 29 | ||||
| -rw-r--r-- | src/gallium/drivers/nv50/nv50_screen.c | 200 | ||||
| -rw-r--r-- | src/gallium/drivers/nv50/nv50_screen.h | 15 | ||||
| -rw-r--r-- | src/gallium/drivers/nv50/nv50_state.c | 15 | ||||
| -rw-r--r-- | src/gallium/drivers/nv50/nv50_state_validate.c | 74 | ||||
| -rw-r--r-- | src/gallium/drivers/nv50/nv50_surface.c | 16 | ||||
| -rw-r--r-- | src/gallium/drivers/nv50/nv50_tex.c | 36 | ||||
| -rw-r--r-- | src/gallium/drivers/nv50/nv50_transfer.c | 82 | ||||
| -rw-r--r-- | src/gallium/drivers/nv50/nv50_vbo.c | 23 |
15 files changed, 1340 insertions, 537 deletions
diff --git a/src/gallium/drivers/nv50/nv50_clear.c b/src/gallium/drivers/nv50/nv50_clear.c index db44a9da0e3..e0b2d2880b0 100644 --- a/src/gallium/drivers/nv50/nv50_clear.c +++ b/src/gallium/drivers/nv50/nv50_clear.c @@ -27,64 +27,42 @@ #include "nv50_context.h" void -nv50_clear(struct pipe_context *pipe, struct pipe_surface *ps, - unsigned clearValue) +nv50_clear(struct pipe_context *pipe, unsigned buffers, + const float *rgba, double depth, unsigned stencil) { struct nv50_context *nv50 = nv50_context(pipe); - struct nouveau_channel *chan = nv50->screen->nvws->channel; + struct nouveau_channel *chan = nv50->screen->base.channel; struct nouveau_grobj *tesla = nv50->screen->tesla; - struct pipe_framebuffer_state fb, s_fb = nv50->framebuffer; - struct pipe_scissor_state sc, s_sc = nv50->scissor; - unsigned dirty = nv50->dirty; + struct pipe_framebuffer_state *fb = &nv50->framebuffer; + unsigned mode = 0, i; - nv50->dirty = 0; + if (!nv50_state_validate(nv50)) + return; - if (ps->format == PIPE_FORMAT_Z24S8_UNORM || - ps->format == PIPE_FORMAT_Z16_UNORM) { - fb.nr_cbufs = 0; - fb.zsbuf = ps; - } else { - fb.nr_cbufs = 1; - fb.cbufs[0] = ps; - fb.zsbuf = NULL; + if (buffers & PIPE_CLEAR_COLOR && fb->nr_cbufs) { + BEGIN_RING(chan, tesla, NV50TCL_CLEAR_COLOR(0), 4); + OUT_RING (chan, fui(rgba[0])); + OUT_RING (chan, fui(rgba[1])); + OUT_RING (chan, fui(rgba[2])); + OUT_RING (chan, fui(rgba[3])); + mode |= 0x3c; } - fb.width = ps->width; - fb.height = ps->height; - pipe->set_framebuffer_state(pipe, &fb); - sc.minx = sc.miny = 0; - sc.maxx = fb.width; - sc.maxy = fb.height; - pipe->set_scissor_state(pipe, &sc); + if (buffers & PIPE_CLEAR_DEPTHSTENCIL) { + BEGIN_RING(chan, tesla, NV50TCL_CLEAR_DEPTH, 1); + OUT_RING (chan, fui(depth)); + BEGIN_RING(chan, tesla, NV50TCL_CLEAR_STENCIL, 1); + OUT_RING (chan, stencil & 0xff); - nv50_state_validate(nv50); - - switch (ps->format) { - case PIPE_FORMAT_A8R8G8B8_UNORM: - BEGIN_RING(chan, tesla, 0x0d80, 4); - OUT_RINGf (chan, ubyte_to_float((clearValue >> 16) & 0xff)); - OUT_RINGf (chan, ubyte_to_float((clearValue >> 8) & 0xff)); - OUT_RINGf (chan, ubyte_to_float((clearValue >> 0) & 0xff)); - OUT_RINGf (chan, ubyte_to_float((clearValue >> 24) & 0xff)); - BEGIN_RING(chan, tesla, 0x19d0, 1); - OUT_RING (chan, 0x3c); - break; - case PIPE_FORMAT_Z24S8_UNORM: - BEGIN_RING(chan, tesla, 0x0d90, 1); - OUT_RINGf (chan, (float)(clearValue >> 8) * (1.0 / 16777215.0)); - BEGIN_RING(chan, tesla, 0x0da0, 1); - OUT_RING (chan, clearValue & 0xff); - BEGIN_RING(chan, tesla, 0x19d0, 1); - OUT_RING (chan, 0x03); - break; - default: - pipe->surface_fill(pipe, ps, 0, 0, ps->width, ps->height, - clearValue); - break; + mode |= 0x03; } - pipe->set_framebuffer_state(pipe, &s_fb); - pipe->set_scissor_state(pipe, &s_sc); - nv50->dirty |= dirty; + BEGIN_RING(chan, tesla, NV50TCL_CLEAR_BUFFERS, 1); + OUT_RING (chan, mode); + + for (i = 1; i < fb->nr_cbufs; i++) { + BEGIN_RING(chan, tesla, NV50TCL_CLEAR_BUFFERS, 1); + OUT_RING (chan, (i << 6) | 0x3c); + } } diff --git a/src/gallium/drivers/nv50/nv50_context.c b/src/gallium/drivers/nv50/nv50_context.c index 565a5da668c..e02afc4be99 100644 --- a/src/gallium/drivers/nv50/nv50_context.c +++ b/src/gallium/drivers/nv50/nv50_context.c @@ -33,7 +33,7 @@ nv50_flush(struct pipe_context *pipe, unsigned flags, { struct nv50_context *nv50 = (struct nv50_context *)pipe; - FIRE_RING(nv50->screen->nvws->channel); + FIRE_RING(nv50->screen->base.channel); } static void @@ -51,6 +51,29 @@ nv50_set_edgeflags(struct pipe_context *pipe, const unsigned *bitfield) { } +static unsigned int +nv50_is_texture_referenced( struct pipe_context *pipe, + struct pipe_texture *texture, + unsigned face, unsigned level) +{ + /** + * FIXME: Optimize. + */ + + return PIPE_REFERENCED_FOR_READ | PIPE_REFERENCED_FOR_WRITE; +} + +static unsigned int +nv50_is_buffer_referenced( struct pipe_context *pipe, + struct pipe_buffer *buf) +{ + /** + * FIXME: Optimize. + */ + + return PIPE_REFERENCED_FOR_READ | PIPE_REFERENCED_FOR_WRITE; +} + struct pipe_context * nv50_create(struct pipe_screen *pscreen, unsigned pctx_id) { @@ -76,6 +99,9 @@ nv50_create(struct pipe_screen *pscreen, unsigned pctx_id) nv50->pipe.flush = nv50_flush; + nv50->pipe.is_texture_referenced = nv50_is_texture_referenced; + nv50->pipe.is_buffer_referenced = nv50_is_buffer_referenced; + nv50_init_surface_functions(nv50); nv50_init_state_functions(nv50); nv50_init_query_functions(nv50); diff --git a/src/gallium/drivers/nv50/nv50_context.h b/src/gallium/drivers/nv50/nv50_context.h index 7b67a754397..9b8cc4d37d0 100644 --- a/src/gallium/drivers/nv50/nv50_context.h +++ b/src/gallium/drivers/nv50/nv50_context.h @@ -63,6 +63,11 @@ struct nv50_rasterizer_stateobj { struct nouveau_stateobj *so; }; +struct nv50_sampler_stateobj { + bool normalized; + unsigned tsc[8]; +}; + struct nv50_miptree_level { int *image_offset; unsigned pitch; @@ -70,7 +75,8 @@ struct nv50_miptree_level { struct nv50_miptree { struct pipe_texture base; - struct pipe_buffer *buffer; + + struct nouveau_bo *bo; struct nv50_miptree_level level[PIPE_MAX_TEXTURE_LEVELS]; int image_nr; @@ -93,13 +99,6 @@ nv50_surface(struct pipe_surface *pt) return (struct nv50_surface *)pt; } -static INLINE struct pipe_buffer * -nv50_surface_buffer(struct pipe_surface *surface) -{ - struct nv50_miptree *mt = (struct nv50_miptree *)surface->texture; - return mt->buffer; -} - struct nv50_state { unsigned dirty; @@ -115,6 +114,7 @@ struct nv50_state { unsigned viewport_bypass; struct nouveau_stateobj *tsc_upload; struct nouveau_stateobj *tic_upload; + unsigned miptree_nr; struct nouveau_stateobj *vertprog; struct nouveau_stateobj *fragprog; struct nouveau_stateobj *vtxfmt; @@ -147,7 +147,7 @@ struct nv50_context { unsigned vtxbuf_nr; struct pipe_vertex_element vtxelt[PIPE_MAX_ATTRIBS]; unsigned vtxelt_nr; - unsigned *sampler[PIPE_MAX_SAMPLERS]; + struct nv50_sampler_stateobj *sampler[PIPE_MAX_SAMPLERS]; unsigned sampler_nr; struct nv50_miptree *miptree[PIPE_MAX_SAMPLERS]; unsigned miptree_nr; diff --git a/src/gallium/drivers/nv50/nv50_miptree.c b/src/gallium/drivers/nv50/nv50_miptree.c index f79a7ca86c9..6b605ba416c 100644 --- a/src/gallium/drivers/nv50/nv50_miptree.c +++ b/src/gallium/drivers/nv50/nv50_miptree.c @@ -29,26 +29,35 @@ static struct pipe_texture * 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; - unsigned usage, width = tmp->width[0], height = tmp->height[0]; + unsigned width = tmp->width[0], height = tmp->height[0]; unsigned depth = tmp->depth[0]; - int i, l; + uint32_t tile_mode, tile_flags, tile_h; + int ret, i, l; mt->base = *tmp; pipe_reference_init(&mt->base.reference, 1); mt->base.screen = pscreen; - usage = PIPE_BUFFER_USAGE_PIXEL; switch (pt->format) { case PIPE_FORMAT_Z24S8_UNORM: case PIPE_FORMAT_Z16_UNORM: - usage |= NOUVEAU_BUFFER_USAGE_ZETA; + tile_flags = 0x2800; break; default: + tile_flags = 0x7000; break; } + if (pt->height[0] > 32) tile_mode = 4; + else if (pt->height[0] > 16) tile_mode = 3; + else if (pt->height[0] > 8) tile_mode = 2; + else if (pt->height[0] > 4) tile_mode = 1; + else tile_mode = 0; + tile_h = 1 << (tile_mode + 2); + switch (pt->target) { case PIPE_TEXTURE_3D: mt->image_nr = pt->depth[0]; @@ -85,7 +94,7 @@ nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *tmp) size = align(pt->width[l], 8) * pt->block.size; size = align(size, 64); - size *= align(pt->height[l], 8) * pt->block.size; + size *= align(pt->height[l], tile_h) * pt->block.size; lvl->image_offset[i] = mt->total_size; @@ -93,12 +102,13 @@ nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *tmp) } } - mt->buffer = pscreen->buffer_create(pscreen, 256, usage, mt->total_size); - if (!mt->buffer) { + ret = nouveau_bo_new_tile(dev, NOUVEAU_BO_VRAM, 256, mt->total_size, + tile_mode, tile_flags, &mt->bo); + if (ret) { FREE(mt); return NULL; } - + return &mt->base; } @@ -106,6 +116,7 @@ static struct pipe_texture * nv50_miptree_blanket(struct pipe_screen *pscreen, const struct pipe_texture *pt, const unsigned *stride, struct pipe_buffer *pb) { + struct nouveau_bo *bo = nouveau_bo(pb); struct nv50_miptree *mt; /* Only supports 2D, non-mipmapped textures for the moment */ @@ -124,7 +135,7 @@ nv50_miptree_blanket(struct pipe_screen *pscreen, const struct pipe_texture *pt, mt->level[0].pitch = *stride; mt->level[0].image_offset = CALLOC(1, sizeof(unsigned)); - pipe_buffer_reference(&mt->buffer, pb); + nouveau_bo_ref(bo, &mt->bo); return &mt->base; } @@ -133,7 +144,7 @@ nv50_miptree_destroy(struct pipe_texture *pt) { struct nv50_miptree *mt = nv50_miptree(pt); - pipe_buffer_reference(&mt->buffer, NULL); + nouveau_bo_ref(NULL, &mt->bo); FREE(mt); } diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c index 2d15868ae84..5f7d06dbecb 100644 --- a/src/gallium/drivers/nv50/nv50_program.c +++ b/src/gallium/drivers/nv50/nv50_program.c @@ -85,6 +85,9 @@ struct nv50_reg { int hw; int neg; + + int rhw; /* result hw for FP outputs, or interpolant index */ + int acc; /* instruction where this reg is last read (first insn == 1) */ }; struct nv50_pc { @@ -108,12 +111,23 @@ struct nv50_pc { struct nv50_reg *temp_temp[16]; unsigned temp_temp_nr; + + unsigned interp_mode[32]; + /* perspective interpolation registers */ + struct nv50_reg *iv_p; + struct nv50_reg *iv_c; + + /* current instruction and total number of insns */ + unsigned insn_cur; + unsigned insn_nr; + + boolean allow32; }; static void alloc_reg(struct nv50_pc *pc, struct nv50_reg *reg) { - int i; + int i = 0; if (reg->type == P_RESULT) { if (pc->p->cfg.high_result < (reg->hw + 1)) @@ -131,7 +145,22 @@ alloc_reg(struct nv50_pc *pc, struct nv50_reg *reg) return; } - for (i = 0; i < NV50_SU_MAX_TEMP; i++) { + if (reg->rhw != -1) { + /* try to allocate temporary with index rhw first */ + if (!(pc->r_temp[reg->rhw])) { + pc->r_temp[reg->rhw] = reg; + reg->hw = reg->rhw; + if (pc->p->cfg.high_temp < (reg->rhw + 1)) + pc->p->cfg.high_temp = reg->rhw + 1; + return; + } + /* make sure we don't get things like $r0 needs to go + * in $r1 and $r1 in $r0 + */ + i = pc->result_nr * 4; + } + + for (; i < NV50_SU_MAX_TEMP; i++) { if (!(pc->r_temp[i])) { pc->r_temp[i] = reg; reg->hw = i; @@ -159,6 +188,7 @@ alloc_temp(struct nv50_pc *pc, struct nv50_reg *dst) r->type = P_TEMP; r->index = -1; r->hw = i; + r->rhw = -1; pc->r_temp[i] = r; return r; } @@ -168,6 +198,38 @@ alloc_temp(struct nv50_pc *pc, struct nv50_reg *dst) return NULL; } +/* Assign the hw of the discarded temporary register src + * to the tgsi register dst and free src. + */ +static void +assimilate_temp(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src) +{ + assert(src->index == -1 && src->hw != -1); + + if (dst->hw != -1) + pc->r_temp[dst->hw] = NULL; + pc->r_temp[src->hw] = dst; + dst->hw = src->hw; + + FREE(src); +} + +/* release the hardware resource held by r */ +static void +release_hw(struct nv50_pc *pc, struct nv50_reg *r) +{ + assert(r->type == P_TEMP); + if (r->hw == -1) + return; + + assert(pc->r_temp[r->hw] == r); + pc->r_temp[r->hw] = NULL; + + r->acc = 0; + if (r->index == -1) + FREE(r); +} + static void free_temp(struct nv50_pc *pc, struct nv50_reg *r) { @@ -250,7 +312,13 @@ alloc_immd(struct nv50_pc *pc, float f) struct nv50_reg *r = CALLOC_STRUCT(nv50_reg); unsigned hw; - hw = ctor_immd(pc, f, 0, 0, 0) * 4; + for (hw = 0; hw < pc->immd_nr * 4; hw++) + if (pc->immd_buf[hw] == f) + break; + + if (hw == pc->immd_nr * 4) + hw = ctor_immd(pc, f, -f, 0.5 * f, 0) * 4; + r->type = P_IMMD; r->hw = hw; r->index = -1; @@ -341,7 +409,8 @@ set_dst(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_program_exec *e) static INLINE void set_immd(struct nv50_pc *pc, struct nv50_reg *imm, struct nv50_program_exec *e) { - unsigned val = fui(pc->immd_buf[imm->hw]); /* XXX */ + float f = pc->immd_buf[imm->hw]; + unsigned val = fui(imm->neg ? -f : f); set_long(pc, e); /*XXX: can't be predicated - bits overlap.. catch cases where both @@ -354,20 +423,35 @@ set_immd(struct nv50_pc *pc, struct nv50_reg *imm, struct nv50_program_exec *e) e->inst[1] |= (val >> 6) << 2; } + +#define INTERP_LINEAR 0 +#define INTERP_FLAT 1 +#define INTERP_PERSPECTIVE 2 +#define INTERP_CENTROID 4 + +/* interpolant index has been stored in dst->rhw */ static void -emit_interp(struct nv50_pc *pc, struct nv50_reg *dst, - struct nv50_reg *src, struct nv50_reg *iv) +emit_interp(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *iv, + unsigned mode) { + assert(dst->rhw != -1); struct nv50_program_exec *e = exec(pc); e->inst[0] |= 0x80000000; set_dst(pc, dst, e); - alloc_reg(pc, src); - e->inst[0] |= (src->hw << 16); - if (iv) { - e->inst[0] |= (1 << 25); - alloc_reg(pc, iv); - e->inst[0] |= (iv->hw << 9); + e->inst[0] |= (dst->rhw << 16); + + if (mode & INTERP_FLAT) { + e->inst[0] |= (1 << 8); + } else { + if (mode & INTERP_PERSPECTIVE) { + e->inst[0] |= (1 << 25); + alloc_reg(pc, iv); + e->inst[0] |= (iv->hw << 9); + } + + if (mode & INTERP_CENTROID) + e->inst[0] |= (1 << 24); } emit(pc, e); @@ -378,22 +462,12 @@ set_data(struct nv50_pc *pc, struct nv50_reg *src, unsigned m, unsigned s, struct nv50_program_exec *e) { set_long(pc, e); -#if 1 - e->inst[1] |= (1 << 22); -#else - if (src->type == P_IMMD) { - e->inst[1] |= (NV50_CB_PMISC << 22); - } else { - if (pc->p->type == PIPE_SHADER_VERTEX) - e->inst[1] |= (NV50_CB_PVP << 22); - else - e->inst[1] |= (NV50_CB_PFP << 22); - } -#endif e->param.index = src->hw; e->param.shift = s; e->param.mask = m << (s % 32); + + e->inst[1] |= (((src->type == P_IMMD) ? 0 : 1) << 22); } static void @@ -405,12 +479,11 @@ emit_mov(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src) set_dst(pc, dst, e); - if (0 && dst->type != P_RESULT && src->type == P_IMMD) { + if (pc->allow32 && dst->type != P_RESULT && src->type == P_IMMD) { set_immd(pc, src, e); /*XXX: 32-bit, but steals part of "half" reg space - need to * catch and handle this case if/when we do half-regs */ - e->inst[0] |= 0x00008000; } else if (src->type == P_IMMD || src->type == P_CONST) { set_long(pc, e); @@ -426,18 +499,25 @@ emit_mov(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src) e->inst[0] |= (src->hw << 9); } - /* We really should support "half" instructions here at some point, - * but I don't feel confident enough about them yet. - */ - set_long(pc, e); if (is_long(e) && !is_immd(e)) { e->inst[1] |= 0x04000000; /* 32-bit */ - e->inst[1] |= 0x0003c000; /* "subsubop" 0xf == mov */ - } + e->inst[1] |= 0x0000c000; /* "subsubop" 0x3 */ + if (!(e->inst[1] & 0x20000000)) + e->inst[1] |= 0x00030000; /* "subsubop" 0xf */ + } else + e->inst[0] |= 0x00008000; emit(pc, e); } +static INLINE void +emit_mov_immdval(struct nv50_pc *pc, struct nv50_reg *dst, float f) +{ + struct nv50_reg *imm = alloc_immd(pc, f); + emit_mov(pc, dst, imm); + FREE(imm); +} + static boolean check_swap_src_0_1(struct nv50_pc *pc, struct nv50_reg **s0, struct nv50_reg **s1) @@ -541,12 +621,26 @@ emit_mul(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src0, struct nv50_program_exec *e = exec(pc); e->inst[0] |= 0xc0000000; - set_long(pc, e); + + if (!pc->allow32) + set_long(pc, e); check_swap_src_0_1(pc, &src0, &src1); set_dst(pc, dst, e); set_src_0(pc, src0, e); - set_src_1(pc, src1, e); + if (src1->type == P_IMMD && !is_long(e)) { + if (src0->neg) + e->inst[0] |= 0x00008000; + set_immd(pc, src1, e); + } else { + set_src_1(pc, src1, e); + if (src0->neg ^ src1->neg) { + if (is_long(e)) + e->inst[1] |= 0x08000000; + else + e->inst[0] |= 0x00008000; + } + } emit(pc, e); } @@ -560,11 +654,20 @@ emit_add(struct nv50_pc *pc, struct nv50_reg *dst, e->inst[0] |= 0xb0000000; check_swap_src_0_1(pc, &src0, &src1); + + if (!pc->allow32 || src0->neg || src1->neg) { + set_long(pc, e); + e->inst[1] |= (src0->neg << 26) | (src1->neg << 27); + } + set_dst(pc, dst, e); set_src_0(pc, src0, e); - if (is_long(e)) + if (src1->type == P_CONST || src1->type == P_ATTR || is_long(e)) set_src_2(pc, src1, e); else + if (src1->type == P_IMMD) + set_immd(pc, src1, e); + else set_src_1(pc, src1, e); emit(pc, e); @@ -588,25 +691,13 @@ emit_minmax(struct nv50_pc *pc, unsigned sub, struct nv50_reg *dst, emit(pc, e); } -static void +static INLINE void emit_sub(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src0, struct nv50_reg *src1) { - struct nv50_program_exec *e = exec(pc); - - e->inst[0] |= 0xb0000000; - - set_long(pc, e); - if (check_swap_src_0_1(pc, &src0, &src1)) - e->inst[1] |= 0x04000000; - else - e->inst[1] |= 0x08000000; - - set_dst(pc, dst, e); - set_src_0(pc, src0, e); - set_src_2(pc, src1, e); - - emit(pc, e); + src1->neg ^= 1; + emit_add(pc, dst, src0, src1); + src1->neg ^= 1; } static void @@ -623,26 +714,21 @@ emit_mad(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src0, set_src_1(pc, src1, e); set_src_2(pc, src2, e); + if (src0->neg ^ src1->neg) + e->inst[1] |= 0x04000000; + if (src2->neg) + e->inst[1] |= 0x08000000; + emit(pc, e); } -static void +static INLINE void emit_msb(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src0, struct nv50_reg *src1, struct nv50_reg *src2) { - struct nv50_program_exec *e = exec(pc); - - e->inst[0] |= 0xe0000000; - set_long(pc, e); - e->inst[1] |= 0x08000000; /* src0 * src1 - src2 */ - - check_swap_src_0_1(pc, &src0, &src1); - set_dst(pc, dst, e); - set_src_0(pc, src0, e); - set_src_1(pc, src1, e); - set_src_2(pc, src2, e); - - emit(pc, e); + src2->neg ^= 1; + emit_mad(pc, dst, src0, src1, src2); + src2->neg ^= 1; } static void @@ -693,6 +779,48 @@ emit_precossin(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src) emit(pc, e); } +#define CVTOP_RN 0x01 +#define CVTOP_FLOOR 0x03 +#define CVTOP_CEIL 0x05 +#define CVTOP_TRUNC 0x07 +#define CVTOP_SAT 0x08 +#define CVTOP_ABS 0x10 + +#define CVT_F32_F32 0xc4 +#define CVT_F32_S32 0x44 +#define CVT_F32_U32 0x64 +#define CVT_S32_F32 0x8c +#define CVT_S32_S32 0x0c +#define CVT_F32_F32_ROP 0xcc + +static void +emit_cvt(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src, + int wp, unsigned cop, unsigned fmt) +{ + struct nv50_program_exec *e; + + e = exec(pc); + set_long(pc, e); + + e->inst[0] |= 0xa0000000; + e->inst[1] |= 0x00004000; + e->inst[1] |= (cop << 16); + e->inst[1] |= (fmt << 24); + set_src_0(pc, src, e); + + if (wp >= 0) + set_pred_wr(pc, 1, wp, e); + + if (dst) + set_dst(pc, dst, e); + else { + e->inst[0] |= 0x000001fc; + e->inst[1] |= 0x00000008; + } + + emit(pc, e); +} + static void emit_set(struct nv50_pc *pc, unsigned c_op, struct nv50_reg *dst, struct nv50_reg *src0, struct nv50_reg *src1) @@ -736,22 +864,10 @@ emit_set(struct nv50_pc *pc, unsigned c_op, struct nv50_reg *dst, free_temp(pc, dst); } -static void +static INLINE void emit_flr(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src) { - struct nv50_program_exec *e = exec(pc); - - e->inst[0] = 0xa0000000; /* cvt */ - set_long(pc, e); - e->inst[1] |= (6 << 29); /* cvt */ - e->inst[1] |= 0x08000000; /* integer mode */ - e->inst[1] |= 0x04000000; /* 32 bit */ - e->inst[1] |= ((0x1 << 3)) << 14; /* .rn */ - e->inst[1] |= (1 << 14); /* src .f32 */ - set_dst(pc, dst, e); - set_src_0(pc, src, e); - - emit(pc, e); + emit_cvt(pc, dst, src, -1, CVTOP_FLOOR, CVT_F32_F32_ROP); } static void @@ -768,21 +884,10 @@ emit_pow(struct nv50_pc *pc, struct nv50_reg *dst, free_temp(pc, temp); } -static void +static INLINE void emit_abs(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src) { - struct nv50_program_exec *e = exec(pc); - - e->inst[0] = 0xa0000000; /* cvt */ - set_long(pc, e); - e->inst[1] |= (6 << 29); /* cvt */ - e->inst[1] |= 0x04000000; /* 32 bit */ - e->inst[1] |= (1 << 14); /* src .f32 */ - e->inst[1] |= ((1 << 6) << 14); /* .abs */ - set_dst(pc, dst, e); - set_src_0(pc, src, e); - - emit(pc, e); + emit_cvt(pc, dst, src, -1, CVTOP_ABS, CVT_F32_F32); } static void @@ -794,18 +899,12 @@ emit_lit(struct nv50_pc *pc, struct nv50_reg **dst, unsigned mask, struct nv50_reg *neg128 = alloc_immd(pc, -127.999999); struct nv50_reg *pos128 = alloc_immd(pc, 127.999999); struct nv50_reg *tmp[4]; + boolean allow32 = pc->allow32; - if (mask & (1 << 0)) - emit_mov(pc, dst[0], one); - - if (mask & (1 << 3)) - emit_mov(pc, dst[3], one); + pc->allow32 = FALSE; if (mask & (3 << 1)) { - if (mask & (1 << 1)) - tmp[0] = dst[1]; - else - tmp[0] = temp_temp(pc); + tmp[0] = alloc_temp(pc, NULL); emit_minmax(pc, 4, tmp[0], src[0], zero); } @@ -823,6 +922,26 @@ emit_lit(struct nv50_pc *pc, struct nv50_reg **dst, unsigned mask, emit_mov(pc, dst[2], zero); set_pred(pc, 3, 0, pc->p->exec_tail); } + + if (mask & (1 << 1)) + assimilate_temp(pc, dst[1], tmp[0]); + else + if (mask & (1 << 2)) + free_temp(pc, tmp[0]); + + pc->allow32 = allow32; + + /* do this last, in case src[i,j] == dst[0,3] */ + if (mask & (1 << 0)) + emit_mov(pc, dst[0], one); + + if (mask & (1 << 3)) + emit_mov(pc, dst[3], one); + + FREE(pos128); + FREE(neg128); + FREE(zero); + FREE(one); } static void @@ -853,6 +972,8 @@ emit_kil(struct nv50_pc *pc, struct nv50_reg *src) e->inst[1] = 0xc4014788; set_src_0(pc, src, e); set_pred_wr(pc, 1, r_pred, e); + if (src->neg) + e->inst[1] |= 0x20000000; emit(pc, e); /* This is probably KILP */ @@ -863,6 +984,180 @@ emit_kil(struct nv50_pc *pc, struct nv50_reg *src) emit(pc, e); } +static void +emit_tex(struct nv50_pc *pc, struct nv50_reg **dst, unsigned mask, + struct nv50_reg **src, unsigned unit, unsigned type, boolean proj) +{ + struct nv50_reg *temp, *t[4]; + struct nv50_program_exec *e; + + unsigned c, mode, dim; + + switch (type) { + case TGSI_TEXTURE_1D: + dim = 1; + break; + case TGSI_TEXTURE_UNKNOWN: + case TGSI_TEXTURE_2D: + case TGSI_TEXTURE_SHADOW1D: /* XXX: x, z */ + case TGSI_TEXTURE_RECT: + dim = 2; + break; + case TGSI_TEXTURE_3D: + case TGSI_TEXTURE_CUBE: + case TGSI_TEXTURE_SHADOW2D: + case TGSI_TEXTURE_SHADOWRECT: /* XXX */ + dim = 3; + break; + default: + assert(0); + break; + } + + alloc_temp4(pc, t, 0); + + if (proj) { + if (src[0]->type == P_TEMP && src[0]->rhw != -1) { + mode = pc->interp_mode[src[0]->index]; + + t[3]->rhw = src[3]->rhw; + emit_interp(pc, t[3], NULL, (mode & INTERP_CENTROID)); + emit_flop(pc, 0, t[3], t[3]); + + for (c = 0; c < dim; c++) { + t[c]->rhw = src[c]->rhw; + emit_interp(pc, t[c], t[3], + (mode | INTERP_PERSPECTIVE)); + } + } else { + emit_flop(pc, 0, t[3], src[3]); + for (c = 0; c < dim; c++) + emit_mul(pc, t[c], src[c], t[3]); + + /* XXX: for some reason the blob sometimes uses MAD: + * emit_mad(pc, t[c], src[0][c], t[3], t[3]) + * pc->p->exec_tail->inst[1] |= 0x080fc000; + */ + } + } else { + if (type == TGSI_TEXTURE_CUBE) { + temp = temp_temp(pc); + emit_minmax(pc, 4, temp, src[0], src[1]); + emit_minmax(pc, 4, temp, temp, src[2]); + emit_flop(pc, 0, temp, temp); + for (c = 0; c < 3; c++) + emit_mul(pc, t[c], src[c], temp); + } else { + for (c = 0; c < dim; c++) + emit_mov(pc, t[c], src[c]); + } + } + + e = exec(pc); + set_long(pc, e); + e->inst[0] |= 0xf0000000; + e->inst[1] |= 0x00000004; + set_dst(pc, t[0], e); + e->inst[0] |= (unit << 9); + + if (dim == 2) + e->inst[0] |= 0x00400000; + else + if (dim == 3) + e->inst[0] |= 0x00800000; + + e->inst[0] |= (mask & 0x3) << 25; + e->inst[1] |= (mask & 0xc) << 12; + + emit(pc, e); + +#if 1 + if (mask & 1) emit_mov(pc, dst[0], t[0]); + if (mask & 2) emit_mov(pc, dst[1], t[1]); + if (mask & 4) emit_mov(pc, dst[2], t[2]); + if (mask & 8) emit_mov(pc, dst[3], t[3]); + + free_temp4(pc, t); +#else + /* XXX: if p.e. MUL is used directly after TEX, it would still use + * the texture coordinates, not the fetched values: latency ? */ + + for (c = 0; c < 4; c++) { + if (mask & (1 << c)) + assimilate_temp(pc, dst[c], t[c]); + else + free_temp(pc, t[c]); + } +#endif +} + +static void +convert_to_long(struct nv50_pc *pc, struct nv50_program_exec *e) +{ + unsigned q = 0, m = ~0; + + assert(!is_long(e)); + + switch (e->inst[0] >> 28) { + case 0x1: + /* MOV */ + q = 0x0403c000; + m = 0xffff7fff; + break; + case 0x8: + /* INTERP */ + m = ~0x02000000; + if (e->inst[0] & 0x02000000) + q = 0x00020000; + break; + case 0x9: + /* RCP */ + break; + case 0xB: + /* ADD */ + m = ~(127 << 16); + q = ((e->inst[0] & (~m)) >> 2); + break; + case 0xC: + /* MUL */ + m = ~0x00008000; + q = ((e->inst[0] & (~m)) << 12); + break; + case 0xE: + /* MAD (if src2 == dst) */ + q = ((e->inst[0] & 0x1fc) << 12); + break; + default: + assert(0); + break; + } + + set_long(pc, e); + pc->p->exec_size++; + + e->inst[0] &= m; + e->inst[1] |= q; +} + +static boolean +negate_supported(const struct tgsi_full_instruction *insn, int i) +{ + switch (insn->Instruction.Opcode) { + case TGSI_OPCODE_DP3: + case TGSI_OPCODE_DP4: + case TGSI_OPCODE_MUL: + case TGSI_OPCODE_KIL: + case TGSI_OPCODE_ADD: + case TGSI_OPCODE_SUB: + case TGSI_OPCODE_MAD: + return TRUE; + case TGSI_OPCODE_POW: + return (i == 1) ? TRUE : FALSE; + default: + return FALSE; + } +} + static struct nv50_reg * tgsi_dst(struct nv50_pc *pc, int c, const struct tgsi_full_dst_register *dst) { @@ -881,11 +1176,14 @@ tgsi_dst(struct nv50_pc *pc, int c, const struct tgsi_full_dst_register *dst) } static struct nv50_reg * -tgsi_src(struct nv50_pc *pc, int chan, const struct tgsi_full_src_register *src) +tgsi_src(struct nv50_pc *pc, int chan, const struct tgsi_full_src_register *src, + boolean neg) { struct nv50_reg *r = NULL; struct nv50_reg *temp; - unsigned c; + unsigned sgn, c; + + sgn = tgsi_util_get_full_src_register_sign_mode(src, chan); c = tgsi_util_get_full_src_register_extswizzle(src, chan); switch (c) { @@ -915,16 +1213,17 @@ tgsi_src(struct nv50_pc *pc, int chan, const struct tgsi_full_src_register *src) break; case TGSI_EXTSWIZZLE_ZERO: r = alloc_immd(pc, 0.0); - break; + return r; case TGSI_EXTSWIZZLE_ONE: - r = alloc_immd(pc, 1.0); - break; + if (sgn == TGSI_UTIL_SIGN_TOGGLE || sgn == TGSI_UTIL_SIGN_SET) + return alloc_immd(pc, -1.0); + return alloc_immd(pc, 1.0); default: assert(0); break; } - switch (tgsi_util_get_full_src_register_sign_mode(src, chan)) { + switch (sgn) { case TGSI_UTIL_SIGN_KEEP: break; case TGSI_UTIL_SIGN_CLEAR: @@ -933,14 +1232,21 @@ tgsi_src(struct nv50_pc *pc, int chan, const struct tgsi_full_src_register *src) r = temp; break; case TGSI_UTIL_SIGN_TOGGLE: - temp = temp_temp(pc); - emit_neg(pc, temp, r); - r = temp; + if (neg) + r->neg = 1; + else { + temp = temp_temp(pc); + emit_neg(pc, temp, r); + r = temp; + } break; case TGSI_UTIL_SIGN_SET: temp = temp_temp(pc); emit_abs(pc, temp, r); - emit_neg(pc, temp, r); + if (neg) + temp->neg = 1; + else + emit_neg(pc, temp, temp); r = temp; break; default: @@ -951,12 +1257,40 @@ tgsi_src(struct nv50_pc *pc, int chan, const struct tgsi_full_src_register *src) return r; } +/* returns TRUE if instruction can overwrite sources before they're read */ +static boolean +direct2dest_op(const struct tgsi_full_instruction *insn) +{ + if (insn->Instruction.Saturate) + return FALSE; + + switch (insn->Instruction.Opcode) { + case TGSI_OPCODE_COS: + case TGSI_OPCODE_DP3: + case TGSI_OPCODE_DP4: + case TGSI_OPCODE_DPH: + case TGSI_OPCODE_KIL: + case TGSI_OPCODE_LIT: + case TGSI_OPCODE_POW: + case TGSI_OPCODE_RCP: + case TGSI_OPCODE_RSQ: + case TGSI_OPCODE_SCS: + case TGSI_OPCODE_SIN: + case TGSI_OPCODE_TEX: + case TGSI_OPCODE_TXP: + return FALSE; + default: + return TRUE; + } +} + static boolean nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) { const struct tgsi_full_instruction *inst = &tok->FullInstruction; struct nv50_reg *rdst[4], *dst[4], *src[3][4], *temp; unsigned mask, sat, unit; + boolean assimilate = FALSE; int i, c; mask = inst->FullDstRegisters[0].DstRegister.WriteMask; @@ -967,6 +1301,10 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) dst[c] = tgsi_dst(pc, c, &inst->FullDstRegisters[0]); else dst[c] = NULL; + rdst[c] = NULL; + src[0][c] = NULL; + src[1][c] = NULL; + src[2][c] = NULL; } for (i = 0; i < inst->Instruction.NumSrcRegs; i++) { @@ -976,7 +1314,8 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) unit = fs->SrcRegister.Index; for (c = 0; c < 4; c++) - src[i][c] = tgsi_src(pc, c, fs); + src[i][c] = tgsi_src(pc, c, fs, + negate_supported(inst, i)); } if (sat) { @@ -984,6 +1323,25 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) rdst[c] = dst[c]; dst[c] = temp_temp(pc); } + } else + if (direct2dest_op(inst)) { + for (c = 0; c < 4; c++) { + if (!dst[c] || dst[c]->type != P_TEMP) + continue; + + for (i = c + 1; i < 4; i++) { + if (dst[c] == src[0][i] || + dst[c] == src[1][i] || + dst[c] == src[2][i]) + break; + } + if (i == 4) + continue; + + assimilate = TRUE; + rdst[c] = dst[c]; + dst[c] = alloc_temp(pc, NULL); + } } switch (inst->Instruction.Opcode) { @@ -1002,7 +1360,7 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) } break; case TGSI_OPCODE_COS: - temp = alloc_temp(pc, NULL); + temp = temp_temp(pc); emit_precossin(pc, temp, src[0][0]); emit_flop(pc, 5, temp, temp); for (c = 0; c < 4; c++) { @@ -1012,7 +1370,7 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) } break; case TGSI_OPCODE_DP3: - temp = alloc_temp(pc, NULL); + temp = temp_temp(pc); emit_mul(pc, temp, src[0][0], src[1][0]); emit_mad(pc, temp, src[0][1], src[1][1], temp); emit_mad(pc, temp, src[0][2], src[1][2], temp); @@ -1021,10 +1379,9 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) continue; emit_mov(pc, dst[c], temp); } - free_temp(pc, temp); break; case TGSI_OPCODE_DP4: - temp = alloc_temp(pc, NULL); + temp = temp_temp(pc); emit_mul(pc, temp, src[0][0], src[1][0]); emit_mad(pc, temp, src[0][1], src[1][1], temp); emit_mad(pc, temp, src[0][2], src[1][2], temp); @@ -1034,10 +1391,9 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) continue; emit_mov(pc, dst[c], temp); } - free_temp(pc, temp); break; case TGSI_OPCODE_DPH: - temp = alloc_temp(pc, NULL); + temp = temp_temp(pc); emit_mul(pc, temp, src[0][0], src[1][0]); emit_mad(pc, temp, src[0][1], src[1][1], temp); emit_mad(pc, temp, src[0][2], src[1][2], temp); @@ -1047,7 +1403,6 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) continue; emit_mov(pc, dst[c], temp); } - free_temp(pc, temp); break; case TGSI_OPCODE_DST: { @@ -1064,7 +1419,7 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) } break; case TGSI_OPCODE_EX2: - temp = alloc_temp(pc, NULL); + temp = temp_temp(pc); emit_preex2(pc, temp, src[0][0]); emit_flop(pc, 6, temp, temp); for (c = 0; c < 4; c++) { @@ -1072,7 +1427,6 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) continue; emit_mov(pc, dst[c], temp); } - free_temp(pc, temp); break; case TGSI_OPCODE_FLR: for (c = 0; c < 4; c++) { @@ -1082,26 +1436,26 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) } break; case TGSI_OPCODE_FRC: - temp = alloc_temp(pc, NULL); + temp = temp_temp(pc); for (c = 0; c < 4; c++) { if (!(mask & (1 << c))) continue; emit_flr(pc, temp, src[0][c]); emit_sub(pc, dst[c], src[0][c], temp); } - free_temp(pc, temp); break; case TGSI_OPCODE_KIL: emit_kil(pc, src[0][0]); emit_kil(pc, src[0][1]); emit_kil(pc, src[0][2]); emit_kil(pc, src[0][3]); + pc->p->cfg.fp.regs[2] |= 0x00100000; break; case TGSI_OPCODE_LIT: emit_lit(pc, &dst[0], mask, &src[0][0]); break; case TGSI_OPCODE_LG2: - temp = alloc_temp(pc, NULL); + temp = temp_temp(pc); emit_flop(pc, 3, temp, src[0][0]); for (c = 0; c < 4; c++) { if (!(mask & (1 << c))) @@ -1110,15 +1464,12 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) } break; case TGSI_OPCODE_LRP: + temp = temp_temp(pc); for (c = 0; c < 4; c++) { if (!(mask & (1 << c))) continue; - /*XXX: we can do better than this */ - temp = alloc_temp(pc, NULL); - emit_neg(pc, temp, src[0][c]); - emit_mad(pc, temp, temp, src[2][c], src[2][c]); - emit_mad(pc, dst[c], src[0][c], src[1][c], temp); - free_temp(pc, temp); + emit_sub(pc, temp, src[1][c], src[2][c]); + emit_mad(pc, dst[c], temp, src[0][c], src[2][c]); } break; case TGSI_OPCODE_MAD: @@ -1157,36 +1508,39 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) } break; case TGSI_OPCODE_POW: - temp = alloc_temp(pc, NULL); + temp = temp_temp(pc); emit_pow(pc, temp, src[0][0], src[1][0]); for (c = 0; c < 4; c++) { if (!(mask & (1 << c))) continue; emit_mov(pc, dst[c], temp); } - free_temp(pc, temp); break; case TGSI_OPCODE_RCP: - for (c = 0; c < 4; c++) { + for (c = 3; c >= 0; c--) { if (!(mask & (1 << c))) continue; emit_flop(pc, 0, dst[c], src[0][0]); } break; case TGSI_OPCODE_RSQ: - for (c = 0; c < 4; c++) { + for (c = 3; c >= 0; c--) { if (!(mask & (1 << c))) continue; emit_flop(pc, 2, dst[c], src[0][0]); } break; case TGSI_OPCODE_SCS: - temp = alloc_temp(pc, NULL); + temp = temp_temp(pc); emit_precossin(pc, temp, src[0][0]); if (mask & (1 << 0)) emit_flop(pc, 5, dst[0], temp); if (mask & (1 << 1)) emit_flop(pc, 4, dst[1], temp); + if (mask & (1 << 2)) + emit_mov_immdval(pc, dst[2], 0.0); + if (mask & (1 << 3)) + emit_mov_immdval(pc, dst[3], 1.0); break; case TGSI_OPCODE_SGE: for (c = 0; c < 4; c++) { @@ -1196,7 +1550,7 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) } break; case TGSI_OPCODE_SIN: - temp = alloc_temp(pc, NULL); + temp = temp_temp(pc); emit_precossin(pc, temp, src[0][0]); emit_flop(pc, 4, temp, temp); for (c = 0; c < 4; c++) { @@ -1220,33 +1574,15 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) } break; case TGSI_OPCODE_TEX: + emit_tex(pc, dst, mask, src[0], unit, + inst->InstructionExtTexture.Texture, FALSE); + break; case TGSI_OPCODE_TXP: - { - struct nv50_reg *t[4]; - struct nv50_program_exec *e; - - alloc_temp4(pc, t, 0); - emit_mov(pc, t[0], src[0][0]); - emit_mov(pc, t[1], src[0][1]); - - e = exec(pc); - e->inst[0] = 0xf6400000; - e->inst[0] |= (unit << 9); - set_long(pc, e); - e->inst[1] |= 0x0000c004; - set_dst(pc, t[0], e); - emit(pc, e); - - if (mask & (1 << 0)) emit_mov(pc, dst[0], t[0]); - if (mask & (1 << 1)) emit_mov(pc, dst[1], t[1]); - if (mask & (1 << 2)) emit_mov(pc, dst[2], t[2]); - if (mask & (1 << 3)) emit_mov(pc, dst[3], t[3]); - - free_temp4(pc, t); - } + emit_tex(pc, dst, mask, src[0], unit, + inst->InstructionExtTexture.Texture, TRUE); break; case TGSI_OPCODE_XPD: - temp = alloc_temp(pc, NULL); + temp = temp_temp(pc); if (mask & (1 << 0)) { emit_mul(pc, temp, src[0][2], src[1][1]); emit_msb(pc, dst[0], src[0][1], src[1][2], temp); @@ -1259,7 +1595,8 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) emit_mul(pc, temp, src[0][1], src[1][0]); emit_msb(pc, dst[2], src[0][0], src[1][1], temp); } - free_temp(pc, temp); + if (mask & (1 << 3)) + emit_mov_immdval(pc, dst[3], 1.0); break; case TGSI_OPCODE_END: break; @@ -1270,21 +1607,26 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) if (sat) { for (c = 0; c < 4; c++) { - struct nv50_program_exec *e; - if (!(mask & (1 << c))) continue; - e = exec(pc); + emit_cvt(pc, rdst[c], dst[c], -1, CVTOP_SAT, + CVT_F32_F32); + } + } else if (assimilate) { + for (c = 0; c < 4; c++) + if (rdst[c]) + assimilate_temp(pc, rdst[c], dst[c]); + } - e->inst[0] = 0xa0000000; /* cvt */ - set_long(pc, e); - e->inst[1] |= (6 << 29); /* cvt */ - e->inst[1] |= 0x04000000; /* 32 bit */ - e->inst[1] |= (1 << 14); /* src .f32 */ - e->inst[1] |= ((1 << 5) << 14); /* .sat */ - set_dst(pc, rdst[c], e); - set_src_0(pc, dst[c], e); - emit(pc, e); + for (i = 0; i < inst->Instruction.NumSrcRegs; i++) { + for (c = 0; c < 4; c++) { + if (!src[i][c]) + continue; + if (src[i][c]->index == -1 && src[i][c]->type == P_IMMD) + FREE(src[i][c]); + else + if (src[i][c]->acc == pc->insn_cur) + release_hw(pc, src[i][c]); } } @@ -1292,12 +1634,169 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) return TRUE; } +/* Adjust a bitmask that indicates what components of a source are used, + * we use this in tx_prep so we only load interpolants that are needed. + */ +static void +insn_adjust_mask(const struct tgsi_full_instruction *insn, unsigned *mask) +{ + const struct tgsi_instruction_ext_texture *tex; + + switch (insn->Instruction.Opcode) { + case TGSI_OPCODE_DP3: + *mask = 0x7; + break; + case TGSI_OPCODE_DP4: + case TGSI_OPCODE_DPH: + *mask = 0xF; + break; + case TGSI_OPCODE_LIT: + *mask = 0xB; + break; + case TGSI_OPCODE_RCP: + case TGSI_OPCODE_RSQ: + *mask = 0x1; + break; + case TGSI_OPCODE_TEX: + case TGSI_OPCODE_TXP: + assert(insn->Instruction.Extended); + tex = &insn->InstructionExtTexture; + + *mask = 0x7; + if (tex->Texture == TGSI_TEXTURE_1D) + *mask = 0x1; + else + if (tex->Texture == TGSI_TEXTURE_2D) + *mask = 0x3; + + if (insn->Instruction.Opcode == TGSI_OPCODE_TXP) + *mask |= 0x8; + break; + default: + break; + } +} + +static void +prep_inspect_insn(struct nv50_pc *pc, const union tgsi_full_token *tok, + unsigned *r_usage[2]) +{ + const struct tgsi_full_instruction *insn; + const struct tgsi_full_src_register *src; + const struct tgsi_dst_register *dst; + + unsigned i, c, k, n, mask, *acc_p; + + insn = &tok->FullInstruction; + dst = &insn->FullDstRegisters[0].DstRegister; + mask = dst->WriteMask; + + if (!r_usage[0]) + r_usage[0] = CALLOC(pc->temp_nr * 4, sizeof(unsigned)); + if (!r_usage[1]) + r_usage[1] = CALLOC(pc->attr_nr * 4, sizeof(unsigned)); + + if (dst->File == TGSI_FILE_TEMPORARY) { + for (c = 0; c < 4; c++) { + if (!(mask & (1 << c))) + continue; + r_usage[0][dst->Index * 4 + c] = pc->insn_nr; + } + } + + for (i = 0; i < insn->Instruction.NumSrcRegs; i++) { + src = &insn->FullSrcRegisters[i]; + + switch (src->SrcRegister.File) { + case TGSI_FILE_TEMPORARY: + acc_p = r_usage[0]; + break; + case TGSI_FILE_INPUT: + acc_p = r_usage[1]; + break; + default: + continue; + } + + insn_adjust_mask(insn, &mask); + + for (c = 0; c < 4; c++) { + if (!(mask & (1 << c))) + continue; + + k = tgsi_util_get_full_src_register_extswizzle(src, c); + switch (k) { + case TGSI_EXTSWIZZLE_X: + case TGSI_EXTSWIZZLE_Y: + case TGSI_EXTSWIZZLE_Z: + case TGSI_EXTSWIZZLE_W: + n = src->SrcRegister.Index * 4 + k; + acc_p[n] = pc->insn_nr; + break; + default: + break; + } + } + } +} + +static unsigned +load_fp_attrib(struct nv50_pc *pc, int i, unsigned *acc, int *mid, + int *aid, int *p_oid) +{ + struct nv50_reg *iv; + int oid, c, n; + unsigned mask = 0; + + iv = (pc->interp_mode[i] & INTERP_CENTROID) ? pc->iv_c : pc->iv_p; + + for (c = 0, n = i * 4; c < 4; c++, n++) { + oid = (*p_oid)++; + pc->attr[n].type = P_TEMP; + pc->attr[n].index = i; + + if (pc->attr[n].acc == acc[n]) + continue; + mask |= (1 << c); + + pc->attr[n].acc = acc[n]; + pc->attr[n].rhw = pc->attr[n].hw = -1; + alloc_reg(pc, &pc->attr[n]); + + pc->attr[n].rhw = (*aid)++; + emit_interp(pc, &pc->attr[n], iv, pc->interp_mode[i]); + + pc->p->cfg.fp.map[(*mid) / 4] |= oid << (8 * ((*mid) % 4)); + (*mid)++; + pc->p->cfg.fp.regs[1] += 0x00010001; + } + + return mask; +} + static boolean nv50_program_tx_prep(struct nv50_pc *pc) { struct tgsi_parse_context p; boolean ret = FALSE; unsigned i, c; + unsigned fcol, bcol, fcrd, depr; + + /* count (centroid) perspective interpolations */ + unsigned centroid_loads = 0; + unsigned perspect_loads = 0; + + /* track register access for temps and attrs */ + unsigned *r_usage[2]; + r_usage[0] = NULL; + r_usage[1] = NULL; + + depr = fcol = bcol = fcrd = 0xffff; + + if (pc->p->type == PIPE_SHADER_FRAGMENT) { + pc->p->cfg.fp.regs[0] = 0x01000404; + pc->p->cfg.fp.regs[1] = 0x00000400; + } tgsi_parse_init(&p, pc->p->pipe.tokens); while (!tgsi_parse_end_of_tokens(&p)) { @@ -1319,9 +1818,10 @@ nv50_program_tx_prep(struct nv50_pc *pc) case TGSI_TOKEN_TYPE_DECLARATION: { const struct tgsi_full_declaration *d; - unsigned last; + unsigned last, first, mode; d = &p.FullToken.FullDeclaration; + first = d->DeclarationRange.First; last = d->DeclarationRange.Last; switch (d->Declaration.File) { @@ -1332,10 +1832,69 @@ nv50_program_tx_prep(struct nv50_pc *pc) case TGSI_FILE_OUTPUT: if (pc->result_nr < (last + 1)) pc->result_nr = last + 1; + + if (!d->Declaration.Semantic) + break; + + switch (d->Semantic.SemanticName) { + case TGSI_SEMANTIC_POSITION: + depr = first; + pc->p->cfg.fp.regs[2] |= 0x00000100; + pc->p->cfg.fp.regs[3] |= 0x00000011; + break; + default: + break; + } + break; case TGSI_FILE_INPUT: + { if (pc->attr_nr < (last + 1)) pc->attr_nr = last + 1; + + if (pc->p->type != PIPE_SHADER_FRAGMENT) + break; + + switch (d->Declaration.Interpolate) { + case TGSI_INTERPOLATE_CONSTANT: + mode = INTERP_FLAT; + break; + case TGSI_INTERPOLATE_PERSPECTIVE: + mode = INTERP_PERSPECTIVE; + break; + default: + mode = INTERP_LINEAR; + break; + } + + if (d->Declaration.Semantic) { + switch (d->Semantic.SemanticName) { + case TGSI_SEMANTIC_POSITION: + fcrd = first; + break; + case TGSI_SEMANTIC_COLOR: + fcol = first; + mode = INTERP_PERSPECTIVE; + break; + case TGSI_SEMANTIC_BCOLOR: + bcol = first; + mode = INTERP_PERSPECTIVE; + break; + } + } + + if (d->Declaration.Centroid) { + mode |= INTERP_CENTROID; + if (mode & INTERP_PERSPECTIVE) + centroid_loads++; + } else + if (mode & INTERP_PERSPECTIVE) + perspect_loads++; + + assert(last < 32); + for (i = first; i <= last; i++) + pc->interp_mode[i] = mode; + } break; case TGSI_FILE_CONSTANT: if (pc->param_nr < (last + 1)) @@ -1351,6 +1910,8 @@ nv50_program_tx_prep(struct nv50_pc *pc) } break; case TGSI_TOKEN_TYPE_INSTRUCTION: + pc->insn_nr++; + prep_inspect_insn(pc, tok, r_usage); break; default: break; @@ -1366,56 +1927,95 @@ nv50_program_tx_prep(struct nv50_pc *pc) for (c = 0; c < 4; c++) { pc->temp[i*4+c].type = P_TEMP; pc->temp[i*4+c].hw = -1; + pc->temp[i*4+c].rhw = -1; pc->temp[i*4+c].index = i; + pc->temp[i*4+c].acc = r_usage[0][i*4+c]; } } } if (pc->attr_nr) { - struct nv50_reg *iv = NULL; - int aid = 0; + int oid = 4, mid = 4, aid = 0; + /* oid = VP output id + * aid = FP attribute/interpolant id + * mid = VP output mapping field ID + */ pc->attr = CALLOC(pc->attr_nr * 4, sizeof(struct nv50_reg)); if (!pc->attr) goto out_err; if (pc->p->type == PIPE_SHADER_FRAGMENT) { - iv = alloc_temp(pc, NULL); - emit_interp(pc, iv, iv, NULL); - emit_flop(pc, 0, iv, iv); - aid++; - } + /* position should be loaded first */ + if (fcrd != 0xffff) { + unsigned mask; + mid = 0; + mask = load_fp_attrib(pc, fcrd, r_usage[1], + &mid, &aid, &oid); + oid = 0; + pc->p->cfg.fp.regs[1] |= (mask << 24); + pc->p->cfg.fp.map[0] = 0x04040404 * fcrd; + } + pc->p->cfg.fp.map[0] += 0x03020100; - for (i = 0; i < pc->attr_nr; i++) { - struct nv50_reg *a = &pc->attr[i*4]; + /* should do MAD fcrd.xy, fcrd, SOME_CONST, fcrd */ - for (c = 0; c < 4; c++) { - if (pc->p->type == PIPE_SHADER_FRAGMENT) { - struct nv50_reg *at = - alloc_temp(pc, NULL); - pc->attr[i*4+c].type = at->type; - pc->attr[i*4+c].hw = at->hw; - pc->attr[i*4+c].index = at->index; + if (perspect_loads) { + pc->iv_p = alloc_temp(pc, NULL); + + if (!(pc->p->cfg.fp.regs[1] & 0x08000000)) { + pc->p->cfg.fp.regs[1] |= 0x08000000; + pc->iv_p->rhw = aid++; + emit_interp(pc, pc->iv_p, NULL, + INTERP_LINEAR); + emit_flop(pc, 0, pc->iv_p, pc->iv_p); } else { - pc->p->cfg.vp.attr[aid/32] |= - (1 << (aid % 32)); - pc->attr[i*4+c].type = P_ATTR; - pc->attr[i*4+c].hw = aid++; - pc->attr[i*4+c].index = i; + pc->iv_p->rhw = aid - 1; + emit_flop(pc, 0, pc->iv_p, + &pc->attr[fcrd * 4 + 3]); } } - if (pc->p->type != PIPE_SHADER_FRAGMENT) - continue; + if (centroid_loads) { + pc->iv_c = alloc_temp(pc, NULL); + pc->iv_c->rhw = pc->iv_p ? aid - 1 : aid++; + emit_interp(pc, pc->iv_c, NULL, + INTERP_CENTROID); + emit_flop(pc, 0, pc->iv_c, pc->iv_c); + pc->p->cfg.fp.regs[1] |= 0x08000000; + } - emit_interp(pc, &a[0], &a[0], iv); - emit_interp(pc, &a[1], &a[1], iv); - emit_interp(pc, &a[2], &a[2], iv); - emit_interp(pc, &a[3], &a[3], iv); - } + for (c = 0; c < 4; c++) { + /* I don't know what these values do, but + * let's set them like the blob does: + */ + if (fcol != 0xffff && r_usage[1][fcol * 4 + c]) + pc->p->cfg.fp.regs[0] += 0x00010000; + if (bcol != 0xffff && r_usage[1][bcol * 4 + c]) + pc->p->cfg.fp.regs[0] += 0x00010000; + } - if (iv) - free_temp(pc, iv); + for (i = 0; i < pc->attr_nr; i++) + load_fp_attrib(pc, i, r_usage[1], + &mid, &aid, &oid); + + if (pc->iv_p) + free_temp(pc, pc->iv_p); + if (pc->iv_c) + free_temp(pc, pc->iv_c); + + pc->p->cfg.fp.high_map = (mid / 4); + pc->p->cfg.fp.high_map += ((mid % 4) ? 1 : 0); + } else { + /* vertex program */ + for (i = 0; i < pc->attr_nr * 4; i++) { + pc->p->cfg.vp.attr[aid / 32] |= + (1 << (aid % 32)); + pc->attr[i].type = P_ATTR; + pc->attr[i].hw = aid++; + pc->attr[i].index = i / 4; + } + } } if (pc->result_nr) { @@ -1430,12 +2030,20 @@ nv50_program_tx_prep(struct nv50_pc *pc) if (pc->p->type == PIPE_SHADER_FRAGMENT) { pc->result[i*4+c].type = P_TEMP; pc->result[i*4+c].hw = -1; + pc->result[i*4+c].rhw = (i == depr) ? + -1 : rid++; } else { pc->result[i*4+c].type = P_RESULT; pc->result[i*4+c].hw = rid++; } pc->result[i*4+c].index = i; } + + if (pc->p->type == PIPE_SHADER_FRAGMENT && + depr != 0xffff) { + pc->result[depr * 4 + 2].rhw = + (pc->result_nr - 1) * 4; + } } } @@ -1456,7 +2064,7 @@ nv50_program_tx_prep(struct nv50_pc *pc) } if (pc->immd_nr) { - int rid = pc->param_nr * 4; + int rid = 0; pc->immd = CALLOC(pc->immd_nr * 4, sizeof(struct nv50_reg)); if (!pc->immd) @@ -1473,15 +2081,38 @@ nv50_program_tx_prep(struct nv50_pc *pc) ret = TRUE; out_err: + if (r_usage[0]) + FREE(r_usage[0]); + if (r_usage[1]) + FREE(r_usage[1]); + tgsi_parse_free(&p); return ret; } +static void +free_nv50_pc(struct nv50_pc *pc) +{ + if (pc->immd) + FREE(pc->immd); + if (pc->param) + FREE(pc->param); + if (pc->result) + FREE(pc->result); + if (pc->attr) + FREE(pc->attr); + if (pc->temp) + FREE(pc->temp); + + FREE(pc); +} + static boolean nv50_program_tx(struct nv50_program *p) { struct tgsi_parse_context parse; struct nv50_pc *pc; + unsigned k; boolean ret; pc = CALLOC_STRUCT(nv50_pc); @@ -1498,10 +2129,16 @@ nv50_program_tx(struct nv50_program *p) while (!tgsi_parse_end_of_tokens(&parse)) { const union tgsi_full_token *tok = &parse.FullToken; + /* don't allow half insn/immd on first and last instruction */ + pc->allow32 = TRUE; + if (pc->insn_cur == 0 || pc->insn_cur + 2 == pc->insn_nr) + pc->allow32 = FALSE; + tgsi_parse_token(&parse); switch (tok->Token.Type) { case TGSI_TOKEN_TYPE_INSTRUCTION: + ++pc->insn_cur; ret = nv50_program_tx_insn(pc, tok); if (ret == FALSE) goto out_err; @@ -1515,8 +2152,40 @@ nv50_program_tx(struct nv50_program *p) struct nv50_reg out; out.type = P_TEMP; - for (out.hw = 0; out.hw < pc->result_nr * 4; out.hw++) - emit_mov(pc, &out, &pc->result[out.hw]); + for (k = 0; k < pc->result_nr * 4; k++) { + if (pc->result[k].rhw == -1) + continue; + if (pc->result[k].hw != pc->result[k].rhw) { + out.hw = pc->result[k].rhw; + emit_mov(pc, &out, &pc->result[k]); + } + if (pc->p->cfg.high_result < (pc->result[k].rhw + 1)) + pc->p->cfg.high_result = pc->result[k].rhw + 1; + } + } + + /* look for single half instructions and make them long */ + struct nv50_program_exec *e, *e_prev; + + for (k = 0, e = pc->p->exec_head, e_prev = NULL; e; e = e->next) { + if (!is_long(e)) + k++; + + if (!e->next || is_long(e->next)) { + if (k & 1) + convert_to_long(pc, e); + k = 0; + } + + if (e->next) + e_prev = e; + } + + if (!is_long(pc->p->exec_tail)) { + /* this may occur if moving FP results */ + assert(e_prev && !is_long(e_prev)); + convert_to_long(pc, e_prev); + convert_to_long(pc, pc->p->exec_tail); } assert(is_long(pc->p->exec_tail) && !is_immd(pc->p->exec_head)); @@ -1530,6 +2199,7 @@ out_err: tgsi_parse_free(&parse); out_cleanup: + free_nv50_pc(pc); return ret; } @@ -1543,16 +2213,16 @@ nv50_program_validate(struct nv50_context *nv50, struct nv50_program *p) static void nv50_program_upload_data(struct nv50_context *nv50, float *map, - unsigned start, unsigned count) + unsigned start, unsigned count, unsigned cbuf) { - struct nouveau_channel *chan = nv50->screen->nvws->channel; + struct nouveau_channel *chan = nv50->screen->base.channel; struct nouveau_grobj *tesla = nv50->screen->tesla; while (count) { unsigned nr = count > 2047 ? 2047 : count; BEGIN_RING(chan, tesla, 0x00000f00, 1); - OUT_RING (chan, (NV50_CB_PMISC << 0) | (start << 8)); + OUT_RING (chan, (cbuf << 0) | (start << 8)); BEGIN_RING(chan, tesla, 0x40000f04, nr); OUT_RINGp (chan, map, nr); @@ -1565,70 +2235,93 @@ nv50_program_upload_data(struct nv50_context *nv50, float *map, static void nv50_program_validate_data(struct nv50_context *nv50, struct nv50_program *p) { - struct nouveau_winsys *nvws = nv50->screen->nvws; - struct pipe_winsys *ws = nv50->pipe.winsys; - unsigned nr = p->param_nr + p->immd_nr; + struct pipe_screen *pscreen = nv50->pipe.screen; - if (!p->data && nr) { - struct nouveau_resource *heap = nv50->screen->vp_data_heap; + if (!p->data[0] && p->immd_nr) { + struct nouveau_resource *heap = nv50->screen->immd_heap[0]; - if (nvws->res_alloc(heap, nr, p, &p->data)) { - while (heap->next && heap->size < nr) { + if (nouveau_resource_alloc(heap, p->immd_nr, p, &p->data[0])) { + while (heap->next && heap->size < p->immd_nr) { struct nv50_program *evict = heap->next->priv; - nvws->res_free(&evict->data); + nouveau_resource_free(&evict->data[0]); } - if (nvws->res_alloc(heap, nr, p, &p->data)) + if (nouveau_resource_alloc(heap, p->immd_nr, p, + &p->data[0])) assert(0); } + + /* immediates only need to be uploaded again when freed */ + nv50_program_upload_data(nv50, p->immd, p->data[0]->start, + p->immd_nr, NV50_CB_PMISC); } - if (p->param_nr) { - float *map = ws->buffer_map(ws, nv50->constbuf[p->type], - PIPE_BUFFER_USAGE_CPU_READ); - nv50_program_upload_data(nv50, map, p->data->start, - p->param_nr); - ws->buffer_unmap(ws, nv50->constbuf[p->type]); + if (!p->data[1] && p->param_nr) { + struct nouveau_resource *heap = + nv50->screen->parm_heap[p->type]; + + if (nouveau_resource_alloc(heap, p->param_nr, p, &p->data[1])) { + while (heap->next && heap->size < p->param_nr) { + struct nv50_program *evict = heap->next->priv; + nouveau_resource_free(&evict->data[1]); + } + + if (nouveau_resource_alloc(heap, p->param_nr, p, + &p->data[1])) + assert(0); + } } - if (p->immd_nr) { - nv50_program_upload_data(nv50, p->immd, - p->data->start + p->param_nr, - p->immd_nr); + if (p->param_nr) { + unsigned cbuf = NV50_CB_PVP; + float *map = pipe_buffer_map(pscreen, nv50->constbuf[p->type], + PIPE_BUFFER_USAGE_CPU_READ); + if (p->type == PIPE_SHADER_FRAGMENT) + cbuf = NV50_CB_PFP; + nv50_program_upload_data(nv50, map, p->data[1]->start, + p->param_nr, cbuf); + pipe_buffer_unmap(pscreen, nv50->constbuf[p->type]); } } static void nv50_program_validate_code(struct nv50_context *nv50, struct nv50_program *p) { - struct nouveau_channel *chan = nv50->screen->nvws->channel; + struct nouveau_channel *chan = nv50->screen->base.channel; struct nouveau_grobj *tesla = nv50->screen->tesla; - struct pipe_screen *screen = nv50->pipe.screen; struct nv50_program_exec *e; struct nouveau_stateobj *so; const unsigned flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_WR; unsigned start, count, *up, *ptr; boolean upload = FALSE; - if (!p->buffer) { - p->buffer = screen->buffer_create(screen, 0x100, 0, p->exec_size * 4); + if (!p->bo) { + nouveau_bo_new(chan->device, NOUVEAU_BO_VRAM, 0x100, + p->exec_size * 4, &p->bo); upload = TRUE; } - if (p->data && p->data->start != p->data_start) { + if ((p->data[0] && p->data[0]->start != p->data_start[0]) || + (p->data[1] && p->data[1]->start != p->data_start[1])) { for (e = p->exec_head; e; e = e->next) { - unsigned ei, ci; + unsigned ei, ci, bs; if (e->param.index < 0) continue; + bs = (e->inst[1] >> 22) & 0x07; + assert(bs < 2); ei = e->param.shift >> 5; - ci = e->param.index + p->data->start; + ci = e->param.index + p->data[bs]->start; e->inst[ei] &= ~e->param.mask; e->inst[ei] |= (ci << e->param.shift); } - p->data_start = p->data->start; + if (p->data[0]) + p->data_start[0] = p->data[0]->start; + if (p->data[1]) + p->data_start[1] = p->data[1]->start; + upload = TRUE; } @@ -1637,13 +2330,11 @@ nv50_program_validate_code(struct nv50_context *nv50, struct nv50_program *p) #ifdef NV50_PROGRAM_DUMP NOUVEAU_ERR("-------\n"); - up = ptr = MALLOC(p->exec_size * 4); for (e = p->exec_head; e; e = e->next) { NOUVEAU_ERR("0x%08x\n", e->inst[0]); if (is_long(e)) NOUVEAU_ERR("0x%08x\n", e->inst[1]); } - #endif up = ptr = MALLOC(p->exec_size * 4); @@ -1655,20 +2346,20 @@ nv50_program_validate_code(struct nv50_context *nv50, struct nv50_program *p) so = so_new(4,2); so_method(so, nv50->screen->tesla, 0x1280, 3); - so_reloc (so, p->buffer, 0, flags | NOUVEAU_BO_HIGH, 0, 0); - so_reloc (so, p->buffer, 0, flags | NOUVEAU_BO_LOW, 0, 0); + so_reloc (so, p->bo, 0, flags | NOUVEAU_BO_HIGH, 0, 0); + so_reloc (so, p->bo, 0, flags | NOUVEAU_BO_LOW, 0, 0); so_data (so, (NV50_CB_PUPLOAD << 16) | 0x0800); //(p->exec_size * 4)); start = 0; count = p->exec_size; while (count) { - struct nouveau_winsys *nvws = nv50->screen->nvws; + struct nouveau_channel *chan = nv50->screen->base.channel; unsigned nr; - so_emit(nvws, so); + so_emit(chan, so); nr = MIN2(count, 2047); - nr = MIN2(nvws->channel->pushbuf->remaining, nr); - if (nvws->channel->pushbuf->remaining < (nr + 3)) { + nr = MIN2(chan->pushbuf->remaining, nr); + if (chan->pushbuf->remaining < (nr + 3)) { FIRE_RING(chan); continue; } @@ -1704,10 +2395,10 @@ nv50_vertprog_validate(struct nv50_context *nv50) so = so_new(13, 2); so_method(so, tesla, NV50TCL_VP_ADDRESS_HIGH, 2); - so_reloc (so, p->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | - NOUVEAU_BO_HIGH, 0, 0); - so_reloc (so, p->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | - NOUVEAU_BO_LOW, 0, 0); + so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | + NOUVEAU_BO_HIGH, 0, 0); + so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | + NOUVEAU_BO_LOW, 0, 0); so_method(so, tesla, 0x1650, 2); so_data (so, p->cfg.vp.attr[0]); so_data (so, p->cfg.vp.attr[1]); @@ -1728,6 +2419,7 @@ nv50_fragprog_validate(struct nv50_context *nv50) struct nouveau_grobj *tesla = nv50->screen->tesla; struct nv50_program *p = nv50->fragprog; struct nouveau_stateobj *so; + unsigned i; if (!p->translated) { nv50_program_validate(nv50, p); @@ -1740,22 +2432,27 @@ nv50_fragprog_validate(struct nv50_context *nv50) so = so_new(64, 2); so_method(so, tesla, NV50TCL_FP_ADDRESS_HIGH, 2); - so_reloc (so, p->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | - NOUVEAU_BO_HIGH, 0, 0); - so_reloc (so, p->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | - NOUVEAU_BO_LOW, 0, 0); + so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | + NOUVEAU_BO_HIGH, 0, 0); + so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | + NOUVEAU_BO_LOW, 0, 0); so_method(so, tesla, 0x1904, 4); - so_data (so, 0x00040404); /* p: 0x01000404 */ + so_data (so, p->cfg.fp.regs[0]); /* 0x01000404 / 0x00040404 */ so_data (so, 0x00000004); so_data (so, 0x00000000); so_data (so, 0x00000000); - so_method(so, tesla, 0x16bc, 3); /*XXX: fixme */ - so_data (so, 0x03020100); - so_data (so, 0x07060504); - so_data (so, 0x0b0a0908); + so_method(so, tesla, 0x16bc, p->cfg.fp.high_map); + for (i = 0; i < p->cfg.fp.high_map; i++) + so_data(so, p->cfg.fp.map[i]); so_method(so, tesla, 0x1988, 2); - so_data (so, 0x08080408); //0x08040404); /* p: 0x0f000401 */ + so_data (so, p->cfg.fp.regs[1]); /* 0x08040404 / 0x0f000401 */ so_data (so, p->cfg.high_temp); + so_method(so, tesla, 0x1298, 1); + so_data (so, p->cfg.high_result); + so_method(so, tesla, 0x19a8, 1); + so_data (so, p->cfg.fp.regs[2]); + so_method(so, tesla, 0x196c, 1); + so_data (so, p->cfg.fp.regs[3]); so_method(so, tesla, 0x1414, 1); so_data (so, 0); /* program start offset */ so_ref(so, &nv50->state.fragprog); @@ -1765,8 +2462,6 @@ nv50_fragprog_validate(struct nv50_context *nv50) void nv50_program_destroy(struct nv50_context *nv50, struct nv50_program *p) { - struct pipe_screen *pscreen = nv50->pipe.screen; - while (p->exec_head) { struct nv50_program_exec *e = p->exec_head; @@ -1776,10 +2471,10 @@ nv50_program_destroy(struct nv50_context *nv50, struct nv50_program *p) p->exec_tail = NULL; p->exec_size = 0; - if (p->buffer) - pipe_buffer_reference(&p->buffer, NULL); + nouveau_bo_ref(NULL, &p->bo); - nv50->screen->nvws->res_free(&p->data); + nouveau_resource_free(&p->data[0]); + nouveau_resource_free(&p->data[1]); p->translated = 0; } diff --git a/src/gallium/drivers/nv50/nv50_program.h b/src/gallium/drivers/nv50/nv50_program.h index 78deed6a384..096e0476aab 100644 --- a/src/gallium/drivers/nv50/nv50_program.h +++ b/src/gallium/drivers/nv50/nv50_program.h @@ -24,10 +24,10 @@ struct nv50_program { struct nv50_program_exec *exec_head; struct nv50_program_exec *exec_tail; unsigned exec_size; - struct nouveau_resource *data; - unsigned data_start; + struct nouveau_resource *data[2]; + unsigned data_start[2]; - struct pipe_buffer *buffer; + struct nouveau_bo *bo; float *immd; unsigned immd_nr; @@ -39,6 +39,11 @@ struct nv50_program { struct { unsigned attr[2]; } vp; + struct { + unsigned regs[4]; + unsigned map[5]; + unsigned high_map; + } fp; } cfg; }; diff --git a/src/gallium/drivers/nv50/nv50_query.c b/src/gallium/drivers/nv50/nv50_query.c index 35cebdbdc32..940e04365f2 100644 --- a/src/gallium/drivers/nv50/nv50_query.c +++ b/src/gallium/drivers/nv50/nv50_query.c @@ -26,7 +26,7 @@ #include "nv50_context.h" struct nv50_query { - struct pipe_buffer *buffer; + struct nouveau_bo *bo; unsigned type; boolean ready; uint64_t result; @@ -41,14 +41,16 @@ nv50_query(struct pipe_query *pipe) static struct pipe_query * nv50_query_create(struct pipe_context *pipe, unsigned type) { - struct pipe_screen *screen = pipe->screen; + struct nouveau_device *dev = nouveau_screen(pipe->screen)->device; struct nv50_query *q = CALLOC_STRUCT(nv50_query); + int ret; assert (q->type == PIPE_QUERY_OCCLUSION_COUNTER); q->type = type; - q->buffer = screen->buffer_create(screen, 256, 0, 16); - if (!q->buffer) { + ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM | NOUVEAU_BO_MAP, 256, + 16, &q->bo); + if (ret) { FREE(q); return NULL; } @@ -62,7 +64,7 @@ nv50_query_destroy(struct pipe_context *pipe, struct pipe_query *pq) struct nv50_query *q = nv50_query(pq); if (q) { - pipe_buffer_reference(&q->buffer, NULL); + nouveau_bo_ref(NULL, &q->bo); FREE(q); } } @@ -71,7 +73,7 @@ static void nv50_query_begin(struct pipe_context *pipe, struct pipe_query *pq) { struct nv50_context *nv50 = nv50_context(pipe); - struct nouveau_channel *chan = nv50->screen->nvws->channel; + struct nouveau_channel *chan = nv50->screen->base.channel; struct nouveau_grobj *tesla = nv50->screen->tesla; struct nv50_query *q = nv50_query(pq); @@ -87,15 +89,14 @@ static void nv50_query_end(struct pipe_context *pipe, struct pipe_query *pq) { struct nv50_context *nv50 = nv50_context(pipe); - struct nouveau_channel *chan = nv50->screen->nvws->channel; + struct nouveau_channel *chan = nv50->screen->base.channel; struct nouveau_grobj *tesla = nv50->screen->tesla; struct nv50_query *q = nv50_query(pq); - struct nouveau_bo *bo = nv50->screen->nvws->get_bo(q->buffer); WAIT_RING (chan, 5); BEGIN_RING(chan, tesla, 0x1b00, 4); - OUT_RELOCh(chan, bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR); - OUT_RELOCl(chan, bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR); + OUT_RELOCh(chan, q->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR); + OUT_RELOCl(chan, q->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR); OUT_RING (chan, 0x00000000); OUT_RING (chan, 0x0100f002); FIRE_RING (chan); @@ -105,7 +106,6 @@ static boolean nv50_query_result(struct pipe_context *pipe, struct pipe_query *pq, boolean wait, uint64_t *result) { - struct pipe_winsys *ws = pipe->winsys; struct nv50_query *q = nv50_query(pq); /*XXX: Want to be able to return FALSE here instead of blocking @@ -113,11 +113,10 @@ nv50_query_result(struct pipe_context *pipe, struct pipe_query *pq, */ if (!q->ready) { - uint32_t *map = ws->buffer_map(ws, q->buffer, - PIPE_BUFFER_USAGE_CPU_READ); - q->result = map[1]; + nouveau_bo_map(q->bo, NOUVEAU_BO_RD); + q->result = ((uint32_t *)q->bo->map)[1]; q->ready = TRUE; - ws->buffer_unmap(ws, q->buffer); + nouveau_bo_unmap(q->bo); } *result = q->result; diff --git a/src/gallium/drivers/nv50/nv50_screen.c b/src/gallium/drivers/nv50/nv50_screen.c index 29805645948..fd39fa738b7 100644 --- a/src/gallium/drivers/nv50/nv50_screen.c +++ b/src/gallium/drivers/nv50/nv50_screen.c @@ -22,8 +22,6 @@ #include "pipe/p_screen.h" -#include "util/u_simple_screen.h" - #include "nv50_context.h" #include "nv50_screen.h" @@ -68,23 +66,6 @@ nv50_screen_is_format_supported(struct pipe_screen *pscreen, return FALSE; } -static const char * -nv50_screen_get_name(struct pipe_screen *pscreen) -{ - struct nv50_screen *screen = nv50_screen(pscreen); - struct nouveau_device *dev = screen->nvws->channel->device; - static char buffer[128]; - - snprintf(buffer, sizeof(buffer), "NV%02X", dev->chipset); - return buffer; -} - -static const char * -nv50_screen_get_vendor(struct pipe_screen *pscreen) -{ - return "nouveau"; -} - static int nv50_screen_get_param(struct pipe_screen *pscreen, int param) { @@ -153,37 +134,64 @@ nv50_screen_get_paramf(struct pipe_screen *pscreen, int param) static void nv50_screen_destroy(struct pipe_screen *pscreen) { - FREE(pscreen); + struct nv50_screen *screen = nv50_screen(pscreen); + + nouveau_notifier_free(&screen->sync); + nouveau_grobj_free(&screen->tesla); + nouveau_grobj_free(&screen->eng2d); + nouveau_grobj_free(&screen->m2mf); + nouveau_screen_fini(&screen->base); + FREE(screen); } struct pipe_screen * -nv50_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws) +nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev) { struct nv50_screen *screen = CALLOC_STRUCT(nv50_screen); + struct nouveau_channel *chan; + struct pipe_screen *pscreen; struct nouveau_stateobj *so; - unsigned tesla_class = 0, ret; - unsigned chipset = nvws->channel->device->chipset; - int i; + unsigned chipset = dev->chipset; + unsigned tesla_class = 0; + int ret, i; if (!screen) return NULL; - screen->nvws = nvws; + pscreen = &screen->base.base; + + ret = nouveau_screen_init(&screen->base, dev); + if (ret) { + nv50_screen_destroy(pscreen); + return NULL; + } + chan = screen->base.channel; + + pscreen->winsys = ws; + pscreen->destroy = nv50_screen_destroy; + pscreen->get_param = nv50_screen_get_param; + pscreen->get_paramf = nv50_screen_get_paramf; + pscreen->is_format_supported = nv50_screen_is_format_supported; + + nv50_screen_init_miptree_functions(pscreen); + nv50_transfer_init_screen_functions(pscreen); /* DMA engine object */ - ret = nvws->grobj_alloc(nvws, 0x5039, &screen->m2mf); + ret = nouveau_grobj_alloc(chan, 0xbeef5039, 0x5039, &screen->m2mf); if (ret) { NOUVEAU_ERR("Error creating M2MF object: %d\n", ret); - nv50_screen_destroy(&screen->pipe); + nv50_screen_destroy(pscreen); return NULL; } + BIND_RING(chan, screen->m2mf, 1); /* 2D object */ - ret = nvws->grobj_alloc(nvws, NV50_2D, &screen->eng2d); + ret = nouveau_grobj_alloc(chan, 0xbeef502d, 0x502d, &screen->eng2d); if (ret) { NOUVEAU_ERR("Error creating 2D object: %d\n", ret); - nv50_screen_destroy(&screen->pipe); + nv50_screen_destroy(pscreen); return NULL; } + BIND_RING(chan, screen->eng2d, 2); /* 3D object */ switch (chipset & 0xf0) { @@ -199,70 +207,55 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws) break; default: NOUVEAU_ERR("Not a known NV50 chipset: NV%02x\n", chipset); - nv50_screen_destroy(&screen->pipe); + nv50_screen_destroy(pscreen); return NULL; } if (tesla_class == 0) { NOUVEAU_ERR("Unknown G8x chipset: NV%02x\n", chipset); - nv50_screen_destroy(&screen->pipe); + nv50_screen_destroy(pscreen); return NULL; } - ret = nvws->grobj_alloc(nvws, tesla_class, &screen->tesla); + ret = nouveau_grobj_alloc(chan, 0xbeef5097, tesla_class, &screen->tesla); if (ret) { NOUVEAU_ERR("Error creating 3D object: %d\n", ret); - nv50_screen_destroy(&screen->pipe); + nv50_screen_destroy(pscreen); return NULL; } + BIND_RING(chan, screen->tesla, 3); /* Sync notifier */ - ret = nvws->notifier_alloc(nvws, 1, &screen->sync); + ret = nouveau_notifier_alloc(chan, 0xbeef0301, 1, &screen->sync); if (ret) { NOUVEAU_ERR("Error creating notifier object: %d\n", ret); - nv50_screen_destroy(&screen->pipe); + nv50_screen_destroy(pscreen); return NULL; } - /* Setup the pipe */ - screen->pipe.winsys = ws; - - screen->pipe.destroy = nv50_screen_destroy; - - screen->pipe.get_name = nv50_screen_get_name; - screen->pipe.get_vendor = nv50_screen_get_vendor; - screen->pipe.get_param = nv50_screen_get_param; - screen->pipe.get_paramf = nv50_screen_get_paramf; - - screen->pipe.is_format_supported = nv50_screen_is_format_supported; - - nv50_screen_init_miptree_functions(&screen->pipe); - nv50_transfer_init_screen_functions(&screen->pipe); - u_simple_screen_init(&screen->pipe); - /* Static M2MF init */ so = so_new(32, 0); so_method(so, screen->m2mf, 0x0180, 3); so_data (so, screen->sync->handle); - so_data (so, screen->nvws->channel->vram->handle); - so_data (so, screen->nvws->channel->vram->handle); - so_emit(nvws, so); + so_data (so, chan->vram->handle); + so_data (so, chan->vram->handle); + so_emit(chan, so); so_ref (NULL, &so); /* Static 2D init */ so = so_new(64, 0); so_method(so, screen->eng2d, NV50_2D_DMA_NOTIFY, 4); so_data (so, screen->sync->handle); - so_data (so, screen->nvws->channel->vram->handle); - so_data (so, screen->nvws->channel->vram->handle); - so_data (so, screen->nvws->channel->vram->handle); + so_data (so, chan->vram->handle); + so_data (so, chan->vram->handle); + so_data (so, chan->vram->handle); so_method(so, screen->eng2d, NV50_2D_OPERATION, 1); so_data (so, NV50_2D_OPERATION_SRCCOPY); so_method(so, screen->eng2d, 0x0290, 1); so_data (so, 0); so_method(so, screen->eng2d, 0x0888, 1); so_data (so, 1); - so_emit(nvws, so); + so_emit(chan, so); so_ref(NULL, &so); /* Static tesla init */ @@ -275,11 +268,11 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws) so_method(so, screen->tesla, NV50TCL_DMA_UNK0(0), NV50TCL_DMA_UNK0__SIZE); for (i = 0; i < NV50TCL_DMA_UNK0__SIZE; i++) - so_data(so, nvws->channel->vram->handle); + so_data(so, chan->vram->handle); so_method(so, screen->tesla, NV50TCL_DMA_UNK1(0), NV50TCL_DMA_UNK1__SIZE); for (i = 0; i < NV50TCL_DMA_UNK1__SIZE; i++) - so_data(so, nvws->channel->vram->handle); + so_data(so, chan->vram->handle); so_method(so, screen->tesla, 0x121c, 1); so_data (so, 1); @@ -290,27 +283,81 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws) so_method(so, screen->tesla, 0x16b8, 1); so_data (so, 8); - /* Shared constant buffer */ - screen->constbuf = screen->pipe.buffer_create(&screen->pipe, 0, 0, 128 * 4 * 4); - if (nvws->res_init(&screen->vp_data_heap, 0, 128)) { - NOUVEAU_ERR("Error initialising constant buffer\n"); - nv50_screen_destroy(&screen->pipe); + /* constant buffers for immediates and VP/FP parameters */ + ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 0, 128*4*4, + &screen->constbuf_misc[0]); + if (ret) { + nv50_screen_destroy(pscreen); return NULL; } + for (i = 0; i < 2; i++) { + ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 0, 128*4*4, + &screen->constbuf_parm[i]); + if (ret) { + nv50_screen_destroy(pscreen); + return NULL; + } + } + + if (nouveau_resource_init(&screen->immd_heap[0], 0, 128) || + nouveau_resource_init(&screen->parm_heap[0], 0, 128) || + nouveau_resource_init(&screen->parm_heap[1], 0, 128)) + { + NOUVEAU_ERR("Error initialising constant buffers.\n"); + nv50_screen_destroy(pscreen); + return NULL; + } + + /* + // map constant buffers: + // B = buffer ID (maybe more than 1 byte) + // N = CB index used in shader instruction + // P = program type (0 = VP, 2 = GP, 3 = FP) + so_method(so, screen->tesla, 0x1694, 1); + so_data (so, 0x000BBNP1); + */ + so_method(so, screen->tesla, 0x1280, 3); - so_reloc (so, screen->constbuf, 0, NOUVEAU_BO_VRAM | + so_reloc (so, screen->constbuf_misc[0], 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); - so_reloc (so, screen->constbuf, 0, NOUVEAU_BO_VRAM | + so_reloc (so, screen->constbuf_misc[0], 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); - so_data (so, (NV50_CB_PMISC << 16) | 0x00001000); + so_data (so, (NV50_CB_PMISC << 16) | 0x00000800); + so_method(so, screen->tesla, 0x1694, 1); + so_data (so, 0x00000001 | (NV50_CB_PMISC << 12)); + so_method(so, screen->tesla, 0x1694, 1); + so_data (so, 0x00000031 | (NV50_CB_PMISC << 12)); + + so_method(so, screen->tesla, 0x1280, 3); + so_reloc (so, screen->constbuf_parm[0], 0, NOUVEAU_BO_VRAM | + NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); + so_reloc (so, screen->constbuf_parm[0], 0, NOUVEAU_BO_VRAM | + NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); + so_data (so, (NV50_CB_PVP << 16) | 0x00000800); + so_method(so, screen->tesla, 0x1694, 1); + so_data (so, 0x00000101 | (NV50_CB_PVP << 12)); + + so_method(so, screen->tesla, 0x1280, 3); + so_reloc (so, screen->constbuf_parm[1], 0, NOUVEAU_BO_VRAM | + NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); + so_reloc (so, screen->constbuf_parm[1], 0, NOUVEAU_BO_VRAM | + NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); + so_data (so, (NV50_CB_PFP << 16) | 0x00000800); + so_method(so, screen->tesla, 0x1694, 1); + so_data (so, 0x00000131 | (NV50_CB_PFP << 12)); /* Texture sampler/image unit setup - we abuse the constant buffer * upload mechanism for the moment to upload data to the tex config * blocks. At some point we *may* want to go the NVIDIA way of doing * things? */ - screen->tic = screen->pipe.buffer_create(&screen->pipe, 0, 0, 32 * 8 * 4); + ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 0, 32*8*4, &screen->tic); + if (ret) { + nv50_screen_destroy(pscreen); + return NULL; + } + so_method(so, screen->tesla, 0x1280, 3); so_reloc (so, screen->tic, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); @@ -324,7 +371,12 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws) NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); so_data (so, 0x00000800); - screen->tsc = screen->pipe.buffer_create(&screen->pipe, 0, 0, 32 * 8 * 4); + ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 0, 32*8*4, &screen->tsc); + if (ret) { + nv50_screen_destroy(pscreen); + return NULL; + } + so_method(so, screen->tesla, 0x1280, 3); so_reloc (so, screen->tsc, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); @@ -352,14 +404,12 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws) so_method(so, screen->tesla, 0x1234, 1); so_data (so, 1); - so_method(so, screen->tesla, 0x1458, 1); - so_data (so, 1); - so_emit(nvws, so); + so_emit(chan, so); so_ref (so, &screen->static_init); so_ref (NULL, &so); - nvws->push_flush(nvws, 0, NULL); + nouveau_pushbuf_flush(chan, 0); - return &screen->pipe; + return pscreen; } diff --git a/src/gallium/drivers/nv50/nv50_screen.h b/src/gallium/drivers/nv50/nv50_screen.h index db567aaac89..61e24a5b571 100644 --- a/src/gallium/drivers/nv50/nv50_screen.h +++ b/src/gallium/drivers/nv50/nv50_screen.h @@ -1,10 +1,10 @@ #ifndef __NV50_SCREEN_H__ #define __NV50_SCREEN_H__ -#include "pipe/p_screen.h" +#include "nouveau/nouveau_screen.h" struct nv50_screen { - struct pipe_screen pipe; + struct nouveau_screen base; struct nouveau_winsys *nvws; @@ -15,11 +15,14 @@ struct nv50_screen { struct nouveau_grobj *m2mf; struct nouveau_notifier *sync; - struct pipe_buffer *constbuf; - struct nouveau_resource *vp_data_heap; + struct nouveau_bo *constbuf_misc[1]; + struct nouveau_bo *constbuf_parm[2]; - struct pipe_buffer *tic; - struct pipe_buffer *tsc; + struct nouveau_resource *immd_heap[1]; + struct nouveau_resource *parm_heap[2]; + + struct nouveau_bo *tic; + struct nouveau_bo *tsc; struct nouveau_stateobj *static_init; }; diff --git a/src/gallium/drivers/nv50/nv50_state.c b/src/gallium/drivers/nv50/nv50_state.c index ba852194cdd..116866a8e78 100644 --- a/src/gallium/drivers/nv50/nv50_state.c +++ b/src/gallium/drivers/nv50/nv50_state.c @@ -136,9 +136,11 @@ static void * nv50_sampler_state_create(struct pipe_context *pipe, const struct pipe_sampler_state *cso) { - unsigned *tsc = CALLOC(8, sizeof(unsigned)); + struct nv50_sampler_stateobj *sso = CALLOC(1, sizeof(*sso)); + unsigned *tsc = sso->tsc; + float limit; - tsc[0] = (0x00024000 | + tsc[0] = (0x00026000 | (wrap_mode(cso->wrap_s) << 0) | (wrap_mode(cso->wrap_t) << 3) | (wrap_mode(cso->wrap_r) << 6)); @@ -202,7 +204,14 @@ nv50_sampler_state_create(struct pipe_context *pipe, tsc[0] |= (nvgl_comparison_op(cso->compare_func) & 0x7); } - return (void *)tsc; + limit = CLAMP(cso->lod_bias, -16.0, 15.0); + tsc[1] |= ((int)(limit * 256.0) & 0x1fff) << 11; + + tsc[2] |= ((int)CLAMP(cso->max_lod, 0.0, 15.0) << 20) | + ((int)CLAMP(cso->min_lod, 0.0, 15.0) << 8); + + sso->normalized = cso->normalized_coords; + return (void *)sso; } static void diff --git a/src/gallium/drivers/nv50/nv50_state_validate.c b/src/gallium/drivers/nv50/nv50_state_validate.c index c13d3de1cb9..0caf4b4e914 100644 --- a/src/gallium/drivers/nv50/nv50_state_validate.c +++ b/src/gallium/drivers/nv50/nv50_state_validate.c @@ -32,6 +32,9 @@ nv50_state_validate_fb(struct nv50_context *nv50) unsigned i, w, h, gw = 0; for (i = 0; i < fb->nr_cbufs; i++) { + struct pipe_texture *pt = fb->cbufs[i]->texture; + struct nouveau_bo *bo = nv50_miptree(pt)->bo; + if (!gw) { w = fb->cbufs[i]->width; h = fb->cbufs[i]->height; @@ -46,12 +49,10 @@ nv50_state_validate_fb(struct nv50_context *nv50) so_data (so, fb->cbufs[i]->height); so_method(so, tesla, NV50TCL_RT_ADDRESS_HIGH(i), 5); - so_reloc (so, nv50_surface_buffer(fb->cbufs[i]), fb->cbufs[i]->offset, - NOUVEAU_BO_VRAM | NOUVEAU_BO_HIGH | - NOUVEAU_BO_RDWR, 0, 0); - so_reloc (so, nv50_surface_buffer(fb->cbufs[i]), fb->cbufs[i]->offset, - NOUVEAU_BO_VRAM | NOUVEAU_BO_LOW | - NOUVEAU_BO_RDWR, 0, 0); + so_reloc (so, bo, fb->cbufs[i]->offset, NOUVEAU_BO_VRAM | + NOUVEAU_BO_HIGH | NOUVEAU_BO_RDWR, 0, 0); + so_reloc (so, bo, fb->cbufs[i]->offset, NOUVEAU_BO_VRAM | + NOUVEAU_BO_LOW | NOUVEAU_BO_RDWR, 0, 0); switch (fb->cbufs[i]->format) { case PIPE_FORMAT_A8R8G8B8_UNORM: so_data(so, 0xcf); @@ -65,7 +66,7 @@ nv50_state_validate_fb(struct nv50_context *nv50) so_data(so, 0xe6); break; } - so_data(so, 0x00000000); + so_data(so, bo->tile_mode << 4); so_data(so, 0x00000000); so_method(so, tesla, 0x1224, 1); @@ -73,6 +74,9 @@ nv50_state_validate_fb(struct nv50_context *nv50) } if (fb->zsbuf) { + struct pipe_texture *pt = fb->zsbuf->texture; + struct nouveau_bo *bo = nv50_miptree(pt)->bo; + if (!gw) { w = fb->zsbuf->width; h = fb->zsbuf->height; @@ -83,12 +87,10 @@ nv50_state_validate_fb(struct nv50_context *nv50) } so_method(so, tesla, NV50TCL_ZETA_ADDRESS_HIGH, 5); - so_reloc (so, nv50_surface_buffer(fb->zsbuf), fb->zsbuf->offset, - NOUVEAU_BO_VRAM | NOUVEAU_BO_HIGH | - NOUVEAU_BO_RDWR, 0, 0); - so_reloc (so, nv50_surface_buffer(fb->zsbuf), fb->zsbuf->offset, - NOUVEAU_BO_VRAM | NOUVEAU_BO_LOW | - NOUVEAU_BO_RDWR, 0, 0); + so_reloc (so, bo, fb->zsbuf->offset, NOUVEAU_BO_VRAM | + NOUVEAU_BO_HIGH | NOUVEAU_BO_RDWR, 0, 0); + so_reloc (so, bo, fb->zsbuf->offset, NOUVEAU_BO_VRAM | + NOUVEAU_BO_LOW | NOUVEAU_BO_RDWR, 0, 0); switch (fb->zsbuf->format) { case PIPE_FORMAT_Z24S8_UNORM: so_data(so, 0x16); @@ -102,7 +104,7 @@ nv50_state_validate_fb(struct nv50_context *nv50) so_data(so, 0x16); break; } - so_data(so, 0x00000000); + so_data(so, bo->tile_mode << 4); so_data(so, 0x00000000); so_method(so, tesla, 0x1538, 1); @@ -131,7 +133,7 @@ static void nv50_state_emit(struct nv50_context *nv50) { struct nv50_screen *screen = nv50->screen; - struct nouveau_winsys *nvws = screen->nvws; + struct nouveau_channel *chan = screen->base.channel; if (nv50->pctx_id != screen->cur_pctx) { nv50->state.dirty |= 0xffffffff; @@ -139,40 +141,40 @@ nv50_state_emit(struct nv50_context *nv50) } if (nv50->state.dirty & NV50_NEW_FRAMEBUFFER) - so_emit(nvws, nv50->state.fb); + so_emit(chan, nv50->state.fb); if (nv50->state.dirty & NV50_NEW_BLEND) - so_emit(nvws, nv50->state.blend); + so_emit(chan, nv50->state.blend); if (nv50->state.dirty & NV50_NEW_ZSA) - so_emit(nvws, nv50->state.zsa); + so_emit(chan, nv50->state.zsa); if (nv50->state.dirty & NV50_NEW_VERTPROG) - so_emit(nvws, nv50->state.vertprog); + so_emit(chan, nv50->state.vertprog); if (nv50->state.dirty & NV50_NEW_FRAGPROG) - so_emit(nvws, nv50->state.fragprog); + so_emit(chan, nv50->state.fragprog); if (nv50->state.dirty & NV50_NEW_RASTERIZER) - so_emit(nvws, nv50->state.rast); + so_emit(chan, nv50->state.rast); if (nv50->state.dirty & NV50_NEW_BLEND_COLOUR) - so_emit(nvws, nv50->state.blend_colour); + so_emit(chan, nv50->state.blend_colour); if (nv50->state.dirty & NV50_NEW_STIPPLE) - so_emit(nvws, nv50->state.stipple); + so_emit(chan, nv50->state.stipple); if (nv50->state.dirty & NV50_NEW_SCISSOR) - so_emit(nvws, nv50->state.scissor); + so_emit(chan, nv50->state.scissor); if (nv50->state.dirty & NV50_NEW_VIEWPORT) - so_emit(nvws, nv50->state.viewport); + so_emit(chan, nv50->state.viewport); if (nv50->state.dirty & NV50_NEW_SAMPLER) - so_emit(nvws, nv50->state.tsc_upload); + so_emit(chan, nv50->state.tsc_upload); if (nv50->state.dirty & NV50_NEW_TEXTURE) - so_emit(nvws, nv50->state.tic_upload); + so_emit(chan, nv50->state.tic_upload); if (nv50->state.dirty & NV50_NEW_ARRAYS) { - so_emit(nvws, nv50->state.vtxfmt); - so_emit(nvws, nv50->state.vtxbuf); + so_emit(chan, nv50->state.vtxfmt); + so_emit(chan, nv50->state.vtxbuf); } nv50->state.dirty = 0; - so_emit_reloc_markers(nvws, nv50->state.fb); - so_emit_reloc_markers(nvws, nv50->state.vertprog); - so_emit_reloc_markers(nvws, nv50->state.fragprog); - so_emit_reloc_markers(nvws, nv50->state.vtxbuf); - so_emit_reloc_markers(nvws, nv50->screen->static_init); + so_emit_reloc_markers(chan, nv50->state.fb); + so_emit_reloc_markers(chan, nv50->state.vertprog); + so_emit_reloc_markers(chan, nv50->state.fragprog); + so_emit_reloc_markers(chan, nv50->state.vtxbuf); + so_emit_reloc_markers(chan, nv50->screen->static_init); } boolean @@ -293,12 +295,12 @@ viewport_uptodate: so_data (so, NV50_CB_TSC); so_method(so, tesla, 0x40000f04, nv50->sampler_nr * 8); for (i = 0; i < nv50->sampler_nr; i++) - so_datap (so, nv50->sampler[i], 8); + so_datap (so, nv50->sampler[i]->tsc, 8); so_ref(so, &nv50->state.tsc_upload); so_ref(NULL, &so); } - if (nv50->dirty & NV50_NEW_TEXTURE) + if (nv50->dirty & (NV50_NEW_TEXTURE | NV50_NEW_SAMPLER)) nv50_tex_validate(nv50); if (nv50->dirty & NV50_NEW_ARRAYS) diff --git a/src/gallium/drivers/nv50/nv50_surface.c b/src/gallium/drivers/nv50/nv50_surface.c index 0cc5168144d..8db3b6d344d 100644 --- a/src/gallium/drivers/nv50/nv50_surface.c +++ b/src/gallium/drivers/nv50/nv50_surface.c @@ -52,21 +52,17 @@ static int nv50_surface_set(struct nv50_screen *screen, struct pipe_surface *ps, int dst) { struct nv50_miptree *mt = nv50_miptree(ps->texture); - struct nouveau_channel *chan = screen->nvws->channel; + struct nouveau_channel *chan = screen->eng2d->channel; struct nouveau_grobj *eng2d = screen->eng2d; - struct nouveau_bo *bo; + struct nouveau_bo *bo = nv50_miptree(ps->texture)->bo; int format, mthd = dst ? NV50_2D_DST_FORMAT : NV50_2D_SRC_FORMAT; int flags = NOUVEAU_BO_VRAM | (dst ? NOUVEAU_BO_WR : NOUVEAU_BO_RD); - - bo = screen->nvws->get_bo(nv50_miptree(ps->texture)->buffer); - if (!bo) - return 1; format = nv50_format(ps->format); if (format < 0) return 1; - if (!bo->tiled) { + if (!bo->tile_flags) { BEGIN_RING(chan, eng2d, mthd, 2); OUT_RING (chan, format); OUT_RING (chan, 1); @@ -80,7 +76,7 @@ nv50_surface_set(struct nv50_screen *screen, struct pipe_surface *ps, int dst) BEGIN_RING(chan, eng2d, mthd, 5); OUT_RING (chan, format); OUT_RING (chan, 0); - OUT_RING (chan, 0); + OUT_RING (chan, bo->tile_mode << 4); OUT_RING (chan, 1); OUT_RING (chan, 0); BEGIN_RING(chan, eng2d, mthd + 0x18, 4); @@ -108,7 +104,7 @@ nv50_surface_do_copy(struct nv50_screen *screen, struct pipe_surface *dst, int dx, int dy, struct pipe_surface *src, int sx, int sy, int w, int h) { - struct nouveau_channel *chan = screen->nvws->channel; + struct nouveau_channel *chan = screen->eng2d->channel; struct nouveau_grobj *eng2d = screen->eng2d; int ret; @@ -165,7 +161,7 @@ nv50_surface_fill(struct pipe_context *pipe, struct pipe_surface *dest, { struct nv50_context *nv50 = (struct nv50_context *)pipe; struct nv50_screen *screen = nv50->screen; - struct nouveau_channel *chan = screen->nvws->channel; + struct nouveau_channel *chan = screen->eng2d->channel; struct nouveau_grobj *eng2d = screen->eng2d; int format, ret; diff --git a/src/gallium/drivers/nv50/nv50_tex.c b/src/gallium/drivers/nv50/nv50_tex.c index 223c8a3a456..ff40c2ad81b 100644 --- a/src/gallium/drivers/nv50/nv50_tex.c +++ b/src/gallium/drivers/nv50/nv50_tex.c @@ -26,7 +26,8 @@ #include "nouveau/nouveau_stateobj.h" static int -nv50_tex_construct(struct nouveau_stateobj *so, struct nv50_miptree *mt) +nv50_tex_construct(struct nv50_context *nv50, struct nouveau_stateobj *so, + struct nv50_miptree *mt, int unit) { switch (mt->base.format) { case PIPE_FORMAT_A8R8G8B8_UNORM: @@ -117,15 +118,18 @@ nv50_tex_construct(struct nouveau_stateobj *so, struct nv50_miptree *mt) return 1; } - so_reloc(so, mt->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_LOW | + so_reloc(so, mt->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_LOW | NOUVEAU_BO_RD, 0, 0); - so_data (so, 0xd0005000); + if (nv50->sampler[unit]->normalized) + so_data (so, 0xd0005000 | mt->bo->tile_mode << 22); + else + so_data (so, 0x5001d000 | mt->bo->tile_mode << 22); so_data (so, 0x00300000); so_data (so, mt->base.width[0]); - so_data (so, (mt->base.depth[0] << 16) | mt->base.height[0]); + so_data (so, (mt->base.last_level << 28) | + (mt->base.depth[0] << 16) | mt->base.height[0]); so_data (so, 0x03000000); - so_reloc(so, mt->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_HIGH | - NOUVEAU_BO_RD, 0, 0); + so_data (so, mt->base.last_level << 4); return 0; } @@ -135,23 +139,35 @@ nv50_tex_validate(struct nv50_context *nv50) { struct nouveau_grobj *tesla = nv50->screen->tesla; struct nouveau_stateobj *so; - int unit; + int unit, push; + + push = nv50->miptree_nr * 9 + 2; + push += MAX2(nv50->miptree_nr, nv50->state.miptree_nr) * 2; - so = so_new(nv50->miptree_nr * 8 + 3, nv50->miptree_nr * 2); + so = so_new(push, nv50->miptree_nr * 2); so_method(so, tesla, 0x0f00, 1); so_data (so, NV50_CB_TIC); - so_method(so, tesla, 0x40000f04, nv50->miptree_nr * 8); for (unit = 0; unit < nv50->miptree_nr; unit++) { struct nv50_miptree *mt = nv50->miptree[unit]; - if (nv50_tex_construct(so, mt)) { + so_method(so, tesla, 0x40000f04, 8); + if (nv50_tex_construct(nv50, so, mt, unit)) { NOUVEAU_ERR("failed tex validate\n"); so_ref(NULL, &so); return; } + + so_method(so, tesla, 0x1458, 1); + so_data (so, (unit << 9) | (unit << 1) | 1); + } + + for (; unit < nv50->state.miptree_nr; unit++) { + so_method(so, tesla, 0x1458, 1); + so_data (so, (unit << 1) | 0); } so_ref(so, &nv50->state.tic_upload); so_ref(NULL, &so); + nv50->state.miptree_nr = nv50->miptree_nr; } diff --git a/src/gallium/drivers/nv50/nv50_transfer.c b/src/gallium/drivers/nv50/nv50_transfer.c index 747195b4f63..d0b7f0bef43 100644 --- a/src/gallium/drivers/nv50/nv50_transfer.c +++ b/src/gallium/drivers/nv50/nv50_transfer.c @@ -6,8 +6,8 @@ struct nv50_transfer { struct pipe_transfer base; - struct pipe_buffer *buffer; - struct nv50_miptree_level *level; + struct nouveau_bo *bo; + unsigned level_offset; int level_pitch; int level_width; int level_height; @@ -16,51 +16,48 @@ struct nv50_transfer { }; static void -nv50_transfer_rect_m2mf(struct pipe_screen *pscreen, struct pipe_buffer *src, - int src_pitch, int sx, int sy, int sw, int sh, - struct pipe_buffer *dst, int dst_pitch, int dx, int dy, +nv50_transfer_rect_m2mf(struct pipe_screen *pscreen, struct nouveau_bo *src_bo, + unsigned src_offset, int src_pitch, int sx, int sy, + int sw, int sh, struct nouveau_bo *dst_bo, + unsigned dst_offset, int dst_pitch, int dx, int dy, int dw, int dh, int cpp, int width, int height, unsigned src_reloc, unsigned dst_reloc) { struct nv50_screen *screen = nv50_screen(pscreen); - struct nouveau_winsys *nvws = screen->nvws; - struct nouveau_channel *chan = nvws->channel; + struct nouveau_channel *chan = screen->m2mf->channel; struct nouveau_grobj *m2mf = screen->m2mf; - struct nouveau_bo *src_bo = nvws->get_bo(src); - struct nouveau_bo *dst_bo = nvws->get_bo(dst); - unsigned src_offset = 0, dst_offset = 0; src_reloc |= NOUVEAU_BO_RD; dst_reloc |= NOUVEAU_BO_WR; WAIT_RING (chan, 14); - if (!src_bo->tiled) { + if (!src_bo->tile_flags) { BEGIN_RING(chan, m2mf, 0x0200, 1); OUT_RING (chan, 1); BEGIN_RING(chan, m2mf, 0x0314, 1); OUT_RING (chan, src_pitch); - src_offset = (sy * src_pitch) + (sx * cpp); + src_offset += (sy * src_pitch) + (sx * cpp); } else { BEGIN_RING(chan, m2mf, 0x0200, 6); OUT_RING (chan, 0); - OUT_RING (chan, 0); + OUT_RING (chan, src_bo->tile_mode << 4); OUT_RING (chan, sw * cpp); OUT_RING (chan, sh); OUT_RING (chan, 1); OUT_RING (chan, 0); } - if (!dst_bo->tiled) { + if (!dst_bo->tile_flags) { BEGIN_RING(chan, m2mf, 0x021c, 1); OUT_RING (chan, 1); BEGIN_RING(chan, m2mf, 0x0318, 1); OUT_RING (chan, dst_pitch); - dst_offset = (dy * dst_pitch) + (dx * cpp); + dst_offset += (dy * dst_pitch) + (dx * cpp); } else { BEGIN_RING(chan, m2mf, 0x021c, 6); OUT_RING (chan, 0); - OUT_RING (chan, 0); + OUT_RING (chan, dst_bo->tile_mode << 4); OUT_RING (chan, dw * cpp); OUT_RING (chan, dh); OUT_RING (chan, 1); @@ -77,13 +74,13 @@ nv50_transfer_rect_m2mf(struct pipe_screen *pscreen, struct pipe_buffer *src, BEGIN_RING(chan, m2mf, 0x030c, 2); OUT_RELOCl(chan, src_bo, src_offset, src_reloc); OUT_RELOCl(chan, dst_bo, dst_offset, dst_reloc); - if (src_bo->tiled) { + if (src_bo->tile_flags) { BEGIN_RING(chan, m2mf, 0x0218, 1); OUT_RING (chan, (dy << 16) | sx); } else { src_offset += (line_count * src_pitch); } - if (dst_bo->tiled) { + if (dst_bo->tile_flags) { BEGIN_RING(chan, m2mf, 0x0234, 1); OUT_RING (chan, (sy << 16) | dx); } else { @@ -108,10 +105,12 @@ nv50_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt, enum pipe_transfer_usage usage, unsigned x, unsigned y, unsigned w, unsigned h) { + struct nouveau_device *dev = nouveau_screen(pscreen)->device; struct nv50_miptree *mt = nv50_miptree(pt); struct nv50_miptree_level *lvl = &mt->level[level]; struct nv50_transfer *tx; unsigned image = 0; + int ret; if (pt->target == PIPE_TEXTURE_CUBE) image = face; @@ -133,20 +132,24 @@ nv50_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt, tx->base.stride = (w * pt->block.size); tx->base.usage = usage; - tx->level = lvl; tx->level_pitch = lvl->pitch; tx->level_width = mt->base.width[level]; tx->level_height = mt->base.height[level]; + tx->level_offset = lvl->image_offset[image]; tx->level_x = x; tx->level_y = y; - tx->buffer = - pipe_buffer_create(pscreen, 0, NOUVEAU_BUFFER_USAGE_TRANSFER, - w * tx->base.block.size * h); + ret = nouveau_bo_new(dev, NOUVEAU_BO_GART | NOUVEAU_BO_MAP, 0, + w * pt->block.size * h, &tx->bo); + if (ret) { + FREE(tx); + return NULL; + } if (usage != PIPE_TRANSFER_WRITE) { - nv50_transfer_rect_m2mf(pscreen, mt->buffer, tx->level_pitch, - x, y, tx->level_width, tx->level_height, - tx->buffer, tx->base.stride, 0, 0, + nv50_transfer_rect_m2mf(pscreen, mt->bo, tx->level_offset, + tx->level_pitch, x, y, tx->level_width, + tx->level_height, tx->bo, 0, + tx->base.stride, 0, 0, tx->base.width, tx->base.height, tx->base.block.size, w, h, NOUVEAU_BO_VRAM | NOUVEAU_BO_GART, @@ -164,17 +167,18 @@ nv50_transfer_del(struct pipe_transfer *ptx) if (ptx->usage != PIPE_TRANSFER_READ) { struct pipe_screen *pscreen = ptx->texture->screen; - nv50_transfer_rect_m2mf(pscreen, tx->buffer, tx->base.stride, + nv50_transfer_rect_m2mf(pscreen, tx->bo, 0, tx->base.stride, 0, 0, tx->base.width, tx->base.height, - mt->buffer, tx->level_pitch, - tx->level_x, tx->level_y, - tx->level_width, tx->level_height, - tx->base.block.size, tx->base.width, - tx->base.height, NOUVEAU_BO_GART, - NOUVEAU_BO_VRAM | NOUVEAU_BO_GART); + mt->bo, tx->level_offset, + tx->level_pitch, tx->level_x, + tx->level_y, tx->level_width, + tx->level_height, tx->base.block.size, + tx->base.width, tx->base.height, + NOUVEAU_BO_GART, NOUVEAU_BO_VRAM | + NOUVEAU_BO_GART); } - pipe_buffer_reference(&tx->buffer, NULL); + nouveau_bo_ref(NULL, &tx->bo); pipe_texture_reference(&ptx->texture, NULL); FREE(ptx); } @@ -184,13 +188,17 @@ nv50_transfer_map(struct pipe_screen *pscreen, struct pipe_transfer *ptx) { struct nv50_transfer *tx = (struct nv50_transfer *)ptx; unsigned flags = 0; + int ret; if (ptx->usage & PIPE_TRANSFER_WRITE) - flags |= PIPE_BUFFER_USAGE_CPU_WRITE; + flags |= NOUVEAU_BO_WR; if (ptx->usage & PIPE_TRANSFER_READ) - flags |= PIPE_BUFFER_USAGE_CPU_READ; + flags |= NOUVEAU_BO_RD; - return pipe_buffer_map(pscreen, tx->buffer, flags); + ret = nouveau_bo_map(tx->bo, flags); + if (ret) + return NULL; + return tx->bo->map; } static void @@ -198,7 +206,7 @@ nv50_transfer_unmap(struct pipe_screen *pscreen, struct pipe_transfer *ptx) { struct nv50_transfer *tx = (struct nv50_transfer *)ptx; - pipe_buffer_unmap(pscreen, tx->buffer); + nouveau_bo_unmap(tx->bo); } void diff --git a/src/gallium/drivers/nv50/nv50_vbo.c b/src/gallium/drivers/nv50/nv50_vbo.c index 0749c906914..f81929f2387 100644 --- a/src/gallium/drivers/nv50/nv50_vbo.c +++ b/src/gallium/drivers/nv50/nv50_vbo.c @@ -22,6 +22,7 @@ #include "pipe/p_context.h" #include "pipe/p_state.h" +#include "pipe/p_inlines.h" #include "nv50_context.h" @@ -53,7 +54,7 @@ nv50_draw_arrays(struct pipe_context *pipe, unsigned mode, unsigned start, unsigned count) { struct nv50_context *nv50 = nv50_context(pipe); - struct nouveau_channel *chan = nv50->screen->nvws->channel; + struct nouveau_channel *chan = nv50->screen->tesla->channel; struct nouveau_grobj *tesla = nv50->screen->tesla; nv50_state_validate(nv50); @@ -83,7 +84,7 @@ static INLINE void nv50_draw_elements_inline_u08(struct nv50_context *nv50, uint8_t *map, unsigned start, unsigned count) { - struct nouveau_channel *chan = nv50->screen->nvws->channel; + struct nouveau_channel *chan = nv50->screen->tesla->channel; struct nouveau_grobj *tesla = nv50->screen->tesla; map += start; @@ -112,7 +113,7 @@ static INLINE void nv50_draw_elements_inline_u16(struct nv50_context *nv50, uint16_t *map, unsigned start, unsigned count) { - struct nouveau_channel *chan = nv50->screen->nvws->channel; + struct nouveau_channel *chan = nv50->screen->tesla->channel; struct nouveau_grobj *tesla = nv50->screen->tesla; map += start; @@ -141,7 +142,7 @@ static INLINE void nv50_draw_elements_inline_u32(struct nv50_context *nv50, uint8_t *map, unsigned start, unsigned count) { - struct nouveau_channel *chan = nv50->screen->nvws->channel; + struct nouveau_channel *chan = nv50->screen->tesla->channel; struct nouveau_grobj *tesla = nv50->screen->tesla; map += start; @@ -163,10 +164,12 @@ nv50_draw_elements(struct pipe_context *pipe, unsigned mode, unsigned start, unsigned count) { struct nv50_context *nv50 = nv50_context(pipe); - struct nouveau_channel *chan = nv50->screen->nvws->channel; + struct nouveau_channel *chan = nv50->screen->tesla->channel; struct nouveau_grobj *tesla = nv50->screen->tesla; - struct pipe_winsys *ws = pipe->winsys; - void *map = ws->buffer_map(ws, indexBuffer, PIPE_BUFFER_USAGE_CPU_READ); + struct pipe_screen *pscreen = pipe->screen; + void *map; + + map = pipe_buffer_map(pscreen, indexBuffer, PIPE_BUFFER_USAGE_CPU_READ); nv50_state_validate(nv50); @@ -193,6 +196,7 @@ nv50_draw_elements(struct pipe_context *pipe, BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1); OUT_RING (chan, 0); + pipe_buffer_unmap(pscreen, indexBuffer); pipe->flush(pipe, 0, NULL); return TRUE; } @@ -212,6 +216,7 @@ nv50_vbo_validate(struct nv50_context *nv50) struct pipe_vertex_element *ve = &nv50->vtxelt[i]; struct pipe_vertex_buffer *vb = &nv50->vtxbuf[ve->vertex_buffer_index]; + struct nouveau_bo *bo = nouveau_bo(vb->buffer); switch (ve->src_format) { case PIPE_FORMAT_R32G32B32A32_FLOAT: @@ -240,10 +245,10 @@ nv50_vbo_validate(struct nv50_context *nv50) so_method(vtxbuf, tesla, 0x900 + (i * 16), 3); so_data (vtxbuf, 0x20000000 | vb->stride); - so_reloc (vtxbuf, vb->buffer, vb->buffer_offset + + so_reloc (vtxbuf, bo, vb->buffer_offset + ve->src_offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); - so_reloc (vtxbuf, vb->buffer, vb->buffer_offset + + so_reloc (vtxbuf, bo, vb->buffer_offset + ve->src_offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); } |