From 2abc1b3641e435e0b68490fa6b0a7ffa7c030c76 Mon Sep 17 00:00:00 2001 From: Zack Rusin Date: Thu, 1 May 2008 12:38:51 -0400 Subject: abstract fetching elts --- src/gallium/auxiliary/draw/draw_pt_varray.c | 17 +++++++++++------ src/gallium/auxiliary/draw/draw_pt_varray_tmp.h | 6 ++---- 2 files changed, 13 insertions(+), 10 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt_varray.c b/src/gallium/auxiliary/draw/draw_pt_varray.c index 355093f945..b0bd2b983e 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray.c +++ b/src/gallium/auxiliary/draw/draw_pt_varray.c @@ -43,6 +43,8 @@ struct varray_frontend { unsigned draw_count; unsigned fetch_count; + unsigned fetch_start; + struct draw_pt_middle_end *middle; unsigned input_prim; @@ -68,15 +70,18 @@ static void varray_flush(struct varray_frontend *varray) varray->draw_count = 0; } -#if 0 -static void varray_check_flush(struct varray_frontend *varray) +static INLINE void fetch_init(struct varray_frontend *varray, + unsigned current_count, + unsigned count) { - if (varray->draw_count + 6 >= DRAW_MAX/* || - varray->fetch_count + 4 >= FETCH_MAX*/) { - varray_flush(varray); + unsigned idx; + const unsigned end = MIN2(FETCH_MAX, count - current_count); + for (idx = 0; idx < end; ++idx) { + varray->fetch_elts[idx] = varray->fetch_start + idx; } + varray->fetch_start += idx; + varray->fetch_count = idx; } -#endif static INLINE void add_draw_el(struct varray_frontend *varray, int idx, ushort flags) diff --git a/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h index b9a319b253..a3509613f5 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h +++ b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h @@ -20,10 +20,8 @@ static void FUNC(struct draw_pt_front_end *frontend, start, count); #endif - for (i = 0; i < count; ++i) { - varray->fetch_elts[i] = start + i; - } - varray->fetch_count = count; + varray->fetch_start = start; + fetch_init(varray, 0, count); switch (varray->input_prim) { case PIPE_PRIM_POINTS: -- cgit v1.2.3 From 90a46ed277cc887d49c8d8c627174c3bd693ecf7 Mon Sep 17 00:00:00 2001 From: Zack Rusin Date: Thu, 1 May 2008 23:54:39 -0400 Subject: split larger primitives in the simple varray pt --- src/gallium/auxiliary/draw/draw_pt.c | 3 +- src/gallium/auxiliary/draw/draw_pt_varray.c | 62 ++++++++- src/gallium/auxiliary/draw/draw_pt_varray_tmp.h | 169 +++++++++++++++++------- 3 files changed, 180 insertions(+), 54 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c index c9c5d18313..bccde6c5fd 100644 --- a/src/gallium/auxiliary/draw/draw_pt.c +++ b/src/gallium/auxiliary/draw/draw_pt.c @@ -80,8 +80,7 @@ draw_pt_arrays(struct draw_context *draw, /* Pick the right frontend */ - if (draw->pt.user.elts || - count >= 256) { + if (draw->pt.user.elts) { frontend = draw->pt.front.vcache; } else { frontend = draw->pt.front.varray; diff --git a/src/gallium/auxiliary/draw/draw_pt_varray.c b/src/gallium/auxiliary/draw/draw_pt_varray.c index b0bd2b983e..c9843bded0 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray.c +++ b/src/gallium/auxiliary/draw/draw_pt_varray.c @@ -58,6 +58,11 @@ static void varray_flush(struct varray_frontend *varray) debug_printf("FLUSH fc = %d, dc = %d\n", varray->fetch_count, varray->draw_count); + debug_printf("\telt0 = %d, eltx = %d, draw0 = %d, drawx = %d\n", + varray->fetch_elts[0], + varray->fetch_elts[varray->fetch_count-1], + varray->draw_elts[0], + varray->draw_elts[varray->draw_count-1]); #endif varray->middle->run(varray->middle, varray->fetch_elts, @@ -71,18 +76,69 @@ static void varray_flush(struct varray_frontend *varray) } static INLINE void fetch_init(struct varray_frontend *varray, - unsigned current_count, unsigned count) { unsigned idx; - const unsigned end = MIN2(FETCH_MAX, count - current_count); - for (idx = 0; idx < end; ++idx) { +#if 0 + debug_printf("FETCH INIT c = %d, fs = %d\n", + count, + varray->fetch_start); +#endif + for (idx = 0; idx < count; ++idx) { varray->fetch_elts[idx] = varray->fetch_start + idx; } varray->fetch_start += idx; varray->fetch_count = idx; } + +static boolean split_prim_inplace(unsigned prim, unsigned *first, unsigned *incr) +{ + switch (prim) { + case PIPE_PRIM_POINTS: + *first = 1; + *incr = 1; + return TRUE; + case PIPE_PRIM_LINES: + *first = 2; + *incr = 2; + return TRUE; + case PIPE_PRIM_LINE_STRIP: + *first = 2; + *incr = 1; + return TRUE; + case PIPE_PRIM_TRIANGLES: + *first = 3; + *incr = 3; + return TRUE; + case PIPE_PRIM_TRIANGLE_STRIP: + *first = 3; + *incr = 1; + return TRUE; + case PIPE_PRIM_TRIANGLE_FAN: + *first = 3; + *incr = 1; + return TRUE; + case PIPE_PRIM_QUADS: + *first = 4; + *incr = 4; + return TRUE; + case PIPE_PRIM_QUAD_STRIP: + *first = 4; + *incr = 2; + return TRUE; + case PIPE_PRIM_POLYGON: + *first = 3; + *incr = 1; + return TRUE; + default: + *first = 0; + *incr = 1; /* set to one so that count % incr works */ + return FALSE; + } +} + + static INLINE void add_draw_el(struct varray_frontend *varray, int idx, ushort flags) { diff --git a/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h index a3509613f5..073c1aadbf 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h +++ b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h @@ -10,30 +10,42 @@ static void FUNC(struct draw_pt_front_end *frontend, boolean flatfirst = (draw->rasterizer->flatshade && draw->rasterizer->flatshade_first); - unsigned i, flags; + unsigned i, j, flags; + unsigned first, incr; -#if 0 - debug_printf("%s (%d) %d/%d\n", __FUNCTION__, draw->prim, start, count); -#endif -#if 0 - debug_printf("INPUT PRIM = %d (start = %d, count = %d)\n", varray->input_prim, + varray->fetch_start = start; + + split_prim_inplace(varray->input_prim, &first, &incr); + +#if 1 + debug_printf("%s (%d) %d/%d\n", __FUNCTION__, + varray->input_prim, start, count); #endif - varray->fetch_start = start; - fetch_init(varray, 0, count); - switch (varray->input_prim) { case PIPE_PRIM_POINTS: - for (i = 0; i < count; i ++) { - POINT(varray, i + 0); + for (j = 0; j + first <= count; j += i) { + unsigned end = MIN2(FETCH_MAX, count - j); + end -= (end % incr); + for (i = 0; i < count; i ++) { + POINT(varray, i + 0); + } + fetch_init(varray, end); + varray_flush(varray); } break; case PIPE_PRIM_LINES: - for (i = 0; i+1 < count; i += 2) { - LINE(varray, DRAW_PIPE_RESET_STIPPLE, - i + 0, i + 1); + for (j = 0; j + first <= count; j += i) { + unsigned end = MIN2(FETCH_MAX, count - j); + end -= (end % incr); + for (i = 0; i+1 < end; i += 2) { + LINE(varray, DRAW_PIPE_RESET_STIPPLE, + i + 0, i + 1); + } + fetch_init(varray, end); + varray_flush(varray); } break; @@ -41,38 +53,68 @@ static void FUNC(struct draw_pt_front_end *frontend, if (count >= 2) { flags = DRAW_PIPE_RESET_STIPPLE; - for (i = 1; i < count; i++, flags = 0) { - LINE(varray, flags, i - 1, i); + for (j = 0; j + first <= count; j += i) { + unsigned end = MIN2(FETCH_MAX, count - j); + end -= (end % incr); + for (i = 1; i < end; i++, flags = 0) { + LINE(varray, flags, i - 1, i); + } + LINE(varray, flags, i - 1, 0); + fetch_init(varray, end); + varray_flush(varray); } - LINE(varray, flags, i - 1, 0); } break; case PIPE_PRIM_LINE_STRIP: flags = DRAW_PIPE_RESET_STIPPLE; - for (i = 1; i < count; i++, flags = 0) { - LINE(varray, flags, i - 1, i); + for (j = 0; j + first <= count; j += i) { + unsigned end = MIN2(FETCH_MAX, count - j); + end -= (end % incr); + for (i = 1; i < end; i++, flags = 0) { + LINE(varray, flags, i - 1, i); + } + fetch_init(varray, end); + varray_flush(varray); } break; case PIPE_PRIM_TRIANGLES: - for (i = 0; i+2 < count; i += 3) { - TRIANGLE(varray, DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, - i + 0, i + 1, i + 2); + for (j = 0; j + first <= count; j += i) { + unsigned end = MIN2(FETCH_MAX, count - j); + end -= (end % incr); + for (i = 0; i+2 < end; i += 3) { + TRIANGLE(varray, DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, + i + 0, i + 1, i + 2); + } + fetch_init(varray, end); + varray_flush(varray); } break; case PIPE_PRIM_TRIANGLE_STRIP: if (flatfirst) { - for (i = 0; i+2 < count; i++) { - TRIANGLE(varray, DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, - i + 0, i + 1 + (i&1), i + 2 - (i&1)); + for (j = 0; j + first <= count; j += i) { + unsigned end = MIN2(FETCH_MAX, count - j); + end -= (end % incr); + for (i = 0; i+2 < end; i++) { + TRIANGLE(varray, DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, + i + 0, i + 1 + (i&1), i + 2 - (i&1)); + } + fetch_init(varray, end); + varray_flush(varray); } } else { - for (i = 0; i+2 < count; i++) { - TRIANGLE(varray, DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, - i + 0 + (i&1), i + 1 - (i&1), i + 2); + for (j = 0; j + first <= count; j += i) { + unsigned end = MIN2(FETCH_MAX, count - j); + end -= (end % incr); + for (i = 0; i+2 < end; i++) { + TRIANGLE(varray, DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, + i + 0 + (i&1), i + 1 - (i&1), i + 2); + } + fetch_init(varray, end); + varray_flush(varray); } } break; @@ -81,51 +123,80 @@ static void FUNC(struct draw_pt_front_end *frontend, if (count >= 3) { if (flatfirst) { flags = DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL; - for (i = 0; i+2 < count; i++) { - TRIANGLE(varray, flags, i + 1, i + 2, 0); + for (j = 0; j + first <= count; j += i) { + unsigned end = MIN2(FETCH_MAX, count - j); + end -= (end % incr); + for (i = 0; i+2 < end; i++) { + TRIANGLE(varray, flags, i + 1, i + 2, 0); + } + fetch_init(varray, end); + varray_flush(varray); } } else { flags = DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL; - for (i = 0; i+2 < count; i++) { - TRIANGLE(varray, flags, 0, i + 1, i + 2); + for (j = 0; j + first <= count; j += i) { + unsigned end = MIN2(FETCH_MAX, count - j); + end -= (end % incr); + for (i = 0; i+2 < end; i++) { + TRIANGLE(varray, flags, 0, i + 1, i + 2); + } + fetch_init(varray, end); + varray_flush(varray); } } } break; case PIPE_PRIM_QUADS: - for (i = 0; i+3 < count; i += 4) { - QUAD(varray, i + 0, i + 1, i + 2, i + 3); + for (j = 0; j + first <= count; j += i) { + unsigned end = MIN2(FETCH_MAX, count - j); + end -= (end % incr); + for (i = 0; i+3 < end; i += 4) { + QUAD(varray, i + 0, i + 1, i + 2, i + 3); + } + fetch_init(varray, end); + varray_flush(varray); } break; case PIPE_PRIM_QUAD_STRIP: - for (i = 0; i+3 < count; i += 2) { - QUAD(varray, i + 2, i + 0, i + 1, i + 3); + for (j = 0; j + first <= count; j += i) { + unsigned end = MIN2(FETCH_MAX, count - j); + end -= (end % incr); + for (i = 0; i+3 < end; i += 2) { + QUAD(varray, i + 2, i + 0, i + 1, i + 3); + } + fetch_init(varray, end); + varray_flush(varray); } break; case PIPE_PRIM_POLYGON: { - /* These bitflags look a little odd because we submit the - * vertices as (1,2,0) to satisfy flatshade requirements. - */ - const unsigned edge_first = DRAW_PIPE_EDGE_FLAG_2; - const unsigned edge_middle = DRAW_PIPE_EDGE_FLAG_0; - const unsigned edge_last = DRAW_PIPE_EDGE_FLAG_1; - - flags = DRAW_PIPE_RESET_STIPPLE | edge_first | edge_middle; - - for (i = 0; i+2 < count; i++, flags = edge_middle) { + /* These bitflags look a little odd because we submit the + * vertices as (1,2,0) to satisfy flatshade requirements. + */ + const unsigned edge_first = DRAW_PIPE_EDGE_FLAG_2; + const unsigned edge_middle = DRAW_PIPE_EDGE_FLAG_0; + const unsigned edge_last = DRAW_PIPE_EDGE_FLAG_1; + + flags = DRAW_PIPE_RESET_STIPPLE | edge_first | edge_middle; + for (j = 0; j + first <= count; j += i) { + unsigned end = MIN2(FETCH_MAX, count - j); + end -= (end % incr); + for (i = 0; i+2 < end; i++, flags = edge_middle) { if (i + 3 == count) flags |= edge_last; - TRIANGLE(varray, flags, i + 1, i + 2, 0); - } + TRIANGLE(varray, flags, i + 1, i + 2, 0); + } + fetch_init(varray, end); + varray_flush(varray); } - break; + } + break; default: assert(0); -- cgit v1.2.3 From abb08e9335b5d7cb004dc9e6cec390ab6968abe5 Mon Sep 17 00:00:00 2001 From: Zack Rusin Date: Sat, 3 May 2008 22:32:17 -0400 Subject: implement linear emition and fetching and plug it in the varray paths --- src/gallium/auxiliary/draw/draw_pt.h | 18 +++++ src/gallium/auxiliary/draw/draw_pt_emit.c | 45 +++++++++++ src/gallium/auxiliary/draw/draw_pt_fetch.c | 36 +++++++++ .../auxiliary/draw/draw_pt_fetch_shade_pipeline.c | 88 ++++++++++++++++++++-- src/gallium/auxiliary/draw/draw_pt_varray.c | 26 ++++++- src/gallium/auxiliary/draw/draw_pt_varray_tmp.h | 4 +- 6 files changed, 208 insertions(+), 9 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt.h b/src/gallium/auxiliary/draw/draw_pt.h index 2dec376cee..2f96ceaf00 100644 --- a/src/gallium/auxiliary/draw/draw_pt.h +++ b/src/gallium/auxiliary/draw/draw_pt.h @@ -92,6 +92,12 @@ struct draw_pt_middle_end { const ushort *draw_elts, unsigned draw_count ); + void (*run_linear)(struct draw_pt_middle_end *, + unsigned fetch_start, + unsigned fetch_count, + const ushort *draw_elts, + unsigned draw_count); + void (*finish)( struct draw_pt_middle_end * ); void (*destroy)( struct draw_pt_middle_end * ); }; @@ -152,6 +158,13 @@ void draw_pt_emit( struct pt_emit *emit, const ushort *elts, unsigned count ); +void draw_pt_emit_linear( struct pt_emit *emit, + const float (*vertex_data)[4], + unsigned vertex_count, + unsigned stride, + unsigned start, + unsigned count ); + void draw_pt_emit_destroy( struct pt_emit *emit ); struct pt_emit *draw_pt_emit_create( struct draw_context *draw ); @@ -170,6 +183,11 @@ void draw_pt_fetch_run( struct pt_fetch *fetch, unsigned count, char *verts ); +void draw_pt_fetch_run_linear( struct pt_fetch *fetch, + unsigned start, + unsigned count, + char *verts ); + void draw_pt_fetch_destroy( struct pt_fetch *fetch ); struct pt_fetch *draw_pt_fetch_create( struct draw_context *draw ); diff --git a/src/gallium/auxiliary/draw/draw_pt_emit.c b/src/gallium/auxiliary/draw/draw_pt_emit.c index f9ac16786e..2a961b7088 100644 --- a/src/gallium/auxiliary/draw/draw_pt_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_emit.c @@ -179,6 +179,51 @@ void draw_pt_emit( struct pt_emit *emit, } +void draw_pt_emit_linear(struct pt_emit *emit, + const float (*vertex_data)[4], + unsigned vertex_count, + unsigned stride, + unsigned start, + unsigned count) +{ + struct draw_context *draw = emit->draw; + struct translate *translate = emit->translate; + struct vbuf_render *render = draw->render; + void *hw_verts; + + debug_printf("XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n"); + /* XXX: need to flush to get prim_vbuf.c to release its allocation?? + */ + draw_do_flush( draw, DRAW_FLUSH_BACKEND ); + + hw_verts = render->allocate_vertices(render, + (ushort)translate->key.output_stride, + (ushort)count); + if (!hw_verts) { + assert(0); + return; + } + + translate->set_buffer(translate, 0, + vertex_data, stride); + + translate->set_buffer(translate, 1, + &draw->rasterizer->point_size, + 0); + + translate->run(translate, + 0, + vertex_count, + hw_verts); + + render->draw_arrays(render, start, count); + + render->release_vertices(render, + hw_verts, + translate->key.output_stride, + vertex_count); +} + struct pt_emit *draw_pt_emit_create( struct draw_context *draw ) { struct pt_emit *emit = CALLOC_STRUCT(pt_emit); diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch.c b/src/gallium/auxiliary/draw/draw_pt_fetch.c index 1f765b73ad..d7cc1807d7 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch.c @@ -167,6 +167,42 @@ void draw_pt_fetch_run( struct pt_fetch *fetch, } +void draw_pt_fetch_run_linear( struct pt_fetch *fetch, + unsigned start, + unsigned count, + char *verts ) +{ + struct draw_context *draw = fetch->draw; + struct translate *translate = fetch->translate; + unsigned i; + + for (i = 0; i < draw->pt.nr_vertex_buffers; i++) { + translate->set_buffer(translate, + i, + ((char *)draw->pt.user.vbuffer[i] + + draw->pt.vertex_buffer[i].buffer_offset), + draw->pt.vertex_buffer[i].pitch ); + } + + translate->run( translate, + start, + count, + verts ); + + /* Edgeflags are hard to fit into a translate program, populate + * them separately if required. In the setup above they are + * defaulted to one, so only need this if there is reason to change + * that default: + */ + if (fetch->need_edgeflags) { + for (i = 0; i < count; i++) { + struct vertex_header *vh = (struct vertex_header *)(verts + i * fetch->vertex_size); + vh->edgeflag = draw_pt_get_edgeflag( draw, start + i ); + } + } +} + + struct pt_fetch *draw_pt_fetch_create( struct draw_context *draw ) { struct pt_fetch *fetch = CALLOC_STRUCT(pt_fetch); diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c index 4ec20493c4..b1e08a8f40 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c @@ -162,7 +162,7 @@ static void fetch_pipeline_run( struct draw_pt_middle_end *middle, fpme->vertex_size, draw_elts, draw_count ); - } + } else { draw_pt_emit( fpme->emit, (const float (*)[4])pipeline_verts->data, @@ -177,6 +177,83 @@ static void fetch_pipeline_run( struct draw_pt_middle_end *middle, } +static void fetch_pipeline_linear_run( struct draw_pt_middle_end *middle, + unsigned fetch_start, + unsigned fetch_count, + const ushort *draw_elts, + unsigned draw_count ) +{ + struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle; + struct draw_context *draw = fpme->draw; + struct draw_vertex_shader *shader = draw->vertex_shader; + unsigned opt = fpme->opt; + unsigned alloc_count = align_int( fetch_count, 4 ); + + struct vertex_header *pipeline_verts = + (struct vertex_header *)MALLOC(fpme->vertex_size * alloc_count); + + if (!pipeline_verts) { + /* Not much we can do here - just skip the rendering. + */ + assert(0); + return; + } + + /* Fetch into our vertex buffer + */ + draw_pt_fetch_run_linear( fpme->fetch, + fetch_start, + fetch_count, + (char *)pipeline_verts ); + + /* Run the shader, note that this overwrites the data[] parts of + * the pipeline verts. If there is no shader, ie a bypass shader, + * then the inputs == outputs, and are already in the correct + * place. + */ + if (opt & PT_SHADE) + { + shader->run_linear(shader, + (const float (*)[4])pipeline_verts->data, + ( float (*)[4])pipeline_verts->data, + (const float (*)[4])draw->pt.user.constants, + fetch_count, + fpme->vertex_size, + fpme->vertex_size); + } + + if (draw_pt_post_vs_run( fpme->post_vs, + pipeline_verts, + fetch_count, + fpme->vertex_size )) + { + opt |= PT_PIPELINE; + } + + /* Do we need to run the pipeline? + */ + if (opt & PT_PIPELINE) { + draw_pipeline_run( fpme->draw, + fpme->prim, + pipeline_verts, + fetch_count, + fpme->vertex_size, + draw_elts, + draw_count ); + } + else { + draw_pt_emit_linear( fpme->emit, + (const float (*)[4])pipeline_verts->data, + fetch_count, + fpme->vertex_size, + 0, /*start*/ + draw_count ); + } + + FREE(pipeline_verts); +} + + static void fetch_pipeline_finish( struct draw_pt_middle_end *middle ) { @@ -206,10 +283,11 @@ struct draw_pt_middle_end *draw_pt_fetch_pipeline_or_emit( struct draw_context * if (!fpme) goto fail; - fpme->base.prepare = fetch_pipeline_prepare; - fpme->base.run = fetch_pipeline_run; - fpme->base.finish = fetch_pipeline_finish; - fpme->base.destroy = fetch_pipeline_destroy; + fpme->base.prepare = fetch_pipeline_prepare; + fpme->base.run = fetch_pipeline_run; + fpme->base.run_linear = fetch_pipeline_linear_run; + fpme->base.finish = fetch_pipeline_finish; + fpme->base.destroy = fetch_pipeline_destroy; fpme->draw = draw; diff --git a/src/gallium/auxiliary/draw/draw_pt_varray.c b/src/gallium/auxiliary/draw/draw_pt_varray.c index c9843bded0..916373acc8 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray.c +++ b/src/gallium/auxiliary/draw/draw_pt_varray.c @@ -75,6 +75,28 @@ static void varray_flush(struct varray_frontend *varray) varray->draw_count = 0; } +static void varray_flush_linear(struct varray_frontend *varray) +{ + if (varray->draw_count) { + debug_printf("FLUSH LINEAR fc = %d, dc = %d\n", + varray->fetch_count, + varray->draw_count); + debug_printf("\telt0 = %d, eltx = %d, draw0 = %d, drawx = %d\n", + varray->fetch_elts[0], + varray->fetch_elts[varray->fetch_count-1], + varray->draw_elts[0], + varray->draw_elts[varray->draw_count-1]); + varray->middle->run_linear(varray->middle, + varray->fetch_elts[0], + varray->fetch_count, + varray->draw_elts, + varray->draw_count); + } + + varray->fetch_count = 0; + varray->draw_count = 0; +} + static INLINE void fetch_init(struct varray_frontend *varray, unsigned count) { @@ -265,8 +287,8 @@ static void varray_prepare(struct draw_pt_front_end *frontend, if (opt & PT_PIPELINE) { varray->base.run = varray_run_extras; - } - else + } + else { varray->base.run = varray_run; } diff --git a/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h index 073c1aadbf..67baafa3be 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h +++ b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h @@ -17,7 +17,7 @@ static void FUNC(struct draw_pt_front_end *frontend, split_prim_inplace(varray->input_prim, &first, &incr); -#if 1 +#if 0 debug_printf("%s (%d) %d/%d\n", __FUNCTION__, varray->input_prim, start, count); @@ -88,7 +88,7 @@ static void FUNC(struct draw_pt_front_end *frontend, i + 0, i + 1, i + 2); } fetch_init(varray, end); - varray_flush(varray); + varray_flush_linear(varray); } break; -- cgit v1.2.3 From ff1fee2cae9fabb47d6a2eb1f9f8094fec3c377f Mon Sep 17 00:00:00 2001 From: Zack Rusin Date: Sun, 4 May 2008 00:44:27 -0400 Subject: don't fill in linear fetch_elts --- src/gallium/auxiliary/draw/draw_pt_varray.c | 4 +++- src/gallium/auxiliary/draw/draw_pt_varray_tmp.h | 3 ++- 2 files changed, 5 insertions(+), 2 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt_varray.c b/src/gallium/auxiliary/draw/draw_pt_varray.c index 916373acc8..fb1b59d53e 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray.c +++ b/src/gallium/auxiliary/draw/draw_pt_varray.c @@ -78,6 +78,7 @@ static void varray_flush(struct varray_frontend *varray) static void varray_flush_linear(struct varray_frontend *varray) { if (varray->draw_count) { +#if 0 debug_printf("FLUSH LINEAR fc = %d, dc = %d\n", varray->fetch_count, varray->draw_count); @@ -86,8 +87,9 @@ static void varray_flush_linear(struct varray_frontend *varray) varray->fetch_elts[varray->fetch_count-1], varray->draw_elts[0], varray->draw_elts[varray->draw_count-1]); +#endif varray->middle->run_linear(varray->middle, - varray->fetch_elts[0], + varray->fetch_start, varray->fetch_count, varray->draw_elts, varray->draw_count); diff --git a/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h index 67baafa3be..a9f844a357 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h +++ b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h @@ -87,8 +87,9 @@ static void FUNC(struct draw_pt_front_end *frontend, TRIANGLE(varray, DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, i + 0, i + 1, i + 2); } - fetch_init(varray, end); + varray->fetch_count = end; varray_flush_linear(varray); + varray->fetch_start += end; } break; -- cgit v1.2.3 From a24cb269e1ba5434acf8c94abd03517c149b9c51 Mon Sep 17 00:00:00 2001 From: Zack Rusin Date: Sun, 4 May 2008 01:23:01 -0400 Subject: implement linear path for fetch_emit pipeline --- src/gallium/auxiliary/draw/draw_pt_fetch_emit.c | 64 +++++++++++++++++++++++-- src/gallium/auxiliary/draw/draw_pt_varray.c | 1 + 2 files changed, 61 insertions(+), 4 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c index a4de341df8..6d5a54cf0e 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c @@ -257,6 +257,61 @@ static void fetch_emit_run( struct draw_pt_middle_end *middle, } +static void fetch_emit_run_linear( struct draw_pt_middle_end *middle, + unsigned fetch_start, + unsigned fetch_count, + const ushort *draw_elts, + unsigned draw_count ) +{ + struct fetch_emit_middle_end *feme = (struct fetch_emit_middle_end *)middle; + struct draw_context *draw = feme->draw; + void *hw_verts; + + /* XXX: need to flush to get prim_vbuf.c to release its allocation?? + */ + draw_do_flush( draw, DRAW_FLUSH_BACKEND ); + + hw_verts = draw->render->allocate_vertices( draw->render, + (ushort)feme->translate->key.output_stride, + (ushort)fetch_count ); + if (!hw_verts) { + assert(0); + return; + } + + /* Single routine to fetch vertices and emit HW verts. + */ + feme->translate->run( feme->translate, + fetch_start, + fetch_count, + hw_verts ); + + if (0) { + unsigned i; + for (i = 0; i < fetch_count; i++) { + debug_printf("\n\nvertex %d:\n", i); + draw_dump_emitted_vertex( feme->vinfo, + (const uint8_t *)hw_verts + feme->vinfo->size * 4 * i ); + } + } + + /* XXX: Draw arrays path to avoid re-emitting index list again and + * again. + */ + draw->render->draw_arrays( draw->render, + 0, /*start*/ + draw_count ); + + /* Done -- that was easy, wasn't it: + */ + draw->render->release_vertices( draw->render, + hw_verts, + feme->translate->key.output_stride, + fetch_count ); + +} + + static void fetch_emit_finish( struct draw_pt_middle_end *middle ) { @@ -285,10 +340,11 @@ struct draw_pt_middle_end *draw_pt_fetch_emit( struct draw_context *draw ) return NULL; } - fetch_emit->base.prepare = fetch_emit_prepare; - fetch_emit->base.run = fetch_emit_run; - fetch_emit->base.finish = fetch_emit_finish; - fetch_emit->base.destroy = fetch_emit_destroy; + fetch_emit->base.prepare = fetch_emit_prepare; + fetch_emit->base.run = fetch_emit_run; + fetch_emit->base.run_linear = fetch_emit_run_linear; + fetch_emit->base.finish = fetch_emit_finish; + fetch_emit->base.destroy = fetch_emit_destroy; fetch_emit->draw = draw; diff --git a/src/gallium/auxiliary/draw/draw_pt_varray.c b/src/gallium/auxiliary/draw/draw_pt_varray.c index fb1b59d53e..e7e21e4bf6 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray.c +++ b/src/gallium/auxiliary/draw/draw_pt_varray.c @@ -88,6 +88,7 @@ static void varray_flush_linear(struct varray_frontend *varray) varray->draw_elts[0], varray->draw_elts[varray->draw_count-1]); #endif + assert(varray->middle->run_linear); varray->middle->run_linear(varray->middle, varray->fetch_start, varray->fetch_count, -- cgit v1.2.3 From 66d72f176de2568f053c6dc54e93d423723ae8aa Mon Sep 17 00:00:00 2001 From: Zack Rusin Date: Sun, 4 May 2008 01:37:32 -0400 Subject: silence debugging output --- src/gallium/auxiliary/draw/draw_pt_emit.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt_emit.c b/src/gallium/auxiliary/draw/draw_pt_emit.c index 2a961b7088..776ca32cfa 100644 --- a/src/gallium/auxiliary/draw/draw_pt_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_emit.c @@ -191,7 +191,9 @@ void draw_pt_emit_linear(struct pt_emit *emit, struct vbuf_render *render = draw->render; void *hw_verts; - debug_printf("XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n"); +#if 0 + debug_printf("Linear emit\n"); +#endif /* XXX: need to flush to get prim_vbuf.c to release its allocation?? */ draw_do_flush( draw, DRAW_FLUSH_BACKEND ); -- cgit v1.2.3 From e897fd6cd35c6b9e398e1903d2e79678fe85708a Mon Sep 17 00:00:00 2001 From: Zack Rusin Date: Mon, 5 May 2008 12:49:40 -0400 Subject: fix the regressions --- src/gallium/auxiliary/draw/draw_pt_varray_tmp.h | 14 +++++++++++++- 1 file changed, 13 insertions(+), 1 deletion(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h index a9f844a357..10ac08ea30 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h +++ b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h @@ -28,9 +28,10 @@ static void FUNC(struct draw_pt_front_end *frontend, for (j = 0; j + first <= count; j += i) { unsigned end = MIN2(FETCH_MAX, count - j); end -= (end % incr); - for (i = 0; i < count; i ++) { + for (i = 0; i < count; i++) { POINT(varray, i + 0); } + i = end; fetch_init(varray, end); varray_flush(varray); } @@ -44,6 +45,7 @@ static void FUNC(struct draw_pt_front_end *frontend, LINE(varray, DRAW_PIPE_RESET_STIPPLE, i + 0, i + 1); } + i = end; fetch_init(varray, end); varray_flush(varray); } @@ -60,6 +62,7 @@ static void FUNC(struct draw_pt_front_end *frontend, LINE(varray, flags, i - 1, i); } LINE(varray, flags, i - 1, 0); + i = end; fetch_init(varray, end); varray_flush(varray); } @@ -74,6 +77,7 @@ static void FUNC(struct draw_pt_front_end *frontend, for (i = 1; i < end; i++, flags = 0) { LINE(varray, flags, i - 1, i); } + i = end; fetch_init(varray, end); varray_flush(varray); } @@ -87,6 +91,7 @@ static void FUNC(struct draw_pt_front_end *frontend, TRIANGLE(varray, DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, i + 0, i + 1, i + 2); } + i = end; varray->fetch_count = end; varray_flush_linear(varray); varray->fetch_start += end; @@ -102,6 +107,7 @@ static void FUNC(struct draw_pt_front_end *frontend, TRIANGLE(varray, DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, i + 0, i + 1 + (i&1), i + 2 - (i&1)); } + i = end; fetch_init(varray, end); varray_flush(varray); } @@ -114,6 +120,7 @@ static void FUNC(struct draw_pt_front_end *frontend, TRIANGLE(varray, DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, i + 0 + (i&1), i + 1 - (i&1), i + 2); } + i = end; fetch_init(varray, end); varray_flush(varray); } @@ -130,6 +137,7 @@ static void FUNC(struct draw_pt_front_end *frontend, for (i = 0; i+2 < end; i++) { TRIANGLE(varray, flags, i + 1, i + 2, 0); } + i = end; fetch_init(varray, end); varray_flush(varray); } @@ -142,6 +150,7 @@ static void FUNC(struct draw_pt_front_end *frontend, for (i = 0; i+2 < end; i++) { TRIANGLE(varray, flags, 0, i + 1, i + 2); } + i = end; fetch_init(varray, end); varray_flush(varray); } @@ -156,6 +165,7 @@ static void FUNC(struct draw_pt_front_end *frontend, for (i = 0; i+3 < end; i += 4) { QUAD(varray, i + 0, i + 1, i + 2, i + 3); } + i = end; fetch_init(varray, end); varray_flush(varray); } @@ -168,6 +178,7 @@ static void FUNC(struct draw_pt_front_end *frontend, for (i = 0; i+3 < end; i += 2) { QUAD(varray, i + 2, i + 0, i + 1, i + 3); } + i = end; fetch_init(varray, end); varray_flush(varray); } @@ -193,6 +204,7 @@ static void FUNC(struct draw_pt_front_end *frontend, TRIANGLE(varray, flags, i + 1, i + 2, 0); } + i = end; fetch_init(varray, end); varray_flush(varray); } -- cgit v1.2.3 From fe586f8612dd517b9a1f0d87fbaf3a75e3caf588 Mon Sep 17 00:00:00 2001 From: Zack Rusin Date: Tue, 6 May 2008 18:59:45 -0400 Subject: redo the linear paths --- src/gallium/auxiliary/draw/draw_pipe.c | 39 ++++ src/gallium/auxiliary/draw/draw_private.h | 6 + src/gallium/auxiliary/draw/draw_pt.h | 6 +- src/gallium/auxiliary/draw/draw_pt_fetch_emit.c | 18 +- .../auxiliary/draw/draw_pt_fetch_shade_pipeline.c | 36 ++-- src/gallium/auxiliary/draw/draw_pt_varray.c | 27 +-- src/gallium/auxiliary/draw/draw_pt_varray_tmp.h | 5 +- .../auxiliary/draw/draw_pt_varray_tmp_linear.h | 198 +++++++++++++++++++++ 8 files changed, 279 insertions(+), 56 deletions(-) create mode 100644 src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pipe.c b/src/gallium/auxiliary/draw/draw_pipe.c index 46afb0f41f..cb97f955b2 100644 --- a/src/gallium/auxiliary/draw/draw_pipe.c +++ b/src/gallium/auxiliary/draw/draw_pipe.c @@ -212,6 +212,45 @@ void draw_pipeline_run( struct draw_context *draw, draw->pipeline.vertex_count = 0; } +void draw_pipeline_run_linear( struct draw_context *draw, + unsigned prim, + struct vertex_header *vertices, + unsigned count, + unsigned stride ) +{ + char *verts = (char *)vertices; + unsigned i; + + draw->pipeline.verts = verts; + draw->pipeline.vertex_stride = stride; + draw->pipeline.vertex_count = count; + + switch (prim) { + case PIPE_PRIM_POINTS: + for (i = 0; i < count; i++) + do_point( draw, + verts + stride * i ); + break; + case PIPE_PRIM_LINES: + for (i = 0; i+1 < count; i += 2) + do_line( draw, + i+0, /* flags */ + verts + stride * ((i+0) & ~DRAW_PIPE_FLAG_MASK), + verts + stride * (i+1)); + break; + case PIPE_PRIM_TRIANGLES: + for (i = 0; i+2 < count; i += 3) + do_triangle( draw, + (i+0), /* flags */ + verts + stride * ((i+0) & ~DRAW_PIPE_FLAG_MASK), + verts + stride * (i+1), + verts + stride * (i+2)); + break; + } + + draw->pipeline.verts = NULL; + draw->pipeline.vertex_count = 0; +} void draw_pipeline_flush( struct draw_context *draw, diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h index cee58bbf73..e036d498b8 100644 --- a/src/gallium/auxiliary/draw/draw_private.h +++ b/src/gallium/auxiliary/draw/draw_private.h @@ -247,6 +247,12 @@ void draw_pipeline_run( struct draw_context *draw, const ushort *elts, unsigned count ); +void draw_pipeline_run_linear( struct draw_context *draw, + unsigned prim, + struct vertex_header *vertices, + unsigned count, + unsigned stride ); + void draw_pipeline_flush( struct draw_context *draw, diff --git a/src/gallium/auxiliary/draw/draw_pt.h b/src/gallium/auxiliary/draw/draw_pt.h index 2f96ceaf00..312fdbe4f4 100644 --- a/src/gallium/auxiliary/draw/draw_pt.h +++ b/src/gallium/auxiliary/draw/draw_pt.h @@ -93,10 +93,8 @@ struct draw_pt_middle_end { unsigned draw_count ); void (*run_linear)(struct draw_pt_middle_end *, - unsigned fetch_start, - unsigned fetch_count, - const ushort *draw_elts, - unsigned draw_count); + unsigned start, + unsigned count); void (*finish)( struct draw_pt_middle_end * ); void (*destroy)( struct draw_pt_middle_end * ); diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c index 6d5a54cf0e..8df4241b82 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c @@ -258,10 +258,8 @@ static void fetch_emit_run( struct draw_pt_middle_end *middle, static void fetch_emit_run_linear( struct draw_pt_middle_end *middle, - unsigned fetch_start, - unsigned fetch_count, - const ushort *draw_elts, - unsigned draw_count ) + unsigned start, + unsigned count ) { struct fetch_emit_middle_end *feme = (struct fetch_emit_middle_end *)middle; struct draw_context *draw = feme->draw; @@ -273,7 +271,7 @@ static void fetch_emit_run_linear( struct draw_pt_middle_end *middle, hw_verts = draw->render->allocate_vertices( draw->render, (ushort)feme->translate->key.output_stride, - (ushort)fetch_count ); + (ushort)count ); if (!hw_verts) { assert(0); return; @@ -282,13 +280,13 @@ static void fetch_emit_run_linear( struct draw_pt_middle_end *middle, /* Single routine to fetch vertices and emit HW verts. */ feme->translate->run( feme->translate, - fetch_start, - fetch_count, + start, + count, hw_verts ); if (0) { unsigned i; - for (i = 0; i < fetch_count; i++) { + for (i = 0; i < count; i++) { debug_printf("\n\nvertex %d:\n", i); draw_dump_emitted_vertex( feme->vinfo, (const uint8_t *)hw_verts + feme->vinfo->size * 4 * i ); @@ -300,14 +298,14 @@ static void fetch_emit_run_linear( struct draw_pt_middle_end *middle, */ draw->render->draw_arrays( draw->render, 0, /*start*/ - draw_count ); + count ); /* Done -- that was easy, wasn't it: */ draw->render->release_vertices( draw->render, hw_verts, feme->translate->key.output_stride, - fetch_count ); + count ); } diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c index b1e08a8f40..dad54690a5 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c @@ -178,18 +178,16 @@ static void fetch_pipeline_run( struct draw_pt_middle_end *middle, static void fetch_pipeline_linear_run( struct draw_pt_middle_end *middle, - unsigned fetch_start, - unsigned fetch_count, - const ushort *draw_elts, - unsigned draw_count ) + unsigned start, + unsigned count) { struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle; struct draw_context *draw = fpme->draw; struct draw_vertex_shader *shader = draw->vertex_shader; unsigned opt = fpme->opt; - unsigned alloc_count = align_int( fetch_count, 4 ); + unsigned alloc_count = align_int( count, 4 ); - struct vertex_header *pipeline_verts = + struct vertex_header *pipeline_verts = (struct vertex_header *)MALLOC(fpme->vertex_size * alloc_count); if (!pipeline_verts) { @@ -202,8 +200,8 @@ static void fetch_pipeline_linear_run( struct draw_pt_middle_end *middle, /* Fetch into our vertex buffer */ draw_pt_fetch_run_linear( fpme->fetch, - fetch_start, - fetch_count, + start, + count, (char *)pipeline_verts ); /* Run the shader, note that this overwrites the data[] parts of @@ -213,18 +211,18 @@ static void fetch_pipeline_linear_run( struct draw_pt_middle_end *middle, */ if (opt & PT_SHADE) { - shader->run_linear(shader, + shader->run_linear(shader, (const float (*)[4])pipeline_verts->data, ( float (*)[4])pipeline_verts->data, (const float (*)[4])draw->pt.user.constants, - fetch_count, + count, fpme->vertex_size, fpme->vertex_size); } if (draw_pt_post_vs_run( fpme->post_vs, pipeline_verts, - fetch_count, + count, fpme->vertex_size )) { opt |= PT_PIPELINE; @@ -233,21 +231,19 @@ static void fetch_pipeline_linear_run( struct draw_pt_middle_end *middle, /* Do we need to run the pipeline? */ if (opt & PT_PIPELINE) { - draw_pipeline_run( fpme->draw, - fpme->prim, - pipeline_verts, - fetch_count, - fpme->vertex_size, - draw_elts, - draw_count ); + draw_pipeline_run_linear( fpme->draw, + fpme->prim, + pipeline_verts, + count, + fpme->vertex_size); } else { draw_pt_emit_linear( fpme->emit, (const float (*)[4])pipeline_verts->data, - fetch_count, + count, fpme->vertex_size, 0, /*start*/ - draw_count ); + count ); } FREE(pipeline_verts); diff --git a/src/gallium/auxiliary/draw/draw_pt_varray.c b/src/gallium/auxiliary/draw/draw_pt_varray.c index e7e21e4bf6..59a9569270 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray.c +++ b/src/gallium/auxiliary/draw/draw_pt_varray.c @@ -75,29 +75,18 @@ static void varray_flush(struct varray_frontend *varray) varray->draw_count = 0; } -static void varray_flush_linear(struct varray_frontend *varray) +static void varray_flush_linear(struct varray_frontend *varray, + unsigned start, unsigned count) { - if (varray->draw_count) { + if (count) { #if 0 - debug_printf("FLUSH LINEAR fc = %d, dc = %d\n", - varray->fetch_count, - varray->draw_count); - debug_printf("\telt0 = %d, eltx = %d, draw0 = %d, drawx = %d\n", - varray->fetch_elts[0], - varray->fetch_elts[varray->fetch_count-1], - varray->draw_elts[0], - varray->draw_elts[varray->draw_count-1]); + debug_printf("FLUSH LINEAR start = %d, count = %d\n", + start, + count); #endif assert(varray->middle->run_linear); - varray->middle->run_linear(varray->middle, - varray->fetch_start, - varray->fetch_count, - varray->draw_elts, - varray->draw_count); + varray->middle->run_linear(varray->middle, start, count); } - - varray->fetch_count = 0; - varray->draw_count = 0; } static INLINE void fetch_init(struct varray_frontend *varray, @@ -261,7 +250,7 @@ static INLINE void varray_ef_quad( struct varray_frontend *varray, #define LINE(vc,flags,i0,i1) varray_line(vc,i0,i1) #define POINT(vc,i0) varray_point(vc,i0) #define FUNC varray_run -#include "draw_pt_varray_tmp.h" +#include "draw_pt_varray_tmp_linear.h" diff --git a/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h index 10ac08ea30..335c4c89ca 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h +++ b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h @@ -92,9 +92,8 @@ static void FUNC(struct draw_pt_front_end *frontend, i + 0, i + 1, i + 2); } i = end; - varray->fetch_count = end; - varray_flush_linear(varray); - varray->fetch_start += end; + fetch_init(varray, end); + varray_flush(varray); } break; diff --git a/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h b/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h new file mode 100644 index 0000000000..dfa4338407 --- /dev/null +++ b/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h @@ -0,0 +1,198 @@ + +static void FUNC(struct draw_pt_front_end *frontend, + pt_elt_func get_elt, + const void *elts, + unsigned count) +{ + struct varray_frontend *varray = (struct varray_frontend *)frontend; + struct draw_context *draw = varray->draw; + unsigned start = (unsigned)elts; + + boolean flatfirst = (draw->rasterizer->flatshade && + draw->rasterizer->flatshade_first); + unsigned i, j, flags; + unsigned first, incr; + + varray->fetch_start = start; + + split_prim_inplace(varray->input_prim, &first, &incr); + +#if 0 + debug_printf("%s (%d) %d/%d\n", __FUNCTION__, + varray->input_prim, + start, count); +#endif + + switch (varray->input_prim) { + case PIPE_PRIM_POINTS: + case PIPE_PRIM_LINES: + case PIPE_PRIM_TRIANGLES: + j = 0; + while (j + first <= count) { + unsigned end = MIN2(FETCH_MAX, count - j); + end -= (end % incr); + varray_flush_linear(varray, start + j, end); + j += end; + } + break; + + case PIPE_PRIM_LINE_LOOP: + if (count >= 2) { + flags = DRAW_PIPE_RESET_STIPPLE; + + for (j = 0; j + first <= count; j += i) { + unsigned end = MIN2(FETCH_MAX, count - j); + end -= (end % incr); + for (i = 1; i < end; i++, flags = 0) { + LINE(varray, flags, i - 1, i); + } + LINE(varray, flags, i - 1, 0); + i = end; + fetch_init(varray, end); + varray_flush(varray); + } + } + break; + + case PIPE_PRIM_LINE_STRIP: + flags = DRAW_PIPE_RESET_STIPPLE; + for (j = 0; j + first <= count; j += i) { + unsigned end = MIN2(FETCH_MAX, count - j); + end -= (end % incr); + for (i = 1; i < end; i++, flags = 0) { + LINE(varray, flags, i - 1, i); + } + i = end; + fetch_init(varray, end); + varray_flush(varray); + } + break; + + case PIPE_PRIM_TRIANGLE_STRIP: + if (flatfirst) { + for (j = 0; j + first <= count; j += i) { + unsigned end = MIN2(FETCH_MAX, count - j); + end -= (end % incr); + for (i = 0; i+2 < end; i++) { + TRIANGLE(varray, DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, + i + 0, i + 1 + (i&1), i + 2 - (i&1)); + } + i = end; + fetch_init(varray, end); + varray_flush(varray); + } + } + else { + for (j = 0; j + first <= count;) { + unsigned end = MIN2(FETCH_MAX, count - j); + //end -= (end % incr); + for (i = 0; i+2 < end; i++) { + TRIANGLE(varray, DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, + i + 0 + (i&1), i + 1 - (i&1), i + 2); + } + fetch_init(varray, end); + varray_flush(varray); + j += end; + if (j <= count) + j -= incr; + } + } + break; + + case PIPE_PRIM_TRIANGLE_FAN: + if (count >= 3) { + if (flatfirst) { + flags = DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL; + for (j = 0; j + first <= count; j += i) { + unsigned end = MIN2(FETCH_MAX, count - j); + end -= (end % incr); + for (i = 0; i+2 < end; i++) { + TRIANGLE(varray, flags, i + 1, i + 2, 0); + } + i = end; + fetch_init(varray, end); + varray_flush(varray); + } + } + else { + flags = DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL; + for (j = 0; j + first <= count; j += i) { + unsigned end = MIN2(FETCH_MAX, count - j); + end -= (end % incr); + for (i = 0; i+2 < end; i++) { + TRIANGLE(varray, flags, 0, i + 1, i + 2); + } + i = end; + fetch_init(varray, end); + varray_flush(varray); + } + } + } + break; + + case PIPE_PRIM_QUADS: + for (j = 0; j + first <= count; j += i) { + unsigned end = MIN2(FETCH_MAX, count - j); + end -= (end % incr); + for (i = 0; i+3 < end; i += 4) { + QUAD(varray, i + 0, i + 1, i + 2, i + 3); + } + i = end; + fetch_init(varray, end); + varray_flush(varray); + } + break; + + case PIPE_PRIM_QUAD_STRIP: + for (j = 0; j + first <= count; j += i) { + unsigned end = MIN2(FETCH_MAX, count - j); + end -= (end % incr); + for (i = 0; i+3 < end; i += 2) { + QUAD(varray, i + 2, i + 0, i + 1, i + 3); + } + i = end; + fetch_init(varray, end); + varray_flush(varray); + } + break; + + case PIPE_PRIM_POLYGON: + { + /* These bitflags look a little odd because we submit the + * vertices as (1,2,0) to satisfy flatshade requirements. + */ + const unsigned edge_first = DRAW_PIPE_EDGE_FLAG_2; + const unsigned edge_middle = DRAW_PIPE_EDGE_FLAG_0; + const unsigned edge_last = DRAW_PIPE_EDGE_FLAG_1; + + flags = DRAW_PIPE_RESET_STIPPLE | edge_first | edge_middle; + for (j = 0; j + first <= count; j += i) { + unsigned end = MIN2(FETCH_MAX, count - j); + end -= (end % incr); + for (i = 0; i+2 < end; i++, flags = edge_middle) { + + if (i + 3 == count) + flags |= edge_last; + + TRIANGLE(varray, flags, i + 1, i + 2, 0); + } + i = end; + fetch_init(varray, end); + varray_flush(varray); + } + } + break; + + default: + assert(0); + break; + } + + varray_flush(varray); +} + +#undef TRIANGLE +#undef QUAD +#undef POINT +#undef LINE +#undef FUNC -- cgit v1.2.3 From 22323af525d00022a1fa06fab7ee84df5ef2d1f0 Mon Sep 17 00:00:00 2001 From: Zack Rusin Date: Wed, 7 May 2008 19:34:12 -0400 Subject: fix silly mistakes --- src/gallium/auxiliary/draw/draw_pt_varray_tmp.h | 2 +- src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h | 4 +--- 2 files changed, 2 insertions(+), 4 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h index 335c4c89ca..fb49452d8b 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h +++ b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h @@ -28,7 +28,7 @@ static void FUNC(struct draw_pt_front_end *frontend, for (j = 0; j + first <= count; j += i) { unsigned end = MIN2(FETCH_MAX, count - j); end -= (end % incr); - for (i = 0; i < count; i++) { + for (i = 0; i < end; i++) { POINT(varray, i + 0); } i = end; diff --git a/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h b/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h index dfa4338407..ab28859c35 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h +++ b/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h @@ -85,7 +85,7 @@ static void FUNC(struct draw_pt_front_end *frontend, else { for (j = 0; j + first <= count;) { unsigned end = MIN2(FETCH_MAX, count - j); - //end -= (end % incr); + end -= (end % incr); for (i = 0; i+2 < end; i++) { TRIANGLE(varray, DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, i + 0 + (i&1), i + 1 - (i&1), i + 2); @@ -93,8 +93,6 @@ static void FUNC(struct draw_pt_front_end *frontend, fetch_init(varray, end); varray_flush(varray); j += end; - if (j <= count) - j -= incr; } } break; -- cgit v1.2.3 From 8d709ae1595047b45a81f2fbd22850887fdbfea0 Mon Sep 17 00:00:00 2001 From: Zack Rusin Date: Thu, 8 May 2008 12:10:24 -0400 Subject: fix triangle strips --- src/gallium/auxiliary/draw/draw_pt_varray_tmp.h | 12 ++++++++++-- src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h | 12 ++++++++++-- 2 files changed, 20 insertions(+), 4 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h index fb49452d8b..d137a758e2 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h +++ b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h @@ -109,19 +109,27 @@ static void FUNC(struct draw_pt_front_end *frontend, i = end; fetch_init(varray, end); varray_flush(varray); + if (j + first + i <= count) { + varray->fetch_start -= 2; + i -= 2; + } } } else { for (j = 0; j + first <= count; j += i) { unsigned end = MIN2(FETCH_MAX, count - j); - end -= (end % incr); - for (i = 0; i+2 < end; i++) { + end -= (end % incr); + for (i = 0; i + 2 < end; i++) { TRIANGLE(varray, DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, i + 0 + (i&1), i + 1 - (i&1), i + 2); } i = end; fetch_init(varray, end); varray_flush(varray); + if (j + first + i <= count) { + varray->fetch_start -= 2; + i -= 2; + } } } break; diff --git a/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h b/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h index ab28859c35..4bf04fa62b 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h +++ b/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h @@ -80,19 +80,27 @@ static void FUNC(struct draw_pt_front_end *frontend, i = end; fetch_init(varray, end); varray_flush(varray); + if (j + first + i <= count) { + varray->fetch_start -= 2; + i -= 2; + } } } else { - for (j = 0; j + first <= count;) { + for (j = 0; j + first <= count; j += i) { unsigned end = MIN2(FETCH_MAX, count - j); end -= (end % incr); for (i = 0; i+2 < end; i++) { TRIANGLE(varray, DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, i + 0 + (i&1), i + 1 - (i&1), i + 2); } + i = end; fetch_init(varray, end); varray_flush(varray); - j += end; + if (j + first + i <= count) { + varray->fetch_start -= 2; + i -= 2; + } } } break; -- cgit v1.2.3 From 8ea6106f01f38853e9c0f1029da55eb449109aea Mon Sep 17 00:00:00 2001 From: Zack Rusin Date: Thu, 8 May 2008 15:11:16 -0400 Subject: fix quad strips --- src/gallium/auxiliary/draw/draw_pt_varray_tmp.h | 4 ++++ src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h | 4 ++++ 2 files changed, 8 insertions(+) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h index d137a758e2..1395275897 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h +++ b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h @@ -188,6 +188,10 @@ static void FUNC(struct draw_pt_front_end *frontend, i = end; fetch_init(varray, end); varray_flush(varray); + if (j + first + i <= count) { + varray->fetch_start -= 2; + i -= 2; + } } break; diff --git a/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h b/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h index 4bf04fa62b..6e2b16d9be 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h +++ b/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h @@ -159,6 +159,10 @@ static void FUNC(struct draw_pt_front_end *frontend, i = end; fetch_init(varray, end); varray_flush(varray); + if (j + first + i <= count) { + varray->fetch_start -= 2; + i -= 2; + } } break; -- cgit v1.2.3 From 501be9c7dd0cc5f985c708fa0e5f35d7fd20deb4 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Fri, 9 May 2008 15:02:59 +0100 Subject: draw: fix translate double-free, minor cleanups --- src/gallium/auxiliary/draw/draw_pipe_vbuf.c | 3 --- src/gallium/auxiliary/draw/draw_pt_fetch.c | 3 ++- src/gallium/auxiliary/draw/draw_pt_fetch_emit.c | 3 ++- 3 files changed, 4 insertions(+), 5 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c index 2a19e6916a..64a9f9084f 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c +++ b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c @@ -427,9 +427,6 @@ static void vbuf_destroy( struct draw_stage *stage ) if(vbuf->indices) align_free( vbuf->indices ); - if(vbuf->translate) - vbuf->translate->release( vbuf->translate ); - if (vbuf->render) vbuf->render->destroy( vbuf->render ); diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch.c b/src/gallium/auxiliary/draw/draw_pt_fetch.c index d62cd9358b..034e0eccb2 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch.c @@ -223,7 +223,8 @@ struct pt_fetch *draw_pt_fetch_create( struct draw_context *draw ) void draw_pt_fetch_destroy( struct pt_fetch *fetch ) { - translate_cache_destroy(fetch->cache); + if (fetch->cache) + translate_cache_destroy(fetch->cache); FREE(fetch); } diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c index 8df4241b82..bdbb039f9e 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c @@ -320,7 +320,8 @@ static void fetch_emit_destroy( struct draw_pt_middle_end *middle ) { struct fetch_emit_middle_end *feme = (struct fetch_emit_middle_end *)middle; - translate_cache_destroy(feme->cache); + if (feme->cache) + translate_cache_destroy(feme->cache); FREE(middle); } -- cgit v1.2.3 From c0a6040f568e0c9be07797b2dc2fdd8a3624ec34 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Fri, 9 May 2008 13:09:58 +0100 Subject: translate: helper functions for mimizing cost of key compares --- src/gallium/auxiliary/translate/translate.h | 21 +++++++++++++++++++++ 1 file changed, 21 insertions(+) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/translate/translate.h b/src/gallium/auxiliary/translate/translate.h index de6f09d18a..b8210af50c 100644 --- a/src/gallium/auxiliary/translate/translate.h +++ b/src/gallium/auxiliary/translate/translate.h @@ -95,6 +95,27 @@ struct translate *translate_lookup_or_create( struct translate_context *tctx, struct translate *translate_create( const struct translate_key *key ); +static INLINE int translate_keysize( const struct translate_key *key ) +{ + return 2 * sizeof(int) + key->nr_elements * sizeof(struct translate_element); +} + +static INLINE int translate_key_compare( const struct translate_key *a, + const struct translate_key *b ) +{ + int keysize = translate_keysize(a); + return memcmp(a, b, keysize); +} + + +static INLINE void translate_key_sanitize( struct translate_key *a ) +{ + int keysize = translate_keysize(a); + char *ptr = (char *)a; + memset(ptr + keysize, 0, sizeof(*a) - keysize); +} + + /******************************************************************************* * Private: */ -- cgit v1.2.3 From 7ddb925b8bc6c18eba953e34d2b630a3a6593f05 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Fri, 9 May 2008 13:10:15 +0100 Subject: draw: mimize cost of translate key compares, use cache universally --- src/gallium/auxiliary/draw/draw_pipe_vbuf.c | 20 ++++++++++++++------ src/gallium/auxiliary/draw/draw_pt_emit.c | 10 +++++----- src/gallium/auxiliary/draw/draw_pt_fetch.c | 7 ++----- src/gallium/auxiliary/draw/draw_pt_fetch_emit.c | 3 ++- 4 files changed, 23 insertions(+), 17 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c index 64a9f9084f..67b9a9503d 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c +++ b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c @@ -42,6 +42,7 @@ #include "draw_vertex.h" #include "draw_pipe.h" #include "translate/translate.h" +#include "translate/translate_cache.h" /** @@ -75,6 +76,8 @@ struct vbuf_stage { /* Cache point size somewhere it's address won't change: */ float point_size; + + struct translate_cache *cache; }; @@ -220,7 +223,6 @@ vbuf_set_prim( struct vbuf_stage *vbuf, uint prim ) /* Translate from pipeline vertices to hw vertices. */ dst_offset = 0; - memset(&hw_key, 0, sizeof(hw_key)); for (i = 0; i < vbuf->vinfo->num_attribs; i++) { unsigned emit_sz = 0; @@ -277,12 +279,10 @@ vbuf_set_prim( struct vbuf_stage *vbuf, uint prim ) /* Don't bother with caching at this stage: */ if (!vbuf->translate || - memcmp(&vbuf->translate->key, &hw_key, sizeof(hw_key)) != 0) + translate_key_compare(&vbuf->translate->key, &hw_key) != 0) { - if (vbuf->translate) - vbuf->translate->release(vbuf->translate); - - vbuf->translate = translate_create( &hw_key ); + translate_key_sanitize(&hw_key); + vbuf->translate = translate_cache_find(vbuf->cache, &hw_key); vbuf->translate->set_buffer(vbuf->translate, 1, &vbuf->point_size, 0); } @@ -430,6 +430,9 @@ static void vbuf_destroy( struct draw_stage *stage ) if (vbuf->render) vbuf->render->destroy( vbuf->render ); + if (vbuf->cache) + translate_cache_destroy(vbuf->cache); + FREE( stage ); } @@ -460,6 +463,11 @@ struct draw_stage *draw_vbuf_stage( struct draw_context *draw, 16 ); if (!vbuf->indices) goto fail; + + vbuf->cache = translate_cache_create(); + if (!vbuf->cache) + goto fail; + vbuf->vertices = NULL; vbuf->vertex_ptr = vbuf->vertices; diff --git a/src/gallium/auxiliary/draw/draw_pt_emit.c b/src/gallium/auxiliary/draw/draw_pt_emit.c index 999b2007a2..4a854f4362 100644 --- a/src/gallium/auxiliary/draw/draw_pt_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_emit.c @@ -49,7 +49,6 @@ void draw_pt_emit_prepare( struct pt_emit *emit, const struct vertex_info *vinfo; unsigned dst_offset; struct translate_key hw_key; - unsigned keysize; unsigned i; boolean ok; @@ -62,7 +61,7 @@ void draw_pt_emit_prepare( struct pt_emit *emit, /* Must do this after set_primitive() above: */ vinfo = draw->render->get_vertex_info(draw->render); - keysize = 2*4 + vinfo->num_attribs * sizeof(hw_key.element[0]); + /* Translate from pipeline vertices to hw vertices. */ @@ -121,9 +120,9 @@ void draw_pt_emit_prepare( struct pt_emit *emit, hw_key.output_stride = vinfo->size * 4; if (!emit->translate || - memcmp(&emit->translate->key, &hw_key, keysize) != 0) + translate_key_compare(&emit->translate->key, &hw_key) != 0) { - memset((char *)&hw_key + keysize, 0, sizeof(hw_key) - keysize); + translate_key_sanitize(&hw_key); emit->translate = translate_cache_find(emit->cache, &hw_key); } } @@ -244,7 +243,8 @@ struct pt_emit *draw_pt_emit_create( struct draw_context *draw ) void draw_pt_emit_destroy( struct pt_emit *emit ) { - translate_cache_destroy(emit->cache); + if (emit->cache) + translate_cache_destroy(emit->cache); FREE(emit); } diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch.c b/src/gallium/auxiliary/draw/draw_pt_fetch.c index 034e0eccb2..07f4c99164 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch.c @@ -62,11 +62,8 @@ void draw_pt_fetch_prepare( struct pt_fetch *fetch, unsigned i, nr = 0; unsigned dst_offset = 0; struct translate_key key; - unsigned keysize; fetch->vertex_size = vertex_size; - keysize = (2*4 + - (draw->pt.nr_vertex_elements + 1) * sizeof(key.element[0])); /* Always emit/leave space for a vertex header. * @@ -111,9 +108,9 @@ void draw_pt_fetch_prepare( struct pt_fetch *fetch, if (!fetch->translate || - memcmp(&fetch->translate->key, &key, keysize) != 0) + translate_key_compare(&fetch->translate->key, &key) != 0) { - memset((char *)&key + keysize, 0, sizeof(key) - keysize); + translate_key_sanitize(&key); fetch->translate = translate_cache_find(fetch->cache, &key); { diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c index bdbb039f9e..a1d041a74f 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c @@ -174,8 +174,9 @@ static void fetch_emit_prepare( struct draw_pt_middle_end *middle, /* Don't bother with caching at this stage: */ if (!feme->translate || - memcmp(&feme->translate->key, &key, sizeof(key)) != 0) + translate_key_compare(&feme->translate->key, &key) != 0) { + translate_key_sanitize(&key); feme->translate = translate_cache_find(feme->cache, &key); -- cgit v1.2.3 From b5e5369da5fc50d63a6ece931fac44b555eb0314 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Mon, 12 May 2008 15:20:38 +0100 Subject: draw: add fetch-shade-emit path Enable with TEST_FSE=t. Performs fetch from API-provided vertex buffers, transformation with one of three (two working) hard-coded shaders, and final emit to hardware vertices all in a single pass. Currently only really useful for profiling in conjunction with SP_NO_RAST=t. --- src/gallium/auxiliary/draw/Makefile | 1 + src/gallium/auxiliary/draw/draw_private.h | 3 + src/gallium/auxiliary/draw/draw_pt.c | 25 +- src/gallium/auxiliary/draw/draw_pt.h | 3 + .../auxiliary/draw/draw_pt_fetch_shade_emit.c | 714 +++++++++++++++++++++ 5 files changed, 742 insertions(+), 4 deletions(-) create mode 100644 src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/Makefile b/src/gallium/auxiliary/draw/Makefile index da7eded21f..68e7744cc5 100644 --- a/src/gallium/auxiliary/draw/Makefile +++ b/src/gallium/auxiliary/draw/Makefile @@ -26,6 +26,7 @@ C_SOURCES = \ draw_pt_emit.c \ draw_pt_fetch.c \ draw_pt_fetch_emit.c \ + draw_pt_fetch_shade_emit.c \ draw_pt_fetch_shade_pipeline.c \ draw_pt_post_vs.c \ draw_pt_varray.c \ diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h index e036d498b8..cbe64cd290 100644 --- a/src/gallium/auxiliary/draw/draw_private.h +++ b/src/gallium/auxiliary/draw/draw_private.h @@ -124,12 +124,14 @@ struct draw_context struct { struct { struct draw_pt_middle_end *fetch_emit; + /*struct draw_pt_middle_end *fetch_shade_emit;*/ struct draw_pt_middle_end *general; } middle; struct { struct draw_pt_front_end *vcache; struct draw_pt_front_end *varray; + struct draw_pt_front_end *fetch_shade_emit; /* temp hack */ } front; struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS]; @@ -154,6 +156,7 @@ struct draw_context const void *constants; } user; + boolean test_fse; } pt; struct { diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c index bccde6c5fd..448deef98c 100644 --- a/src/gallium/auxiliary/draw/draw_pt.c +++ b/src/gallium/auxiliary/draw/draw_pt.c @@ -64,7 +64,7 @@ draw_pt_arrays(struct draw_context *draw, opt |= PT_PIPELINE; } - if (!draw->bypass_clipping) { + if (!draw->bypass_clipping && !draw->pt.test_fse) { opt |= PT_CLIPTEST; } @@ -72,16 +72,20 @@ draw_pt_arrays(struct draw_context *draw, opt |= PT_SHADE; } - if (opt) - middle = draw->pt.middle.general; - else + + if (opt == 0) middle = draw->pt.middle.fetch_emit; + else + middle = draw->pt.middle.general; /* Pick the right frontend */ if (draw->pt.user.elts) { frontend = draw->pt.front.vcache; + } else if (opt == PT_SHADE && draw->pt.test_fse) { + /* should be a middle end.. */ + frontend = draw->pt.front.fetch_shade_emit; } else { frontend = draw->pt.front.varray; } @@ -113,6 +117,14 @@ boolean draw_pt_init( struct draw_context *draw ) if (!draw->pt.middle.fetch_emit) return FALSE; + draw->pt.test_fse = GETENV("DRAW_FSE") != NULL; + if (draw->pt.test_fse) { + draw->pt.front.fetch_shade_emit = draw_pt_fetch_shade_emit( draw ); + if (!draw->pt.front.fetch_shade_emit) + return FALSE; + } + + draw->pt.middle.general = draw_pt_fetch_pipeline_or_emit( draw ); if (!draw->pt.middle.general) return FALSE; @@ -133,6 +145,11 @@ void draw_pt_destroy( struct draw_context *draw ) draw->pt.middle.fetch_emit = NULL; } + if (draw->pt.front.fetch_shade_emit) { + draw->pt.front.fetch_shade_emit->destroy( draw->pt.front.fetch_shade_emit ); + draw->pt.front.fetch_shade_emit = NULL; + } + if (draw->pt.front.vcache) { draw->pt.front.vcache->destroy( draw->pt.front.vcache ); draw->pt.front.vcache = NULL; diff --git a/src/gallium/auxiliary/draw/draw_pt.h b/src/gallium/auxiliary/draw/draw_pt.h index 312fdbe4f4..bcd89f6bd6 100644 --- a/src/gallium/auxiliary/draw/draw_pt.h +++ b/src/gallium/auxiliary/draw/draw_pt.h @@ -121,6 +121,8 @@ const void *draw_pt_elt_ptr( struct draw_context *draw, struct draw_pt_front_end *draw_pt_vcache( struct draw_context *draw ); struct draw_pt_front_end *draw_pt_varray(struct draw_context *draw); +struct draw_pt_front_end *draw_pt_fetch_shade_emit( struct draw_context *draw ); + /* Middle-ends: * * Currently one general-purpose case which can do all possibilities, @@ -132,6 +134,7 @@ struct draw_pt_front_end *draw_pt_varray(struct draw_context *draw); * vertex_elements. */ struct draw_pt_middle_end *draw_pt_fetch_emit( struct draw_context *draw ); +//struct draw_pt_middle_end *draw_pt_fetch_shade_emit( struct draw_context *draw ); struct draw_pt_middle_end *draw_pt_fetch_pipeline_or_emit(struct draw_context *draw); diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c new file mode 100644 index 0000000000..9e1d1add36 --- /dev/null +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c @@ -0,0 +1,714 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + /* + * Authors: + * Keith Whitwell + */ + + +#include "pipe/p_util.h" +#include "draw/draw_context.h" +#include "draw/draw_private.h" +#include "draw/draw_vbuf.h" +#include "draw/draw_vertex.h" +#include "draw/draw_pt.h" +#include "draw/draw_vs.h" + +#include "translate/translate.h" + +struct fetch_shade_emit; + +struct fse_shader { + struct translate_key key; + + void (*run_linear)( const struct fetch_shade_emit *fse, + unsigned start, + unsigned count, + char *buffer ); +}; + +/* Prototype fetch, shade, emit-hw-verts all in one go. + */ +struct fetch_shade_emit { + struct draw_pt_front_end base; + + struct draw_context *draw; + + struct translate_key key; + + /* Temporaries: + */ + const float *constants; + unsigned pitch[PIPE_MAX_ATTRIBS]; + const ubyte *src[PIPE_MAX_ATTRIBS]; + unsigned prim; + + /* Points to one of the three hardwired example shaders, below: + */ + struct fse_shader *active; + + /* Temporary: A list of hard-wired shaders. Of course the plan + * would be to generate these for a given (vertex-shader, + * translate-key) pair... + */ + struct fse_shader shader[10]; + int nr_shaders; +}; + + + +/* Not quite passthrough yet -- we're still running the 'shader' here, + * inlined into the vertex fetch function. + */ +static void fetch_xyz_rgb_st( const struct fetch_shade_emit *fse, + unsigned start, + unsigned count, + char *buffer ) +{ + unsigned i; + + const float *m = fse->constants; + const float m0 = m[0], m4 = m[4], m8 = m[8], m12 = m[12]; + const float m1 = m[1], m5 = m[5], m9 = m[9], m13 = m[13]; + const float m2 = m[2], m6 = m[6], m10 = m[10], m14 = m[14]; + const float m3 = m[3], m7 = m[7], m11 = m[11], m15 = m[15]; + + const ubyte *xyz = fse->src[0] + start * fse->pitch[0]; + const ubyte *st = fse->src[2] + start * fse->pitch[2]; + + float *out = (float *)buffer; + + + assert(fse->pitch[1] == 0); + + /* loop over vertex attributes (vertex shader inputs) + */ + for (i = 0; i < count; i++) { + { + const float *in = (const float *)xyz; + const float ix = in[0], iy = in[1], iz = in[2]; + + out[0] = m0 * ix + m4 * iy + m8 * iz + m12; + out[1] = m1 * ix + m5 * iy + m9 * iz + m13; + out[2] = m2 * ix + m6 * iy + m10 * iz + m14; + out[3] = m3 * ix + m7 * iy + m11 * iz + m15; + xyz += fse->pitch[0]; + } + + { + out[4] = 1.0f; + out[5] = 1.0f; + out[6] = 1.0f; + out[7] = 1.0f; + } + + { + const float *in = (const float *)st; st += fse->pitch[2]; + out[8] = in[0]; + out[9] = in[1]; + out[10] = 0.0f; + out[11] = 1.0f; + } + + out += 12; + } +} + + + +static void fetch_xyz_rgb( const struct fetch_shade_emit *fse, + unsigned start, + unsigned count, + char *buffer ) +{ + unsigned i; + + const float *m = (const float *)fse->constants; + const float m0 = m[0], m4 = m[4], m8 = m[8], m12 = m[12]; + const float m1 = m[1], m5 = m[5], m9 = m[9], m13 = m[13]; + const float m2 = m[2], m6 = m[6], m10 = m[10], m14 = m[14]; + const float m3 = m[3], m7 = m[7], m11 = m[11], m15 = m[15]; + + const ubyte *xyz = fse->src[0] + start * fse->pitch[0]; + const ubyte *rgb = fse->src[1] + start * fse->pitch[1]; + + float *out = (float *)buffer; + +// debug_printf("rgb %f %f %f\n", rgb[0], rgb[1], rgb[2]); + + + for (i = 0; i < count; i++) { + { + const float *in = (const float *)xyz; + const float ix = in[0], iy = in[1], iz = in[2]; + + out[0] = m0 * ix + m4 * iy + m8 * iz + m12; + out[1] = m1 * ix + m5 * iy + m9 * iz + m13; + out[2] = m2 * ix + m6 * iy + m10 * iz + m14; + out[3] = m3 * ix + m7 * iy + m11 * iz + m15; + xyz += fse->pitch[0]; + } + + { + const float *in = (const float *)rgb; + out[4] = in[0]; + out[5] = in[1]; + out[6] = in[2]; + out[7] = 1.0f; + rgb += fse->pitch[1]; + } + + out += 8; + } +} + + + + +static void fetch_xyz_rgb_psiz( const struct fetch_shade_emit *fse, + unsigned start, + unsigned count, + char *buffer ) +{ + unsigned i; + + const float *m = (const float *)fse->constants; + const float m0 = m[0], m4 = m[4], m8 = m[8], m12 = m[12]; + const float m1 = m[1], m5 = m[5], m9 = m[9], m13 = m[13]; + const float m2 = m[2], m6 = m[6], m10 = m[10], m14 = m[14]; + const float m3 = m[3], m7 = m[7], m11 = m[11], m15 = m[15]; + + const ubyte *xyz = fse->src[0] + start * fse->pitch[0]; + const float *rgb = (const float *)(fse->src[1] + start * fse->pitch[1]); + const float psiz = 1.0; + + float *out = (float *)buffer; + + + assert(fse->pitch[1] == 0); + + for (i = 0; i < count; i++) { + { + const float *in = (const float *)xyz; + const float ix = in[0], iy = in[1], iz = in[2]; + + out[0] = m0 * ix + m4 * iy + m8 * iz + m12; + out[1] = m1 * ix + m5 * iy + m9 * iz + m13; + out[2] = m2 * ix + m6 * iy + m10 * iz + m14; + out[3] = m3 * ix + m7 * iy + m11 * iz + m15; + xyz += fse->pitch[0]; + } + + { + out[4] = rgb[0]; + out[5] = rgb[1]; + out[6] = rgb[2]; + out[7] = 1.0f; + } + + { + out[8] = psiz; + } + + out += 9; + } +} + + + + +static boolean set_prim( struct fetch_shade_emit *fse, + unsigned prim, + unsigned count ) +{ + struct draw_context *draw = fse->draw; + + fse->prim = prim; + + switch (prim) { + case PIPE_PRIM_LINE_LOOP: + if (count > 1024) + return FALSE; + draw->render->set_primitive( draw->render, PIPE_PRIM_LINE_STRIP ); + break; + + case PIPE_PRIM_TRIANGLE_FAN: + case PIPE_PRIM_POLYGON: + if (count > 1024) + return FALSE; + draw->render->set_primitive( draw->render, prim ); + break; + + case PIPE_PRIM_QUADS: + case PIPE_PRIM_QUAD_STRIP: + draw->render->set_primitive( draw->render, PIPE_PRIM_TRIANGLES ); + break; + + default: + draw->render->set_primitive( draw->render, prim ); + break; + } + + return TRUE; +} + + + + + + +static void fse_prepare( struct draw_pt_front_end *fe, + unsigned prim, + struct draw_pt_middle_end *unused, + unsigned opt ) +{ + struct fetch_shade_emit *fse = (struct fetch_shade_emit *)fe; + struct draw_context *draw = fse->draw; + unsigned num_vs_inputs = draw->vertex_shader->info.num_inputs; + unsigned num_vs_outputs = draw->vertex_shader->info.num_outputs; + const struct vertex_info *vinfo; + unsigned i; + boolean need_psize = 0; + + + if (draw->pt.user.elts) { + assert(0); + return ; + } + + if (!set_prim(fse, prim, /*count*/1022 )) { + assert(0); + return ; + } + + /* Must do this after set_primitive() above: + */ + vinfo = draw->render->get_vertex_info(draw->render); + + + + fse->key.nr_elements = MAX2(num_vs_outputs, /* outputs - translate to hw format */ + num_vs_inputs); /* inputs - fetch from api format */ + + fse->key.output_stride = vinfo->size * 4; + memset(fse->key.element, 0, + fse->key.nr_elements * sizeof(fse->key.element[0])); + + for (i = 0; i < num_vs_inputs; i++) { + const struct pipe_vertex_element *src = &draw->pt.vertex_element[i]; + fse->key.element[i].input_format = src->src_format; + + /* Consider ignoring these at this point, ie make generated + * programs independent of this state: + */ + fse->key.element[i].input_buffer = 0; //src->vertex_buffer_index; + fse->key.element[i].input_offset = 0; //src->src_offset; + } + + + { + unsigned dst_offset = 0; + + for (i = 0; i < vinfo->num_attribs; i++) { + unsigned emit_sz = 0; + unsigned output_format = PIPE_FORMAT_NONE; + unsigned vs_output = vinfo->src_index[i]; + + switch (vinfo->emit[i]) { + case EMIT_4F: + output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; + emit_sz = 4 * sizeof(float); + break; + case EMIT_3F: + output_format = PIPE_FORMAT_R32G32B32_FLOAT; + emit_sz = 3 * sizeof(float); + break; + case EMIT_2F: + output_format = PIPE_FORMAT_R32G32_FLOAT; + emit_sz = 2 * sizeof(float); + break; + case EMIT_1F: + output_format = PIPE_FORMAT_R32_FLOAT; + emit_sz = 1 * sizeof(float); + break; + case EMIT_1F_PSIZE: + need_psize = 1; + output_format = PIPE_FORMAT_R32_FLOAT; + emit_sz = 1 * sizeof(float); + vs_output = num_vs_outputs + 1; + + break; + default: + assert(0); + break; + } + + /* The elements in the key correspond to vertex shader output + * numbers, not to positions in the hw vertex description -- + * that's handled by the output_offset field. + */ + fse->key.element[vs_output].output_format = output_format; + fse->key.element[vs_output].output_offset = dst_offset; + + dst_offset += emit_sz; + assert(fse->key.output_stride >= dst_offset); + } + } + + /* To make psize work, really need to tell the vertex shader to + * copy that value from input->output. For 'translate' this was + * implicit for all elements. + */ +#if 0 + if (need_psize) { + unsigned input = num_vs_inputs + 1; + const struct pipe_vertex_element *src = &draw->pt.vertex_element[i]; + fse->key.element[i].input_format = PIPE_FORMAT_R32_FLOAT; + fse->key.element[i].input_buffer = 0; //nr_buffers + 1; + fse->key.element[i].input_offset = 0; + + fse->key.nr_elements += 1; + + } +#endif + + fse->constants = draw->pt.user.constants; + + /* Would normally look up a vertex shader and peruse its list of + * varients somehow. We omitted that step and put all the + * hardcoded "shaders" into an array. We're just making the + * assumption that this happens to be a matching shader... ie + * you're running isosurf, aren't you? + */ + fse->active = NULL; + for (i = 0; i < fse->nr_shaders; i++) { + if (translate_key_compare( &fse->key, &fse->shader[i].key) == 0) + fse->active = &fse->shader[i]; + } + + if (!fse->active) { + assert(0); + return ; + } + + /* Now set buffer pointers: + */ + for (i = 0; i < num_vs_inputs; i++) { + unsigned buf = draw->pt.vertex_element[i].vertex_buffer_index; + + fse->src[i] = ((const ubyte *) draw->pt.user.vbuffer[buf] + + draw->pt.vertex_buffer[buf].buffer_offset + + draw->pt.vertex_element[i].src_offset); + + fse->pitch[i] = draw->pt.vertex_buffer[buf].pitch; + + } + + + //return TRUE; +} + + +static boolean split_prim_inplace(unsigned prim, unsigned *first, unsigned *incr) +{ + switch (prim) { + case PIPE_PRIM_POINTS: + *first = 1; + *incr = 1; + return TRUE; + case PIPE_PRIM_LINES: + *first = 2; + *incr = 2; + return TRUE; + case PIPE_PRIM_LINE_STRIP: + *first = 2; + *incr = 1; + return TRUE; + case PIPE_PRIM_TRIANGLES: + *first = 3; + *incr = 3; + return TRUE; + case PIPE_PRIM_TRIANGLE_STRIP: + *first = 3; + *incr = 1; + return TRUE; + case PIPE_PRIM_QUADS: + *first = 4; + *incr = 4; + return TRUE; + case PIPE_PRIM_QUAD_STRIP: + *first = 4; + *incr = 2; + return TRUE; + default: + *first = 0; + *incr = 1; /* set to one so that count % incr works */ + return FALSE; + } +} + + + + +#define INDEX(i) (start + (i)) +static void fse_render_linear( struct vbuf_render *render, + unsigned prim, + unsigned start, + unsigned length ) +{ + ushort *tmp = NULL; + unsigned i, j; + + switch (prim) { + case PIPE_PRIM_LINE_LOOP: + tmp = MALLOC( sizeof(ushort) * (length + 1) ); + + for (i = 0; i < length; i++) + tmp[i] = INDEX(i); + tmp[length] = 0; + + render->draw( render, + tmp, + length+1 ); + break; + + + case PIPE_PRIM_QUAD_STRIP: + tmp = MALLOC( sizeof(ushort) * (length / 2 * 6) ); + + for (j = i = 0; i + 3 < length; i += 2, j += 6) { + tmp[j+0] = INDEX(i+0); + tmp[j+1] = INDEX(i+1); + tmp[j+2] = INDEX(i+3); + + tmp[j+3] = INDEX(i+2); + tmp[j+4] = INDEX(i+0); + tmp[j+5] = INDEX(i+3); + } + + if (j) + render->draw( render, tmp, j ); + break; + + case PIPE_PRIM_QUADS: + tmp = MALLOC( sizeof(int) * (length / 4 * 6) ); + + for (j = i = 0; i + 3 < length; i += 4, j += 6) { + tmp[j+0] = INDEX(i+0); + tmp[j+1] = INDEX(i+1); + tmp[j+2] = INDEX(i+3); + + tmp[j+3] = INDEX(i+1); + tmp[j+4] = INDEX(i+2); + tmp[j+5] = INDEX(i+3); + } + + if (j) + render->draw( render, tmp, j ); + break; + + default: + render->draw_arrays( render, + start, + length ); + break; + } + + if (tmp) + FREE(tmp); +} + + + +static boolean do_draw( struct fetch_shade_emit *fse, + unsigned start, unsigned count ) +{ + struct draw_context *draw = fse->draw; + + char *hw_verts = + draw->render->allocate_vertices( draw->render, + (ushort)fse->key.output_stride, + (ushort)count ); + + if (!hw_verts) + return FALSE; + + /* Single routine to fetch vertices, run shader and emit HW verts. + * Clipping and viewport transformation are done on hardware. + */ + fse->active->run_linear( fse, + start, count, + hw_verts ); + + /* Draw arrays path to avoid re-emitting index list again and + * again. + */ + fse_render_linear( draw->render, + fse->prim, + 0, + count ); + + + draw->render->release_vertices( draw->render, + hw_verts, + fse->key.output_stride, + count ); + + return TRUE; +} + + +static void +fse_run(struct draw_pt_front_end *fe, + pt_elt_func elt_func, + const void *elt_ptr, + unsigned count) +{ + struct fetch_shade_emit *fse = (struct fetch_shade_emit *)fe; + unsigned i = 0; + unsigned first, incr; + unsigned start = elt_func(elt_ptr, 0); + + //debug_printf("%s prim %d start %d count %d\n", __FUNCTION__, prim, start, count); + + split_prim_inplace(fse->prim, &first, &incr); + + count -= (count - first) % incr; + + while (i + first <= count) { + int nr = MIN2( count - i, 1024 ); + + /* snap to prim boundary + */ + nr -= (nr - first) % incr; + + if (!do_draw( fse, start + i, nr )) { + assert(0); + return ; + } + + /* increment allowing for repeated vertices + */ + i += nr - (first - incr); + } + + //return TRUE; +} + + +static void fse_finish( struct draw_pt_front_end *frontend ) +{ +} + + +static void +fse_destroy( struct draw_pt_front_end *frontend ) +{ + FREE(frontend); +} + +struct draw_pt_front_end *draw_pt_fetch_shade_emit( struct draw_context *draw ) +{ + struct fetch_shade_emit *fse = CALLOC_STRUCT(fetch_shade_emit); + if (!fse) + return NULL; + + fse->base.prepare = fse_prepare; + fse->base.run = fse_run; + fse->base.finish = fse_finish; + fse->base.destroy = fse_destroy; + fse->draw = draw; + + fse->shader[0].run_linear = fetch_xyz_rgb_st; + fse->shader[0].key.nr_elements = 3; + fse->shader[0].key.output_stride = 12 * sizeof(float); + + fse->shader[0].key.element[0].input_format = PIPE_FORMAT_R32G32B32_FLOAT; + fse->shader[0].key.element[0].input_buffer = 0; + fse->shader[0].key.element[0].input_offset = 0; + fse->shader[0].key.element[0].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; + fse->shader[0].key.element[0].output_offset = 0; + + fse->shader[0].key.element[1].input_format = PIPE_FORMAT_R32G32B32_FLOAT; + fse->shader[0].key.element[1].input_buffer = 0; + fse->shader[0].key.element[1].input_offset = 0; + fse->shader[0].key.element[1].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; + fse->shader[0].key.element[1].output_offset = 16; + + fse->shader[0].key.element[1].input_format = PIPE_FORMAT_R32G32_FLOAT; + fse->shader[0].key.element[1].input_buffer = 0; + fse->shader[0].key.element[1].input_offset = 0; + fse->shader[0].key.element[1].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; + fse->shader[0].key.element[1].output_offset = 32; + + fse->shader[1].run_linear = fetch_xyz_rgb; + fse->shader[1].key.nr_elements = 2; + fse->shader[1].key.output_stride = 8 * sizeof(float); + + fse->shader[1].key.element[0].input_format = PIPE_FORMAT_R32G32B32_FLOAT; + fse->shader[1].key.element[0].input_buffer = 0; + fse->shader[1].key.element[0].input_offset = 0; + fse->shader[1].key.element[0].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; + fse->shader[1].key.element[0].output_offset = 0; + + fse->shader[1].key.element[1].input_format = PIPE_FORMAT_R32G32B32_FLOAT; + fse->shader[1].key.element[1].input_buffer = 0; + fse->shader[1].key.element[1].input_offset = 0; + fse->shader[1].key.element[1].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; + fse->shader[1].key.element[1].output_offset = 16; + + fse->shader[2].run_linear = fetch_xyz_rgb_psiz; + fse->shader[2].key.nr_elements = 3; + fse->shader[2].key.output_stride = 9 * sizeof(float); + + fse->shader[2].key.element[0].input_format = PIPE_FORMAT_R32G32B32_FLOAT; + fse->shader[2].key.element[0].input_buffer = 0; + fse->shader[2].key.element[0].input_offset = 0; + fse->shader[2].key.element[0].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; + fse->shader[2].key.element[0].output_offset = 0; + + fse->shader[2].key.element[1].input_format = PIPE_FORMAT_R32G32B32_FLOAT; + fse->shader[2].key.element[1].input_buffer = 0; + fse->shader[2].key.element[1].input_offset = 0; + fse->shader[2].key.element[1].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; + fse->shader[2].key.element[1].output_offset = 16; + + /* psize is special + * -- effectively add it here as another input!?! + * -- who knows how to add it as a buffer? + */ + fse->shader[2].key.element[2].input_format = PIPE_FORMAT_R32_FLOAT; + fse->shader[2].key.element[2].input_buffer = 0; + fse->shader[2].key.element[2].input_offset = 0; + fse->shader[2].key.element[2].output_format = PIPE_FORMAT_R32_FLOAT; + fse->shader[2].key.element[2].output_offset = 32; + + fse->nr_shaders = 3; + + return &fse->base; +} -- cgit v1.2.3 From 44463b2997826cd14def00abf724a7a65a4fc7cb Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Mon, 12 May 2008 17:36:35 +0100 Subject: draw: streamline the varray path - drop support for running the pipeline (ie. don't populate the flags values) - pass through all split-able primitives intact to the middle end - only primitives that can't be split are shunted on the draw-element path --- src/gallium/auxiliary/draw/draw_pt.c | 2 +- src/gallium/auxiliary/draw/draw_pt_varray.c | 106 +++---------- .../auxiliary/draw/draw_pt_varray_tmp_linear.h | 165 +++------------------ 3 files changed, 44 insertions(+), 229 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c index 448deef98c..d9e73a2396 100644 --- a/src/gallium/auxiliary/draw/draw_pt.c +++ b/src/gallium/auxiliary/draw/draw_pt.c @@ -81,7 +81,7 @@ draw_pt_arrays(struct draw_context *draw, /* Pick the right frontend */ - if (draw->pt.user.elts) { + if (draw->pt.user.elts || (opt & PT_PIPELINE)) { frontend = draw->pt.front.vcache; } else if (opt == PT_SHADE && draw->pt.test_fse) { /* should be a middle end.. */ diff --git a/src/gallium/auxiliary/draw/draw_pt_varray.c b/src/gallium/auxiliary/draw/draw_pt_varray.c index 59a9569270..d92ad4fda1 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray.c +++ b/src/gallium/auxiliary/draw/draw_pt_varray.c @@ -154,9 +154,9 @@ static boolean split_prim_inplace(unsigned prim, unsigned *first, unsigned *incr static INLINE void add_draw_el(struct varray_frontend *varray, - int idx, ushort flags) + int idx) { - varray->draw_elts[varray->draw_count++] = idx | flags; + varray->draw_elts[varray->draw_count++] = idx; } @@ -165,106 +165,47 @@ static INLINE void varray_triangle( struct varray_frontend *varray, unsigned i1, unsigned i2 ) { - add_draw_el(varray, i0, 0); - add_draw_el(varray, i1, 0); - add_draw_el(varray, i2, 0); -} - -static INLINE void varray_triangle_flags( struct varray_frontend *varray, - ushort flags, - unsigned i0, - unsigned i1, - unsigned i2 ) -{ - add_draw_el(varray, i0, flags); - add_draw_el(varray, i1, 0); - add_draw_el(varray, i2, 0); + add_draw_el(varray, i0); + add_draw_el(varray, i1); + add_draw_el(varray, i2); } static INLINE void varray_line( struct varray_frontend *varray, unsigned i0, unsigned i1 ) { - add_draw_el(varray, i0, 0); - add_draw_el(varray, i1, 0); -} - - -static INLINE void varray_line_flags( struct varray_frontend *varray, - ushort flags, - unsigned i0, - unsigned i1 ) -{ - add_draw_el(varray, i0, flags); - add_draw_el(varray, i1, 0); + add_draw_el(varray, i0); + add_draw_el(varray, i1); } static INLINE void varray_point( struct varray_frontend *varray, unsigned i0 ) { - add_draw_el(varray, i0, 0); -} - -static INLINE void varray_quad( struct varray_frontend *varray, - unsigned i0, - unsigned i1, - unsigned i2, - unsigned i3 ) -{ - varray_triangle( varray, i0, i1, i3 ); - varray_triangle( varray, i1, i2, i3 ); + add_draw_el(varray, i0); } -static INLINE void varray_ef_quad( struct varray_frontend *varray, - unsigned i0, - unsigned i1, - unsigned i2, - unsigned i3 ) -{ - const unsigned omitEdge1 = DRAW_PIPE_EDGE_FLAG_0 | DRAW_PIPE_EDGE_FLAG_2; - const unsigned omitEdge2 = DRAW_PIPE_EDGE_FLAG_0 | DRAW_PIPE_EDGE_FLAG_1; - - varray_triangle_flags( varray, - DRAW_PIPE_RESET_STIPPLE | omitEdge1, - i0, i1, i3 ); - - varray_triangle_flags( varray, - omitEdge2, - i1, i2, i3 ); -} -/* At least for now, we're back to using a template include file for - * this. The two paths aren't too different though - it may be - * possible to reunify them. - */ -#define TRIANGLE(vc,flags,i0,i1,i2) varray_triangle_flags(vc,flags,i0,i1,i2) -#define QUAD(vc,i0,i1,i2,i3) varray_ef_quad(vc,i0,i1,i2,i3) -#define LINE(vc,flags,i0,i1) varray_line_flags(vc,flags,i0,i1) -#define POINT(vc,i0) varray_point(vc,i0) -#define FUNC varray_run_extras -#include "draw_pt_varray_tmp.h" -#define TRIANGLE(vc,flags,i0,i1,i2) varray_triangle(vc,i0,i1,i2) -#define QUAD(vc,i0,i1,i2,i3) varray_quad(vc,i0,i1,i2,i3) -#define LINE(vc,flags,i0,i1) varray_line(vc,i0,i1) +#define TRIANGLE(vc,i0,i1,i2) varray_triangle(vc,i0,i1,i2) +#define LINE(vc,i0,i1) varray_line(vc,i0,i1) #define POINT(vc,i0) varray_point(vc,i0) #define FUNC varray_run #include "draw_pt_varray_tmp_linear.h" -static unsigned reduced_prim[PIPE_PRIM_POLYGON + 1] = { +static unsigned decompose_prim[PIPE_PRIM_POLYGON + 1] = { PIPE_PRIM_POINTS, PIPE_PRIM_LINES, - PIPE_PRIM_LINES, - PIPE_PRIM_LINES, + PIPE_PRIM_LINE_STRIP, + PIPE_PRIM_LINES, /* decomposed */ PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES + PIPE_PRIM_TRIANGLE_STRIP, + PIPE_PRIM_TRIANGLES, /* decomposed */ + PIPE_PRIM_QUADS, + PIPE_PRIM_QUAD_STRIP, + PIPE_PRIM_TRIANGLES /* decomposed */ }; @@ -276,17 +217,10 @@ static void varray_prepare(struct draw_pt_front_end *frontend, { struct varray_frontend *varray = (struct varray_frontend *)frontend; - if (opt & PT_PIPELINE) - { - varray->base.run = varray_run_extras; - } - else - { - varray->base.run = varray_run; - } + varray->base.run = varray_run; varray->input_prim = prim; - varray->output_prim = reduced_prim[prim]; + varray->output_prim = decompose_prim[prim]; varray->middle = middle; middle->prepare(middle, varray->output_prim, opt); diff --git a/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h b/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h index 6e2b16d9be..b6f1f0cadc 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h +++ b/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h @@ -1,3 +1,7 @@ +static unsigned trim( unsigned count, unsigned first, unsigned incr ) +{ + return count - (count - first) % incr; +} static void FUNC(struct draw_pt_front_end *frontend, pt_elt_func get_elt, @@ -5,12 +9,9 @@ static void FUNC(struct draw_pt_front_end *frontend, unsigned count) { struct varray_frontend *varray = (struct varray_frontend *)frontend; - struct draw_context *draw = varray->draw; unsigned start = (unsigned)elts; - boolean flatfirst = (draw->rasterizer->flatshade && - draw->rasterizer->flatshade_first); - unsigned i, j, flags; + unsigned i, j; unsigned first, incr; varray->fetch_start = start; @@ -27,26 +28,30 @@ static void FUNC(struct draw_pt_front_end *frontend, case PIPE_PRIM_POINTS: case PIPE_PRIM_LINES: case PIPE_PRIM_TRIANGLES: - j = 0; - while (j + first <= count) { - unsigned end = MIN2(FETCH_MAX, count - j); - end -= (end % incr); - varray_flush_linear(varray, start + j, end); - j += end; + case PIPE_PRIM_LINE_STRIP: + case PIPE_PRIM_TRIANGLE_STRIP: + case PIPE_PRIM_QUADS: + case PIPE_PRIM_QUAD_STRIP: + + for (j = 0; j < count;) { + unsigned remaining = count - j; + unsigned nr = trim( MIN2(FETCH_MAX, remaining), first, incr ); + varray_flush_linear(varray, start + j, nr); + j += nr; + if (nr != remaining) + j -= (first - incr); } break; case PIPE_PRIM_LINE_LOOP: if (count >= 2) { - flags = DRAW_PIPE_RESET_STIPPLE; - for (j = 0; j + first <= count; j += i) { unsigned end = MIN2(FETCH_MAX, count - j); end -= (end % incr); - for (i = 1; i < end; i++, flags = 0) { - LINE(varray, flags, i - 1, i); + for (i = 1; i < end; i++) { + LINE(varray, i - 1, i); } - LINE(varray, flags, i - 1, 0); + LINE(varray, i - 1, 0); i = end; fetch_init(varray, end); varray_flush(varray); @@ -54,145 +59,21 @@ static void FUNC(struct draw_pt_front_end *frontend, } break; - case PIPE_PRIM_LINE_STRIP: - flags = DRAW_PIPE_RESET_STIPPLE; - for (j = 0; j + first <= count; j += i) { - unsigned end = MIN2(FETCH_MAX, count - j); - end -= (end % incr); - for (i = 1; i < end; i++, flags = 0) { - LINE(varray, flags, i - 1, i); - } - i = end; - fetch_init(varray, end); - varray_flush(varray); - } - break; - - case PIPE_PRIM_TRIANGLE_STRIP: - if (flatfirst) { - for (j = 0; j + first <= count; j += i) { - unsigned end = MIN2(FETCH_MAX, count - j); - end -= (end % incr); - for (i = 0; i+2 < end; i++) { - TRIANGLE(varray, DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, - i + 0, i + 1 + (i&1), i + 2 - (i&1)); - } - i = end; - fetch_init(varray, end); - varray_flush(varray); - if (j + first + i <= count) { - varray->fetch_start -= 2; - i -= 2; - } - } - } - else { - for (j = 0; j + first <= count; j += i) { - unsigned end = MIN2(FETCH_MAX, count - j); - end -= (end % incr); - for (i = 0; i+2 < end; i++) { - TRIANGLE(varray, DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, - i + 0 + (i&1), i + 1 - (i&1), i + 2); - } - i = end; - fetch_init(varray, end); - varray_flush(varray); - if (j + first + i <= count) { - varray->fetch_start -= 2; - i -= 2; - } - } - } - break; + case PIPE_PRIM_POLYGON: case PIPE_PRIM_TRIANGLE_FAN: - if (count >= 3) { - if (flatfirst) { - flags = DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL; - for (j = 0; j + first <= count; j += i) { - unsigned end = MIN2(FETCH_MAX, count - j); - end -= (end % incr); - for (i = 0; i+2 < end; i++) { - TRIANGLE(varray, flags, i + 1, i + 2, 0); - } - i = end; - fetch_init(varray, end); - varray_flush(varray); - } - } - else { - flags = DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL; - for (j = 0; j + first <= count; j += i) { - unsigned end = MIN2(FETCH_MAX, count - j); - end -= (end % incr); - for (i = 0; i+2 < end; i++) { - TRIANGLE(varray, flags, 0, i + 1, i + 2); - } - i = end; - fetch_init(varray, end); - varray_flush(varray); - } - } - } - break; - - case PIPE_PRIM_QUADS: - for (j = 0; j + first <= count; j += i) { - unsigned end = MIN2(FETCH_MAX, count - j); - end -= (end % incr); - for (i = 0; i+3 < end; i += 4) { - QUAD(varray, i + 0, i + 1, i + 2, i + 3); - } - i = end; - fetch_init(varray, end); - varray_flush(varray); - } - break; - - case PIPE_PRIM_QUAD_STRIP: for (j = 0; j + first <= count; j += i) { unsigned end = MIN2(FETCH_MAX, count - j); end -= (end % incr); - for (i = 0; i+3 < end; i += 2) { - QUAD(varray, i + 2, i + 0, i + 1, i + 3); + for (i = 2; i < end; i++) { + TRIANGLE(varray, 0, i - 1, i); } i = end; fetch_init(varray, end); varray_flush(varray); - if (j + first + i <= count) { - varray->fetch_start -= 2; - i -= 2; - } } break; - case PIPE_PRIM_POLYGON: - { - /* These bitflags look a little odd because we submit the - * vertices as (1,2,0) to satisfy flatshade requirements. - */ - const unsigned edge_first = DRAW_PIPE_EDGE_FLAG_2; - const unsigned edge_middle = DRAW_PIPE_EDGE_FLAG_0; - const unsigned edge_last = DRAW_PIPE_EDGE_FLAG_1; - - flags = DRAW_PIPE_RESET_STIPPLE | edge_first | edge_middle; - for (j = 0; j + first <= count; j += i) { - unsigned end = MIN2(FETCH_MAX, count - j); - end -= (end % incr); - for (i = 0; i+2 < end; i++, flags = edge_middle) { - - if (i + 3 == count) - flags |= edge_last; - - TRIANGLE(varray, flags, i + 1, i + 2, 0); - } - i = end; - fetch_init(varray, end); - varray_flush(varray); - } - } - break; - default: assert(0); break; -- cgit v1.2.3 From bbda45ec769120324f44febf00c6bb170f594f23 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Mon, 12 May 2008 19:40:20 +0100 Subject: draw: turn fse path into a middle end Also add some util functions in pt_util.c --- src/gallium/auxiliary/draw/Makefile | 2 + src/gallium/auxiliary/draw/draw_private.h | 2 +- src/gallium/auxiliary/draw/draw_pt.c | 34 +- src/gallium/auxiliary/draw/draw_pt.h | 9 +- .../auxiliary/draw/draw_pt_fetch_shade_emit.c | 39 +- src/gallium/auxiliary/draw/draw_pt_middle_fse.c | 705 +++++++++++++++++++++ src/gallium/auxiliary/draw/draw_pt_util.c | 103 +++ src/gallium/auxiliary/draw/draw_pt_varray.c | 45 -- src/gallium/auxiliary/draw/draw_pt_varray_tmp.h | 2 +- .../auxiliary/draw/draw_pt_varray_tmp_linear.h | 9 +- src/gallium/auxiliary/draw/draw_pt_vcache.c | 15 +- 11 files changed, 847 insertions(+), 118 deletions(-) create mode 100644 src/gallium/auxiliary/draw/draw_pt_middle_fse.c create mode 100644 src/gallium/auxiliary/draw/draw_pt_util.c (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/Makefile b/src/gallium/auxiliary/draw/Makefile index 68e7744cc5..67d78bdbbd 100644 --- a/src/gallium/auxiliary/draw/Makefile +++ b/src/gallium/auxiliary/draw/Makefile @@ -27,8 +27,10 @@ C_SOURCES = \ draw_pt_fetch.c \ draw_pt_fetch_emit.c \ draw_pt_fetch_shade_emit.c \ + draw_pt_middle_fse.c \ draw_pt_fetch_shade_pipeline.c \ draw_pt_post_vs.c \ + draw_pt_util.c \ draw_pt_varray.c \ draw_pt_vcache.c \ draw_vertex.c \ diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h index cbe64cd290..86b901a3c8 100644 --- a/src/gallium/auxiliary/draw/draw_private.h +++ b/src/gallium/auxiliary/draw/draw_private.h @@ -124,7 +124,7 @@ struct draw_context struct { struct { struct draw_pt_middle_end *fetch_emit; - /*struct draw_pt_middle_end *fetch_shade_emit;*/ + struct draw_pt_middle_end *fetch_shade_emit; struct draw_pt_middle_end *general; } middle; diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c index d9e73a2396..91e35db819 100644 --- a/src/gallium/auxiliary/draw/draw_pt.c +++ b/src/gallium/auxiliary/draw/draw_pt.c @@ -75,6 +75,8 @@ draw_pt_arrays(struct draw_context *draw, if (opt == 0) middle = draw->pt.middle.fetch_emit; + else if (opt == PT_SHADE && draw->pt.test_fse) + middle = draw->pt.middle.fetch_shade_emit; else middle = draw->pt.middle.general; @@ -83,9 +85,11 @@ draw_pt_arrays(struct draw_context *draw, */ if (draw->pt.user.elts || (opt & PT_PIPELINE)) { frontend = draw->pt.front.vcache; +#if 0 } else if (opt == PT_SHADE && draw->pt.test_fse) { /* should be a middle end.. */ frontend = draw->pt.front.fetch_shade_emit; +#endif } else { frontend = draw->pt.front.varray; } @@ -105,6 +109,8 @@ draw_pt_arrays(struct draw_context *draw, boolean draw_pt_init( struct draw_context *draw ) { + draw->pt.test_fse = GETENV("DRAW_FSE") != NULL; + draw->pt.front.vcache = draw_pt_vcache( draw ); if (!draw->pt.front.vcache) return FALSE; @@ -117,8 +123,11 @@ boolean draw_pt_init( struct draw_context *draw ) if (!draw->pt.middle.fetch_emit) return FALSE; - draw->pt.test_fse = GETENV("DRAW_FSE") != NULL; if (draw->pt.test_fse) { + draw->pt.middle.fetch_shade_emit = draw_pt_middle_fse( draw ); + if (!draw->pt.middle.fetch_shade_emit) + return FALSE; + draw->pt.front.fetch_shade_emit = draw_pt_fetch_shade_emit( draw ); if (!draw->pt.front.fetch_shade_emit) return FALSE; @@ -145,6 +154,11 @@ void draw_pt_destroy( struct draw_context *draw ) draw->pt.middle.fetch_emit = NULL; } + if (draw->pt.middle.fetch_shade_emit) { + draw->pt.middle.fetch_shade_emit->destroy( draw->pt.middle.fetch_shade_emit ); + draw->pt.middle.fetch_shade_emit = NULL; + } + if (draw->pt.front.fetch_shade_emit) { draw->pt.front.fetch_shade_emit->destroy( draw->pt.front.fetch_shade_emit ); draw->pt.front.fetch_shade_emit = NULL; @@ -163,19 +177,6 @@ void draw_pt_destroy( struct draw_context *draw ) -static unsigned reduced_prim[PIPE_PRIM_POLYGON + 1] = { - PIPE_PRIM_POINTS, - PIPE_PRIM_LINES, - PIPE_PRIM_LINES, - PIPE_PRIM_LINES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES -}; - /** * Draw vertex arrays @@ -188,9 +189,10 @@ void draw_arrays(struct draw_context *draw, unsigned prim, unsigned start, unsigned count) { - if (reduced_prim[prim] != draw->reduced_prim) { + unsigned reduced_prim = draw_pt_reduced_prim(prim); + if (reduced_prim != draw->reduced_prim) { draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE ); - draw->reduced_prim = reduced_prim[prim]; + draw->reduced_prim = reduced_prim; } /* drawing done here: */ diff --git a/src/gallium/auxiliary/draw/draw_pt.h b/src/gallium/auxiliary/draw/draw_pt.h index bcd89f6bd6..cdae46b8d2 100644 --- a/src/gallium/auxiliary/draw/draw_pt.h +++ b/src/gallium/auxiliary/draw/draw_pt.h @@ -134,7 +134,7 @@ struct draw_pt_front_end *draw_pt_fetch_shade_emit( struct draw_context *draw ); * vertex_elements. */ struct draw_pt_middle_end *draw_pt_fetch_emit( struct draw_context *draw ); -//struct draw_pt_middle_end *draw_pt_fetch_shade_emit( struct draw_context *draw ); +struct draw_pt_middle_end *draw_pt_middle_fse( struct draw_context *draw ); struct draw_pt_middle_end *draw_pt_fetch_pipeline_or_emit(struct draw_context *draw); @@ -213,4 +213,11 @@ struct pt_post_vs *draw_pt_post_vs_create( struct draw_context *draw ); void draw_pt_post_vs_destroy( struct pt_post_vs *pvs ); +/******************************************************************************* + * Utils: + */ +void draw_pt_split_prim(unsigned prim, unsigned *first, unsigned *incr); +unsigned draw_pt_reduced_prim(unsigned prim); + + #endif diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c index 9e1d1add36..f756d3e0bb 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c @@ -434,43 +434,6 @@ static void fse_prepare( struct draw_pt_front_end *fe, } -static boolean split_prim_inplace(unsigned prim, unsigned *first, unsigned *incr) -{ - switch (prim) { - case PIPE_PRIM_POINTS: - *first = 1; - *incr = 1; - return TRUE; - case PIPE_PRIM_LINES: - *first = 2; - *incr = 2; - return TRUE; - case PIPE_PRIM_LINE_STRIP: - *first = 2; - *incr = 1; - return TRUE; - case PIPE_PRIM_TRIANGLES: - *first = 3; - *incr = 3; - return TRUE; - case PIPE_PRIM_TRIANGLE_STRIP: - *first = 3; - *incr = 1; - return TRUE; - case PIPE_PRIM_QUADS: - *first = 4; - *incr = 4; - return TRUE; - case PIPE_PRIM_QUAD_STRIP: - *first = 4; - *incr = 2; - return TRUE; - default: - *first = 0; - *incr = 1; /* set to one so that count % incr works */ - return FALSE; - } -} @@ -596,7 +559,7 @@ fse_run(struct draw_pt_front_end *fe, //debug_printf("%s prim %d start %d count %d\n", __FUNCTION__, prim, start, count); - split_prim_inplace(fse->prim, &first, &incr); + draw_pt_split_prim(fse->prim, &first, &incr); count -= (count - first) % incr; diff --git a/src/gallium/auxiliary/draw/draw_pt_middle_fse.c b/src/gallium/auxiliary/draw/draw_pt_middle_fse.c new file mode 100644 index 0000000000..cdb7d260da --- /dev/null +++ b/src/gallium/auxiliary/draw/draw_pt_middle_fse.c @@ -0,0 +1,705 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + /* + * Authors: + * Keith Whitwell + */ + + +#include "pipe/p_util.h" +#include "draw/draw_context.h" +#include "draw/draw_private.h" +#include "draw/draw_vbuf.h" +#include "draw/draw_vertex.h" +#include "draw/draw_pt.h" +#include "draw/draw_vs.h" + +#include "translate/translate.h" + +struct fetch_shade_emit; + +struct fse_shader { + struct translate_key key; + + void (*run_linear)( const struct fetch_shade_emit *fse, + unsigned start, + unsigned count, + char *buffer ); + + void (*run_elts)( const struct fetch_shade_emit *fse, + const unsigned *fetch_elts, + unsigned fetch_count, + char *buffer ); + +}; + +/* Prototype fetch, shade, emit-hw-verts all in one go. + */ +struct fetch_shade_emit { + struct draw_pt_middle_end base; + struct draw_context *draw; + + struct translate_key key; + + /* Temporaries: + */ + const float *constants; + unsigned pitch[PIPE_MAX_ATTRIBS]; + const ubyte *src[PIPE_MAX_ATTRIBS]; + unsigned prim; + + /* Points to one of the three hardwired example shaders, below: + */ + struct fse_shader *active; + + /* Temporary: A list of hard-wired shaders. Of course the plan + * would be to generate these for a given (vertex-shader, + * translate-key) pair... + */ + struct fse_shader shader[10]; + int nr_shaders; +}; + + + +/* Not quite passthrough yet -- we're still running the 'shader' here, + * inlined into the vertex fetch function. + */ +static void shader0_run_linear( const struct fetch_shade_emit *fse, + unsigned start, + unsigned count, + char *buffer ) +{ + unsigned i; + + const float *m = fse->constants; + const ubyte *xyz = fse->src[0] + start * fse->pitch[0]; + const ubyte *rgb = fse->src[1] + start * fse->pitch[1]; + const ubyte *st = fse->src[2] + start * fse->pitch[2]; + + float *out = (float *)buffer; + + /* loop over vertex attributes (vertex shader inputs) + */ + for (i = 0; i < count; i++) { + { + const float *in = (const float *)xyz; + const float ix = in[0], iy = in[1], iz = in[2]; + + out[0] = m[0] * ix + m[4] * iy + m[8] * iz + m[12]; + out[1] = m[1] * ix + m[5] * iy + m[9] * iz + m[13]; + out[2] = m[2] * ix + m[6] * iy + m[10] * iz + m[14]; + out[3] = m[3] * ix + m[7] * iy + m[11] * iz + m[15]; + xyz += fse->pitch[0]; + } + + { + const float *in = (const float *)rgb; + out[4] = in[0]; + out[5] = in[1]; + out[6] = in[2]; + out[7] = 1.0f; + rgb += fse->pitch[1]; + } + + { + const float *in = (const float *)st; + out[8] = in[0]; + out[9] = in[1]; + out[10] = 0.0f; + out[11] = 1.0f; + st += fse->pitch[2]; + } + + out += 12; + } +} + + + +static void shader1_run_linear( const struct fetch_shade_emit *fse, + unsigned start, + unsigned count, + char *buffer ) +{ + unsigned i; + const float *m = (const float *)fse->constants; + const ubyte *xyz = fse->src[0] + start * fse->pitch[0]; + const ubyte *rgb = fse->src[1] + start * fse->pitch[1]; + float *out = (float *)buffer; + +// debug_printf("rgb %f %f %f\n", rgb[0], rgb[1], rgb[2]); + + + for (i = 0; i < count; i++) { + { + const float *in = (const float *)xyz; + const float ix = in[0], iy = in[1], iz = in[2]; + + out[0] = m[0] * ix + m[4] * iy + m[8] * iz + m[12]; + out[1] = m[1] * ix + m[5] * iy + m[9] * iz + m[13]; + out[2] = m[2] * ix + m[6] * iy + m[10] * iz + m[14]; + out[3] = m[3] * ix + m[7] * iy + m[11] * iz + m[15]; + xyz += fse->pitch[0]; + } + + { + const float *in = (const float *)rgb; + out[4] = in[0]; + out[5] = in[1]; + out[6] = in[2]; + out[7] = 1.0f; + rgb += fse->pitch[1]; + } + + out += 8; + } +} + + + + +static void shader2_run_linear( const struct fetch_shade_emit *fse, + unsigned start, + unsigned count, + char *buffer ) +{ + unsigned i; + const float *m = (const float *)fse->constants; + const ubyte *xyz = fse->src[0] + start * fse->pitch[0]; + const ubyte *rgb = fse->src[1] + start * fse->pitch[1]; + const float psiz = 1.0; + float *out = (float *)buffer; + + + assert(fse->pitch[1] == 0); + + for (i = 0; i < count; i++) { + { + const float *in = (const float *)xyz; + const float ix = in[0], iy = in[1], iz = in[2]; + + out[0] = m[0] * ix + m[4] * iy + m[8] * iz + m[12]; + out[1] = m[1] * ix + m[5] * iy + m[9] * iz + m[13]; + out[2] = m[2] * ix + m[6] * iy + m[10] * iz + m[14]; + out[3] = m[3] * ix + m[7] * iy + m[11] * iz + m[15]; + xyz += fse->pitch[0]; + } + + { + const float *in = (const float *)rgb; + out[4] = in[0]; + out[5] = in[1]; + out[6] = in[2]; + out[7] = 1.0f; + rgb += fse->pitch[1]; + } + + { + out[8] = psiz; + } + + out += 9; + } +} + + + + +static void shader0_run_elts( const struct fetch_shade_emit *fse, + const unsigned *elts, + unsigned count, + char *buffer ) +{ + unsigned i; + const float *m = fse->constants; + float *out = (float *)buffer; + + + /* loop over vertex attributes (vertex shader inputs) + */ + for (i = 0; i < count; i++) { + unsigned elt = elts[i]; + { + const ubyte *xyz = fse->src[0] + elt * fse->pitch[0]; + const float *in = (const float *)xyz; + const float ix = in[0], iy = in[1], iz = in[2]; + + out[0] = m[0] * ix + m[4] * iy + m[8] * iz + m[12]; + out[1] = m[1] * ix + m[5] * iy + m[9] * iz + m[13]; + out[2] = m[2] * ix + m[6] * iy + m[10] * iz + m[14]; + out[3] = m[3] * ix + m[7] * iy + m[11] * iz + m[15]; + } + + { + const ubyte *rgb = fse->src[1] + elt * fse->pitch[1]; + const float *in = (const float *)rgb; + out[4] = in[0]; + out[5] = in[1]; + out[6] = in[2]; + out[7] = 1.0f; + } + + { + const ubyte *st = fse->src[2] + elt * fse->pitch[2]; + const float *in = (const float *)st; + out[8] = in[0]; + out[9] = in[1]; + out[10] = 0.0f; + out[11] = 1.0f; + } + + out += 12; + } +} + + + +static void shader1_run_elts( const struct fetch_shade_emit *fse, + const unsigned *elts, + unsigned count, + char *buffer ) +{ + unsigned i; + const float *m = (const float *)fse->constants; + float *out = (float *)buffer; + + for (i = 0; i < count; i++) { + unsigned elt = elts[i]; + + { + const ubyte *xyz = fse->src[0] + elt * fse->pitch[0]; + const float *in = (const float *)xyz; + const float ix = in[0], iy = in[1], iz = in[2]; + + out[0] = m[0] * ix + m[4] * iy + m[8] * iz + m[12]; + out[1] = m[1] * ix + m[5] * iy + m[9] * iz + m[13]; + out[2] = m[2] * ix + m[6] * iy + m[10] * iz + m[14]; + out[3] = m[3] * ix + m[7] * iy + m[11] * iz + m[15]; + xyz += fse->pitch[0]; + } + + { + const ubyte *rgb = fse->src[1] + elt * fse->pitch[1]; + const float *in = (const float *)rgb; + out[4] = in[0]; + out[5] = in[1]; + out[6] = in[2]; + out[7] = 1.0f; + rgb += fse->pitch[1]; + } + + out += 8; + } +} + + + + +static void shader2_run_elts( const struct fetch_shade_emit *fse, + const unsigned *elts, + unsigned count, + char *buffer ) +{ + unsigned i; + const float *m = (const float *)fse->constants; + const float psiz = 1.0; + float *out = (float *)buffer; + + for (i = 0; i < count; i++) { + unsigned elt = elts[i]; + { + const ubyte *xyz = fse->src[0] + elt * fse->pitch[0]; + const float *in = (const float *)xyz; + const float ix = in[0], iy = in[1], iz = in[2]; + + out[0] = m[0] * ix + m[4] * iy + m[8] * iz + m[12]; + out[1] = m[1] * ix + m[5] * iy + m[9] * iz + m[13]; + out[2] = m[2] * ix + m[6] * iy + m[10] * iz + m[14]; + out[3] = m[3] * ix + m[7] * iy + m[11] * iz + m[15]; + } + + { + const ubyte *rgb = fse->src[1] + elt * fse->pitch[1]; + out[4] = rgb[0]; + out[5] = rgb[1]; + out[6] = rgb[2]; + out[7] = 1.0f; + } + + { + out[8] = psiz; + } + + out += 9; + } +} + + + +static void fse_prepare( struct draw_pt_middle_end *middle, + unsigned prim, + unsigned opt ) +{ + struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle; + struct draw_context *draw = fse->draw; + unsigned num_vs_inputs = draw->vertex_shader->info.num_inputs; + unsigned num_vs_outputs = draw->vertex_shader->info.num_outputs; + const struct vertex_info *vinfo; + unsigned i; + boolean need_psize = 0; + + + if (draw->pt.user.elts) { + assert(0); + return ; + } + + if (!draw->render->set_primitive( draw->render, + prim )) { + assert(0); + return; + } + + /* Must do this after set_primitive() above: + */ + vinfo = draw->render->get_vertex_info(draw->render); + + + + fse->key.nr_elements = MAX2(num_vs_outputs, /* outputs - translate to hw format */ + num_vs_inputs); /* inputs - fetch from api format */ + + fse->key.output_stride = vinfo->size * 4; + memset(fse->key.element, 0, + fse->key.nr_elements * sizeof(fse->key.element[0])); + + for (i = 0; i < num_vs_inputs; i++) { + const struct pipe_vertex_element *src = &draw->pt.vertex_element[i]; + fse->key.element[i].input_format = src->src_format; + + /* Consider ignoring these at this point, ie make generated + * programs independent of this state: + */ + fse->key.element[i].input_buffer = 0; //src->vertex_buffer_index; + fse->key.element[i].input_offset = 0; //src->src_offset; + } + + + { + unsigned dst_offset = 0; + + for (i = 0; i < vinfo->num_attribs; i++) { + unsigned emit_sz = 0; + unsigned output_format = PIPE_FORMAT_NONE; + unsigned vs_output = vinfo->src_index[i]; + + switch (vinfo->emit[i]) { + case EMIT_4F: + output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; + emit_sz = 4 * sizeof(float); + break; + case EMIT_3F: + output_format = PIPE_FORMAT_R32G32B32_FLOAT; + emit_sz = 3 * sizeof(float); + break; + case EMIT_2F: + output_format = PIPE_FORMAT_R32G32_FLOAT; + emit_sz = 2 * sizeof(float); + break; + case EMIT_1F: + output_format = PIPE_FORMAT_R32_FLOAT; + emit_sz = 1 * sizeof(float); + break; + case EMIT_1F_PSIZE: + need_psize = 1; + output_format = PIPE_FORMAT_R32_FLOAT; + emit_sz = 1 * sizeof(float); + vs_output = num_vs_outputs + 1; + + break; + default: + assert(0); + break; + } + + /* The elements in the key correspond to vertex shader output + * numbers, not to positions in the hw vertex description -- + * that's handled by the output_offset field. + */ + fse->key.element[vs_output].output_format = output_format; + fse->key.element[vs_output].output_offset = dst_offset; + + dst_offset += emit_sz; + assert(fse->key.output_stride >= dst_offset); + } + } + + /* To make psize work, really need to tell the vertex shader to + * copy that value from input->output. For 'translate' this was + * implicit for all elements. + */ +#if 0 + if (need_psize) { + unsigned input = num_vs_inputs + 1; + const struct pipe_vertex_element *src = &draw->pt.vertex_element[i]; + fse->key.element[i].input_format = PIPE_FORMAT_R32_FLOAT; + fse->key.element[i].input_buffer = 0; //nr_buffers + 1; + fse->key.element[i].input_offset = 0; + + fse->key.nr_elements += 1; + + } +#endif + + fse->constants = draw->pt.user.constants; + + /* Would normally look up a vertex shader and peruse its list of + * varients somehow. We omitted that step and put all the + * hardcoded "shaders" into an array. We're just making the + * assumption that this happens to be a matching shader... ie + * you're running isosurf, aren't you? + */ + fse->active = NULL; + for (i = 0; i < fse->nr_shaders; i++) { + if (translate_key_compare( &fse->key, &fse->shader[i].key) == 0) + fse->active = &fse->shader[i]; + } + + if (!fse->active) { + assert(0); + return ; + } + + /* Now set buffer pointers: + */ + for (i = 0; i < num_vs_inputs; i++) { + unsigned buf = draw->pt.vertex_element[i].vertex_buffer_index; + + fse->src[i] = ((const ubyte *) draw->pt.user.vbuffer[buf] + + draw->pt.vertex_buffer[buf].buffer_offset + + draw->pt.vertex_element[i].src_offset); + + fse->pitch[i] = draw->pt.vertex_buffer[buf].pitch; + + } + + + //return TRUE; +} + + + + + + + +static void fse_run_linear( struct draw_pt_middle_end *middle, + unsigned start, + unsigned count ) +{ + struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle; + struct draw_context *draw = fse->draw; + + char *hw_verts; + + /* XXX: need to flush to get prim_vbuf.c to release its allocation?? + */ + draw_do_flush( draw, DRAW_FLUSH_BACKEND ); + + hw_verts = draw->render->allocate_vertices( draw->render, + (ushort)fse->key.output_stride, + (ushort)count ); + + if (!hw_verts) { + assert(0); + return; + } + + /* Single routine to fetch vertices, run shader and emit HW verts. + * Clipping and viewport transformation are done elsewhere -- + * either by the API or on hardware, or for some other reason not + * required... + */ + fse->active->run_linear( fse, + start, count, + hw_verts ); + + /* Draw arrays path to avoid re-emitting index list again and + * again. + */ + draw->render->draw_arrays( draw->render, + 0, + count ); + + + draw->render->release_vertices( draw->render, + hw_verts, + fse->key.output_stride, + count ); +} + + +static void +fse_run(struct draw_pt_middle_end *middle, + const unsigned *fetch_elts, + unsigned fetch_count, + const ushort *draw_elts, + unsigned draw_count ) +{ + struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle; + struct draw_context *draw = fse->draw; + void *hw_verts; + + /* XXX: need to flush to get prim_vbuf.c to release its allocation?? + */ + draw_do_flush( draw, DRAW_FLUSH_BACKEND ); + + hw_verts = draw->render->allocate_vertices( draw->render, + (ushort)fse->key.output_stride, + (ushort)fetch_count ); + if (!hw_verts) { + assert(0); + return; + } + + + /* Single routine to fetch vertices, run shader and emit HW verts. + */ + fse->active->run_elts( fse, + fetch_elts, + fetch_count, + hw_verts ); + + draw->render->draw( draw->render, + draw_elts, + draw_count ); + + draw->render->release_vertices( draw->render, + hw_verts, + fse->key.output_stride, + fetch_count ); + +} + + +static void fse_finish( struct draw_pt_middle_end *middle ) +{ +} + + +static void +fse_destroy( struct draw_pt_middle_end *middle ) +{ + FREE(middle); +} + +struct draw_pt_middle_end *draw_pt_middle_fse( struct draw_context *draw ) +{ + struct fetch_shade_emit *fse = CALLOC_STRUCT(fetch_shade_emit); + if (!fse) + return NULL; + + fse->base.prepare = fse_prepare; + fse->base.run = fse_run; + fse->base.run_linear = fse_run_linear; + fse->base.finish = fse_finish; + fse->base.destroy = fse_destroy; + fse->draw = draw; + + fse->shader[0].run_linear = shader0_run_linear; + fse->shader[0].run_elts = shader0_run_elts; + fse->shader[0].key.nr_elements = 3; + fse->shader[0].key.output_stride = 12 * sizeof(float); + + fse->shader[0].key.element[0].input_format = PIPE_FORMAT_R32G32B32_FLOAT; + fse->shader[0].key.element[0].input_buffer = 0; + fse->shader[0].key.element[0].input_offset = 0; + fse->shader[0].key.element[0].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; + fse->shader[0].key.element[0].output_offset = 0; + + fse->shader[0].key.element[1].input_format = PIPE_FORMAT_R32G32B32_FLOAT; + fse->shader[0].key.element[1].input_buffer = 0; + fse->shader[0].key.element[1].input_offset = 0; + fse->shader[0].key.element[1].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; + fse->shader[0].key.element[1].output_offset = 16; + + fse->shader[0].key.element[1].input_format = PIPE_FORMAT_R32G32_FLOAT; + fse->shader[0].key.element[1].input_buffer = 0; + fse->shader[0].key.element[1].input_offset = 0; + fse->shader[0].key.element[1].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; + fse->shader[0].key.element[1].output_offset = 32; + + fse->shader[1].run_linear = shader1_run_linear; + fse->shader[1].run_elts = shader1_run_elts; + fse->shader[1].key.nr_elements = 2; + fse->shader[1].key.output_stride = 8 * sizeof(float); + + fse->shader[1].key.element[0].input_format = PIPE_FORMAT_R32G32B32_FLOAT; + fse->shader[1].key.element[0].input_buffer = 0; + fse->shader[1].key.element[0].input_offset = 0; + fse->shader[1].key.element[0].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; + fse->shader[1].key.element[0].output_offset = 0; + + fse->shader[1].key.element[1].input_format = PIPE_FORMAT_R32G32B32_FLOAT; + fse->shader[1].key.element[1].input_buffer = 0; + fse->shader[1].key.element[1].input_offset = 0; + fse->shader[1].key.element[1].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; + fse->shader[1].key.element[1].output_offset = 16; + + fse->shader[2].run_linear = shader2_run_linear; + fse->shader[2].run_elts = shader2_run_elts; + fse->shader[2].key.nr_elements = 3; + fse->shader[2].key.output_stride = 9 * sizeof(float); + + fse->shader[2].key.element[0].input_format = PIPE_FORMAT_R32G32B32_FLOAT; + fse->shader[2].key.element[0].input_buffer = 0; + fse->shader[2].key.element[0].input_offset = 0; + fse->shader[2].key.element[0].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; + fse->shader[2].key.element[0].output_offset = 0; + + fse->shader[2].key.element[1].input_format = PIPE_FORMAT_R32G32B32_FLOAT; + fse->shader[2].key.element[1].input_buffer = 0; + fse->shader[2].key.element[1].input_offset = 0; + fse->shader[2].key.element[1].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; + fse->shader[2].key.element[1].output_offset = 16; + + /* psize is special + * -- effectively add it here as another input!?! + * -- who knows how to add it as a buffer? + */ + fse->shader[2].key.element[2].input_format = PIPE_FORMAT_R32_FLOAT; + fse->shader[2].key.element[2].input_buffer = 0; + fse->shader[2].key.element[2].input_offset = 0; + fse->shader[2].key.element[2].output_format = PIPE_FORMAT_R32_FLOAT; + fse->shader[2].key.element[2].output_offset = 32; + + fse->nr_shaders = 3; + + return &fse->base; +} diff --git a/src/gallium/auxiliary/draw/draw_pt_util.c b/src/gallium/auxiliary/draw/draw_pt_util.c new file mode 100644 index 0000000000..32c8a9632c --- /dev/null +++ b/src/gallium/auxiliary/draw/draw_pt_util.c @@ -0,0 +1,103 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + /* + * Authors: + * Keith Whitwell + */ + +#include "pipe/p_util.h" +#include "draw/draw_context.h" +#include "draw/draw_private.h" +#include "draw/draw_pt.h" + +void draw_pt_split_prim(unsigned prim, unsigned *first, unsigned *incr) +{ + switch (prim) { + case PIPE_PRIM_POINTS: + *first = 1; + *incr = 1; + break; + case PIPE_PRIM_LINES: + *first = 2; + *incr = 2; + break; + case PIPE_PRIM_LINE_STRIP: + case PIPE_PRIM_LINE_LOOP: + *first = 2; + *incr = 1; + break; + case PIPE_PRIM_TRIANGLES: + *first = 3; + *incr = 3; + break; + case PIPE_PRIM_TRIANGLE_STRIP: + case PIPE_PRIM_TRIANGLE_FAN: + case PIPE_PRIM_POLYGON: + *first = 3; + *incr = 1; + break; + case PIPE_PRIM_QUADS: + *first = 4; + *incr = 4; + break; + case PIPE_PRIM_QUAD_STRIP: + *first = 4; + *incr = 2; + break; + default: + assert(0); + *first = 0; + *incr = 1; /* set to one so that count % incr works */ + break; + } +} + + +unsigned draw_pt_reduced_prim(unsigned prim) +{ + switch (prim) { + case PIPE_PRIM_POINTS: + return PIPE_PRIM_POINTS; + case PIPE_PRIM_LINES: + case PIPE_PRIM_LINE_STRIP: + case PIPE_PRIM_LINE_LOOP: + return PIPE_PRIM_LINES; + case PIPE_PRIM_TRIANGLES: + case PIPE_PRIM_TRIANGLE_STRIP: + case PIPE_PRIM_TRIANGLE_FAN: + case PIPE_PRIM_POLYGON: + case PIPE_PRIM_QUADS: + case PIPE_PRIM_QUAD_STRIP: + return PIPE_PRIM_TRIANGLES; + default: + assert(0); + return PIPE_PRIM_POINTS; + } +} + + diff --git a/src/gallium/auxiliary/draw/draw_pt_varray.c b/src/gallium/auxiliary/draw/draw_pt_varray.c index d92ad4fda1..af6e2d5157 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray.c +++ b/src/gallium/auxiliary/draw/draw_pt_varray.c @@ -106,51 +106,6 @@ static INLINE void fetch_init(struct varray_frontend *varray, } -static boolean split_prim_inplace(unsigned prim, unsigned *first, unsigned *incr) -{ - switch (prim) { - case PIPE_PRIM_POINTS: - *first = 1; - *incr = 1; - return TRUE; - case PIPE_PRIM_LINES: - *first = 2; - *incr = 2; - return TRUE; - case PIPE_PRIM_LINE_STRIP: - *first = 2; - *incr = 1; - return TRUE; - case PIPE_PRIM_TRIANGLES: - *first = 3; - *incr = 3; - return TRUE; - case PIPE_PRIM_TRIANGLE_STRIP: - *first = 3; - *incr = 1; - return TRUE; - case PIPE_PRIM_TRIANGLE_FAN: - *first = 3; - *incr = 1; - return TRUE; - case PIPE_PRIM_QUADS: - *first = 4; - *incr = 4; - return TRUE; - case PIPE_PRIM_QUAD_STRIP: - *first = 4; - *incr = 2; - return TRUE; - case PIPE_PRIM_POLYGON: - *first = 3; - *incr = 1; - return TRUE; - default: - *first = 0; - *incr = 1; /* set to one so that count % incr works */ - return FALSE; - } -} static INLINE void add_draw_el(struct varray_frontend *varray, diff --git a/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h index 1395275897..6979f6b544 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h +++ b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h @@ -15,7 +15,7 @@ static void FUNC(struct draw_pt_front_end *frontend, varray->fetch_start = start; - split_prim_inplace(varray->input_prim, &first, &incr); + draw_pt_split_prim(varray->input_prim, &first, &incr); #if 0 debug_printf("%s (%d) %d/%d\n", __FUNCTION__, diff --git a/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h b/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h index b6f1f0cadc..114ed371a0 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h +++ b/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h @@ -16,7 +16,13 @@ static void FUNC(struct draw_pt_front_end *frontend, varray->fetch_start = start; - split_prim_inplace(varray->input_prim, &first, &incr); + draw_pt_split_prim(varray->input_prim, &first, &incr); + + /* Sanitize primitive length: + */ + count = trim(count, first, incr); + if (count < first) + return; #if 0 debug_printf("%s (%d) %d/%d\n", __FUNCTION__, @@ -32,7 +38,6 @@ static void FUNC(struct draw_pt_front_end *frontend, case PIPE_PRIM_TRIANGLE_STRIP: case PIPE_PRIM_QUADS: case PIPE_PRIM_QUAD_STRIP: - for (j = 0; j < count;) { unsigned remaining = count - j; unsigned nr = trim( MIN2(FETCH_MAX, remaining), first, incr ); diff --git a/src/gallium/auxiliary/draw/draw_pt_vcache.c b/src/gallium/auxiliary/draw/draw_pt_vcache.c index 6b3fb1406b..a3495f2a30 100644 --- a/src/gallium/auxiliary/draw/draw_pt_vcache.c +++ b/src/gallium/auxiliary/draw/draw_pt_vcache.c @@ -204,19 +204,6 @@ static void vcache_ef_quad( struct vcache_frontend *vcache, -static unsigned reduced_prim[PIPE_PRIM_POLYGON + 1] = { - PIPE_PRIM_POINTS, - PIPE_PRIM_LINES, - PIPE_PRIM_LINES, - PIPE_PRIM_LINES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES, - PIPE_PRIM_TRIANGLES -}; - static void vcache_prepare( struct draw_pt_front_end *frontend, @@ -236,7 +223,7 @@ static void vcache_prepare( struct draw_pt_front_end *frontend, } vcache->input_prim = prim; - vcache->output_prim = reduced_prim[prim]; + vcache->output_prim = draw_pt_reduced_prim(prim); vcache->middle = middle; middle->prepare( middle, vcache->output_prim, opt ); -- cgit v1.2.3 From 1c624846a81b0218b4a07328f485e295432c6312 Mon Sep 17 00:00:00 2001 From: Zack Rusin Date: Tue, 13 May 2008 16:06:09 -0400 Subject: decomposition from keith, adds decomposition of more prim to the pipeline --- src/gallium/auxiliary/draw/draw_pipe.c | 74 ++++++++---- src/gallium/auxiliary/draw/draw_pt_decompose.h | 153 +++++++++++++++++++++++++ src/gallium/auxiliary/draw/draw_pt_varray.c | 11 +- src/gallium/auxiliary/draw/draw_pt_vcache.c | 117 +++++++++---------- 4 files changed, 263 insertions(+), 92 deletions(-) create mode 100644 src/gallium/auxiliary/draw/draw_pt_decompose.h (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pipe.c b/src/gallium/auxiliary/draw/draw_pipe.c index cb97f955b2..1d26706dee 100644 --- a/src/gallium/auxiliary/draw/draw_pipe.c +++ b/src/gallium/auxiliary/draw/draw_pipe.c @@ -212,6 +212,55 @@ void draw_pipeline_run( struct draw_context *draw, draw->pipeline.vertex_count = 0; } +#define QUAD(i0,i1,i2,i3) \ + do_triangle( draw, \ + ( DRAW_PIPE_RESET_STIPPLE | \ + DRAW_PIPE_EDGE_FLAG_0 | \ + DRAW_PIPE_EDGE_FLAG_2 ), \ + verts + stride * ((i0) & ~DRAW_PIPE_FLAG_MASK), \ + verts + stride * (i1), \ + verts + stride * (i3)); \ + do_triangle( draw, \ + ( DRAW_PIPE_EDGE_FLAG_0 | \ + DRAW_PIPE_EDGE_FLAG_1 ), \ + verts + stride * ((i1) & ~DRAW_PIPE_FLAG_MASK), \ + verts + stride * (i2), \ + verts + stride * (i3)) + +#define TRIANGLE(flags,i0,i1,i2) \ + do_triangle( draw, \ + flags, /* flags */ \ + verts + stride * ((i0) & ~DRAW_PIPE_FLAG_MASK), \ + verts + stride * (i1), \ + verts + stride * (i2)) + +#define LINE(flags,i0,i1) \ + do_line( draw, \ + flags, \ + verts + stride * ((i0) & ~DRAW_PIPE_FLAG_MASK), \ + verts + stride * (i+1)) + +#define POINT(i0) \ + do_point( draw, \ + verts + stride * i0 ) + +#define FUNC pipe_run_linear +#define ARGS \ + struct draw_context *draw, \ + unsigned prim, \ + struct vertex_header *vertices, \ + unsigned stride + +#define LOCAL_VARS \ + char *verts = (char *)vertices; \ + boolean flatfirst = (draw->rasterizer->flatshade && \ + draw->rasterizer->flatshade_first); \ + unsigned i, flags + +#define FLUSH + +#include "draw_pt_decompose.h" + void draw_pipeline_run_linear( struct draw_context *draw, unsigned prim, struct vertex_header *vertices, @@ -219,34 +268,11 @@ void draw_pipeline_run_linear( struct draw_context *draw, unsigned stride ) { char *verts = (char *)vertices; - unsigned i; - draw->pipeline.verts = verts; draw->pipeline.vertex_stride = stride; draw->pipeline.vertex_count = count; - switch (prim) { - case PIPE_PRIM_POINTS: - for (i = 0; i < count; i++) - do_point( draw, - verts + stride * i ); - break; - case PIPE_PRIM_LINES: - for (i = 0; i+1 < count; i += 2) - do_line( draw, - i+0, /* flags */ - verts + stride * ((i+0) & ~DRAW_PIPE_FLAG_MASK), - verts + stride * (i+1)); - break; - case PIPE_PRIM_TRIANGLES: - for (i = 0; i+2 < count; i += 3) - do_triangle( draw, - (i+0), /* flags */ - verts + stride * ((i+0) & ~DRAW_PIPE_FLAG_MASK), - verts + stride * (i+1), - verts + stride * (i+2)); - break; - } + pipe_run_linear(draw, prim, vertices, stride, count); draw->pipeline.verts = NULL; draw->pipeline.vertex_count = 0; diff --git a/src/gallium/auxiliary/draw/draw_pt_decompose.h b/src/gallium/auxiliary/draw/draw_pt_decompose.h new file mode 100644 index 0000000000..dccfde99dd --- /dev/null +++ b/src/gallium/auxiliary/draw/draw_pt_decompose.h @@ -0,0 +1,153 @@ + + +static void FUNC( ARGS, + unsigned count ) +{ + LOCAL_VARS; + + switch (prim) { + case PIPE_PRIM_POINTS: + for (i = 0; i < count; i ++) { + POINT( (i + 0) ); + } + break; + + case PIPE_PRIM_LINES: + for (i = 0; i+1 < count; i += 2) { + LINE( DRAW_PIPE_RESET_STIPPLE, + (i + 0), + (i + 1)); + } + break; + + case PIPE_PRIM_LINE_LOOP: + if (count >= 2) { + flags = DRAW_PIPE_RESET_STIPPLE; + + for (i = 1; i < count; i++, flags = 0) { + LINE( flags, + (i - 1), + (i )); + } + + LINE( flags, + (i - 1), + (0 )); + } + break; + + case PIPE_PRIM_LINE_STRIP: + flags = DRAW_PIPE_RESET_STIPPLE; + for (i = 1; i < count; i++, flags = 0) { + LINE( flags, + (i - 1), + (i )); + } + break; + + case PIPE_PRIM_TRIANGLES: + for (i = 0; i+2 < count; i += 3) { + TRIANGLE( DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, + (i + 0), + (i + 1), + (i + 2 )); + } + break; + + case PIPE_PRIM_TRIANGLE_STRIP: + if (flatfirst) { + for (i = 0; i+2 < count; i++) { + TRIANGLE( DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, + (i + 0), + (i + 1 + (i&1)), + (i + 2 - (i&1))); + } + } + else { + for (i = 0; i+2 < count; i++) { + TRIANGLE( DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, + (i + 0 + (i&1)), + (i + 1 - (i&1)), + (i + 2 )); + } + } + break; + + case PIPE_PRIM_TRIANGLE_FAN: + if (count >= 3) { + if (flatfirst) { + for (i = 0; i+2 < count; i++) { + TRIANGLE( DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, + (i + 1), + (i + 2), + (0 )); + } + } + else { + for (i = 0; i+2 < count; i++) { + TRIANGLE( DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, + (0), + (i + 1), + (i + 2 )); + } + } + } + break; + + + case PIPE_PRIM_QUADS: + for (i = 0; i+3 < count; i += 4) { + QUAD( (i + 0), + (i + 1), + (i + 2), + (i + 3)); + } + break; + + case PIPE_PRIM_QUAD_STRIP: + for (i = 0; i+3 < count; i += 2) { + QUAD( (i + 2), + (i + 0), + (i + 1), + (i + 3)); + } + break; + + case PIPE_PRIM_POLYGON: + { + /* These bitflags look a little odd because we submit the + * vertices as (1,2,0) to satisfy flatshade requirements. + */ + const unsigned edge_first = DRAW_PIPE_EDGE_FLAG_2; + const unsigned edge_middle = DRAW_PIPE_EDGE_FLAG_0; + const unsigned edge_last = DRAW_PIPE_EDGE_FLAG_1; + + flags = DRAW_PIPE_RESET_STIPPLE | edge_first | edge_middle; + + for (i = 0; i+2 < count; i++, flags = edge_middle) { + + if (i + 3 == count) + flags |= edge_last; + + TRIANGLE( flags, + (i + 1), + (i + 2), + (0)); + } + } + break; + + default: + assert(0); + break; + } + + FLUSH; +} + + +#undef TRIANGLE +#undef QUAD +#undef POINT +#undef LINE +#undef FUNC diff --git a/src/gallium/auxiliary/draw/draw_pt_varray.c b/src/gallium/auxiliary/draw/draw_pt_varray.c index af6e2d5157..06fd866ccd 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray.c +++ b/src/gallium/auxiliary/draw/draw_pt_varray.c @@ -141,14 +141,19 @@ static INLINE void varray_point( struct varray_frontend *varray, } - +#if 0 +#define TRIANGLE(flags,i0,i1,i2) varray_triangle(varray,i0,i1,i2) +#define LINE(flags,i0,i1) varray_line(varray,i0,i1) +#define POINT(i0) varray_point(varray,i0) +#define FUNC varray_decompose +#include "draw_pt_decompose.h" +#else #define TRIANGLE(vc,i0,i1,i2) varray_triangle(vc,i0,i1,i2) #define LINE(vc,i0,i1) varray_line(vc,i0,i1) #define POINT(vc,i0) varray_point(vc,i0) #define FUNC varray_run #include "draw_pt_varray_tmp_linear.h" - - +#endif static unsigned decompose_prim[PIPE_PRIM_POLYGON + 1] = { PIPE_PRIM_POINTS, diff --git a/src/gallium/auxiliary/draw/draw_pt_vcache.c b/src/gallium/auxiliary/draw/draw_pt_vcache.c index a3495f2a30..6c17edba34 100644 --- a/src/gallium/auxiliary/draw/draw_pt_vcache.c +++ b/src/gallium/auxiliary/draw/draw_pt_vcache.c @@ -104,22 +104,10 @@ static INLINE void vcache_elt( struct vcache_frontend *vcache, static void vcache_triangle( struct vcache_frontend *vcache, + ushort flags, unsigned i0, unsigned i1, unsigned i2 ) -{ - vcache_elt(vcache, i0, 0); - vcache_elt(vcache, i1, 0); - vcache_elt(vcache, i2, 0); - vcache_check_flush(vcache); -} - - -static void vcache_triangle_flags( struct vcache_frontend *vcache, - ushort flags, - unsigned i0, - unsigned i1, - unsigned i2 ) { vcache_elt(vcache, i0, flags); vcache_elt(vcache, i1, 0); @@ -128,19 +116,9 @@ static void vcache_triangle_flags( struct vcache_frontend *vcache, } static void vcache_line( struct vcache_frontend *vcache, + ushort flags, unsigned i0, unsigned i1 ) -{ - vcache_elt(vcache, i0, 0); - vcache_elt(vcache, i1, 0); - vcache_check_flush(vcache); -} - - -static void vcache_line_flags( struct vcache_frontend *vcache, - ushort flags, - unsigned i0, - unsigned i1 ) { vcache_elt(vcache, i0, flags); vcache_elt(vcache, i1, 0); @@ -161,46 +139,63 @@ static void vcache_quad( struct vcache_frontend *vcache, unsigned i2, unsigned i3 ) { - vcache_triangle( vcache, i0, i1, i3 ); - vcache_triangle( vcache, i1, i2, i3 ); -} - -static void vcache_ef_quad( struct vcache_frontend *vcache, - unsigned i0, - unsigned i1, - unsigned i2, - unsigned i3 ) -{ - const unsigned omitEdge1 = DRAW_PIPE_EDGE_FLAG_0 | DRAW_PIPE_EDGE_FLAG_2; - const unsigned omitEdge2 = DRAW_PIPE_EDGE_FLAG_0 | DRAW_PIPE_EDGE_FLAG_1; - - vcache_triangle_flags( vcache, - DRAW_PIPE_RESET_STIPPLE | omitEdge1, - i0, i1, i3 ); - - vcache_triangle_flags( vcache, - omitEdge2, - i1, i2, i3 ); + vcache_triangle( vcache, + ( DRAW_PIPE_RESET_STIPPLE | + DRAW_PIPE_EDGE_FLAG_0 | + DRAW_PIPE_EDGE_FLAG_2 ), + i0, i1, i3 ); + + vcache_triangle( vcache, + ( DRAW_PIPE_EDGE_FLAG_0 | + DRAW_PIPE_EDGE_FLAG_1 ), + i1, i2, i3 ); } /* At least for now, we're back to using a template include file for * this. The two paths aren't too different though - it may be * possible to reunify them. */ -#define TRIANGLE(vc,flags,i0,i1,i2) vcache_triangle_flags(vc,flags,i0,i1,i2) -#define QUAD(vc,i0,i1,i2,i3) vcache_ef_quad(vc,i0,i1,i2,i3) -#define LINE(vc,flags,i0,i1) vcache_line_flags(vc,flags,i0,i1) -#define POINT(vc,i0) vcache_point(vc,i0) -#define FUNC vcache_run_extras -#include "draw_pt_vcache_tmp.h" - -#define TRIANGLE(vc,flags,i0,i1,i2) vcache_triangle(vc,i0,i1,i2) -#define QUAD(vc,i0,i1,i2,i3) vcache_quad(vc,i0,i1,i2,i3) -#define LINE(vc,flags,i0,i1) vcache_line(vc,i0,i1) -#define POINT(vc,i0) vcache_point(vc,i0) +#define TRIANGLE(flags,i0,i1,i2) \ + vcache_triangle(vcache, \ + flags, \ + get_elt(elts,i0), \ + get_elt(elts,i1), \ + get_elt(elts,i2)) + +#define QUAD(i0,i1,i2,i3) \ + vcache_quad(vcache, \ + get_elt(elts,i0), \ + get_elt(elts,i1), \ + get_elt(elts,i2), \ + get_elt(elts,i3)) + +#define LINE(flags,i0,i1) \ + vcache_line(vcache, \ + flags, \ + get_elt(elts,i0), \ + get_elt(elts,i1)) + +#define POINT(i0) \ + vcache_point(vcache, \ + get_elt(elts,i0)) + #define FUNC vcache_run -#include "draw_pt_vcache_tmp.h" +#define ARGS \ + struct draw_pt_front_end *frontend, \ + pt_elt_func get_elt, \ + const void *elts + +#define LOCAL_VARS \ + struct vcache_frontend *vcache = (struct vcache_frontend *)frontend; \ + struct draw_context *draw = vcache->draw; \ + boolean flatfirst = (draw->rasterizer->flatshade && \ + draw->rasterizer->flatshade_first); \ + unsigned prim = vcache->input_prim; \ + unsigned i, flags; +#define FLUSH vcache_flush( vcache ) + +#include "draw_pt_decompose.h" @@ -213,15 +208,7 @@ static void vcache_prepare( struct draw_pt_front_end *frontend, { struct vcache_frontend *vcache = (struct vcache_frontend *)frontend; - if (opt & PT_PIPELINE) - { - vcache->base.run = vcache_run_extras; - } - else - { - vcache->base.run = vcache_run; - } - + vcache->base.run = vcache_run; vcache->input_prim = prim; vcache->output_prim = draw_pt_reduced_prim(prim); -- cgit v1.2.3 From 8b25b5256fad23e8ea11c6931ecac658ca60c0b0 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 13 May 2008 09:46:53 +0100 Subject: draw: remove disabled non-sse swizzle code --- src/gallium/auxiliary/draw/draw_vs_sse.c | 50 +------------------------------- 1 file changed, 1 insertion(+), 49 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c index e3f4e67472..edf235cddc 100644 --- a/src/gallium/auxiliary/draw/draw_vs_sse.c +++ b/src/gallium/auxiliary/draw/draw_vs_sse.c @@ -47,9 +47,7 @@ #include "tgsi/util/tgsi_parse.h" #define SSE_MAX_VERTICES 4 -#define SSE_SWIZZLES 1 -#if SSE_SWIZZLES typedef void (XSTDCALL *codegen_function) ( const struct tgsi_exec_vector *input, /* 1 */ struct tgsi_exec_vector *output, /* 2 */ @@ -62,14 +60,6 @@ typedef void (XSTDCALL *codegen_function) ( float (*aos_output)[4], /* 9 */ uint num_outputs, /* 10 */ uint output_stride ); /* 11 */ -#else -typedef void (XSTDCALL *codegen_function) ( - const struct tgsi_exec_vector *input, - struct tgsi_exec_vector *output, - float (*constant)[4], - struct tgsi_exec_vector *temporary, - float (*immediates)[4] ); -#endif struct draw_sse_vertex_shader { struct draw_vertex_shader base; @@ -111,7 +101,6 @@ vs_sse_run_linear( struct draw_vertex_shader *base, for (i = 0; i < count; i += MAX_TGSI_VERTICES) { unsigned int max_vertices = MIN2(MAX_TGSI_VERTICES, count - i); -#if SSE_SWIZZLES /* run compiled shader */ shader->func(machine->Inputs, @@ -128,43 +117,6 @@ vs_sse_run_linear( struct draw_vertex_shader *base, input = (const float (*)[4])((const char *)input + input_stride * max_vertices); output = (float (*)[4])((char *)output + output_stride * max_vertices); -#else - unsigned int j, slot; - - /* Swizzle inputs. - */ - for (j = 0; j < max_vertices; j++) { - for (slot = 0; slot < base->info.num_inputs; slot++) { - machine->Inputs[slot].xyzw[0].f[j] = input[slot][0]; - machine->Inputs[slot].xyzw[1].f[j] = input[slot][1]; - machine->Inputs[slot].xyzw[2].f[j] = input[slot][2]; - machine->Inputs[slot].xyzw[3].f[j] = input[slot][3]; - } - - input = (const float (*)[4])((const char *)input + input_stride); - } - - /* run compiled shader - */ - shader->func(machine->Inputs, - machine->Outputs, - (float (*)[4])constants, - machine->Temps, - shader->immediates); - - /* Unswizzle all output results. - */ - for (j = 0; j < max_vertices; j++) { - for (slot = 0; slot < base->info.num_outputs; slot++) { - output[slot][0] = machine->Outputs[slot].xyzw[0].f[j]; - output[slot][1] = machine->Outputs[slot].xyzw[1].f[j]; - output[slot][2] = machine->Outputs[slot].xyzw[2].f[j]; - output[slot][3] = machine->Outputs[slot].xyzw[3].f[j]; - } - - output = (float (*)[4])((char *)output + output_stride); - } -#endif } } @@ -211,7 +163,7 @@ draw_create_vs_sse(struct draw_context *draw, x86_init_func( &vs->sse2_program ); if (!tgsi_emit_sse2( (struct tgsi_token *) vs->base.state.tokens, - &vs->sse2_program, vs->immediates, SSE_SWIZZLES )) + &vs->sse2_program, vs->immediates, TRUE )) goto fail; vs->func = (codegen_function) x86_get_func( &vs->sse2_program ); -- cgit v1.2.3 From b23706454bb165a62888d264e95a98a2e4cf139c Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 13 May 2008 13:35:14 +0100 Subject: draw: get rid of fetch-shade-emit frontend hack The code is now living in it's intended place as a pt middle end. --- src/gallium/auxiliary/draw/Makefile | 1 - src/gallium/auxiliary/draw/draw_private.h | 1 - src/gallium/auxiliary/draw/draw_pt.c | 14 - src/gallium/auxiliary/draw/draw_pt.h | 1 - .../auxiliary/draw/draw_pt_fetch_shade_emit.c | 677 --------------------- 5 files changed, 694 deletions(-) delete mode 100644 src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/Makefile b/src/gallium/auxiliary/draw/Makefile index 67d78bdbbd..3053682da8 100644 --- a/src/gallium/auxiliary/draw/Makefile +++ b/src/gallium/auxiliary/draw/Makefile @@ -26,7 +26,6 @@ C_SOURCES = \ draw_pt_emit.c \ draw_pt_fetch.c \ draw_pt_fetch_emit.c \ - draw_pt_fetch_shade_emit.c \ draw_pt_middle_fse.c \ draw_pt_fetch_shade_pipeline.c \ draw_pt_post_vs.c \ diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h index 86b901a3c8..fd51a57781 100644 --- a/src/gallium/auxiliary/draw/draw_private.h +++ b/src/gallium/auxiliary/draw/draw_private.h @@ -131,7 +131,6 @@ struct draw_context struct { struct draw_pt_front_end *vcache; struct draw_pt_front_end *varray; - struct draw_pt_front_end *fetch_shade_emit; /* temp hack */ } front; struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS]; diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c index 91e35db819..75f44d503e 100644 --- a/src/gallium/auxiliary/draw/draw_pt.c +++ b/src/gallium/auxiliary/draw/draw_pt.c @@ -85,11 +85,6 @@ draw_pt_arrays(struct draw_context *draw, */ if (draw->pt.user.elts || (opt & PT_PIPELINE)) { frontend = draw->pt.front.vcache; -#if 0 - } else if (opt == PT_SHADE && draw->pt.test_fse) { - /* should be a middle end.. */ - frontend = draw->pt.front.fetch_shade_emit; -#endif } else { frontend = draw->pt.front.varray; } @@ -127,10 +122,6 @@ boolean draw_pt_init( struct draw_context *draw ) draw->pt.middle.fetch_shade_emit = draw_pt_middle_fse( draw ); if (!draw->pt.middle.fetch_shade_emit) return FALSE; - - draw->pt.front.fetch_shade_emit = draw_pt_fetch_shade_emit( draw ); - if (!draw->pt.front.fetch_shade_emit) - return FALSE; } @@ -159,11 +150,6 @@ void draw_pt_destroy( struct draw_context *draw ) draw->pt.middle.fetch_shade_emit = NULL; } - if (draw->pt.front.fetch_shade_emit) { - draw->pt.front.fetch_shade_emit->destroy( draw->pt.front.fetch_shade_emit ); - draw->pt.front.fetch_shade_emit = NULL; - } - if (draw->pt.front.vcache) { draw->pt.front.vcache->destroy( draw->pt.front.vcache ); draw->pt.front.vcache = NULL; diff --git a/src/gallium/auxiliary/draw/draw_pt.h b/src/gallium/auxiliary/draw/draw_pt.h index cdae46b8d2..e03816ebbc 100644 --- a/src/gallium/auxiliary/draw/draw_pt.h +++ b/src/gallium/auxiliary/draw/draw_pt.h @@ -121,7 +121,6 @@ const void *draw_pt_elt_ptr( struct draw_context *draw, struct draw_pt_front_end *draw_pt_vcache( struct draw_context *draw ); struct draw_pt_front_end *draw_pt_varray(struct draw_context *draw); -struct draw_pt_front_end *draw_pt_fetch_shade_emit( struct draw_context *draw ); /* Middle-ends: * diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c deleted file mode 100644 index f756d3e0bb..0000000000 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c +++ /dev/null @@ -1,677 +0,0 @@ -/************************************************************************** - * - * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. - * All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the - * "Software"), to deal in the Software without restriction, including - * without limitation the rights to use, copy, modify, merge, publish, - * distribute, sub license, and/or sell copies of the Software, and to - * permit persons to whom the Software is furnished to do so, subject to - * the following conditions: - * - * The above copyright notice and this permission notice (including the - * next paragraph) shall be included in all copies or substantial portions - * of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF - * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. - * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR - * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, - * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE - * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - * - **************************************************************************/ - - /* - * Authors: - * Keith Whitwell - */ - - -#include "pipe/p_util.h" -#include "draw/draw_context.h" -#include "draw/draw_private.h" -#include "draw/draw_vbuf.h" -#include "draw/draw_vertex.h" -#include "draw/draw_pt.h" -#include "draw/draw_vs.h" - -#include "translate/translate.h" - -struct fetch_shade_emit; - -struct fse_shader { - struct translate_key key; - - void (*run_linear)( const struct fetch_shade_emit *fse, - unsigned start, - unsigned count, - char *buffer ); -}; - -/* Prototype fetch, shade, emit-hw-verts all in one go. - */ -struct fetch_shade_emit { - struct draw_pt_front_end base; - - struct draw_context *draw; - - struct translate_key key; - - /* Temporaries: - */ - const float *constants; - unsigned pitch[PIPE_MAX_ATTRIBS]; - const ubyte *src[PIPE_MAX_ATTRIBS]; - unsigned prim; - - /* Points to one of the three hardwired example shaders, below: - */ - struct fse_shader *active; - - /* Temporary: A list of hard-wired shaders. Of course the plan - * would be to generate these for a given (vertex-shader, - * translate-key) pair... - */ - struct fse_shader shader[10]; - int nr_shaders; -}; - - - -/* Not quite passthrough yet -- we're still running the 'shader' here, - * inlined into the vertex fetch function. - */ -static void fetch_xyz_rgb_st( const struct fetch_shade_emit *fse, - unsigned start, - unsigned count, - char *buffer ) -{ - unsigned i; - - const float *m = fse->constants; - const float m0 = m[0], m4 = m[4], m8 = m[8], m12 = m[12]; - const float m1 = m[1], m5 = m[5], m9 = m[9], m13 = m[13]; - const float m2 = m[2], m6 = m[6], m10 = m[10], m14 = m[14]; - const float m3 = m[3], m7 = m[7], m11 = m[11], m15 = m[15]; - - const ubyte *xyz = fse->src[0] + start * fse->pitch[0]; - const ubyte *st = fse->src[2] + start * fse->pitch[2]; - - float *out = (float *)buffer; - - - assert(fse->pitch[1] == 0); - - /* loop over vertex attributes (vertex shader inputs) - */ - for (i = 0; i < count; i++) { - { - const float *in = (const float *)xyz; - const float ix = in[0], iy = in[1], iz = in[2]; - - out[0] = m0 * ix + m4 * iy + m8 * iz + m12; - out[1] = m1 * ix + m5 * iy + m9 * iz + m13; - out[2] = m2 * ix + m6 * iy + m10 * iz + m14; - out[3] = m3 * ix + m7 * iy + m11 * iz + m15; - xyz += fse->pitch[0]; - } - - { - out[4] = 1.0f; - out[5] = 1.0f; - out[6] = 1.0f; - out[7] = 1.0f; - } - - { - const float *in = (const float *)st; st += fse->pitch[2]; - out[8] = in[0]; - out[9] = in[1]; - out[10] = 0.0f; - out[11] = 1.0f; - } - - out += 12; - } -} - - - -static void fetch_xyz_rgb( const struct fetch_shade_emit *fse, - unsigned start, - unsigned count, - char *buffer ) -{ - unsigned i; - - const float *m = (const float *)fse->constants; - const float m0 = m[0], m4 = m[4], m8 = m[8], m12 = m[12]; - const float m1 = m[1], m5 = m[5], m9 = m[9], m13 = m[13]; - const float m2 = m[2], m6 = m[6], m10 = m[10], m14 = m[14]; - const float m3 = m[3], m7 = m[7], m11 = m[11], m15 = m[15]; - - const ubyte *xyz = fse->src[0] + start * fse->pitch[0]; - const ubyte *rgb = fse->src[1] + start * fse->pitch[1]; - - float *out = (float *)buffer; - -// debug_printf("rgb %f %f %f\n", rgb[0], rgb[1], rgb[2]); - - - for (i = 0; i < count; i++) { - { - const float *in = (const float *)xyz; - const float ix = in[0], iy = in[1], iz = in[2]; - - out[0] = m0 * ix + m4 * iy + m8 * iz + m12; - out[1] = m1 * ix + m5 * iy + m9 * iz + m13; - out[2] = m2 * ix + m6 * iy + m10 * iz + m14; - out[3] = m3 * ix + m7 * iy + m11 * iz + m15; - xyz += fse->pitch[0]; - } - - { - const float *in = (const float *)rgb; - out[4] = in[0]; - out[5] = in[1]; - out[6] = in[2]; - out[7] = 1.0f; - rgb += fse->pitch[1]; - } - - out += 8; - } -} - - - - -static void fetch_xyz_rgb_psiz( const struct fetch_shade_emit *fse, - unsigned start, - unsigned count, - char *buffer ) -{ - unsigned i; - - const float *m = (const float *)fse->constants; - const float m0 = m[0], m4 = m[4], m8 = m[8], m12 = m[12]; - const float m1 = m[1], m5 = m[5], m9 = m[9], m13 = m[13]; - const float m2 = m[2], m6 = m[6], m10 = m[10], m14 = m[14]; - const float m3 = m[3], m7 = m[7], m11 = m[11], m15 = m[15]; - - const ubyte *xyz = fse->src[0] + start * fse->pitch[0]; - const float *rgb = (const float *)(fse->src[1] + start * fse->pitch[1]); - const float psiz = 1.0; - - float *out = (float *)buffer; - - - assert(fse->pitch[1] == 0); - - for (i = 0; i < count; i++) { - { - const float *in = (const float *)xyz; - const float ix = in[0], iy = in[1], iz = in[2]; - - out[0] = m0 * ix + m4 * iy + m8 * iz + m12; - out[1] = m1 * ix + m5 * iy + m9 * iz + m13; - out[2] = m2 * ix + m6 * iy + m10 * iz + m14; - out[3] = m3 * ix + m7 * iy + m11 * iz + m15; - xyz += fse->pitch[0]; - } - - { - out[4] = rgb[0]; - out[5] = rgb[1]; - out[6] = rgb[2]; - out[7] = 1.0f; - } - - { - out[8] = psiz; - } - - out += 9; - } -} - - - - -static boolean set_prim( struct fetch_shade_emit *fse, - unsigned prim, - unsigned count ) -{ - struct draw_context *draw = fse->draw; - - fse->prim = prim; - - switch (prim) { - case PIPE_PRIM_LINE_LOOP: - if (count > 1024) - return FALSE; - draw->render->set_primitive( draw->render, PIPE_PRIM_LINE_STRIP ); - break; - - case PIPE_PRIM_TRIANGLE_FAN: - case PIPE_PRIM_POLYGON: - if (count > 1024) - return FALSE; - draw->render->set_primitive( draw->render, prim ); - break; - - case PIPE_PRIM_QUADS: - case PIPE_PRIM_QUAD_STRIP: - draw->render->set_primitive( draw->render, PIPE_PRIM_TRIANGLES ); - break; - - default: - draw->render->set_primitive( draw->render, prim ); - break; - } - - return TRUE; -} - - - - - - -static void fse_prepare( struct draw_pt_front_end *fe, - unsigned prim, - struct draw_pt_middle_end *unused, - unsigned opt ) -{ - struct fetch_shade_emit *fse = (struct fetch_shade_emit *)fe; - struct draw_context *draw = fse->draw; - unsigned num_vs_inputs = draw->vertex_shader->info.num_inputs; - unsigned num_vs_outputs = draw->vertex_shader->info.num_outputs; - const struct vertex_info *vinfo; - unsigned i; - boolean need_psize = 0; - - - if (draw->pt.user.elts) { - assert(0); - return ; - } - - if (!set_prim(fse, prim, /*count*/1022 )) { - assert(0); - return ; - } - - /* Must do this after set_primitive() above: - */ - vinfo = draw->render->get_vertex_info(draw->render); - - - - fse->key.nr_elements = MAX2(num_vs_outputs, /* outputs - translate to hw format */ - num_vs_inputs); /* inputs - fetch from api format */ - - fse->key.output_stride = vinfo->size * 4; - memset(fse->key.element, 0, - fse->key.nr_elements * sizeof(fse->key.element[0])); - - for (i = 0; i < num_vs_inputs; i++) { - const struct pipe_vertex_element *src = &draw->pt.vertex_element[i]; - fse->key.element[i].input_format = src->src_format; - - /* Consider ignoring these at this point, ie make generated - * programs independent of this state: - */ - fse->key.element[i].input_buffer = 0; //src->vertex_buffer_index; - fse->key.element[i].input_offset = 0; //src->src_offset; - } - - - { - unsigned dst_offset = 0; - - for (i = 0; i < vinfo->num_attribs; i++) { - unsigned emit_sz = 0; - unsigned output_format = PIPE_FORMAT_NONE; - unsigned vs_output = vinfo->src_index[i]; - - switch (vinfo->emit[i]) { - case EMIT_4F: - output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; - emit_sz = 4 * sizeof(float); - break; - case EMIT_3F: - output_format = PIPE_FORMAT_R32G32B32_FLOAT; - emit_sz = 3 * sizeof(float); - break; - case EMIT_2F: - output_format = PIPE_FORMAT_R32G32_FLOAT; - emit_sz = 2 * sizeof(float); - break; - case EMIT_1F: - output_format = PIPE_FORMAT_R32_FLOAT; - emit_sz = 1 * sizeof(float); - break; - case EMIT_1F_PSIZE: - need_psize = 1; - output_format = PIPE_FORMAT_R32_FLOAT; - emit_sz = 1 * sizeof(float); - vs_output = num_vs_outputs + 1; - - break; - default: - assert(0); - break; - } - - /* The elements in the key correspond to vertex shader output - * numbers, not to positions in the hw vertex description -- - * that's handled by the output_offset field. - */ - fse->key.element[vs_output].output_format = output_format; - fse->key.element[vs_output].output_offset = dst_offset; - - dst_offset += emit_sz; - assert(fse->key.output_stride >= dst_offset); - } - } - - /* To make psize work, really need to tell the vertex shader to - * copy that value from input->output. For 'translate' this was - * implicit for all elements. - */ -#if 0 - if (need_psize) { - unsigned input = num_vs_inputs + 1; - const struct pipe_vertex_element *src = &draw->pt.vertex_element[i]; - fse->key.element[i].input_format = PIPE_FORMAT_R32_FLOAT; - fse->key.element[i].input_buffer = 0; //nr_buffers + 1; - fse->key.element[i].input_offset = 0; - - fse->key.nr_elements += 1; - - } -#endif - - fse->constants = draw->pt.user.constants; - - /* Would normally look up a vertex shader and peruse its list of - * varients somehow. We omitted that step and put all the - * hardcoded "shaders" into an array. We're just making the - * assumption that this happens to be a matching shader... ie - * you're running isosurf, aren't you? - */ - fse->active = NULL; - for (i = 0; i < fse->nr_shaders; i++) { - if (translate_key_compare( &fse->key, &fse->shader[i].key) == 0) - fse->active = &fse->shader[i]; - } - - if (!fse->active) { - assert(0); - return ; - } - - /* Now set buffer pointers: - */ - for (i = 0; i < num_vs_inputs; i++) { - unsigned buf = draw->pt.vertex_element[i].vertex_buffer_index; - - fse->src[i] = ((const ubyte *) draw->pt.user.vbuffer[buf] + - draw->pt.vertex_buffer[buf].buffer_offset + - draw->pt.vertex_element[i].src_offset); - - fse->pitch[i] = draw->pt.vertex_buffer[buf].pitch; - - } - - - //return TRUE; -} - - - - - - -#define INDEX(i) (start + (i)) -static void fse_render_linear( struct vbuf_render *render, - unsigned prim, - unsigned start, - unsigned length ) -{ - ushort *tmp = NULL; - unsigned i, j; - - switch (prim) { - case PIPE_PRIM_LINE_LOOP: - tmp = MALLOC( sizeof(ushort) * (length + 1) ); - - for (i = 0; i < length; i++) - tmp[i] = INDEX(i); - tmp[length] = 0; - - render->draw( render, - tmp, - length+1 ); - break; - - - case PIPE_PRIM_QUAD_STRIP: - tmp = MALLOC( sizeof(ushort) * (length / 2 * 6) ); - - for (j = i = 0; i + 3 < length; i += 2, j += 6) { - tmp[j+0] = INDEX(i+0); - tmp[j+1] = INDEX(i+1); - tmp[j+2] = INDEX(i+3); - - tmp[j+3] = INDEX(i+2); - tmp[j+4] = INDEX(i+0); - tmp[j+5] = INDEX(i+3); - } - - if (j) - render->draw( render, tmp, j ); - break; - - case PIPE_PRIM_QUADS: - tmp = MALLOC( sizeof(int) * (length / 4 * 6) ); - - for (j = i = 0; i + 3 < length; i += 4, j += 6) { - tmp[j+0] = INDEX(i+0); - tmp[j+1] = INDEX(i+1); - tmp[j+2] = INDEX(i+3); - - tmp[j+3] = INDEX(i+1); - tmp[j+4] = INDEX(i+2); - tmp[j+5] = INDEX(i+3); - } - - if (j) - render->draw( render, tmp, j ); - break; - - default: - render->draw_arrays( render, - start, - length ); - break; - } - - if (tmp) - FREE(tmp); -} - - - -static boolean do_draw( struct fetch_shade_emit *fse, - unsigned start, unsigned count ) -{ - struct draw_context *draw = fse->draw; - - char *hw_verts = - draw->render->allocate_vertices( draw->render, - (ushort)fse->key.output_stride, - (ushort)count ); - - if (!hw_verts) - return FALSE; - - /* Single routine to fetch vertices, run shader and emit HW verts. - * Clipping and viewport transformation are done on hardware. - */ - fse->active->run_linear( fse, - start, count, - hw_verts ); - - /* Draw arrays path to avoid re-emitting index list again and - * again. - */ - fse_render_linear( draw->render, - fse->prim, - 0, - count ); - - - draw->render->release_vertices( draw->render, - hw_verts, - fse->key.output_stride, - count ); - - return TRUE; -} - - -static void -fse_run(struct draw_pt_front_end *fe, - pt_elt_func elt_func, - const void *elt_ptr, - unsigned count) -{ - struct fetch_shade_emit *fse = (struct fetch_shade_emit *)fe; - unsigned i = 0; - unsigned first, incr; - unsigned start = elt_func(elt_ptr, 0); - - //debug_printf("%s prim %d start %d count %d\n", __FUNCTION__, prim, start, count); - - draw_pt_split_prim(fse->prim, &first, &incr); - - count -= (count - first) % incr; - - while (i + first <= count) { - int nr = MIN2( count - i, 1024 ); - - /* snap to prim boundary - */ - nr -= (nr - first) % incr; - - if (!do_draw( fse, start + i, nr )) { - assert(0); - return ; - } - - /* increment allowing for repeated vertices - */ - i += nr - (first - incr); - } - - //return TRUE; -} - - -static void fse_finish( struct draw_pt_front_end *frontend ) -{ -} - - -static void -fse_destroy( struct draw_pt_front_end *frontend ) -{ - FREE(frontend); -} - -struct draw_pt_front_end *draw_pt_fetch_shade_emit( struct draw_context *draw ) -{ - struct fetch_shade_emit *fse = CALLOC_STRUCT(fetch_shade_emit); - if (!fse) - return NULL; - - fse->base.prepare = fse_prepare; - fse->base.run = fse_run; - fse->base.finish = fse_finish; - fse->base.destroy = fse_destroy; - fse->draw = draw; - - fse->shader[0].run_linear = fetch_xyz_rgb_st; - fse->shader[0].key.nr_elements = 3; - fse->shader[0].key.output_stride = 12 * sizeof(float); - - fse->shader[0].key.element[0].input_format = PIPE_FORMAT_R32G32B32_FLOAT; - fse->shader[0].key.element[0].input_buffer = 0; - fse->shader[0].key.element[0].input_offset = 0; - fse->shader[0].key.element[0].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; - fse->shader[0].key.element[0].output_offset = 0; - - fse->shader[0].key.element[1].input_format = PIPE_FORMAT_R32G32B32_FLOAT; - fse->shader[0].key.element[1].input_buffer = 0; - fse->shader[0].key.element[1].input_offset = 0; - fse->shader[0].key.element[1].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; - fse->shader[0].key.element[1].output_offset = 16; - - fse->shader[0].key.element[1].input_format = PIPE_FORMAT_R32G32_FLOAT; - fse->shader[0].key.element[1].input_buffer = 0; - fse->shader[0].key.element[1].input_offset = 0; - fse->shader[0].key.element[1].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; - fse->shader[0].key.element[1].output_offset = 32; - - fse->shader[1].run_linear = fetch_xyz_rgb; - fse->shader[1].key.nr_elements = 2; - fse->shader[1].key.output_stride = 8 * sizeof(float); - - fse->shader[1].key.element[0].input_format = PIPE_FORMAT_R32G32B32_FLOAT; - fse->shader[1].key.element[0].input_buffer = 0; - fse->shader[1].key.element[0].input_offset = 0; - fse->shader[1].key.element[0].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; - fse->shader[1].key.element[0].output_offset = 0; - - fse->shader[1].key.element[1].input_format = PIPE_FORMAT_R32G32B32_FLOAT; - fse->shader[1].key.element[1].input_buffer = 0; - fse->shader[1].key.element[1].input_offset = 0; - fse->shader[1].key.element[1].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; - fse->shader[1].key.element[1].output_offset = 16; - - fse->shader[2].run_linear = fetch_xyz_rgb_psiz; - fse->shader[2].key.nr_elements = 3; - fse->shader[2].key.output_stride = 9 * sizeof(float); - - fse->shader[2].key.element[0].input_format = PIPE_FORMAT_R32G32B32_FLOAT; - fse->shader[2].key.element[0].input_buffer = 0; - fse->shader[2].key.element[0].input_offset = 0; - fse->shader[2].key.element[0].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; - fse->shader[2].key.element[0].output_offset = 0; - - fse->shader[2].key.element[1].input_format = PIPE_FORMAT_R32G32B32_FLOAT; - fse->shader[2].key.element[1].input_buffer = 0; - fse->shader[2].key.element[1].input_offset = 0; - fse->shader[2].key.element[1].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; - fse->shader[2].key.element[1].output_offset = 16; - - /* psize is special - * -- effectively add it here as another input!?! - * -- who knows how to add it as a buffer? - */ - fse->shader[2].key.element[2].input_format = PIPE_FORMAT_R32_FLOAT; - fse->shader[2].key.element[2].input_buffer = 0; - fse->shader[2].key.element[2].input_offset = 0; - fse->shader[2].key.element[2].output_format = PIPE_FORMAT_R32_FLOAT; - fse->shader[2].key.element[2].output_offset = 32; - - fse->nr_shaders = 3; - - return &fse->base; -} -- cgit v1.2.3 From 2f0d1396e4c1626b3b1ac799bd29e86a9530369e Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 13 May 2008 13:40:22 +0100 Subject: draw: move some state into a new 'vs' area --- src/gallium/auxiliary/draw/draw_context.c | 21 ++++--------- src/gallium/auxiliary/draw/draw_pipe.h | 2 +- src/gallium/auxiliary/draw/draw_pipe_aaline.c | 2 +- src/gallium/auxiliary/draw/draw_pipe_aapoint.c | 4 +-- src/gallium/auxiliary/draw/draw_pipe_clip.c | 6 ++-- src/gallium/auxiliary/draw/draw_pipe_flatshade.c | 2 +- src/gallium/auxiliary/draw/draw_pipe_stipple.c | 2 +- src/gallium/auxiliary/draw/draw_pipe_twoside.c | 2 +- src/gallium/auxiliary/draw/draw_pipe_wide_point.c | 4 +-- src/gallium/auxiliary/draw/draw_private.h | 28 +++++++++++------ .../auxiliary/draw/draw_pt_fetch_shade_pipeline.c | 6 ++-- src/gallium/auxiliary/draw/draw_pt_middle_fse.c | 4 +-- src/gallium/auxiliary/draw/draw_vs.c | 35 +++++++++++++++++++--- src/gallium/auxiliary/draw/draw_vs_exec.c | 2 +- src/gallium/auxiliary/draw/draw_vs_sse.c | 2 +- 15 files changed, 75 insertions(+), 47 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_context.c b/src/gallium/auxiliary/draw/draw_context.c index 98e23fa830..2242074965 100644 --- a/src/gallium/auxiliary/draw/draw_context.c +++ b/src/gallium/auxiliary/draw/draw_context.c @@ -56,12 +56,6 @@ struct draw_context *draw_create( void ) draw->reduced_prim = ~0; /* != any of PIPE_PRIM_x */ - tgsi_exec_machine_init(&draw->machine); - - /* FIXME: give this machine thing a proper constructor: - */ - draw->machine.Inputs = align_malloc(PIPE_MAX_ATTRIBS * sizeof(struct tgsi_exec_vector), 16); - draw->machine.Outputs = align_malloc(PIPE_MAX_ATTRIBS * sizeof(struct tgsi_exec_vector), 16); if (!draw_pipeline_init( draw )) goto fail; @@ -69,6 +63,9 @@ struct draw_context *draw_create( void ) if (!draw_pt_init( draw )) goto fail; + if (!draw_vs_init( draw )) + goto fail; + return draw; fail: @@ -83,13 +80,6 @@ void draw_destroy( struct draw_context *draw ) return; - if (draw->machine.Inputs) - align_free(draw->machine.Inputs); - - if (draw->machine.Outputs) - align_free(draw->machine.Outputs); - - tgsi_exec_machine_free_data(&draw->machine); /* Not so fast -- we're just borrowing this at the moment. * @@ -99,6 +89,7 @@ void draw_destroy( struct draw_context *draw ) draw_pipeline_destroy( draw ); draw_pt_destroy( draw ); + draw_vs_destroy( draw ); FREE( draw ); } @@ -295,7 +286,7 @@ int draw_find_vs_output(struct draw_context *draw, uint semantic_name, uint semantic_index) { - const struct draw_vertex_shader *vs = draw->vertex_shader; + const struct draw_vertex_shader *vs = draw->vs.vertex_shader; uint i; for (i = 0; i < vs->info.num_outputs; i++) { if (vs->info.output_semantic_name[i] == semantic_name && @@ -320,7 +311,7 @@ draw_find_vs_output(struct draw_context *draw, uint draw_num_vs_outputs(struct draw_context *draw) { - uint count = draw->vertex_shader->info.num_outputs; + uint count = draw->vs.vertex_shader->info.num_outputs; if (draw->extra_vp_outputs.slot > 0) count++; return count; diff --git a/src/gallium/auxiliary/draw/draw_pipe.h b/src/gallium/auxiliary/draw/draw_pipe.h index f1cb0891ca..dbad8f98ac 100644 --- a/src/gallium/auxiliary/draw/draw_pipe.h +++ b/src/gallium/auxiliary/draw/draw_pipe.h @@ -116,7 +116,7 @@ dup_vert( struct draw_stage *stage, { struct vertex_header *tmp = stage->tmp[idx]; const uint vsize = sizeof(struct vertex_header) - + stage->draw->num_vs_outputs * 4 * sizeof(float); + + stage->draw->vs.num_vs_outputs * 4 * sizeof(float); memcpy(tmp, vert, vsize); tmp->vertex_id = UNDEFINED_VERTEX_ID; return tmp; diff --git a/src/gallium/auxiliary/draw/draw_pipe_aaline.c b/src/gallium/auxiliary/draw/draw_pipe_aaline.c index f501b2aed4..d93708ad3c 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_aaline.c +++ b/src/gallium/auxiliary/draw/draw_pipe_aaline.c @@ -651,7 +651,7 @@ aaline_first_line(struct draw_stage *stage, struct prim_header *header) } /* update vertex attrib info */ - aaline->tex_slot = draw->num_vs_outputs; + aaline->tex_slot = draw->vs.num_vs_outputs; assert(aaline->tex_slot > 0); /* output[0] is vertex pos */ /* advertise the extra post-transformed vertex attribute */ diff --git a/src/gallium/auxiliary/draw/draw_pipe_aapoint.c b/src/gallium/auxiliary/draw/draw_pipe_aapoint.c index 122a48660a..97d74ad693 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_aapoint.c +++ b/src/gallium/auxiliary/draw/draw_pipe_aapoint.c @@ -681,7 +681,7 @@ aapoint_first_point(struct draw_stage *stage, struct prim_header *header) bind_aapoint_fragment_shader(aapoint); /* update vertex attrib info */ - aapoint->tex_slot = draw->num_vs_outputs; + aapoint->tex_slot = draw->vs.num_vs_outputs; assert(aapoint->tex_slot > 0); /* output[0] is vertex pos */ draw->extra_vp_outputs.semantic_name = TGSI_SEMANTIC_GENERIC; @@ -692,7 +692,7 @@ aapoint_first_point(struct draw_stage *stage, struct prim_header *header) aapoint->psize_slot = -1; if (draw->rasterizer->point_size_per_vertex) { /* find PSIZ vertex output */ - const struct draw_vertex_shader *vs = draw->vertex_shader; + const struct draw_vertex_shader *vs = draw->vs.vertex_shader; uint i; for (i = 0; i < vs->info.num_outputs; i++) { if (vs->info.output_semantic_name[i] == TGSI_SEMANTIC_PSIZE) { diff --git a/src/gallium/auxiliary/draw/draw_pipe_clip.c b/src/gallium/auxiliary/draw/draw_pipe_clip.c index ce80c94163..c11ed934a4 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_clip.c +++ b/src/gallium/auxiliary/draw/draw_pipe_clip.c @@ -112,7 +112,7 @@ static void interp( const struct clipper *clip, const struct vertex_header *out, const struct vertex_header *in ) { - const unsigned nr_attrs = clip->stage.draw->num_vs_outputs; + const unsigned nr_attrs = clip->stage.draw->vs.num_vs_outputs; unsigned j; /* Vertex header. @@ -180,7 +180,7 @@ static void emit_poly( struct draw_stage *stage, header.flags |= edge_last; if (0) { - const struct draw_vertex_shader *vs = stage->draw->vertex_shader; + const struct draw_vertex_shader *vs = stage->draw->vs.vertex_shader; uint j, k; debug_printf("Clipped tri:\n"); for (j = 0; j < 3; j++) { @@ -425,7 +425,7 @@ clip_init_state( struct draw_stage *stage ) clipper->flat = stage->draw->rasterizer->flatshade ? TRUE : FALSE; if (clipper->flat) { - const struct draw_vertex_shader *vs = stage->draw->vertex_shader; + const struct draw_vertex_shader *vs = stage->draw->vs.vertex_shader; uint i; clipper->num_color_attribs = 0; diff --git a/src/gallium/auxiliary/draw/draw_pipe_flatshade.c b/src/gallium/auxiliary/draw/draw_pipe_flatshade.c index 09b68c4559..21a9c3b77f 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_flatshade.c +++ b/src/gallium/auxiliary/draw/draw_pipe_flatshade.c @@ -159,7 +159,7 @@ static void flatshade_line_1( struct draw_stage *stage, static void flatshade_init_state( struct draw_stage *stage ) { struct flat_stage *flat = flat_stage(stage); - const struct draw_vertex_shader *vs = stage->draw->vertex_shader; + const struct draw_vertex_shader *vs = stage->draw->vs.vertex_shader; uint i; /* Find which vertex shader outputs are colors, make a list */ diff --git a/src/gallium/auxiliary/draw/draw_pipe_stipple.c b/src/gallium/auxiliary/draw/draw_pipe_stipple.c index 3cbced362e..4673d5dcba 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_stipple.c +++ b/src/gallium/auxiliary/draw/draw_pipe_stipple.c @@ -71,7 +71,7 @@ screen_interp( struct draw_context *draw, const struct vertex_header *v1 ) { uint attr; - for (attr = 0; attr < draw->num_vs_outputs; attr++) { + for (attr = 0; attr < draw->vs.num_vs_outputs; attr++) { const float *val0 = v0->data[attr]; const float *val1 = v1->data[attr]; float *newv = dst->data[attr]; diff --git a/src/gallium/auxiliary/draw/draw_pipe_twoside.c b/src/gallium/auxiliary/draw/draw_pipe_twoside.c index 50872fdbe9..3ac825f565 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_twoside.c +++ b/src/gallium/auxiliary/draw/draw_pipe_twoside.c @@ -105,7 +105,7 @@ static void twoside_first_tri( struct draw_stage *stage, struct prim_header *header ) { struct twoside_stage *twoside = twoside_stage(stage); - const struct draw_vertex_shader *vs = stage->draw->vertex_shader; + const struct draw_vertex_shader *vs = stage->draw->vs.vertex_shader; uint i; twoside->attrib_front0 = 0; diff --git a/src/gallium/auxiliary/draw/draw_pipe_wide_point.c b/src/gallium/auxiliary/draw/draw_pipe_wide_point.c index ed08573382..df92e3f2d0 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_wide_point.c +++ b/src/gallium/auxiliary/draw/draw_pipe_wide_point.c @@ -197,7 +197,7 @@ static void widepoint_first_point( struct draw_stage *stage, if (draw->rasterizer->point_sprite) { /* find vertex shader texcoord outputs */ - const struct draw_vertex_shader *vs = draw->vertex_shader; + const struct draw_vertex_shader *vs = draw->vs.vertex_shader; uint i, j = 0; for (i = 0; i < vs->info.num_outputs; i++) { if (vs->info.output_semantic_name[i] == TGSI_SEMANTIC_GENERIC) { @@ -212,7 +212,7 @@ static void widepoint_first_point( struct draw_stage *stage, wide->psize_slot = -1; if (draw->rasterizer->point_size_per_vertex) { /* find PSIZ vertex output */ - const struct draw_vertex_shader *vs = draw->vertex_shader; + const struct draw_vertex_shader *vs = draw->vs.vertex_shader; uint i; for (i = 0; i < vs->info.num_outputs; i++) { if (vs->info.output_semantic_name[i] == TGSI_SEMANTIC_PSIZE) { diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h index fd51a57781..3418ee2b88 100644 --- a/src/gallium/auxiliary/draw/draw_private.h +++ b/src/gallium/auxiliary/draw/draw_private.h @@ -169,13 +169,24 @@ struct draw_context /* pipe state that we need: */ const struct pipe_rasterizer_state *rasterizer; struct pipe_viewport_state viewport; + boolean identity_viewport; - struct draw_vertex_shader *vertex_shader; + struct { + struct draw_vertex_shader *vertex_shader; + uint num_vs_outputs; /**< convenience, from vertex_shader */ - boolean identity_viewport; - uint num_vs_outputs; /**< convenience, from vertex_shader */ + /** TGSI program interpreter runtime state */ + struct tgsi_exec_machine machine; + + /* This (and the tgsi_exec_machine struct) probably need to be moved somewhere private. + */ + struct gallivm_cpu_engine *engine; + + struct translate_cache *fetch_cache; + struct translate_cache *emit_cache; + } vs; /* Clip derived state: */ @@ -192,16 +203,15 @@ struct draw_context unsigned reduced_prim; - /** TGSI program interpreter runtime state */ - struct tgsi_exec_machine machine; - - /* This (and the tgsi_exec_machine struct) probably need to be moved somewhere private. - */ - struct gallivm_cpu_engine *engine; void *driver_private; }; +/******************************************************************************* + * Vertex shader code: + */ +boolean draw_vs_init( struct draw_context *draw ); +void draw_vs_destroy( struct draw_context *draw ); diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c index dad54690a5..06718779a5 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c @@ -55,7 +55,7 @@ static void fetch_pipeline_prepare( struct draw_pt_middle_end *middle, { struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle; struct draw_context *draw = fpme->draw; - struct draw_vertex_shader *vs = draw->vertex_shader; + struct draw_vertex_shader *vs = draw->vs.vertex_shader; /* Add one to num_outputs because the pipeline occasionally tags on * an additional texcoord, eg for AA lines. @@ -107,7 +107,7 @@ static void fetch_pipeline_run( struct draw_pt_middle_end *middle, { struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle; struct draw_context *draw = fpme->draw; - struct draw_vertex_shader *shader = draw->vertex_shader; + struct draw_vertex_shader *shader = draw->vs.vertex_shader; unsigned opt = fpme->opt; unsigned alloc_count = align_int( fetch_count, 4 ); @@ -183,7 +183,7 @@ static void fetch_pipeline_linear_run( struct draw_pt_middle_end *middle, { struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle; struct draw_context *draw = fpme->draw; - struct draw_vertex_shader *shader = draw->vertex_shader; + struct draw_vertex_shader *shader = draw->vs.vertex_shader; unsigned opt = fpme->opt; unsigned alloc_count = align_int( count, 4 ); diff --git a/src/gallium/auxiliary/draw/draw_pt_middle_fse.c b/src/gallium/auxiliary/draw/draw_pt_middle_fse.c index cdb7d260da..643ea151c1 100644 --- a/src/gallium/auxiliary/draw/draw_pt_middle_fse.c +++ b/src/gallium/auxiliary/draw/draw_pt_middle_fse.c @@ -368,8 +368,8 @@ static void fse_prepare( struct draw_pt_middle_end *middle, { struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle; struct draw_context *draw = fse->draw; - unsigned num_vs_inputs = draw->vertex_shader->info.num_inputs; - unsigned num_vs_outputs = draw->vertex_shader->info.num_outputs; + unsigned num_vs_inputs = draw->vs.vertex_shader->info.num_inputs; + unsigned num_vs_outputs = draw->vs.vertex_shader->info.num_outputs; const struct vertex_info *vinfo; unsigned i; boolean need_psize = 0; diff --git a/src/gallium/auxiliary/draw/draw_vs.c b/src/gallium/auxiliary/draw/draw_vs.c index 03fe00a951..4142dd9589 100644 --- a/src/gallium/auxiliary/draw/draw_vs.c +++ b/src/gallium/auxiliary/draw/draw_vs.c @@ -66,13 +66,13 @@ draw_bind_vertex_shader(struct draw_context *draw, if (dvs) { - draw->vertex_shader = dvs; - draw->num_vs_outputs = dvs->info.num_outputs; + draw->vs.vertex_shader = dvs; + draw->vs.num_vs_outputs = dvs->info.num_outputs; dvs->prepare( dvs, draw ); } else { - draw->vertex_shader = NULL; - draw->num_vs_outputs = 0; + draw->vs.vertex_shader = NULL; + draw->vs.num_vs_outputs = 0; } } @@ -83,3 +83,30 @@ draw_delete_vertex_shader(struct draw_context *draw, { dvs->delete( dvs ); } + + + +boolean +draw_vs_init( struct draw_context *draw ) +{ + tgsi_exec_machine_init(&draw->vs.machine); + /* FIXME: give this machine thing a proper constructor: + */ + draw->vs.machine.Inputs = align_malloc(PIPE_MAX_ATTRIBS * sizeof(struct tgsi_exec_vector), 16); + draw->vs.machine.Outputs = align_malloc(PIPE_MAX_ATTRIBS * sizeof(struct tgsi_exec_vector), 16); + + return TRUE; +} + +void +draw_vs_destroy( struct draw_context *draw ) +{ + if (draw->vs.machine.Inputs) + align_free(draw->vs.machine.Inputs); + + if (draw->vs.machine.Outputs) + align_free(draw->vs.machine.Outputs); + + tgsi_exec_machine_free_data(&draw->vs.machine); + +} diff --git a/src/gallium/auxiliary/draw/draw_vs_exec.c b/src/gallium/auxiliary/draw/draw_vs_exec.c index 7a02f6334b..cb80d008cd 100644 --- a/src/gallium/auxiliary/draw/draw_vs_exec.c +++ b/src/gallium/auxiliary/draw/draw_vs_exec.c @@ -182,7 +182,7 @@ draw_create_vs_exec(struct draw_context *draw, vs->base.prepare = vs_exec_prepare; vs->base.run_linear = vs_exec_run_linear; vs->base.delete = vs_exec_delete; - vs->machine = &draw->machine; + vs->machine = &draw->vs.machine; return &vs->base; } diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c index edf235cddc..13ad032bd3 100644 --- a/src/gallium/auxiliary/draw/draw_vs_sse.c +++ b/src/gallium/auxiliary/draw/draw_vs_sse.c @@ -158,7 +158,7 @@ draw_create_vs_sse(struct draw_context *draw, vs->base.prepare = vs_sse_prepare; vs->base.run_linear = vs_sse_run_linear; vs->base.delete = vs_sse_delete; - vs->machine = &draw->machine; + vs->machine = &draw->vs.machine; x86_init_func( &vs->sse2_program ); -- cgit v1.2.3 From 7c99d7fe60e7bb0b7cf103a851aeef4614278ca6 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 15 May 2008 12:39:08 +0100 Subject: draw: create specialized vs varients incorporating fetch & emit --- src/gallium/auxiliary/draw/Makefile | 3 +- src/gallium/auxiliary/draw/draw_private.h | 2 + .../auxiliary/draw/draw_pt_fetch_shade_emit.c | 338 +++++++++++++++++++++ src/gallium/auxiliary/draw/draw_vs.c | 83 ++++- src/gallium/auxiliary/draw/draw_vs.h | 105 +++++++ src/gallium/auxiliary/draw/draw_vs_exec.c | 2 + src/gallium/auxiliary/draw/draw_vs_llvm.c | 2 + src/gallium/auxiliary/draw/draw_vs_sse.c | 2 + src/gallium/auxiliary/draw/draw_vs_varient.c | 229 ++++++++++++++ 9 files changed, 764 insertions(+), 2 deletions(-) create mode 100644 src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c create mode 100644 src/gallium/auxiliary/draw/draw_vs_varient.c (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/Makefile b/src/gallium/auxiliary/draw/Makefile index 3053682da8..84877994fb 100644 --- a/src/gallium/auxiliary/draw/Makefile +++ b/src/gallium/auxiliary/draw/Makefile @@ -26,7 +26,7 @@ C_SOURCES = \ draw_pt_emit.c \ draw_pt_fetch.c \ draw_pt_fetch_emit.c \ - draw_pt_middle_fse.c \ + draw_pt_fetch_shade_emit.c \ draw_pt_fetch_shade_pipeline.c \ draw_pt_post_vs.c \ draw_pt_util.c \ @@ -34,6 +34,7 @@ C_SOURCES = \ draw_pt_vcache.c \ draw_vertex.c \ draw_vs.c \ + draw_vs_varient.c \ draw_vs_exec.c \ draw_vs_llvm.c \ draw_vs_sse.c diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h index 3418ee2b88..c095bf3d7b 100644 --- a/src/gallium/auxiliary/draw/draw_private.h +++ b/src/gallium/auxiliary/draw/draw_private.h @@ -184,7 +184,9 @@ struct draw_context struct gallivm_cpu_engine *engine; + struct translate *fetch; struct translate_cache *fetch_cache; + struct translate *emit; struct translate_cache *emit_cache; } vs; diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c new file mode 100644 index 0000000000..74945dcfe9 --- /dev/null +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c @@ -0,0 +1,338 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + /* + * Authors: + * Keith Whitwell + */ + + +#include "pipe/p_util.h" +#include "draw/draw_context.h" +#include "draw/draw_private.h" +#include "draw/draw_vbuf.h" +#include "draw/draw_vertex.h" +#include "draw/draw_pt.h" +#include "draw/draw_vs.h" + +#include "translate/translate.h" + +struct fetch_shade_emit; + + +/* Prototype fetch, shade, emit-hw-verts all in one go. + */ +struct fetch_shade_emit { + struct draw_pt_middle_end base; + struct draw_context *draw; + + + /* Temporaries: + */ + const float *constants; + unsigned pitch[PIPE_MAX_ATTRIBS]; + const ubyte *src[PIPE_MAX_ATTRIBS]; + unsigned prim; + + struct draw_vs_varient_key key; + struct draw_vs_varient *active; +}; + + + + +static void fse_prepare( struct draw_pt_middle_end *middle, + unsigned prim, + unsigned opt ) +{ + struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle; + struct draw_context *draw = fse->draw; + unsigned num_vs_inputs = draw->vs.vertex_shader->info.num_inputs; + unsigned num_vs_outputs = draw->vs.vertex_shader->info.num_outputs; + const struct vertex_info *vinfo; + unsigned i; + boolean need_psize = 0; + + + if (draw->pt.user.elts) { + assert(0); + return ; + } + + if (!draw->render->set_primitive( draw->render, + prim )) { + assert(0); + return; + } + + /* Must do this after set_primitive() above: + */ + vinfo = draw->render->get_vertex_info(draw->render); + + + + fse->key.nr_elements = MAX2(num_vs_outputs, /* outputs - translate to hw format */ + num_vs_inputs); /* inputs - fetch from api format */ + + fse->key.output_stride = vinfo->size * 4; + memset(fse->key.element, 0, + fse->key.nr_elements * sizeof(fse->key.element[0])); + + for (i = 0; i < num_vs_inputs; i++) { + const struct pipe_vertex_element *src = &draw->pt.vertex_element[i]; + fse->key.element[i].in.format = src->src_format; + + /* Consider ignoring these, ie make generated programs + * independent of this state: + */ + fse->key.element[i].in.buffer = src->vertex_buffer_index; + fse->key.element[i].in.offset = src->src_offset; + } + + + { + unsigned dst_offset = 0; + + for (i = 0; i < vinfo->num_attribs; i++) { + unsigned emit_sz = 0; + unsigned output_format = PIPE_FORMAT_NONE; + unsigned vs_output = vinfo->src_index[i]; + + switch (vinfo->emit[i]) { + case EMIT_4F: + output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; + emit_sz = 4 * sizeof(float); + break; + case EMIT_3F: + output_format = PIPE_FORMAT_R32G32B32_FLOAT; + emit_sz = 3 * sizeof(float); + break; + case EMIT_2F: + output_format = PIPE_FORMAT_R32G32_FLOAT; + emit_sz = 2 * sizeof(float); + break; + case EMIT_1F: + output_format = PIPE_FORMAT_R32_FLOAT; + emit_sz = 1 * sizeof(float); + break; + case EMIT_1F_PSIZE: + need_psize = 1; + output_format = PIPE_FORMAT_R32_FLOAT; + emit_sz = 1 * sizeof(float); + vs_output = num_vs_outputs + 1; + + break; + default: + assert(0); + break; + } + + /* The elements in the key correspond to vertex shader output + * numbers, not to positions in the hw vertex description -- + * that's handled by the output_offset field. + */ + fse->key.element[vs_output].out.format = output_format; + fse->key.element[vs_output].out.offset = dst_offset; + + dst_offset += emit_sz; + assert(fse->key.output_stride >= dst_offset); + } + } + + /* To make psize work, really need to tell the vertex shader to + * copy that value from input->output. For 'translate' this was + * implicit for all elements. + */ +#if 0 + if (need_psize) { + unsigned input = num_vs_inputs + 1; + const struct pipe_vertex_element *src = &draw->pt.vertex_element[i]; + fse->key.element[i].input_format = PIPE_FORMAT_R32_FLOAT; + fse->key.element[i].input_buffer = 0; //nr_buffers + 1; + fse->key.element[i].input_offset = 0; + + fse->key.nr_elements += 1; + + } +#endif + + /* Would normally look up a vertex shader and peruse its list of + * varients somehow. We omitted that step and put all the + * hardcoded "shaders" into an array. We're just making the + * assumption that this happens to be a matching shader... ie + * you're running isosurf, aren't you? + */ + fse->active = draw_vs_lookup_varient( draw->vs.vertex_shader, + &fse->key ); + + if (!fse->active) { + assert(0); + return ; + } + + /* Now set buffer pointers: + */ + for (i = 0; i < num_vs_inputs; i++) { + unsigned buf = draw->pt.vertex_element[i].vertex_buffer_index; + + fse->active->set_input( fse->active, + i, + + ((const ubyte *) draw->pt.user.vbuffer[buf] + + draw->pt.vertex_buffer[buf].buffer_offset), + + draw->pt.vertex_buffer[buf].pitch ); + } + + fse->active->set_constants( fse->active, + (const float (*)[4])draw->pt.user.constants ); + + //return TRUE; +} + + + + + + + +static void fse_run_linear( struct draw_pt_middle_end *middle, + unsigned start, + unsigned count ) +{ + struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle; + struct draw_context *draw = fse->draw; + unsigned alloc_count = align(count, 4); + char *hw_verts; + + /* XXX: need to flush to get prim_vbuf.c to release its allocation?? + */ + draw_do_flush( draw, DRAW_FLUSH_BACKEND ); + + hw_verts = draw->render->allocate_vertices( draw->render, + (ushort)fse->key.output_stride, + (ushort)alloc_count ); + + if (!hw_verts) { + assert(0); + return; + } + + /* Single routine to fetch vertices, run shader and emit HW verts. + * Clipping and viewport transformation are done elsewhere -- + * either by the API or on hardware, or for some other reason not + * required... + */ + fse->active->run_linear( fse->active, + start, count, + hw_verts ); + + /* Draw arrays path to avoid re-emitting index list again and + * again. + */ + draw->render->draw_arrays( draw->render, + 0, + count ); + + + draw->render->release_vertices( draw->render, + hw_verts, + fse->key.output_stride, + count ); +} + + +static void +fse_run(struct draw_pt_middle_end *middle, + const unsigned *fetch_elts, + unsigned fetch_count, + const ushort *draw_elts, + unsigned draw_count ) +{ + struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle; + struct draw_context *draw = fse->draw; + unsigned alloc_count = align(fetch_count, 4); + void *hw_verts; + + /* XXX: need to flush to get prim_vbuf.c to release its allocation?? + */ + draw_do_flush( draw, DRAW_FLUSH_BACKEND ); + + hw_verts = draw->render->allocate_vertices( draw->render, + (ushort)fse->key.output_stride, + (ushort)alloc_count ); + if (!hw_verts) { + assert(0); + return; + } + + + /* Single routine to fetch vertices, run shader and emit HW verts. + */ + fse->active->run_elts( fse->active, + fetch_elts, + fetch_count, + hw_verts ); + + draw->render->draw( draw->render, + draw_elts, + draw_count ); + + draw->render->release_vertices( draw->render, + hw_verts, + fse->key.output_stride, + fetch_count ); + +} + + +static void fse_finish( struct draw_pt_middle_end *middle ) +{ +} + + +static void +fse_destroy( struct draw_pt_middle_end *middle ) +{ + FREE(middle); +} + +struct draw_pt_middle_end *draw_pt_middle_fse( struct draw_context *draw ) +{ + struct fetch_shade_emit *fse = CALLOC_STRUCT(fetch_shade_emit); + if (!fse) + return NULL; + + fse->base.prepare = fse_prepare; + fse->base.run = fse_run; + fse->base.run_linear = fse_run_linear; + fse->base.finish = fse_finish; + fse->base.destroy = fse_destroy; + fse->draw = draw; + + return &fse->base; +} diff --git a/src/gallium/auxiliary/draw/draw_vs.c b/src/gallium/auxiliary/draw/draw_vs.c index 4142dd9589..9b899d404e 100644 --- a/src/gallium/auxiliary/draw/draw_vs.c +++ b/src/gallium/auxiliary/draw/draw_vs.c @@ -36,6 +36,8 @@ #include "draw_private.h" #include "draw_context.h" #include "draw_vs.h" +#include "translate/translate.h" +#include "translate/translate_cache.h" @@ -90,11 +92,25 @@ boolean draw_vs_init( struct draw_context *draw ) { tgsi_exec_machine_init(&draw->vs.machine); + /* FIXME: give this machine thing a proper constructor: */ draw->vs.machine.Inputs = align_malloc(PIPE_MAX_ATTRIBS * sizeof(struct tgsi_exec_vector), 16); - draw->vs.machine.Outputs = align_malloc(PIPE_MAX_ATTRIBS * sizeof(struct tgsi_exec_vector), 16); + if (!draw->vs.machine.Inputs) + return FALSE; + draw->vs.machine.Outputs = align_malloc(PIPE_MAX_ATTRIBS * sizeof(struct tgsi_exec_vector), 16); + if (!draw->vs.machine.Outputs) + return FALSE; + + draw->vs.emit_cache = translate_cache_create(); + if (!draw->vs.emit_cache) + return FALSE; + + draw->vs.fetch_cache = translate_cache_create(); + if (!draw->vs.fetch_cache) + return FALSE; + return TRUE; } @@ -107,6 +123,71 @@ draw_vs_destroy( struct draw_context *draw ) if (draw->vs.machine.Outputs) align_free(draw->vs.machine.Outputs); + if (draw->vs.fetch_cache) + translate_cache_destroy(draw->vs.fetch_cache); + + if (draw->vs.emit_cache) + translate_cache_destroy(draw->vs.emit_cache); + tgsi_exec_machine_free_data(&draw->vs.machine); } + + +struct draw_vs_varient * +draw_vs_lookup_varient( struct draw_vertex_shader *vs, + const struct draw_vs_varient_key *key ) +{ + struct draw_vs_varient *varient; + unsigned i; + + /* Lookup existing varient: + */ + for (i = 0; i < vs->nr_varients; i++) + if (draw_vs_varient_key_compare(key, &vs->varient[i]->key) == 0) + return vs->varient[i]; + + /* Else have to create a new one: + */ + varient = vs->create_varient( vs, key ); + if (varient == NULL) + return NULL; + + /* Add it to our list: + */ + assert(vs->nr_varients < Elements(vs->varient)); + vs->varient[vs->nr_varients++] = varient; + + /* Done + */ + return varient; +} + + +struct translate * +draw_vs_get_fetch( struct draw_context *draw, + struct translate_key *key ) +{ + if (!draw->vs.fetch || + translate_key_compare(&draw->vs.fetch->key, key) != 0) + { + translate_key_sanitize(key); + draw->vs.fetch = translate_cache_find(draw->vs.fetch_cache, key); + } + + return draw->vs.fetch; +} + +struct translate * +draw_vs_get_emit( struct draw_context *draw, + struct translate_key *key ) +{ + if (!draw->vs.emit || + translate_key_compare(&draw->vs.emit->key, key) != 0) + { + translate_key_sanitize(key); + draw->vs.emit = translate_cache_find(draw->vs.emit_cache, key); + } + + return draw->vs.emit; +} diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h index f9772b83b8..677be0d28d 100644 --- a/src/gallium/auxiliary/draw/draw_vs.h +++ b/src/gallium/auxiliary/draw/draw_vs.h @@ -38,10 +38,63 @@ struct draw_context; struct pipe_shader_state; +struct draw_vs_input +{ + enum pipe_format format; + unsigned buffer; + unsigned offset; +}; + +struct draw_vs_output +{ + enum pipe_format format; + unsigned offset; +}; + +struct draw_vs_element { + struct draw_vs_input in; + struct draw_vs_output out; +}; + +struct draw_vs_varient_key { + unsigned output_stride; + unsigned nr_elements; + struct draw_vs_element element[PIPE_MAX_ATTRIBS]; +}; + +struct draw_vs_varient { + struct draw_vs_varient_key key; + + struct draw_vertex_shader *vs; + + void (*set_input)( struct draw_vs_varient *, + unsigned i, + const void *ptr, + unsigned stride ); + + void (*set_constants)( struct draw_vs_varient *, + const float (*constants)[4] ); + + + void (*run_linear)( struct draw_vs_varient *shader, + unsigned start, + unsigned count, + void *output_buffer ); + + void (*run_elts)( struct draw_vs_varient *shader, + const unsigned *elts, + unsigned count, + void *output_buffer ); + + void (*destroy)( struct draw_vs_varient * ); +}; + + /** * Private version of the compiled vertex_shader */ struct draw_vertex_shader { + struct draw_context *draw; /* This member will disappear shortly: */ @@ -49,6 +102,14 @@ struct draw_vertex_shader { struct tgsi_shader_info info; + /* + */ + struct draw_vs_varient *varient[16]; + unsigned nr_varients; + struct draw_vs_varient *(*create_varient)( struct draw_vertex_shader *shader, + const struct draw_vs_varient_key *key ); + + void (*prepare)( struct draw_vertex_shader *shader, struct draw_context *draw ); @@ -68,6 +129,15 @@ struct draw_vertex_shader { }; +struct draw_vs_varient * +draw_vs_lookup_varient( struct draw_vertex_shader *base, + const struct draw_vs_varient_key *key ); + + +/******************************************************************************** + * Internal functions: + */ + struct draw_vertex_shader * draw_create_vs_exec(struct draw_context *draw, const struct pipe_shader_state *templ); @@ -80,8 +150,43 @@ struct draw_vertex_shader * draw_create_vs_llvm(struct draw_context *draw, const struct pipe_shader_state *templ); +/******************************************************************************** + * Helpers for vs implementations that don't do their own fetch/emit varients. + * Means these can be shared between shaders. + */ +struct translate; +struct translate_key; + +struct translate *draw_vs_get_fetch( struct draw_context *draw, + struct translate_key *key ); + + +struct translate *draw_vs_get_emit( struct draw_context *draw, + struct translate_key *key ); + +struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs, + const struct draw_vs_varient_key *key ); + + + +static INLINE int draw_vs_varient_keysize( const struct draw_vs_varient_key *key ) +{ + return 2 * sizeof(int) + key->nr_elements * sizeof(struct draw_vs_element); +} + +static INLINE int draw_vs_varient_key_compare( const struct draw_vs_varient_key *a, + const struct draw_vs_varient_key *b ) +{ + int keysize = draw_vs_varient_keysize(a); + return memcmp(a, b, keysize); +} + + + + #define MAX_TGSI_VERTICES 4 + #endif diff --git a/src/gallium/auxiliary/draw/draw_vs_exec.c b/src/gallium/auxiliary/draw/draw_vs_exec.c index cb80d008cd..4501877efc 100644 --- a/src/gallium/auxiliary/draw/draw_vs_exec.c +++ b/src/gallium/auxiliary/draw/draw_vs_exec.c @@ -179,9 +179,11 @@ draw_create_vs_exec(struct draw_context *draw, tgsi_scan_shader(state->tokens, &vs->base.info); + vs->base.draw = draw; vs->base.prepare = vs_exec_prepare; vs->base.run_linear = vs_exec_run_linear; vs->base.delete = vs_exec_delete; + vs->base.create_varient = draw_vs_varient_generic; vs->machine = &draw->vs.machine; return &vs->base; diff --git a/src/gallium/auxiliary/draw/draw_vs_llvm.c b/src/gallium/auxiliary/draw/draw_vs_llvm.c index 171da51dd5..621472ec7c 100644 --- a/src/gallium/auxiliary/draw/draw_vs_llvm.c +++ b/src/gallium/auxiliary/draw/draw_vs_llvm.c @@ -114,7 +114,9 @@ draw_create_vs_llvm(struct draw_context *draw, tgsi_scan_shader(vs->base.state.tokens, &vs->base.info); + vs->base.draw = draw; vs->base.prepare = vs_llvm_prepare; + vs->base.create_varient = draw_vs_varient_generic; vs->base.run_linear = vs_llvm_run_linear; vs->base.delete = vs_llvm_delete; vs->machine = &draw->machine; diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c index 13ad032bd3..df94a7e0c7 100644 --- a/src/gallium/auxiliary/draw/draw_vs_sse.c +++ b/src/gallium/auxiliary/draw/draw_vs_sse.c @@ -155,6 +155,8 @@ draw_create_vs_sse(struct draw_context *draw, tgsi_scan_shader(templ->tokens, &vs->base.info); + vs->base.draw = draw; + vs->base.create_varient = draw_vs_varient_generic; vs->base.prepare = vs_sse_prepare; vs->base.run_linear = vs_sse_run_linear; vs->base.delete = vs_sse_delete; diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c new file mode 100644 index 0000000000..d27b0f6187 --- /dev/null +++ b/src/gallium/auxiliary/draw/draw_vs_varient.c @@ -0,0 +1,229 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + /* + * Authors: + * Keith Whitwell + */ + +#include "pipe/p_util.h" +#include "draw/draw_context.h" +#include "draw/draw_private.h" +#include "draw/draw_vbuf.h" +#include "draw/draw_vertex.h" +#include "draw/draw_vs.h" +#include "translate/translate.h" +#include "translate/translate_cache.h" + +/* A first pass at incorporating vertex fetch/emit functionality into + */ +struct draw_vs_varient_generic { + struct draw_vs_varient base; + + + + struct draw_vertex_shader *shader; + struct draw_context *draw; + + /* Basic plan is to run these two translate functions before/after + * the vertex shader's existing run_linear() routine to simulate + * the inclusion of this functionality into the shader... + * + * Next will look at actually including it. + */ + struct translate *fetch; + struct translate *emit; + + const float (*constants)[4]; +}; + + + + +static void vsvg_set_constants( struct draw_vs_varient *varient, + const float (*constants)[4] ) +{ + struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient; + + vsvg->constants = constants; +} + + +static void vsvg_set_input( struct draw_vs_varient *varient, + unsigned buffer, + const void *ptr, + unsigned stride ) +{ + struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient; + + vsvg->fetch->set_buffer(vsvg->fetch, + buffer, + ptr, + stride); +} + + +static void vsvg_run_elts( struct draw_vs_varient *varient, + const unsigned *elts, + unsigned count, + void *output_buffer) +{ + struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient; + + /* Want to do this in small batches for cache locality? + */ + + vsvg->fetch->run_elts( vsvg->fetch, + elts, + count, + output_buffer ); + + //if (!vsvg->base.vs->is_passthrough) + { + vsvg->base.vs->run_linear( vsvg->base.vs, + output_buffer, + output_buffer, + vsvg->constants, + count, + vsvg->base.key.output_stride, + vsvg->base.key.output_stride); + + //if (!vsvg->already_in_emit_format) + + vsvg->emit->set_buffer( vsvg->emit, + 0, + output_buffer, + vsvg->base.key.output_stride ); + + + vsvg->emit->run( vsvg->emit, + 0, count, + output_buffer ); + } +} + + +static void vsvg_run_linear( struct draw_vs_varient *varient, + unsigned start, + unsigned count, + void *output_buffer ) +{ + struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient; + + //debug_printf("%s %d %d\n", __FUNCTION__, start, count); + + + vsvg->fetch->run( vsvg->fetch, + start, + count, + output_buffer ); + + //if (!vsvg->base.vs->is_passthrough) + { + vsvg->base.vs->run_linear( vsvg->base.vs, + output_buffer, + output_buffer, + vsvg->constants, + count, + vsvg->base.key.output_stride, + vsvg->base.key.output_stride); + + //if (!vsvg->already_in_emit_format) + vsvg->emit->set_buffer( vsvg->emit, + 0, + output_buffer, + vsvg->base.key.output_stride ); + + + vsvg->emit->run( vsvg->emit, + 0, count, + output_buffer ); + } +} + + + +static void vsvg_destroy( struct draw_vs_varient *varient ) +{ + FREE(varient); +} + + +struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs, + const struct draw_vs_varient_key *key ) +{ + unsigned i; + struct translate_key fetch, emit; + + struct draw_vs_varient_generic *vsvg = CALLOC_STRUCT( draw_vs_varient_generic ); + if (vsvg == NULL) + return NULL; + + vsvg->base.key = *key; + vsvg->base.vs = vs; + vsvg->base.set_input = vsvg_set_input; + vsvg->base.set_constants = vsvg_set_constants; + vsvg->base.run_elts = vsvg_run_elts; + vsvg->base.run_linear = vsvg_run_linear; + vsvg->base.destroy = vsvg_destroy; + + + + /* OK, have to build a new one: + */ + fetch.nr_elements = vs->info.num_inputs; + fetch.output_stride = 0; + for (i = 0; i < vs->info.num_inputs; i++) { + fetch.element[i].input_format = key->element[i].in.format; + fetch.element[i].input_buffer = key->element[i].in.buffer; + fetch.element[i].input_offset = key->element[i].in.offset; + fetch.element[i].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; + fetch.element[i].output_offset = fetch.output_stride; + fetch.output_stride += 4 * sizeof(float); + } + + + emit.nr_elements = vs->info.num_outputs; + emit.output_stride = key->output_stride; + for (i = 0; i < vs->info.num_outputs; i++) { + emit.element[i].input_format = PIPE_FORMAT_R32G32B32A32_FLOAT; + emit.element[i].input_buffer = 0; + emit.element[i].input_offset = i * 4 * sizeof(float); + emit.element[i].output_format = key->element[i].out.format; + emit.element[i].output_offset = key->element[i].out.offset; + } + + vsvg->fetch = draw_vs_get_fetch( vs->draw, &fetch ); + vsvg->emit = draw_vs_get_emit( vs->draw, &emit ); + + return &vsvg->base; +} + + + + + -- cgit v1.2.3 From 9232f0c023af060b12f77dee5e8b6a533c48e146 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Mon, 19 May 2008 16:28:53 +0100 Subject: rtasm: remove unused struct member --- src/gallium/auxiliary/rtasm/rtasm_x86sse.h | 1 - 1 file changed, 1 deletion(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h index eacaeeaf6f..baa10b7d4a 100644 --- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h +++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h @@ -44,7 +44,6 @@ struct x86_function { unsigned stack_offset; int need_emms; unsigned char error_overflow[4]; - const char *fn; }; enum x86_reg_file { -- cgit v1.2.3 From 8618e6aa16bdba2c8b08124261bbaedaf7e22447 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 20 May 2008 14:34:06 +0100 Subject: translate: remove spurious comment --- src/gallium/auxiliary/translate/translate_sse.c | 8 +------- 1 file changed, 1 insertion(+), 7 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/translate/translate_sse.c b/src/gallium/auxiliary/translate/translate_sse.c index a54ac5a82f..582d6f6466 100644 --- a/src/gallium/auxiliary/translate/translate_sse.c +++ b/src/gallium/auxiliary/translate/translate_sse.c @@ -472,13 +472,7 @@ static boolean build_vertex_emit( struct translate_sse *p, x86_lea(p->func, vertexECX, x86_make_disp(vertexECX, p->translate.key.output_stride)); /* Incr index - */ /* Emit code for each of the attributes. Currently routes - * everything through SSE registers, even when it might be more - * efficient to stick with regular old x86. No optimization or - * other tricks - enough new ground to cover here just getting - * things working. - */ - + */ if (linear) { x86_inc(p->func, idxEBX); } -- cgit v1.2.3 From d3e64caef6f8654af1a84825803e517ab8221c68 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Wed, 21 May 2008 08:28:16 +0100 Subject: rtasm: export debug reg print function --- src/gallium/auxiliary/rtasm/rtasm_x86sse.c | 20 +++++++++----------- src/gallium/auxiliary/rtasm/rtasm_x86sse.h | 3 +++ 2 files changed, 12 insertions(+), 11 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c index 4e036d9032..68ac91ed13 100644 --- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c +++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c @@ -36,11 +36,8 @@ #define DUMP_SSE 0 -#if DUMP_SSE -static void -_print_reg( - struct x86_reg reg ) +void x86_print_reg( struct x86_reg reg ) { if (reg.mod != mod_REG) debug_printf( "[" ); @@ -77,6 +74,7 @@ _print_reg( debug_printf( "]" ); } +#if DUMP_SSE #define DUMP_START() debug_printf( "\n" ) #define DUMP_END() debug_printf( "\n" ) @@ -87,7 +85,7 @@ _print_reg( foo++; \ if (*foo) \ foo++; \ - debug_printf( "\n% 15s ", foo ); \ + debug_printf( "\n% 4x% 15s ", p->csr - p->store, foo ); \ } while (0) #define DUMP_I( I ) do { \ @@ -97,27 +95,27 @@ _print_reg( #define DUMP_R( R0 ) do { \ DUMP(); \ - _print_reg( R0 ); \ + x86_print_reg( R0 ); \ } while( 0 ) #define DUMP_RR( R0, R1 ) do { \ DUMP(); \ - _print_reg( R0 ); \ + x86_print_reg( R0 ); \ debug_printf( ", " ); \ - _print_reg( R1 ); \ + x86_print_reg( R1 ); \ } while( 0 ) #define DUMP_RI( R0, I ) do { \ DUMP(); \ - _print_reg( R0 ); \ + x86_print_reg( R0 ); \ debug_printf( ", %u", I ); \ } while( 0 ) #define DUMP_RRI( R0, R1, I ) do { \ DUMP(); \ - _print_reg( R0 ); \ + x86_print_reg( R0 ); \ debug_printf( ", " ); \ - _print_reg( R1 ); \ + x86_print_reg( R1 ); \ debug_printf( ", %u", I ); \ } while( 0 ) diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h index baa10b7d4a..1e02c6e73b 100644 --- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h +++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h @@ -106,6 +106,9 @@ void x86_init_func_size( struct x86_function *p, unsigned code_size ); void x86_release_func( struct x86_function *p ); void (*x86_get_func( struct x86_function *p ))( void ); +/* Debugging: + */ +void x86_print_reg( struct x86_reg reg ); /* Create and manipulate registers and regmem values: -- cgit v1.2.3 From b5c8b3fba6ac90a0d83e02bfe432142f1adee9e5 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Wed, 21 May 2008 08:29:19 +0100 Subject: translate: mark functions as PIPE_CDECL --- src/gallium/auxiliary/translate/translate.h | 18 +++++++++--------- src/gallium/auxiliary/translate/translate_generic.c | 16 ++++++++-------- src/gallium/auxiliary/translate/translate_sse.c | 20 +++++++------------- 3 files changed, 24 insertions(+), 30 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/translate/translate.h b/src/gallium/auxiliary/translate/translate.h index b8210af50c..c3b754a902 100644 --- a/src/gallium/auxiliary/translate/translate.h +++ b/src/gallium/auxiliary/translate/translate.h @@ -71,15 +71,15 @@ struct translate { const void *ptr, unsigned stride ); - void (*run_elts)( struct translate *, - const unsigned *elts, - unsigned count, - void *output_buffer); - - void (*run)( struct translate *, - unsigned start, - unsigned count, - void *output_buffer); + void (PIPE_CDECL *run_elts)( struct translate *, + const unsigned *elts, + unsigned count, + void *output_buffer); + + void (PIPE_CDECL *run)( struct translate *, + unsigned start, + unsigned count, + void *output_buffer); }; diff --git a/src/gallium/auxiliary/translate/translate_generic.c b/src/gallium/auxiliary/translate/translate_generic.c index 402780ee53..a25d94f2ca 100644 --- a/src/gallium/auxiliary/translate/translate_generic.c +++ b/src/gallium/auxiliary/translate/translate_generic.c @@ -541,10 +541,10 @@ static emit_func get_emit_func( enum pipe_format format ) /** * Fetch vertex attributes for 'count' vertices. */ -static void generic_run_elts( struct translate *translate, - const unsigned *elts, - unsigned count, - void *output_buffer ) +static void PIPE_CDECL generic_run_elts( struct translate *translate, + const unsigned *elts, + unsigned count, + void *output_buffer ) { struct translate_generic *tg = translate_generic(translate); char *vert = output_buffer; @@ -580,10 +580,10 @@ static void generic_run_elts( struct translate *translate, -static void generic_run( struct translate *translate, - unsigned start, - unsigned count, - void *output_buffer ) +static void PIPE_CDECL generic_run( struct translate *translate, + unsigned start, + unsigned count, + void *output_buffer ) { struct translate_generic *tg = translate_generic(translate); char *vert = output_buffer; diff --git a/src/gallium/auxiliary/translate/translate_sse.c b/src/gallium/auxiliary/translate/translate_sse.c index 582d6f6466..2fc8b9d3d0 100644 --- a/src/gallium/auxiliary/translate/translate_sse.c +++ b/src/gallium/auxiliary/translate/translate_sse.c @@ -45,22 +45,16 @@ #define W 3 -#ifdef WIN32 -#define RTASM __cdecl -#else -#define RTASM -#endif - -typedef void (RTASM *run_func)( struct translate *translate, - unsigned start, - unsigned count, - void *output_buffer ); - -typedef void (RTASM *run_elts_func)( struct translate *translate, - const unsigned *elts, +typedef void (PIPE_CDECL *run_func)( struct translate *translate, + unsigned start, unsigned count, void *output_buffer ); +typedef void (PIPE_CDECL *run_elts_func)( struct translate *translate, + const unsigned *elts, + unsigned count, + void *output_buffer ); + struct translate_sse { -- cgit v1.2.3 From ba738a3135415de8b381cd8845cd6c435d5747a8 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Wed, 21 May 2008 09:43:30 +0100 Subject: draw: mark varient functions as PIPE_CDECL --- src/gallium/auxiliary/draw/draw_vs.h | 30 +++++++++++++++++++++--------- 1 file changed, 21 insertions(+), 9 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h index 677be0d28d..6bfc2c8d75 100644 --- a/src/gallium/auxiliary/draw/draw_vs.h +++ b/src/gallium/auxiliary/draw/draw_vs.h @@ -62,6 +62,19 @@ struct draw_vs_varient_key { struct draw_vs_element element[PIPE_MAX_ATTRIBS]; }; +struct draw_vs_varient; + +typedef void (PIPE_CDECL *vsv_run_elts_func)( struct draw_vs_varient *, + const unsigned *elts, + unsigned count, + void *output_buffer); + +typedef void (PIPE_CDECL *vsv_run_linear_func)( struct draw_vs_varient *, + unsigned start, + unsigned count, + void *output_buffer); + + struct draw_vs_varient { struct draw_vs_varient_key key; @@ -75,16 +88,15 @@ struct draw_vs_varient { void (*set_constants)( struct draw_vs_varient *, const float (*constants)[4] ); + void (PIPE_CDECL *run_linear)( struct draw_vs_varient *shader, + unsigned start, + unsigned count, + void *output_buffer ); - void (*run_linear)( struct draw_vs_varient *shader, - unsigned start, - unsigned count, - void *output_buffer ); - - void (*run_elts)( struct draw_vs_varient *shader, - const unsigned *elts, - unsigned count, - void *output_buffer ); + void (PIPE_CDECL *run_elts)( struct draw_vs_varient *shader, + const unsigned *elts, + unsigned count, + void *output_buffer ); void (*destroy)( struct draw_vs_varient * ); }; -- cgit v1.2.3 From 1ba10e5ccf5cd0c990922e982e1e9bc6be48a5e4 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Wed, 21 May 2008 09:44:16 +0100 Subject: draw: add aos vertex shader varient --- src/gallium/auxiliary/draw/Makefile | 2 + src/gallium/auxiliary/draw/draw_vs.h | 10 + src/gallium/auxiliary/draw/draw_vs_aos.c | 1739 +++++++++++++++++++++++++++ src/gallium/auxiliary/draw/draw_vs_aos.h | 181 +++ src/gallium/auxiliary/draw/draw_vs_aos_io.c | 314 +++++ src/gallium/auxiliary/draw/draw_vs_sse.c | 1 + 6 files changed, 2247 insertions(+) create mode 100644 src/gallium/auxiliary/draw/draw_vs_aos.c create mode 100644 src/gallium/auxiliary/draw/draw_vs_aos.h create mode 100644 src/gallium/auxiliary/draw/draw_vs_aos_io.c (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/Makefile b/src/gallium/auxiliary/draw/Makefile index 84877994fb..9a88ecc070 100644 --- a/src/gallium/auxiliary/draw/Makefile +++ b/src/gallium/auxiliary/draw/Makefile @@ -35,6 +35,8 @@ C_SOURCES = \ draw_vertex.c \ draw_vs.c \ draw_vs_varient.c \ + draw_vs_aos.c \ + draw_vs_aos_io.c \ draw_vs_exec.c \ draw_vs_llvm.c \ draw_vs_sse.c diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h index 6bfc2c8d75..5a8d0da06d 100644 --- a/src/gallium/auxiliary/draw/draw_vs.h +++ b/src/gallium/auxiliary/draw/draw_vs.h @@ -162,6 +162,16 @@ struct draw_vertex_shader * draw_create_vs_llvm(struct draw_context *draw, const struct pipe_shader_state *templ); + + +struct draw_vs_varient_key; +struct draw_vertex_shader; + +struct draw_vs_varient *draw_vs_varient_aos_sse( struct draw_vertex_shader *vs, + const struct draw_vs_varient_key *key ); + + + /******************************************************************************** * Helpers for vs implementations that don't do their own fetch/emit varients. * Means these can be shared between shaders. diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c new file mode 100644 index 0000000000..620f5e3592 --- /dev/null +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -0,0 +1,1739 @@ +/* + * Mesa 3-D graphics library + * Version: 6.3 + * + * Copyright (C) 1999-2004 Brian Paul All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included + * in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN + * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +/** + * Translate tgsi vertex programs to x86/x87/SSE/SSE2 machine code + * using the rtasm runtime assembler. Based on the old + * t_vb_arb_program_sse.c + */ + + +#include "pipe/p_util.h" +#include "pipe/p_shader_tokens.h" +#include "tgsi/util/tgsi_parse.h" +#include "tgsi/util/tgsi_util.h" +#include "tgsi/exec/tgsi_exec.h" +#include "tgsi/util/tgsi_dump.h" + +#include "draw_vs.h" +#include "draw_vs_aos.h" + +#include "rtasm/rtasm_x86sse.h" + +#ifdef PIPE_ARCH_X86 + + +#define DISASSEM 0 + + + + + +static INLINE boolean eq( struct x86_reg a, + struct x86_reg b ) +{ + return (a.file == b.file && + a.idx == b.idx && + a.mod == b.mod && + a.disp == b.disp); +} + + +static struct x86_reg get_reg_ptr(struct aos_compilation *cp, + unsigned file, + unsigned idx ) +{ + struct x86_reg ptr = cp->machine_EDX; + + switch (file) { + case TGSI_FILE_INPUT: + return x86_make_disp(ptr, Offset(struct aos_machine, input[idx])); + + case TGSI_FILE_OUTPUT: + return x86_make_disp(ptr, Offset(struct aos_machine, output[idx])); + + case TGSI_FILE_TEMPORARY: + return x86_make_disp(ptr, Offset(struct aos_machine, temp[idx])); + + case TGSI_FILE_IMMEDIATE: + return x86_make_disp(ptr, Offset(struct aos_machine, immediate[idx])); + + case TGSI_FILE_CONSTANT: + return x86_make_disp(ptr, Offset(struct aos_machine, constant[idx])); + + case AOS_FILE_INTERNAL: + return x86_make_disp(ptr, Offset(struct aos_machine, immediate[idx])); + + default: + ERROR(cp, "unknown reg file"); + return x86_make_reg(0,0); + } +} + + +struct x86_reg aos_get_internal( struct aos_compilation *cp, + unsigned imm ) +{ + return get_reg_ptr( cp, + AOS_FILE_INTERNAL, + imm + 1 ); +} + +static void spill( struct aos_compilation *cp, unsigned idx ) +{ + if (!cp->xmm[idx].dirty || + (cp->xmm[idx].file != TGSI_FILE_INPUT && /* inputs are fetched into xmm & set dirty */ + cp->xmm[idx].file != TGSI_FILE_OUTPUT && + cp->xmm[idx].file != TGSI_FILE_TEMPORARY)) { + ERROR(cp, "invalid spill"); + return; + } + else { + struct x86_reg oldval = get_reg_ptr(cp, + cp->xmm[idx].file, + cp->xmm[idx].idx); + + assert(cp->xmm[idx].dirty); + sse_movups(cp->func, oldval, x86_make_reg(file_XMM, idx)); + cp->xmm[idx].dirty = 0; + } +} + +struct x86_reg aos_get_xmm_reg( struct aos_compilation *cp ) +{ + unsigned i; + unsigned oldest = 0; + + for (i = 0; i < 8; i++) + if (cp->xmm[i].last_used < cp->xmm[oldest].last_used) + oldest = i; + + /* Need to write out the old value? + */ + if (cp->xmm[oldest].dirty) + spill(cp, oldest); + + assert(cp->xmm[oldest].last_used != cp->insn_counter); + + cp->xmm[oldest].file = TGSI_FILE_NULL; + cp->xmm[oldest].idx = 0; + cp->xmm[oldest].last_used = cp->insn_counter; + return x86_make_reg(file_XMM, oldest); +} + +void aos_release_xmm_reg( struct aos_compilation *cp, + unsigned idx ) +{ + cp->xmm[idx].file = TGSI_FILE_NULL; + cp->xmm[idx].idx = 0; + cp->xmm[idx].dirty = 0; + cp->xmm[idx].last_used = 0; +} + +static void invalidate_xmm( struct aos_compilation *cp, + unsigned file, unsigned idx ) +{ + unsigned i; + + /* Invalidate any old copy of this register in XMM0-7. + */ + for (i = 0; i < 8; i++) { + if (cp->xmm[i].file == file && cp->xmm[i].idx == idx) { + + if (cp->xmm[i].dirty) + spill(cp, i); + + aos_release_xmm_reg(cp, i); + break; + } + } + + for (; i < 8; i++) { + if (cp->xmm[i].file == file && cp->xmm[i].idx == idx) { + assert(0); + } + } +} + + +void aos_adopt_xmm_reg( struct aos_compilation *cp, + struct x86_reg reg, + unsigned file, + unsigned idx, + unsigned dirty ) +{ + if (reg.file != file_XMM) { + assert(0); + return; + } + + invalidate_xmm(cp, file, idx); + cp->xmm[reg.idx].file = file; + cp->xmm[reg.idx].idx = idx; + cp->xmm[reg.idx].dirty = dirty; +} + + + +static struct x86_reg aos_get_shader_reg_ptr( struct aos_compilation *cp, + unsigned file, + unsigned idx ) +{ + invalidate_xmm( cp, file, idx ); + return get_reg_ptr( cp, file, idx ); +} + + +/* As above, but return a pointer. Note - this pointer may alias + * those returned by get_arg_ptr(). + */ +static struct x86_reg get_dst_ptr( struct aos_compilation *cp, + const struct tgsi_full_dst_register *dst ) +{ + return aos_get_shader_reg_ptr( cp, dst->DstRegister.File, dst->DstRegister.Index ); +} + + + + + +/* Return an XMM reg if the argument is resident, otherwise return a + * base+offset pointer to the saved value. + */ +struct x86_reg aos_get_shader_reg( struct aos_compilation *cp, + unsigned file, + unsigned idx ) +{ + unsigned i; + + for (i = 0; i < 8; i++) { + if (cp->xmm[i].file == file && + cp->xmm[i].idx == idx) + { + cp->xmm[i].last_used = cp->insn_counter; + return x86_make_reg(file_XMM, i); + } + } + + /* If not found in the XMM register file, return an indirect + * reference to the in-memory copy: + */ + return get_reg_ptr( cp, file, idx ); +} + + + + + +/* Emulate pshufd insn in regular SSE, if necessary: + */ +static void emit_pshufd( struct aos_compilation *cp, + struct x86_reg dst, + struct x86_reg arg0, + ubyte shuf ) +{ + if (cp->have_sse2) { + sse2_pshufd(cp->func, dst, arg0, shuf); + } + else { + if (!eq(dst, arg0)) + sse_movups(cp->func, dst, arg0); + + sse_shufps(cp->func, dst, dst, shuf); + } +} + + + + +/* Helper for writemask: + */ +static boolean emit_shuf_copy1( struct aos_compilation *cp, + struct x86_reg dst, + struct x86_reg arg0, + struct x86_reg arg1, + ubyte shuf ) +{ + struct x86_reg tmp = aos_get_xmm_reg(cp); + sse_movups(cp->func, dst, arg1); + emit_pshufd(cp, dst, dst, shuf); + emit_pshufd(cp, tmp, arg0, shuf); + + sse_movss(cp->func, dst, tmp); + + emit_pshufd(cp, dst, dst, shuf); + + aos_release_xmm_reg(cp, tmp.idx); + return TRUE; +} + + +/* Helper for writemask: + */ +static boolean emit_shuf_copy2( struct aos_compilation *cp, + struct x86_reg dst, + struct x86_reg arg0, + struct x86_reg arg1, + ubyte shuf ) +{ + struct x86_reg tmp = aos_get_xmm_reg(cp); + emit_pshufd(cp, dst, arg1, shuf); + emit_pshufd(cp, tmp, arg0, shuf); + + sse_shufps(cp->func, dst, tmp, SHUF(X, Y, Z, W)); + + emit_pshufd(cp, dst, dst, shuf); + + aos_release_xmm_reg(cp, tmp.idx); + return TRUE; +} + +#define SSE_SWIZZLE_NOOP ((0<<0) | (1<<2) | (2<<4) | (3<<6)) + + +/* Locate a source register and perform any required (simple) swizzle. + * + * Just fail on complex swizzles at this point. + */ +static struct x86_reg fetch_src( struct aos_compilation *cp, + const struct tgsi_full_src_register *src ) +{ + struct x86_reg arg0 = aos_get_shader_reg(cp, + src->SrcRegister.File, + src->SrcRegister.Index); + unsigned i; + unsigned swz = 0; + unsigned negs = 0; + unsigned abs = 0; + + for (i = 0; i < 4; i++) { + unsigned swizzle = tgsi_util_get_full_src_register_extswizzle( src, i ); + unsigned neg = tgsi_util_get_full_src_register_sign_mode( src, i ); + + switch (swizzle) { + case TGSI_EXTSWIZZLE_ZERO: + case TGSI_EXTSWIZZLE_ONE: + ERROR(cp, "not supporting full swizzles yet in tgsi_aos_sse2"); + break; + + default: + swz |= (swizzle & 0x3) << (i * 2); + break; + } + + switch (neg) { + case TGSI_UTIL_SIGN_TOGGLE: + negs |= (1<func, dst, arg0); + + aos_release_xmm_reg(cp, tmp.idx); + arg0 = dst; + } + + if (abs && abs != 0xf) { + ERROR(cp, "unsupported partial abs"); + } + + if (abs) { + struct x86_reg neg = aos_get_internal(cp, IMM_NEGS); + struct x86_reg tmp = aos_get_xmm_reg(cp); + + sse_movups(cp->func, tmp, arg0); + sse_mulps(cp->func, tmp, neg); + sse_maxps(cp->func, dst, arg0); + + aos_release_xmm_reg(cp, tmp.idx); + arg0 = dst; + } + } + + return arg0; +} + +static void x87_fld_src( struct aos_compilation *cp, + const struct tgsi_full_src_register *src, + unsigned channel ) +{ + struct x86_reg arg0 = aos_get_shader_reg_ptr(cp, + src->SrcRegister.File, + src->SrcRegister.Index); + + unsigned swizzle = tgsi_util_get_full_src_register_extswizzle( src, channel ); + unsigned neg = tgsi_util_get_full_src_register_sign_mode( src, channel ); + + switch (swizzle) { + case TGSI_EXTSWIZZLE_ZERO: + x87_fldz( cp->func ); + break; + + case TGSI_EXTSWIZZLE_ONE: + x87_fld1( cp->func ); + break; + + default: + x87_fld( cp->func, x86_make_disp(arg0, (swizzle & 3) * sizeof(float)) ); + break; + } + + + switch (neg) { + case TGSI_UTIL_SIGN_TOGGLE: + /* Flip the sign: + */ + x87_fchs( cp->func ); + break; + + case TGSI_UTIL_SIGN_KEEP: + break; + + case TGSI_UTIL_SIGN_CLEAR: + x87_fabs( cp->func ); + break; + + case TGSI_UTIL_SIGN_SET: + x87_fabs( cp->func ); + x87_fchs( cp->func ); + break; + + default: + ERROR(cp, "unsupported sign-mode"); + break; + } +} + + + + + + +/* Used to implement write masking. This and most of the other instructions + * here would be easier to implement if there had been a translation + * to a 2 argument format (dst/arg0, arg1) at the shader level before + * attempting to translate to x86/sse code. + */ +static void store_dest( struct aos_compilation *cp, + const struct tgsi_full_dst_register *reg, + struct x86_reg result ) +{ + if (reg->DstRegister.WriteMask == 0) + { + return; + } + else if (reg->DstRegister.WriteMask == TGSI_WRITEMASK_XYZW) + { + if (result.file == file_XMM) { + aos_adopt_xmm_reg(cp, + result, + reg->DstRegister.File, + reg->DstRegister.Index, + TRUE); + } + else { + struct x86_reg dst = aos_get_xmm_reg(cp); + aos_adopt_xmm_reg(cp, + dst, + reg->DstRegister.File, + reg->DstRegister.Index, + TRUE); + sse_movups(cp->func, dst, result); + } + } + else + { + /* Previous value of the dest register: + */ + struct x86_reg old_dst = aos_get_shader_reg(cp, + reg->DstRegister.File, + reg->DstRegister.Index); + + + /* Alloc an xmm reg to hold the new value of the dest register: + */ + struct x86_reg dst = aos_get_xmm_reg(cp); + + aos_adopt_xmm_reg(cp, + dst, + reg->DstRegister.File, + reg->DstRegister.Index, + TRUE ); + + switch (reg->DstRegister.WriteMask) { + case TGSI_WRITEMASK_X: + if (result.file == file_XMM) { + sse_movups(cp->func, dst, old_dst); + sse_movss(cp->func, dst, result); + } + else { + struct x86_reg tmp = aos_get_xmm_reg(cp); + sse_movups(cp->func, dst, old_dst); + sse_movss(cp->func, tmp, result); + sse_movss(cp->func, dst, tmp); + aos_release_xmm_reg(cp, tmp.idx); + } + break; + + case TGSI_WRITEMASK_XY: + sse_movups(cp->func, dst, old_dst); + sse_shufps(cp->func, dst, result, SHUF(X, Y, Z, W)); + break; + + case TGSI_WRITEMASK_ZW: + sse_movups(cp->func, dst, result); + sse_shufps(cp->func, dst, old_dst, SHUF(X, Y, Z, W)); + break; + + case TGSI_WRITEMASK_YZW: + if (old_dst.file == file_XMM) { + sse_movups(cp->func, dst, result); + sse_movss(cp->func, dst, old_dst); + } + else { + struct x86_reg tmp = aos_get_xmm_reg(cp); + sse_movups(cp->func, dst, result); + sse_movss(cp->func, tmp, old_dst); + sse_movss(cp->func, dst, tmp); + aos_release_xmm_reg(cp, tmp.idx); + } + break; + + case TGSI_WRITEMASK_Y: + emit_shuf_copy1(cp, dst, result, old_dst, SHUF(Y,X,Z,W)); + break; + + case TGSI_WRITEMASK_Z: + emit_shuf_copy1(cp, dst, result, old_dst, SHUF(Z,Y,X,W)); + break; + + case TGSI_WRITEMASK_W: + emit_shuf_copy1(cp, dst, result, old_dst, SHUF(W,Y,Z,X)); + break; + + case TGSI_WRITEMASK_XZ: + emit_shuf_copy2(cp, dst, result, old_dst, SHUF(X,Z,Y,W)); + break; + + case TGSI_WRITEMASK_XW: + emit_shuf_copy2(cp, dst, result, old_dst, SHUF(X,W,Z,Y)); + + case TGSI_WRITEMASK_YZ: + emit_shuf_copy2(cp, dst, result, old_dst, SHUF(Z,Y,X,W)); + break; + + case TGSI_WRITEMASK_YW: + emit_shuf_copy2(cp, dst, result, old_dst, SHUF(W,Y,Z,X)); + break; + + case TGSI_WRITEMASK_XZW: + emit_shuf_copy1(cp, dst, old_dst, result, SHUF(Y,X,Z,W)); + break; + + case TGSI_WRITEMASK_XYW: + emit_shuf_copy1(cp, dst, old_dst, result, SHUF(Z,Y,X,W)); + break; + + case TGSI_WRITEMASK_XYZ: + emit_shuf_copy1(cp, dst, old_dst, result, SHUF(W,Y,Z,X)); + break; + + default: + assert(0); /* not possible */ + break; + } + } +} + + +static void x87_fst_or_nop( struct x86_function *func, + unsigned writemask, + unsigned channel, + struct x86_reg ptr ) +{ + if (writemask & (1<DstRegister.WriteMask; + + x87_fst_or_nop(cp->func, writemask, 0, ptr); + x87_fst_or_nop(cp->func, writemask, 1, ptr); + x87_fst_or_nop(cp->func, writemask, 2, ptr); + x87_fstp_or_pop(cp->func, writemask, 3, ptr); +} + +/* Save current x87 state and put it into single precision mode. + */ +static void save_fpu_state( struct aos_compilation *cp ) +{ +#if 0 + x87_fnstcw( cp->func, x86_make_disp(regEDX, get_offset(m, &m->fpucntl_restore))); + x87_fldcw( cp->func, ); +#endif +} + +static void restore_fpu_state( struct aos_compilation *cp ) +{ +#if 0 + x87_fnclex(cp->func); + x87_fldcw(cp->func, x86_make_disp(regEDX, get_offset(m, &m->fpucntl_restore))); +#endif +} + +static void set_fpu_round_neg_inf( struct aos_compilation *cp ) +{ +#if 0 + if (cp->fpucntl != RND_NEG_FPU) { + struct x86_reg regEDX = x86_make_reg(file_REG32, reg_DX); + struct arb_vp_machine *m = NULL; + + cp->fpucntl = RND_NEG_FPU; + x87_fnclex(cp->func); + x87_fldcw(cp->func, x86_make_disp(regEDX, get_offset(m, &m->fpucntl_rnd_neg))); + } +#endif +} + +static void set_fpu_round_nearest( struct aos_compilation *cp ) +{ +#if 0 +#endif +} + + +static void emit_x87_ex2( struct aos_compilation *cp ) +{ + struct x86_reg st0 = x86_make_reg(file_x87, 0); + struct x86_reg st1 = x86_make_reg(file_x87, 1); + struct x86_reg st3 = x86_make_reg(file_x87, 3); + + set_fpu_round_neg_inf( cp ); + + x87_fld(cp->func, st0); /* a a */ + x87_fprndint( cp->func ); /* int(a) a */ + x87_fld(cp->func, st0); /* int(a) int(a) a */ + x87_fstp(cp->func, st3); /* int(a) a int(a)*/ + x87_fsubp(cp->func, st1); /* frac(a) int(a) */ + x87_f2xm1(cp->func); /* (2^frac(a))-1 int(a)*/ + x87_fld1(cp->func); /* 1 (2^frac(a))-1 int(a)*/ + x87_faddp(cp->func, st1); /* 2^frac(a) int(a) */ + x87_fscale(cp->func); /* 2^a */ +} + + + +/** + * The traditional instructions. All operate on internal registers + * and ignore write masks and swizzling issues. + */ + +static boolean emit_ABS( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); + struct x86_reg dst = aos_get_xmm_reg(cp); + struct x86_reg neg = aos_get_internal(cp, IMM_NEGS); + + sse_movups(cp->func, dst, arg0); + sse_mulps(cp->func, dst, neg); + sse_maxps(cp->func, dst, arg0); + + store_dest(cp, &op->FullDstRegisters[0], dst); + return TRUE; +} + +static boolean emit_ADD( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); + struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); + struct x86_reg dst = aos_get_xmm_reg(cp); + + sse_movups(cp->func, dst, arg0); + sse_addps(cp->func, dst, arg1); + + store_dest(cp, &op->FullDstRegisters[0], dst); + return TRUE; +} + +static boolean emit_COS( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + x87_fld_src(cp, &op->FullSrcRegisters[0], 0); + x87_fcos(cp->func); + x87_fstp_dest4(cp, &op->FullDstRegisters[0]); + return TRUE; +} + + +/* The dotproduct instructions don't really do that well in sse: + */ +static boolean emit_DP3( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); + struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); + struct x86_reg dst = aos_get_xmm_reg(cp); + struct x86_reg tmp = aos_get_xmm_reg(cp); + + sse_movups(cp->func, dst, arg0); + sse_mulps(cp->func, dst, arg1); + + /* Now the hard bit: sum the first 3 values: + */ + sse_movhlps(cp->func, tmp, dst); + sse_addss(cp->func, dst, tmp); /* a*x+c*z, b*y, ?, ? */ + emit_pshufd(cp, tmp, dst, SHUF(Y,X,W,Z)); + sse_addss(cp->func, dst, tmp); + sse_shufps(cp->func, dst, dst, SHUF(X, X, X, X)); + + aos_release_xmm_reg(cp, tmp.idx); + store_dest(cp, &op->FullDstRegisters[0], dst); + return TRUE; +} + + + +static boolean emit_DP4( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); + struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); + struct x86_reg dst = aos_get_xmm_reg(cp); + struct x86_reg tmp = aos_get_xmm_reg(cp); + + sse_movups(cp->func, dst, arg0); + sse_mulps(cp->func, dst, arg1); + + /* Now the hard bit: sum the values: + */ + sse_movhlps(cp->func, tmp, dst); + sse_addps(cp->func, dst, tmp); /* a*x+c*z, b*y+d*w, a*x+c*z, b*y+d*w */ + emit_pshufd(cp, tmp, dst, SHUF(Y,X,W,Z)); + sse_addss(cp->func, dst, tmp); + sse_shufps(cp->func, dst, dst, SHUF(X, X, X, X)); + + aos_release_xmm_reg(cp, tmp.idx); + store_dest(cp, &op->FullDstRegisters[0], dst); + return TRUE; +} + +static boolean emit_DPH( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); + struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); + struct x86_reg dst = aos_get_xmm_reg(cp); + struct x86_reg tmp = aos_get_xmm_reg(cp); + + sse_movups(cp->func, dst, arg0); + sse_mulps(cp->func, dst, arg1); + + /* Now the hard bit: sum the values (from DP3): + */ + sse_movhlps(cp->func, tmp, dst); + sse_addss(cp->func, dst, tmp); /* a*x+c*z, b*y, ?, ? */ + emit_pshufd(cp, tmp, dst, SHUF(Y,X,W,Z)); + sse_addss(cp->func, dst, tmp); + emit_pshufd(cp, tmp, arg1, SHUF(W,W,W,W)); + sse_addss(cp->func, dst, tmp); + sse_shufps(cp->func, dst, dst, SHUF(X, X, X, X)); + + aos_release_xmm_reg(cp, tmp.idx); + store_dest(cp, &op->FullDstRegisters[0], dst); + return TRUE; +} + +static boolean emit_DST( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); + struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); + struct x86_reg dst = aos_get_xmm_reg(cp); + struct x86_reg tmp = aos_get_xmm_reg(cp); + struct x86_reg ones = aos_get_internal(cp, IMM_ONES); + +/* dst[0] = 1.0 * 1.0F; */ +/* dst[1] = arg0[1] * arg1[1]; */ +/* dst[2] = arg0[2] * 1.0; */ +/* dst[3] = 1.0 * arg1[3]; */ + + emit_shuf_copy2(cp, dst, arg0, ones, SHUF(X,W,Z,Y)); + emit_shuf_copy2(cp, tmp, arg1, ones, SHUF(X,Z,Y,W)); + sse_mulps(cp->func, dst, tmp); + + aos_release_xmm_reg(cp, tmp.idx); + store_dest(cp, &op->FullDstRegisters[0], dst); + return TRUE; +} + +static boolean emit_LG2( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + x87_fld1(cp->func); /* 1 */ + x87_fld_src(cp, &op->FullSrcRegisters[0], 0); /* a0 1 */ + x87_fyl2x(cp->func); /* log2(a0) */ + x87_fstp_dest4(cp, &op->FullDstRegisters[0]); + return TRUE; +} + + +static boolean emit_EX2( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + x87_fld_src(cp, &op->FullSrcRegisters[0], 0); + + emit_x87_ex2(cp); + + x87_fstp_dest4(cp, &op->FullDstRegisters[0]); + return TRUE; +} + +static boolean emit_EXP( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg dst = get_dst_ptr(cp, &op->FullDstRegisters[0]); + struct x86_reg st0 = x86_make_reg(file_x87, 0); + struct x86_reg st1 = x86_make_reg(file_x87, 1); + struct x86_reg st3 = x86_make_reg(file_x87, 3); + unsigned writemask = op->FullDstRegisters[0].DstRegister.WriteMask; + + /* CAUTION: dst may alias arg0! + */ + x87_fld_src(cp, &op->FullSrcRegisters[0], 0); /* arg0.x */ + x87_fld(cp->func, st0); /* arg arg */ + + /* by default, fpu is setup to round-to-nearest. We want to + * change this now, and track the state through to the end of the + * generated function so that it isn't repeated unnecessarily. + * Alternately, could subtract .5 to get round to -inf behaviour. + */ + set_fpu_round_neg_inf( cp ); + x87_fprndint( cp->func ); /* flr(a) a */ + x87_fld(cp->func, st0); /* flr(a) flr(a) a */ + x87_fld1(cp->func); /* 1 floor(a) floor(a) a */ + x87_fst_or_nop(cp->func, writemask, 3, dst); /* stack unchanged */ + + x87_fscale(cp->func); /* 2^floor(a) floor(a) a */ + x87_fst(cp->func, st3); /* 2^floor(a) floor(a) a 2^floor(a)*/ + + x87_fstp_or_pop(cp->func, writemask, 0, dst); /* flr(a) a 2^flr(a) */ + + x87_fsubrp(cp->func, st1); /* frac(a) 2^flr(a) */ + + x87_fst_or_nop(cp->func, writemask, 1, dst); /* frac(a) 2^flr(a) */ + + x87_f2xm1(cp->func); /* (2^frac(a))-1 2^flr(a)*/ + x87_fld1(cp->func); /* 1 (2^frac(a))-1 2^flr(a)*/ + x87_faddp(cp->func, st1); /* 2^frac(a) 2^flr(a) */ + x87_fmulp(cp->func, st1); /* 2^a */ + + x87_fstp_or_pop(cp->func, writemask, 2, dst); + +/* dst[0] = 2^floor(tmp); */ +/* dst[1] = frac(tmp); */ +/* dst[2] = 2^floor(tmp) * 2^frac(tmp); */ +/* dst[3] = 1.0F; */ + return TRUE; +} + +static boolean emit_LOG( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg dst = get_dst_ptr(cp, &op->FullDstRegisters[0]); + struct x86_reg st0 = x86_make_reg(file_x87, 0); + struct x86_reg st1 = x86_make_reg(file_x87, 1); + struct x86_reg st2 = x86_make_reg(file_x87, 2); + unsigned writemask = op->FullDstRegisters[0].DstRegister.WriteMask; + + /* CAUTION: dst may alias arg0! + */ + x87_fld_src(cp, &op->FullSrcRegisters[0], 0); /* arg0.x */ + x87_fabs(cp->func); /* |arg0.x| */ + x87_fxtract(cp->func); /* mantissa(arg0.x), exponent(arg0.x) */ + x87_fst(cp->func, st2); /* mantissa, exponent, mantissa */ + x87_fld1(cp->func); /* 1, mantissa, exponent, mantissa */ + x87_fyl2x(cp->func); /* log2(mantissa), exponent, mantissa */ + x87_fadd(cp->func, st0, st1); /* e+l2(m), e, m */ + + x87_fstp_or_pop(cp->func, writemask, 2, dst); /* e, m */ + + x87_fld1(cp->func); /* 1, e, m */ + x87_fsub(cp->func, st1, st0); /* 1, e-1, m */ + + x87_fstp_or_pop(cp->func, writemask, 3, dst); /* e-1,m */ + x87_fstp_or_pop(cp->func, writemask, 0, dst); /* m */ + + x87_fadd(cp->func, st0, st0); /* 2m */ + + x87_fstp_or_pop( cp->func, writemask, 1, dst ); + + return TRUE; +} + +static boolean emit_FLR( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg dst = get_dst_ptr(cp, &op->FullDstRegisters[0]); + unsigned writemask = op->FullDstRegisters[0].DstRegister.WriteMask; + int i; + + set_fpu_round_neg_inf( cp ); + + /* Load all sources first to avoid aliasing + */ + for (i = 0; i < 4; i++) { + if (writemask & (1<FullSrcRegisters[0], i); + } + } + + for (i = 0; i < 4; i++) { + if (writemask & (1<func ); + x87_fstp(cp->func, x86_make_disp(dst, i*4)); + } + } + + return TRUE; +} + + +static boolean emit_RND( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg dst = get_dst_ptr(cp, &op->FullDstRegisters[0]); + unsigned writemask = op->FullDstRegisters[0].DstRegister.WriteMask; + int i; + + set_fpu_round_nearest( cp ); + + /* Load all sources first to avoid aliasing + */ + for (i = 0; i < 4; i++) { + if (writemask & (1<FullSrcRegisters[0], i); + } + } + + for (i = 0; i < 4; i++) { + if (writemask & (1<func ); + x87_fstp(cp->func, x86_make_disp(dst, i*4)); + } + } + + return TRUE; +} + + +static boolean emit_FRC( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg dst = get_dst_ptr(cp, &op->FullDstRegisters[0]); + struct x86_reg st0 = x86_make_reg(file_x87, 0); + struct x86_reg st1 = x86_make_reg(file_x87, 1); + unsigned writemask = op->FullDstRegisters[0].DstRegister.WriteMask; + int i; + + set_fpu_round_neg_inf( cp ); + + /* suck all the source values onto the stack before writing out any + * dst, which may alias... + */ + for (i = 0; i < 4; i++) { + if (writemask & (1<FullSrcRegisters[0], i); + } + } + + for (i = 0; i < 4; i++) { + if (writemask & (1<func, st0); /* a a */ + x87_fprndint( cp->func ); /* flr(a) a */ + x87_fsubrp(cp->func, st1); /* frc(a) */ + x87_fstp(cp->func, x86_make_disp(dst, i*4)); + } + } + + return TRUE; +} + + + +static boolean emit_LIT( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg dst = get_dst_ptr(cp, &op->FullDstRegisters[0]); + struct x86_reg st1 = x86_make_reg(file_x87, 1); + unsigned fixup1, fixup2; + unsigned writemask = op->FullDstRegisters[0].DstRegister.WriteMask; + + + /* Load the interesting parts of arg0: + */ + x87_fld_src(cp, &op->FullSrcRegisters[0], 3); + x87_fld_src(cp, &op->FullSrcRegisters[0], 1); + x87_fld_src(cp, &op->FullSrcRegisters[0], 0); + + + if (writemask & TGSI_WRITEMASK_XW) { + x87_fld1(cp->func); + x87_fst_or_nop(cp->func, writemask, 0, dst); + x87_fstp_or_pop(cp->func, writemask, 3, dst); + } + + if (writemask & TGSI_WRITEMASK_YZ) { + + /* Pre-zero destinations, may be overwritten later... fixme. + */ + x87_fldz(cp->func); + x87_fst_or_nop(cp->func, writemask, 1, dst); + x87_fstp_or_pop(cp->func, writemask, 2, dst); + + + /* Check arg0[0]: + */ + x87_fldz(cp->func); /* 0 a0 a1 a3 */ + x87_fucomp(cp->func, st1); /* a0 a1 a3 */ + x87_fnstsw(cp->func, cp->tmp_EAX); + x86_sahf(cp->func); + fixup1 = x86_jcc_forward(cp->func, cc_AE); + + x87_fstp_or_pop(cp->func, writemask, 1, dst); /* a1 a3 */ + + /* Check arg0[1]: + */ + x87_fldz(cp->func); /* 0 a1 a3 */ + x87_fucomp(cp->func, st1); /* a1 a3 */ + x87_fnstsw(cp->func, cp->tmp_EAX); + x86_sahf(cp->func); + fixup2 = x86_jcc_forward(cp->func, cc_AE); + + /* Compute pow(a1, a3) + */ + x87_fyl2x(cp->func); /* a3*log2(a1) */ + + emit_x87_ex2( cp ); /* 2^(a3*log2(a1)) */ + + x87_fstp_or_pop(cp->func, writemask, 2, dst); + + /* Land jumps: + */ + x86_fixup_fwd_jump(cp->func, fixup1); + x86_fixup_fwd_jump(cp->func, fixup2); + } + + return TRUE; +} + + + +static boolean emit_MAX( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); + struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); + struct x86_reg dst = aos_get_xmm_reg(cp); + + sse_movups(cp->func, dst, arg0); + sse_maxps(cp->func, dst, arg1); + + store_dest(cp, &op->FullDstRegisters[0], dst); + return TRUE; +} + + +static boolean emit_MIN( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); + struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); + struct x86_reg dst = aos_get_xmm_reg(cp); + + sse_movups(cp->func, dst, arg0); + sse_minps(cp->func, dst, arg1); + + store_dest(cp, &op->FullDstRegisters[0], dst); + return TRUE; +} + +static boolean emit_MOV( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); + struct x86_reg dst = aos_get_xmm_reg(cp); + + sse_movups(cp->func, dst, arg0); + + store_dest(cp, &op->FullDstRegisters[0], dst); + return TRUE; +} + +static boolean emit_MUL( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); + struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); + struct x86_reg dst = aos_get_xmm_reg(cp); + + sse_movups(cp->func, dst, arg0); + sse_mulps(cp->func, dst, arg1); + + store_dest(cp, &op->FullDstRegisters[0], dst); + return TRUE; +} + + +static boolean emit_MAD( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); + struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); + struct x86_reg arg2 = fetch_src(cp, &op->FullSrcRegisters[2]); + struct x86_reg dst = aos_get_xmm_reg(cp); + + sse_movups(cp->func, dst, arg0); + sse_mulps(cp->func, dst, arg1); + sse_addps(cp->func, dst, arg2); + + store_dest(cp, &op->FullDstRegisters[0], dst); + return TRUE; +} + + +static boolean emit_POW( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + x87_fld_src(cp, &op->FullSrcRegisters[1], 0); /* a1.x */ + x87_fld_src(cp, &op->FullSrcRegisters[0], 0); /* a0.x a1.x */ + x87_fyl2x(cp->func); /* a1*log2(a0) */ + + emit_x87_ex2( cp ); /* 2^(a1*log2(a0)) */ + + x87_fstp_dest4(cp, &op->FullDstRegisters[0]); + return TRUE; +} + + +static boolean emit_RCP( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); + struct x86_reg dst = aos_get_xmm_reg(cp); + + if (cp->have_sse2) { + sse2_rcpss(cp->func, dst, arg0); + /* extend precision here... + */ + } + else { + struct x86_reg ones = aos_get_internal(cp, IMM_ONES); + sse_movss(cp->func, dst, ones); + sse_divss(cp->func, dst, arg0); + } + + sse_shufps(cp->func, dst, dst, SHUF(X, X, X, X)); + + store_dest(cp, &op->FullDstRegisters[0], dst); + return TRUE; +} + +static boolean emit_RSQ( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); + struct x86_reg dst = aos_get_xmm_reg(cp); + + sse_rsqrtss(cp->func, dst, arg0); + + /* Extend precision here... + */ + + sse_shufps(cp->func, dst, dst, SHUF(X, X, X, X)); + + store_dest(cp, &op->FullDstRegisters[0], dst); + return TRUE; +} + + +static boolean emit_SGE( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); + struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); + struct x86_reg dst = aos_get_xmm_reg(cp); + struct x86_reg ones = aos_get_internal(cp, IMM_ONES); + + sse_movups(cp->func, dst, arg0); + sse_cmpps(cp->func, dst, arg1, cc_NotLessThan); + sse_andps(cp->func, dst, ones); + + store_dest(cp, &op->FullDstRegisters[0], dst); + return TRUE; +} + +static boolean emit_SIN( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + x87_fld_src(cp, &op->FullSrcRegisters[0], 0); + x87_fsin(cp->func); + x87_fstp_dest4(cp, &op->FullDstRegisters[0]); + return TRUE; +} + + + +static boolean emit_SLT( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); + struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); + struct x86_reg dst = aos_get_xmm_reg(cp); + struct x86_reg ones = aos_get_internal(cp, IMM_ONES); + + sse_movups(cp->func, dst, arg0); + sse_cmpps(cp->func, dst, arg1, cc_LessThan); + sse_andps(cp->func, dst, ones); + + store_dest(cp, &op->FullDstRegisters[0], dst); + return TRUE; +} + +static boolean emit_SUB( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); + struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); + struct x86_reg dst = aos_get_xmm_reg(cp); + + sse_movups(cp->func, dst, arg0); + sse_subps(cp->func, dst, arg1); + + store_dest(cp, &op->FullDstRegisters[0], dst); + return TRUE; +} + + +static boolean emit_XPD( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); + struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); + struct x86_reg dst = aos_get_xmm_reg(cp); + struct x86_reg tmp0 = aos_get_xmm_reg(cp); + struct x86_reg tmp1 = aos_get_xmm_reg(cp); + + /* Could avoid tmp0, tmp1 if we overwrote arg0, arg1. Need a way + * to invalidate registers. This will come with better analysis + * (liveness analysis) of the incoming program. + */ + emit_pshufd(cp, dst, arg0, SHUF(Y, Z, X, W)); + emit_pshufd(cp, tmp1, arg1, SHUF(Z, X, Y, W)); + sse_mulps(cp->func, dst, tmp1); + emit_pshufd(cp, tmp0, arg0, SHUF(Z, X, Y, W)); + emit_pshufd(cp, tmp1, arg1, SHUF(Y, Z, X, W)); + sse_mulps(cp->func, tmp0, tmp1); + sse_subps(cp->func, dst, tmp0); + +/* dst[0] = arg0[1] * arg1[2] - arg0[2] * arg1[1]; */ +/* dst[1] = arg0[2] * arg1[0] - arg0[0] * arg1[2]; */ +/* dst[2] = arg0[0] * arg1[1] - arg0[1] * arg1[0]; */ +/* dst[3] is undef */ + + + aos_release_xmm_reg(cp, tmp0.idx); + aos_release_xmm_reg(cp, tmp1.idx); + store_dest(cp, &op->FullDstRegisters[0], dst); + return TRUE; +} + + + +static boolean +emit_instruction( struct aos_compilation *cp, + struct tgsi_full_instruction *inst ) +{ + switch( inst->Instruction.Opcode ) { + case TGSI_OPCODE_MOV: + return emit_MOV( cp, inst ); + + case TGSI_OPCODE_LIT: + return emit_LIT(cp, inst); + + case TGSI_OPCODE_RCP: + return emit_RCP(cp, inst); + + case TGSI_OPCODE_RSQ: + return emit_RSQ(cp, inst); + + case TGSI_OPCODE_EXP: + return emit_EXP(cp, inst); + + case TGSI_OPCODE_LOG: + return emit_LOG(cp, inst); + + case TGSI_OPCODE_MUL: + return emit_MUL(cp, inst); + + case TGSI_OPCODE_ADD: + return emit_ADD(cp, inst); + + case TGSI_OPCODE_DP3: + return emit_DP3(cp, inst); + + case TGSI_OPCODE_DP4: + return emit_DP4(cp, inst); + + case TGSI_OPCODE_DST: + return emit_DST(cp, inst); + + case TGSI_OPCODE_MIN: + return emit_MIN(cp, inst); + + case TGSI_OPCODE_MAX: + return emit_MAX(cp, inst); + + case TGSI_OPCODE_SLT: + return emit_SLT(cp, inst); + + case TGSI_OPCODE_SGE: + return emit_SGE(cp, inst); + + case TGSI_OPCODE_MAD: + return emit_MAD(cp, inst); + + case TGSI_OPCODE_SUB: + return emit_SUB(cp, inst); + + case TGSI_OPCODE_LERP: +// return emit_LERP(cp, inst); + return FALSE; + + case TGSI_OPCODE_FRAC: + return emit_FRC(cp, inst); + + case TGSI_OPCODE_CLAMP: +// return emit_CLAMP(cp, inst); + return FALSE; + + case TGSI_OPCODE_FLOOR: + return emit_FLR(cp, inst); + + case TGSI_OPCODE_ROUND: + return emit_RND(cp, inst); + + case TGSI_OPCODE_EXPBASE2: + return emit_EX2(cp, inst); + + case TGSI_OPCODE_LOGBASE2: + return emit_LG2(cp, inst); + + case TGSI_OPCODE_POWER: + return emit_POW(cp, inst); + + case TGSI_OPCODE_CROSSPRODUCT: + return emit_XPD(cp, inst); + + case TGSI_OPCODE_ABS: + return emit_ABS(cp, inst); + + case TGSI_OPCODE_DPH: + return emit_DPH(cp, inst); + + case TGSI_OPCODE_COS: + return emit_COS(cp, inst); + + case TGSI_OPCODE_SIN: + return emit_SIN(cp, inst); + + case TGSI_OPCODE_END: + return TRUE; + + default: + return FALSE; + } +} + +static boolean note_immediate( struct aos_compilation *cp, + struct tgsi_full_immediate *imm ) +{ + unsigned pos = cp->num_immediates++; + unsigned j; + + for (j = 0; j < imm->Immediate.Size; j++) { + cp->vaos->machine->immediate[pos][j] = imm->u.ImmediateFloat32[j].Float; + } + + return TRUE; +} + + + + +static void find_last_write_outputs( struct aos_compilation *cp ) +{ + struct tgsi_parse_context parse; + unsigned this_instruction = 0; + unsigned i; + + tgsi_parse_init( &parse, cp->vaos->base.vs->state.tokens ); + + while (!tgsi_parse_end_of_tokens( &parse )) { + + tgsi_parse_token( &parse ); + + if (parse.FullToken.Token.Type != TGSI_TOKEN_TYPE_INSTRUCTION) + continue; + + for (i = 0; i < TGSI_FULL_MAX_DST_REGISTERS; i++) { + if (parse.FullToken.FullInstruction.FullDstRegisters[i].DstRegister.File == + TGSI_FILE_OUTPUT) + { + unsigned idx = parse.FullToken.FullInstruction.FullDstRegisters[i].DstRegister.Index; + cp->output_last_write[idx] = this_instruction; + } + } + + this_instruction++; + } + + tgsi_parse_free( &parse ); +} + + +#define ARG_VARIENT 1 +#define ARG_START_ELTS 2 +#define ARG_COUNT 3 +#define ARG_OUTBUF 4 + + +static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient, + boolean linear ) +{ + struct tgsi_parse_context parse; + struct aos_compilation cp; + unsigned fixup, label; + + tgsi_parse_init( &parse, varient->base.vs->state.tokens ); + + memset(&cp, 0, sizeof(cp)); + + cp.insn_counter = 1; + cp.vaos = varient; + cp.have_sse2 = 1; + cp.func = &varient->func[ linear ? 0 : 1 ]; + + cp.tmp_EAX = x86_make_reg(file_REG32, reg_AX); + cp.idx_EBX = x86_make_reg(file_REG32, reg_BX); + cp.outbuf_ECX = x86_make_reg(file_REG32, reg_CX); + cp.machine_EDX = x86_make_reg(file_REG32, reg_DX); + cp.count_ESI = x86_make_reg(file_REG32, reg_SI); + + x86_init_func(cp.func); + + find_last_write_outputs(&cp); + + x86_push(cp.func, cp.idx_EBX); + x86_push(cp.func, cp.count_ESI); + + + /* Load arguments into regs: + */ + x86_mov(cp.func, cp.machine_EDX, x86_fn_arg(cp.func, ARG_VARIENT)); + x86_mov(cp.func, cp.idx_EBX, x86_fn_arg(cp.func, ARG_START_ELTS)); + x86_mov(cp.func, cp.count_ESI, x86_fn_arg(cp.func, ARG_COUNT)); + x86_mov(cp.func, cp.outbuf_ECX, x86_fn_arg(cp.func, ARG_OUTBUF)); + + + /* Compare count to zero and possibly bail. + */ + x86_xor(cp.func, cp.tmp_EAX, cp.tmp_EAX); + x86_cmp(cp.func, cp.count_ESI, cp.tmp_EAX); + fixup = x86_jcc_forward(cp.func, cc_E); + + /* Dig out the machine pointer from inside the varient arg + */ + x86_mov(cp.func, cp.machine_EDX, + x86_make_disp(cp.machine_EDX, + Offset( struct draw_vs_varient_aos_sse, machine ))); + + save_fpu_state( &cp ); + + /* Note address for loop jump + */ + label = x86_get_label(cp.func); + { + /* Fetch inputs... TODO: fetch lazily... + */ + if (!aos_fetch_inputs( &cp, linear )) + goto fail; + + /* Emit the shader: + */ + while( !tgsi_parse_end_of_tokens( &parse ) && !cp.error ) + { + tgsi_parse_token( &parse ); + + switch (parse.FullToken.Token.Type) { + case TGSI_TOKEN_TYPE_IMMEDIATE: + if (!note_immediate( &cp, &parse.FullToken.FullImmediate )) + goto fail; + break; + + case TGSI_TOKEN_TYPE_INSTRUCTION: + if (!emit_instruction( &cp, &parse.FullToken.FullInstruction )) + goto fail; + break; + } + + cp.insn_counter++; + debug_printf("\n"); + } + + if (cp.error) + goto fail; + + /* Emit output... TODO: do this eagerly after the last write to a + * given output. + */ + if (!aos_emit_outputs( &cp )) + goto fail; + + + /* Next vertex: + */ + x86_lea(cp.func, + cp.outbuf_ECX, + x86_make_disp(cp.outbuf_ECX, + cp.vaos->base.key.output_stride)); + + /* Incr index + */ + if (linear) { + x86_inc(cp.func, cp.idx_EBX); + } + else { + x86_lea(cp.func, cp.idx_EBX, x86_make_disp(cp.idx_EBX, 4)); + } + + } + /* decr count, loop if not zero + */ + x86_dec(cp.func, cp.count_ESI); +/* x86_test(cp.func, cp.count_ESI, cp.count_ESI); */ + x86_jcc(cp.func, cc_NZ, label); + + restore_fpu_state(&cp); + + /* Land forward jump here: + */ + x86_fixup_fwd_jump(cp.func, fixup); + + /* Exit mmx state? + */ + if (cp.func->need_emms) + mmx_emms(cp.func); + + x86_pop(cp.func, cp.count_ESI); + x86_pop(cp.func, cp.idx_EBX); + + x86_ret(cp.func); + + tgsi_parse_free( &parse ); + return !cp.error; + + fail: + tgsi_parse_free( &parse ); + return FALSE; +} + + + +static void vaos_set_buffer( struct draw_vs_varient *varient, + unsigned buf, + const void *ptr, + unsigned stride ) +{ + struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient; + unsigned i; + + for (i = 0; i < vaos->base.vs->info.num_inputs; i++) { + if (vaos->base.key.element[i].in.buffer == buf) { + vaos->machine->attrib[i].input_ptr = ((char *)ptr + + vaos->base.key.element[i].in.offset); + vaos->machine->attrib[i].input_stride = stride; + } + } +} + + +static void vaos_destroy( struct draw_vs_varient *varient ) +{ + struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient; + + if (vaos->machine) + align_free( vaos->machine ); + + x86_release_func( &vaos->func[0] ); + x86_release_func( &vaos->func[1] ); + + FREE(vaos); +} + +static void vaos_run_elts( struct draw_vs_varient *varient, + const unsigned *elts, + unsigned count, + void *output_buffer ) +{ + struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient; + + vaos->gen_run_elts( varient, + elts, + count, + output_buffer ); +} + +static void vaos_run_linear( struct draw_vs_varient *varient, + unsigned start, + unsigned count, + void *output_buffer ) +{ + struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient; + + vaos->gen_run_linear( varient, + start, + count, + output_buffer ); +} + + +static void vaos_set_constants( struct draw_vs_varient *varient, + const float (*constants)[4] ) +{ + struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient; + + memcpy(vaos->machine->constant, + constants, + (vaos->base.vs->info.file_max[TGSI_FILE_CONSTANT] + 1) * 4 * sizeof(float)); +} + + +static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs, + const struct draw_vs_varient_key *key ) +{ + struct draw_vs_varient_aos_sse *vaos = CALLOC_STRUCT(draw_vs_varient_aos_sse); + + if (!vaos) + goto fail; + + vaos->base.key = *key; + vaos->base.vs = vs; + vaos->base.set_input = vaos_set_buffer; + vaos->base.set_constants = vaos_set_constants; + vaos->base.destroy = vaos_destroy; + vaos->base.run_linear = vaos_run_linear; + vaos->base.run_elts = vaos_run_elts; + + vaos->machine = align_malloc( sizeof(struct aos_machine), 16 ); + if (!vaos->machine) + goto fail; + + memset(vaos->machine, 0, sizeof(struct aos_machine)); + + tgsi_dump(vs->state.tokens, 0); + + if (!build_vertex_program( vaos, TRUE )) + goto fail; + + if (!build_vertex_program( vaos, FALSE )) + goto fail; + + vaos->gen_run_linear = (vsv_run_linear_func)x86_get_func(&vaos->func[0]); + if (!vaos->gen_run_linear) + goto fail; + + vaos->gen_run_elts = (vsv_run_elts_func)x86_get_func(&vaos->func[1]); + if (!vaos->gen_run_elts) + goto fail; + + return &vaos->base; + + fail: + if (vaos->machine) + align_free( vaos->machine ); + + if (vaos) + x86_release_func( &vaos->func[0] ); + + if (vaos) + x86_release_func( &vaos->func[1] ); + + FREE(vaos); + + return NULL; +} + + +struct draw_vs_varient *draw_vs_varient_aos_sse( struct draw_vertex_shader *vs, + const struct draw_vs_varient_key *key ) +{ + struct draw_vs_varient *varient = varient_aos_sse( vs, key ); + + if (varient == NULL) { + assert(0); + varient = draw_vs_varient_generic( vs, key ); + } + + return varient; +} + + + +#endif diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h new file mode 100644 index 0000000000..1d8a055a90 --- /dev/null +++ b/src/gallium/auxiliary/draw/draw_vs_aos.h @@ -0,0 +1,181 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell + */ + +#ifndef DRAW_VS_AOS_H +#define DRAW_VS_AOS_H + + +struct tgsi_token; +struct x86_function; + +#include "pipe/p_state.h" +#include "rtasm/rtasm_x86sse.h" + + + + + +#define X 0 +#define Y 1 +#define Z 2 +#define W 3 + +#define MAX_INPUTS PIPE_MAX_ATTRIBS +#define MAX_OUTPUTS PIPE_MAX_ATTRIBS +#define MAX_TEMPS PIPE_MAX_ATTRIBS /* say */ +#define MAX_CONSTANTS PIPE_MAX_ATTRIBS /* say */ +#define MAX_IMMEDIATES PIPE_MAX_ATTRIBS /* say */ +#define MAX_INTERNALS 4 + +#define AOS_FILE_INTERNAL TGSI_FILE_COUNT + +/* This is the temporary storage used by all the aos_sse vs varients. + * Create one per context and reuse by passing a pointer in at + * vs_varient creation?? + */ +struct aos_machine { + float input [MAX_INPUTS ][4]; + float output [MAX_OUTPUTS ][4]; + float temp [MAX_TEMPS ][4]; + float constant [MAX_CONSTANTS ][4]; /* fixme -- should just be a pointer */ + float immediate[MAX_IMMEDIATES][4]; /* fixme -- should just be a pointer */ + float internal [MAX_INTERNALS ][4]; + + unsigned fpu_round_nearest; + unsigned fpu_round_neg_inf; + + struct { + const void *input_ptr; + unsigned input_stride; + + unsigned output_offset; + } attrib[PIPE_MAX_ATTRIBS]; +}; + + + + +struct aos_compilation { + struct x86_function *func; + struct draw_vs_varient_aos_sse *vaos; + + unsigned insn_counter; + unsigned num_immediates; + + struct { + unsigned idx:16; + unsigned file:8; + unsigned dirty:8; + unsigned last_used; + } xmm[8]; + + + boolean input_fetched[PIPE_MAX_ATTRIBS]; + unsigned output_last_write[PIPE_MAX_ATTRIBS]; + + boolean have_sse2; + boolean error; + short fpucntl; + + /* these are actually known values, but putting them in a struct + * like this is helpful to keep them in sync across the file. + */ + struct x86_reg tmp_EAX; + struct x86_reg idx_EBX; /* either start+i or &elt[i] */ + struct x86_reg outbuf_ECX; + struct x86_reg machine_EDX; + struct x86_reg count_ESI; /* decrements to zero */ +}; + +struct x86_reg aos_get_xmm_reg( struct aos_compilation *cp ); +void aos_release_xmm_reg( struct aos_compilation *cp, unsigned idx ); + +void aos_adopt_xmm_reg( struct aos_compilation *cp, + struct x86_reg reg, + unsigned file, + unsigned idx, + unsigned dirty ); + +struct x86_reg aos_get_shader_reg( struct aos_compilation *cp, + unsigned file, + unsigned idx ); + +boolean aos_fetch_inputs( struct aos_compilation *cp, + boolean linear ); + +boolean aos_emit_outputs( struct aos_compilation *cp ); + + +#define IMM_ONES 0 /* 1, 1,1,1 */ +#define IMM_NEGS 1 /* 1,-1,0,0 */ +#define IMM_IDENTITY 2 /* 0, 0,0,1 */ +#define IMM_INV_255 3 /* 1/255, 1/255, 1/255, 1/255 */ +#define IMM_255 4 /* 255, 255, 255, 255 */ + +struct x86_reg aos_get_internal( struct aos_compilation *cp, + unsigned imm ); + + +#define ERROR(cp, msg) \ +do { \ + debug_printf("%s: x86 translation failed: %s\n", __FUNCTION__, msg); \ + cp->error = 1; \ + assert(0); \ +} while (0) + + + + + + +struct draw_vs_varient_aos_sse { + struct draw_vs_varient base; + struct draw_context *draw; + +#if 0 + struct { + const void *ptr; + unsigned stride; + } attrib[PIPE_MAX_ATTRIBS]; +#endif + + struct aos_machine *machine; /* XXX: temporarily unshared */ + + vsv_run_linear_func gen_run_linear; + vsv_run_elts_func gen_run_elts; + + + struct x86_function func[2]; +}; + + + +#endif + diff --git a/src/gallium/auxiliary/draw/draw_vs_aos_io.c b/src/gallium/auxiliary/draw/draw_vs_aos_io.c new file mode 100644 index 0000000000..72b2b3d11d --- /dev/null +++ b/src/gallium/auxiliary/draw/draw_vs_aos_io.c @@ -0,0 +1,314 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "pipe/p_util.h" +#include "pipe/p_shader_tokens.h" +#include "tgsi/util/tgsi_parse.h" +#include "tgsi/util/tgsi_util.h" +#include "tgsi/exec/tgsi_exec.h" +#include "draw_vs.h" +#include "draw_vs_aos.h" + +#include "rtasm/rtasm_x86sse.h" + +#ifdef PIPE_ARCH_X86 + +/* Note - don't yet have to worry about interacting with the code in + * draw_vs_aos.c as there is no intermingling of generated code... + * That may have to change, we'll see. + */ +static void emit_load_R32G32B32A32( struct aos_compilation *cp, + struct x86_reg data, + struct x86_reg src_ptr ) +{ + sse_movups(cp->func, data, src_ptr); +} + +static void emit_load_R32G32B32( struct aos_compilation *cp, + struct x86_reg data, + struct x86_reg src_ptr ) +{ + sse_movss(cp->func, data, x86_make_disp(src_ptr, 8)); + sse_shufps(cp->func, data, aos_get_internal( cp, IMM_IDENTITY ), SHUF(X,Y,Z,W) ); + sse_shufps(cp->func, data, data, SHUF(Y,Z,X,W) ); + sse_movlps(cp->func, data, src_ptr); +} + +static void emit_load_R32G32( struct aos_compilation *cp, + struct x86_reg data, + struct x86_reg src_ptr ) +{ + sse_movups(cp->func, data, aos_get_internal( cp, IMM_IDENTITY ) ); + sse_movlps(cp->func, data, src_ptr); +} + + +static void emit_load_R32( struct aos_compilation *cp, + struct x86_reg data, + struct x86_reg src_ptr ) +{ + sse_movss(cp->func, data, src_ptr); + sse_orps(cp->func, data, aos_get_internal( cp, IMM_IDENTITY ) ); +} + + +static void emit_load_R8G8B8A8_UNORM( struct aos_compilation *cp, + struct x86_reg data, + struct x86_reg src_ptr ) +{ + sse_movss(cp->func, data, src_ptr); + sse2_punpcklbw(cp->func, data, aos_get_internal( cp, IMM_IDENTITY )); + sse2_punpcklbw(cp->func, data, aos_get_internal( cp, IMM_IDENTITY )); + sse2_cvtdq2ps(cp->func, data, data); + sse_mulps(cp->func, data, aos_get_internal(cp, IMM_INV_255)); +} + + + +static void get_src_ptr( struct x86_function *func, + struct x86_reg src, + struct x86_reg machine, + struct x86_reg elt, + unsigned a ) +{ + struct x86_reg input_ptr = + x86_make_disp(machine, + Offset(struct aos_machine, attrib[a].input_ptr)); + + struct x86_reg input_stride = + x86_make_disp(machine, + Offset(struct aos_machine, attrib[a].input_stride)); + + /* Calculate pointer to current attrib: + */ + x86_mov(func, src, input_stride); + x86_imul(func, src, elt); + x86_add(func, src, input_ptr); +} + + +/* Extended swizzles? Maybe later. + */ +static void emit_swizzle( struct aos_compilation *cp, + struct x86_reg dest, + struct x86_reg src, + unsigned shuffle ) +{ + sse_shufps(cp->func, dest, src, shuffle); +} + + +static boolean load_input( struct aos_compilation *cp, + unsigned idx, + boolean linear ) +{ + unsigned format = cp->vaos->base.key.element[idx].in.format; + struct x86_reg src = cp->tmp_EAX; + struct x86_reg dataXMM = aos_get_xmm_reg(cp); + + /* Figure out source pointer address: + */ + get_src_ptr(cp->func, + src, + cp->machine_EDX, + linear ? cp->idx_EBX : x86_deref(cp->idx_EBX), + idx); + + src = x86_deref(src); + + aos_adopt_xmm_reg( cp, + dataXMM, + TGSI_FILE_INPUT, + idx, + TRUE ); + + switch (format) { + case PIPE_FORMAT_R32_FLOAT: + emit_load_R32(cp, dataXMM, src); + break; + case PIPE_FORMAT_R32G32_FLOAT: + emit_load_R32G32(cp, dataXMM, src); + break; + case PIPE_FORMAT_R32G32B32_FLOAT: + emit_load_R32G32B32(cp, dataXMM, src); + break; + case PIPE_FORMAT_R32G32B32A32_FLOAT: + emit_load_R32G32B32A32(cp, dataXMM, src); + break; + case PIPE_FORMAT_B8G8R8A8_UNORM: + emit_load_R8G8B8A8_UNORM(cp, dataXMM, src); + emit_swizzle(cp, dataXMM, dataXMM, SHUF(Z,Y,X,W)); + break; + case PIPE_FORMAT_R8G8B8A8_UNORM: + emit_load_R8G8B8A8_UNORM(cp, dataXMM, src); + break; + default: + ERROR(cp, "unhandled input format"); + return FALSE; + } + + return TRUE; +} + + +boolean aos_fetch_inputs( struct aos_compilation *cp, boolean linear ) +{ + unsigned i; + + for (i = 0; i < cp->vaos->base.vs->info.num_inputs; i++) { + if (!load_input( cp, i, linear )) + return FALSE; + cp->insn_counter++; + debug_printf("\n"); + } + + return TRUE; +} + + + + + + + +static void emit_store_R32G32B32A32( struct aos_compilation *cp, + struct x86_reg dst_ptr, + struct x86_reg dataXMM ) +{ + sse_movups(cp->func, dst_ptr, dataXMM); +} + +static void emit_store_R32G32B32( struct aos_compilation *cp, + struct x86_reg dst_ptr, + struct x86_reg dataXMM ) +{ + sse_movlps(cp->func, dst_ptr, dataXMM); + sse_shufps(cp->func, dataXMM, dataXMM, SHUF(Z,Z,Z,Z) ); /* NOTE! destructive */ + sse_movss(cp->func, x86_make_disp(dst_ptr,8), dataXMM); +} + +static void emit_store_R32G32( struct aos_compilation *cp, + struct x86_reg dst_ptr, + struct x86_reg dataXMM ) +{ + sse_movlps(cp->func, dst_ptr, dataXMM); +} + +static void emit_store_R32( struct aos_compilation *cp, + struct x86_reg dst_ptr, + struct x86_reg dataXMM ) +{ + sse_movss(cp->func, dst_ptr, dataXMM); +} + + + +static void emit_store_R8G8B8A8_UNORM( struct aos_compilation *cp, + struct x86_reg dst_ptr, + struct x86_reg dataXMM ) +{ + sse_mulps(cp->func, dataXMM, aos_get_internal(cp, IMM_255)); + sse2_cvtps2dq(cp->func, dataXMM, dataXMM); + sse2_packssdw(cp->func, dataXMM, dataXMM); + sse2_packuswb(cp->func, dataXMM, dataXMM); + sse_movss(cp->func, dst_ptr, dataXMM); +} + + + + + +static boolean emit_output( struct aos_compilation *cp, + struct x86_reg ptr, + struct x86_reg dataXMM, + unsigned format ) +{ + switch (format) { + case PIPE_FORMAT_R32_FLOAT: + emit_store_R32(cp, ptr, dataXMM); + break; + case PIPE_FORMAT_R32G32_FLOAT: + emit_store_R32G32(cp, ptr, dataXMM); + break; + case PIPE_FORMAT_R32G32B32_FLOAT: + emit_store_R32G32B32(cp, ptr, dataXMM); + break; + case PIPE_FORMAT_R32G32B32A32_FLOAT: + emit_store_R32G32B32A32(cp, ptr, dataXMM); + break; + case PIPE_FORMAT_B8G8R8A8_UNORM: + emit_swizzle(cp, dataXMM, dataXMM, SHUF(Z,Y,X,W)); + emit_store_R8G8B8A8_UNORM(cp, ptr, dataXMM); + break; + case PIPE_FORMAT_R8G8B8A8_UNORM: + emit_store_R8G8B8A8_UNORM(cp, ptr, dataXMM); + break; + default: + ERROR(cp, "unhandled output format"); + return FALSE; + } + + return TRUE; +} + + + +boolean aos_emit_outputs( struct aos_compilation *cp ) +{ + unsigned i; + + for (i = 0; i < cp->vaos->base.vs->info.num_inputs; i++) { + unsigned format = cp->vaos->base.key.element[i].out.format; + unsigned offset = cp->vaos->base.key.element[i].out.offset; + + struct x86_reg data = aos_get_shader_reg( cp, + TGSI_FILE_OUTPUT, + i ); + + if (data.file != file_XMM) { + struct x86_reg tmp = aos_get_xmm_reg( cp ); + sse_movups(cp->func, tmp, data); + data = tmp; + } + + if (!emit_output( cp, + x86_make_disp( cp->outbuf_ECX, offset ), + data, + format )) + return FALSE; + + aos_release_xmm_reg( cp, data.idx ); + + cp->insn_counter++; + debug_printf("\n"); + } + + return TRUE; +} + +#endif diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c index df94a7e0c7..0581c3042f 100644 --- a/src/gallium/auxiliary/draw/draw_vs_sse.c +++ b/src/gallium/auxiliary/draw/draw_vs_sse.c @@ -157,6 +157,7 @@ draw_create_vs_sse(struct draw_context *draw, vs->base.draw = draw; vs->base.create_varient = draw_vs_varient_generic; +// vs->base.create_varient = draw_vs_varient_aos_sse; vs->base.prepare = vs_sse_prepare; vs->base.run_linear = vs_sse_run_linear; vs->base.delete = vs_sse_delete; -- cgit v1.2.3 From 030af06691bc5bc82ca141a576da7a2edffe9d1c Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Wed, 21 May 2008 20:14:55 +0100 Subject: rtasm: add x87 instructions and debug-check for x87 stack usage --- src/gallium/auxiliary/rtasm/rtasm_x86sse.c | 120 +++++++++++++++++++++++++++++ src/gallium/auxiliary/rtasm/rtasm_x86sse.h | 20 ++++- 2 files changed, 138 insertions(+), 2 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c index 68ac91ed13..a2e8af343b 100644 --- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c +++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c @@ -986,6 +986,26 @@ void sse2_movd( struct x86_function *p, /*********************************************************************** * x87 instructions */ +static void note_x87_pop( struct x86_function *p ) +{ + p->x87_stack--; + assert(p->x87_stack >= 0); + debug_printf("\nstack: %d\n", p->x87_stack); +} + +static void note_x87_push( struct x86_function *p ) +{ + p->x87_stack++; + assert(p->x87_stack <= 7); + debug_printf("\nstack: %d\n", p->x87_stack); +} + +void x87_assert_stack_empty( struct x86_function *p ) +{ + assert (p->x87_stack == 0); +} + + void x87_fist( struct x86_function *p, struct x86_reg dst ) { DUMP_R( dst ); @@ -998,6 +1018,7 @@ void x87_fistp( struct x86_function *p, struct x86_reg dst ) DUMP_R( dst ); emit_1ub(p, 0xdb); emit_modrm_noreg(p, 3, dst); + note_x87_pop(p); } void x87_fild( struct x86_function *p, struct x86_reg arg ) @@ -1005,12 +1026,14 @@ void x87_fild( struct x86_function *p, struct x86_reg arg ) DUMP_R( arg ); emit_1ub(p, 0xdf); emit_modrm_noreg(p, 0, arg); + note_x87_push(p); } void x87_fldz( struct x86_function *p ) { DUMP(); emit_2ub(p, 0xd9, 0xee); + note_x87_push(p); } @@ -1027,18 +1050,21 @@ void x87_fld1( struct x86_function *p ) { DUMP(); emit_2ub(p, 0xd9, 0xe8); + note_x87_push(p); } void x87_fldl2e( struct x86_function *p ) { DUMP(); emit_2ub(p, 0xd9, 0xea); + note_x87_push(p); } void x87_fldln2( struct x86_function *p ) { DUMP(); emit_2ub(p, 0xd9, 0xed); + note_x87_push(p); } void x87_fwait( struct x86_function *p ) @@ -1059,6 +1085,49 @@ void x87_fclex( struct x86_function *p ) x87_fnclex(p); } +void x87_fcmovb( struct x86_function *p, struct x86_reg arg ) +{ + DUMP_R( arg ); + assert(arg.file == file_x87); + emit_2ub(p, 0xda, 0xc0+arg.idx); +} + +void x87_fcmove( struct x86_function *p, struct x86_reg arg ) +{ + DUMP_R( arg ); + assert(arg.file == file_x87); + emit_2ub(p, 0xda, 0xc8+arg.idx); +} + +void x87_fcmovbe( struct x86_function *p, struct x86_reg arg ) +{ + DUMP_R( arg ); + assert(arg.file == file_x87); + emit_2ub(p, 0xda, 0xd0+arg.idx); +} + +void x87_fcmovnb( struct x86_function *p, struct x86_reg arg ) +{ + DUMP_R( arg ); + assert(arg.file == file_x87); + emit_2ub(p, 0xdb, 0xc0+arg.idx); +} + +void x87_fcmovne( struct x86_function *p, struct x86_reg arg ) +{ + DUMP_R( arg ); + assert(arg.file == file_x87); + emit_2ub(p, 0xdb, 0xc8+arg.idx); +} + +void x87_fcmovnbe( struct x86_function *p, struct x86_reg arg ) +{ + DUMP_R( arg ); + assert(arg.file == file_x87); + emit_2ub(p, 0xdb, 0xd0+arg.idx); +} + + static void x87_arith_op( struct x86_function *p, struct x86_reg dst, struct x86_reg arg, unsigned char dst0ub0, @@ -1146,6 +1215,7 @@ void x87_fmulp( struct x86_function *p, struct x86_reg dst ) assert(dst.file == file_x87); assert(dst.idx >= 1); emit_2ub(p, 0xde, 0xc8+dst.idx); + note_x87_pop(p); } void x87_fsubp( struct x86_function *p, struct x86_reg dst ) @@ -1154,6 +1224,7 @@ void x87_fsubp( struct x86_function *p, struct x86_reg dst ) assert(dst.file == file_x87); assert(dst.idx >= 1); emit_2ub(p, 0xde, 0xe8+dst.idx); + note_x87_pop(p); } void x87_fsubrp( struct x86_function *p, struct x86_reg dst ) @@ -1162,6 +1233,7 @@ void x87_fsubrp( struct x86_function *p, struct x86_reg dst ) assert(dst.file == file_x87); assert(dst.idx >= 1); emit_2ub(p, 0xde, 0xe0+dst.idx); + note_x87_pop(p); } void x87_faddp( struct x86_function *p, struct x86_reg dst ) @@ -1170,6 +1242,7 @@ void x87_faddp( struct x86_function *p, struct x86_reg dst ) assert(dst.file == file_x87); assert(dst.idx >= 1); emit_2ub(p, 0xde, 0xc0+dst.idx); + note_x87_pop(p); } void x87_fdivp( struct x86_function *p, struct x86_reg dst ) @@ -1178,6 +1251,7 @@ void x87_fdivp( struct x86_function *p, struct x86_reg dst ) assert(dst.file == file_x87); assert(dst.idx >= 1); emit_2ub(p, 0xde, 0xf8+dst.idx); + note_x87_pop(p); } void x87_fdivrp( struct x86_function *p, struct x86_reg dst ) @@ -1186,6 +1260,13 @@ void x87_fdivrp( struct x86_function *p, struct x86_reg dst ) assert(dst.file == file_x87); assert(dst.idx >= 1); emit_2ub(p, 0xde, 0xf0+dst.idx); + note_x87_pop(p); +} + +void x87_ftst( struct x86_function *p ) +{ + DUMP(); + emit_2ub(p, 0xd9, 0xe4); } void x87_fucom( struct x86_function *p, struct x86_reg arg ) @@ -1200,12 +1281,15 @@ void x87_fucomp( struct x86_function *p, struct x86_reg arg ) DUMP_R( arg ); assert(arg.file == file_x87); emit_2ub(p, 0xdd, 0xe8+arg.idx); + note_x87_pop(p); } void x87_fucompp( struct x86_function *p ) { DUMP(); emit_2ub(p, 0xda, 0xe9); + note_x87_pop(p); /* pop twice */ + note_x87_pop(p); /* pop twice */ } void x87_fxch( struct x86_function *p, struct x86_reg arg ) @@ -1287,6 +1371,7 @@ void x87_fyl2x( struct x86_function *p ) { DUMP(); emit_2ub(p, 0xd9, 0xf1); + note_x87_pop(p); } /* st1 = st1 * log2(st0 + 1.0); @@ -1298,6 +1383,7 @@ void x87_fyl2xp1( struct x86_function *p ) { DUMP(); emit_2ub(p, 0xd9, 0xf9); + note_x87_pop(p); } @@ -1310,6 +1396,7 @@ void x87_fld( struct x86_function *p, struct x86_reg arg ) emit_1ub(p, 0xd9); emit_modrm_noreg(p, 0, arg); } + note_x87_push(p); } void x87_fst( struct x86_function *p, struct x86_reg dst ) @@ -1332,8 +1419,15 @@ void x87_fstp( struct x86_function *p, struct x86_reg dst ) emit_1ub(p, 0xd9); emit_modrm_noreg(p, 3, dst); } + note_x87_pop(p); +} + +void x87_fpop( struct x86_function *p ) +{ + x87_fstp( p, x86_make_reg( file_x87, 0 )); } + void x87_fcom( struct x86_function *p, struct x86_reg dst ) { DUMP_R( dst ); @@ -1345,6 +1439,7 @@ void x87_fcom( struct x86_function *p, struct x86_reg dst ) } } + void x87_fcomp( struct x86_function *p, struct x86_reg dst ) { DUMP_R( dst ); @@ -1354,6 +1449,20 @@ void x87_fcomp( struct x86_function *p, struct x86_reg dst ) emit_1ub(p, 0xd8); emit_modrm_noreg(p, 3, dst); } + note_x87_pop(p); +} + +void x87_fcomi( struct x86_function *p, struct x86_reg arg ) +{ + DUMP_R( arg ); + emit_2ub(p, 0xdb, 0xf0+arg.idx); +} + +void x87_fcomip( struct x86_function *p, struct x86_reg arg ) +{ + DUMP_R( arg ); + emit_2ub(p, 0xdb, 0xf0+arg.idx); + note_x87_pop(p); } @@ -1372,6 +1481,17 @@ void x87_fnstsw( struct x86_function *p, struct x86_reg dst ) } +void x87_fnstcw( struct x86_function *p, struct x86_reg dst ) +{ + DUMP_R( dst ); + assert(dst.file == file_REG32); + + emit_1ub(p, 0x9b); /* WAIT -- needed? */ + emit_1ub(p, 0xd9); + emit_modrm_noreg(p, 7, dst); +} + + /*********************************************************************** diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h index 1e02c6e73b..9f7e31e055 100644 --- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h +++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h @@ -41,8 +41,11 @@ struct x86_function { unsigned size; unsigned char *store; unsigned char *csr; - unsigned stack_offset; - int need_emms; + + unsigned stack_offset:16; + unsigned need_emms:8; + int x87_stack:8; + unsigned char error_overflow[4]; }; @@ -229,13 +232,23 @@ void x86_test( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void x86_xor( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void x86_sahf( struct x86_function *p ); +void x87_assert_stack_empty( struct x86_function *p ); + void x87_f2xm1( struct x86_function *p ); void x87_fabs( struct x86_function *p ); void x87_fadd( struct x86_function *p, struct x86_reg dst, struct x86_reg arg ); void x87_faddp( struct x86_function *p, struct x86_reg dst ); void x87_fchs( struct x86_function *p ); void x87_fclex( struct x86_function *p ); +void x87_fcmovb( struct x86_function *p, struct x86_reg src ); +void x87_fcmovbe( struct x86_function *p, struct x86_reg src ); +void x87_fcmove( struct x86_function *p, struct x86_reg src ); +void x87_fcmovnb( struct x86_function *p, struct x86_reg src ); +void x87_fcmovnbe( struct x86_function *p, struct x86_reg src ); +void x87_fcmovne( struct x86_function *p, struct x86_reg src ); void x87_fcom( struct x86_function *p, struct x86_reg dst ); +void x87_fcomi( struct x86_function *p, struct x86_reg dst ); +void x87_fcomip( struct x86_function *p, struct x86_reg dst ); void x87_fcomp( struct x86_function *p, struct x86_reg dst ); void x87_fcos( struct x86_function *p ); void x87_fdiv( struct x86_function *p, struct x86_reg dst, struct x86_reg arg ); @@ -255,6 +268,7 @@ void x87_fmul( struct x86_function *p, struct x86_reg dst, struct x86_reg arg ); void x87_fmulp( struct x86_function *p, struct x86_reg dst ); void x87_fnclex( struct x86_function *p ); void x87_fprndint( struct x86_function *p ); +void x87_fpop( struct x86_function *p ); void x87_fscale( struct x86_function *p ); void x87_fsin( struct x86_function *p ); void x87_fsincos( struct x86_function *p ); @@ -265,11 +279,13 @@ void x87_fsub( struct x86_function *p, struct x86_reg dst, struct x86_reg arg ); void x87_fsubp( struct x86_function *p, struct x86_reg dst ); void x87_fsubr( struct x86_function *p, struct x86_reg dst, struct x86_reg arg ); void x87_fsubrp( struct x86_function *p, struct x86_reg dst ); +void x87_ftst( struct x86_function *p ); void x87_fxch( struct x86_function *p, struct x86_reg dst ); void x87_fxtract( struct x86_function *p ); void x87_fyl2x( struct x86_function *p ); void x87_fyl2xp1( struct x86_function *p ); void x87_fwait( struct x86_function *p ); +void x87_fnstcw( struct x86_function *p, struct x86_reg dst ); void x87_fnstsw( struct x86_function *p, struct x86_reg dst ); void x87_fucompp( struct x86_function *p ); void x87_fucomp( struct x86_function *p, struct x86_reg arg ); -- cgit v1.2.3 From 889473b3f5a216bd753c357974d6bae29fe3c41d Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Wed, 21 May 2008 20:28:56 +0100 Subject: draw: add viewport to varient state --- .../auxiliary/draw/draw_pt_fetch_shade_emit.c | 9 +++- src/gallium/auxiliary/draw/draw_vs.h | 8 +++- src/gallium/auxiliary/draw/draw_vs_aos.c | 50 ++++++++++++++++++++++ src/gallium/auxiliary/draw/draw_vs_aos.h | 9 +++- src/gallium/auxiliary/draw/draw_vs_sse.c | 4 +- src/gallium/auxiliary/draw/draw_vs_varient.c | 10 +++++ 6 files changed, 84 insertions(+), 6 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c index 74945dcfe9..984fbb6767 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c @@ -95,10 +95,14 @@ static void fse_prepare( struct draw_pt_middle_end *middle, + fse->key.output_stride = vinfo->size * 4; fse->key.nr_elements = MAX2(num_vs_outputs, /* outputs - translate to hw format */ num_vs_inputs); /* inputs - fetch from api format */ - fse->key.output_stride = vinfo->size * 4; + fse->key.viewport = 1; + fse->key.clip = 0; + fse->key.pad = 0; + memset(fse->key.element, 0, fse->key.nr_elements * sizeof(fse->key.element[0])); @@ -211,6 +215,9 @@ static void fse_prepare( struct draw_pt_middle_end *middle, fse->active->set_constants( fse->active, (const float (*)[4])draw->pt.user.constants ); + fse->active->set_viewport( fse->active, + &draw->viewport ); + //return TRUE; } diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h index 5a8d0da06d..ff3e19b2a8 100644 --- a/src/gallium/auxiliary/draw/draw_vs.h +++ b/src/gallium/auxiliary/draw/draw_vs.h @@ -58,7 +58,10 @@ struct draw_vs_element { struct draw_vs_varient_key { unsigned output_stride; - unsigned nr_elements; + unsigned nr_elements:16; + unsigned viewport:1; + unsigned clip:1; + unsigned pad:14; struct draw_vs_element element[PIPE_MAX_ATTRIBS]; }; @@ -88,6 +91,9 @@ struct draw_vs_varient { void (*set_constants)( struct draw_vs_varient *, const float (*constants)[4] ); + void (*set_viewport)( struct draw_vs_varient *, + const struct pipe_viewport_state * ); + void (PIPE_CDECL *run_linear)( struct draw_vs_varient *shader, unsigned start, unsigned count, diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index 620f5e3592..b8e66e8b78 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -1401,6 +1401,37 @@ emit_instruction( struct aos_compilation *cp, } } + +static boolean emit_viewport( struct aos_compilation *cp ) +{ + struct x86_reg pos = aos_get_shader_reg(cp, + TGSI_FILE_OUTPUT, + 0); + + struct x86_reg scale = x86_make_disp(cp->machine_EDX, + Offset(struct aos_machine, scale)); + + struct x86_reg translate = x86_make_disp(cp->machine_EDX, + Offset(struct aos_machine, translate)); + + if (pos.file != file_XMM) { + struct x86_reg dst = aos_get_xmm_reg(cp); + sse_movups(cp->func, dst, pos); + pos = dst; + } + + sse_mulps(cp->func, pos, scale); + sse_addps(cp->func, pos, translate); + + aos_adopt_xmm_reg( cp, + pos, + TGSI_FILE_OUTPUT, + 0, + TRUE ); + return TRUE; +} + + static boolean note_immediate( struct aos_compilation *cp, struct tgsi_full_immediate *imm ) { @@ -1540,6 +1571,10 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient, if (cp.error) goto fail; + if (cp.vaos->base.key.viewport) { + emit_viewport(&cp); + } + /* Emit output... TODO: do this eagerly after the last write to a * given output. */ @@ -1665,11 +1700,25 @@ static void vaos_set_constants( struct draw_vs_varient *varient, } +static void vaos_set_viewport( struct draw_vs_varient *varient, + const struct pipe_viewport_state *viewport ) +{ + struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient; + + memcpy(vaos->machine->scale, viewport->scale, 4 * sizeof(float)); + memcpy(vaos->machine->translate, viewport->translate, 4 * sizeof(float)); +} + + + static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs, const struct draw_vs_varient_key *key ) { struct draw_vs_varient_aos_sse *vaos = CALLOC_STRUCT(draw_vs_varient_aos_sse); + if (key->clip) + return NULL; + if (!vaos) goto fail; @@ -1677,6 +1726,7 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs, vaos->base.vs = vs; vaos->base.set_input = vaos_set_buffer; vaos->base.set_constants = vaos_set_constants; + vaos->base.set_viewport = vaos_set_viewport; vaos->base.destroy = vaos_destroy; vaos->base.run_linear = vaos_run_linear; vaos->base.run_elts = vaos_run_elts; diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h index 1d8a055a90..16fef6451c 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.h +++ b/src/gallium/auxiliary/draw/draw_vs_aos.h @@ -68,8 +68,13 @@ struct aos_machine { float immediate[MAX_IMMEDIATES][4]; /* fixme -- should just be a pointer */ float internal [MAX_INTERNALS ][4]; - unsigned fpu_round_nearest; - unsigned fpu_round_neg_inf; + float scale[4]; /* viewport */ + float translate[4]; /* viewport */ + + ushort fpu_round_nearest; + ushort fpu_round_neg_inf; + ushort fpu_restore; + ushort fpucntl; /* one of FPU_* above */ struct { const void *input_ptr; diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c index 0581c3042f..7781782ae8 100644 --- a/src/gallium/auxiliary/draw/draw_vs_sse.c +++ b/src/gallium/auxiliary/draw/draw_vs_sse.c @@ -156,8 +156,8 @@ draw_create_vs_sse(struct draw_context *draw, tgsi_scan_shader(templ->tokens, &vs->base.info); vs->base.draw = draw; - vs->base.create_varient = draw_vs_varient_generic; -// vs->base.create_varient = draw_vs_varient_aos_sse; + vs->base.create_varient = draw_vs_varient_aos_sse; +// vs->base.create_varient = draw_vs_varient_generic; vs->base.prepare = vs_sse_prepare; vs->base.run_linear = vs_sse_run_linear; vs->base.delete = vs_sse_delete; diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c index d27b0f6187..f6f621a748 100644 --- a/src/gallium/auxiliary/draw/draw_vs_varient.c +++ b/src/gallium/auxiliary/draw/draw_vs_varient.c @@ -167,6 +167,12 @@ static void vsvg_run_linear( struct draw_vs_varient *varient, + +static void vsvg_set_viewport( struct draw_vs_varient *varient, + const struct pipe_viewport_state *viewport ) +{ +} + static void vsvg_destroy( struct draw_vs_varient *varient ) { FREE(varient); @@ -179,6 +185,9 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs, unsigned i; struct translate_key fetch, emit; + if (key->viewport || key->clip) + return NULL; + struct draw_vs_varient_generic *vsvg = CALLOC_STRUCT( draw_vs_varient_generic ); if (vsvg == NULL) return NULL; @@ -187,6 +196,7 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs, vsvg->base.vs = vs; vsvg->base.set_input = vsvg_set_input; vsvg->base.set_constants = vsvg_set_constants; + vsvg->base.set_viewport = vsvg_set_viewport; vsvg->base.run_elts = vsvg_run_elts; vsvg->base.run_linear = vsvg_run_linear; vsvg->base.destroy = vsvg_destroy; -- cgit v1.2.3 From 194a7be28f6eed502f2475d9a637cb3610ca75f6 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Wed, 21 May 2008 20:31:08 +0100 Subject: draw: fix vs aos internal/machine state --- src/gallium/auxiliary/draw/draw_vs_aos.c | 59 ++++++++++++++++++++++++++++++-- src/gallium/auxiliary/draw/draw_vs_aos.h | 9 +++-- 2 files changed, 63 insertions(+), 5 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index b8e66e8b78..67761f881d 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -83,7 +83,7 @@ static struct x86_reg get_reg_ptr(struct aos_compilation *cp, return x86_make_disp(ptr, Offset(struct aos_machine, constant[idx])); case AOS_FILE_INTERNAL: - return x86_make_disp(ptr, Offset(struct aos_machine, immediate[idx])); + return x86_make_disp(ptr, Offset(struct aos_machine, internal[idx])); default: ERROR(cp, "unknown reg file"); @@ -97,9 +97,63 @@ struct x86_reg aos_get_internal( struct aos_compilation *cp, { return get_reg_ptr( cp, AOS_FILE_INTERNAL, - imm + 1 ); + imm ); +} + +#define X87_CW_EXCEPTION_INV_OP (1<<0) +#define X87_CW_EXCEPTION_DENORM_OP (1<<1) +#define X87_CW_EXCEPTION_ZERO_DIVIDE (1<<2) +#define X87_CW_EXCEPTION_OVERFLOW (1<<3) +#define X87_CW_EXCEPTION_UNDERFLOW (1<<4) +#define X87_CW_EXCEPTION_PRECISION (1<<5) +#define X87_CW_PRECISION_SINGLE (0<<8) +#define X87_CW_PRECISION_RESERVED (1<<8) +#define X87_CW_PRECISION_DOUBLE (2<<8) +#define X87_CW_PRECISION_DOUBLE_EXT (3<<8) +#define X87_CW_PRECISION_MASK (3<<8) +#define X87_CW_ROUND_NEAREST (0<<10) +#define X87_CW_ROUND_DOWN (1<<10) +#define X87_CW_ROUND_UP (2<<10) +#define X87_CW_ROUND_ZERO (3<<10) +#define X87_CW_ROUND_MASK (3<<10) +#define X87_CW_INFINITY (1<<12) + +static void init_internals( struct aos_machine *machine ) +{ + float inv = 1.0f/255.0f; + float f255 = 255.0f; + + ASSIGN_4V(machine->internal[IMM_ONES], 1.0f, 1.0f, 1.0f, 1.0f); + ASSIGN_4V(machine->internal[IMM_NEGS], -1.0f, -1.0f, -1.0f, -1.0f); + ASSIGN_4V(machine->internal[IMM_IDENTITY], 0.0f, 0.0f, 0.0f, 1.0f); + ASSIGN_4V(machine->internal[IMM_INV_255], inv, inv, inv, inv); + ASSIGN_4V(machine->internal[IMM_255], f255, f255, f255, f255); + + + machine->fpu_rnd_nearest = (X87_CW_EXCEPTION_INV_OP | + X87_CW_EXCEPTION_DENORM_OP | + X87_CW_EXCEPTION_ZERO_DIVIDE | + X87_CW_EXCEPTION_OVERFLOW | + X87_CW_EXCEPTION_UNDERFLOW | + X87_CW_EXCEPTION_PRECISION | + (1<<6) | + X87_CW_ROUND_NEAREST | + X87_CW_PRECISION_DOUBLE_EXT); + + assert(machine->fpu_rnd_nearest == 0x37f); + + machine->fpu_rnd_neg_inf = (X87_CW_EXCEPTION_INV_OP | + X87_CW_EXCEPTION_DENORM_OP | + X87_CW_EXCEPTION_ZERO_DIVIDE | + X87_CW_EXCEPTION_OVERFLOW | + X87_CW_EXCEPTION_UNDERFLOW | + X87_CW_EXCEPTION_PRECISION | + (1<<6) | + X87_CW_ROUND_DOWN | + X87_CW_PRECISION_DOUBLE_EXT); } + static void spill( struct aos_compilation *cp, unsigned idx ) { if (!cp->xmm[idx].dirty || @@ -1736,6 +1790,7 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs, goto fail; memset(vaos->machine, 0, sizeof(struct aos_machine)); + init_internals(vaos->machine); tgsi_dump(vs->state.tokens, 0); diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h index 16fef6451c..c2afd4e9a0 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.h +++ b/src/gallium/auxiliary/draw/draw_vs_aos.h @@ -52,10 +52,13 @@ struct x86_function; #define MAX_TEMPS PIPE_MAX_ATTRIBS /* say */ #define MAX_CONSTANTS PIPE_MAX_ATTRIBS /* say */ #define MAX_IMMEDIATES PIPE_MAX_ATTRIBS /* say */ -#define MAX_INTERNALS 4 +#define MAX_INTERNALS 8 #define AOS_FILE_INTERNAL TGSI_FILE_COUNT +#define FPU_RND_NEG 1 +#define FPU_RND_NEAREST 2 + /* This is the temporary storage used by all the aos_sse vs varients. * Create one per context and reuse by passing a pointer in at * vs_varient creation?? @@ -71,8 +74,8 @@ struct aos_machine { float scale[4]; /* viewport */ float translate[4]; /* viewport */ - ushort fpu_round_nearest; - ushort fpu_round_neg_inf; + ushort fpu_rnd_nearest; + ushort fpu_rnd_neg_inf; ushort fpu_restore; ushort fpucntl; /* one of FPU_* above */ -- cgit v1.2.3 From 2302a5d3c1ea2c682dfc034012a054b8327a81de Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Wed, 21 May 2008 20:32:43 +0100 Subject: draw: fix fpu control word manipulations --- src/gallium/auxiliary/draw/draw_vs_aos.c | 33 +++++++++++++++++++------------- 1 file changed, 20 insertions(+), 13 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index 67761f881d..e736990acc 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -690,41 +690,47 @@ static void x87_fstp_dest4( struct aos_compilation *cp, x87_fstp_or_pop(cp->func, writemask, 3, ptr); } +#define FPU_MANIP 1 /* Save current x87 state and put it into single precision mode. */ static void save_fpu_state( struct aos_compilation *cp ) { -#if 0 - x87_fnstcw( cp->func, x86_make_disp(regEDX, get_offset(m, &m->fpucntl_restore))); - x87_fldcw( cp->func, ); +#if FPU_MANIP + x87_fnstcw( cp->func, x86_make_disp(cp->machine_EDX, + Offset(struct aos_machine, fpu_restore))); #endif } static void restore_fpu_state( struct aos_compilation *cp ) { -#if 0 +#if FPU_MANIP x87_fnclex(cp->func); - x87_fldcw(cp->func, x86_make_disp(regEDX, get_offset(m, &m->fpucntl_restore))); + x87_fldcw( cp->func, x86_make_disp(cp->machine_EDX, + Offset(struct aos_machine, fpu_restore))); #endif } static void set_fpu_round_neg_inf( struct aos_compilation *cp ) { -#if 0 - if (cp->fpucntl != RND_NEG_FPU) { - struct x86_reg regEDX = x86_make_reg(file_REG32, reg_DX); - struct arb_vp_machine *m = NULL; - - cp->fpucntl = RND_NEG_FPU; +#if FPU_MANIP + if (cp->fpucntl != FPU_RND_NEG) { + cp->fpucntl = FPU_RND_NEG; x87_fnclex(cp->func); - x87_fldcw(cp->func, x86_make_disp(regEDX, get_offset(m, &m->fpucntl_rnd_neg))); + x87_fldcw( cp->func, x86_make_disp(cp->machine_EDX, + Offset(struct aos_machine, fpu_rnd_neg_inf))); } #endif } static void set_fpu_round_nearest( struct aos_compilation *cp ) { -#if 0 +#if FPU_MANIP + if (cp->fpucntl != FPU_RND_NEAREST) { + cp->fpucntl = FPU_RND_NEAREST; + x87_fnclex(cp->func); + x87_fldcw( cp->func, x86_make_disp(cp->machine_EDX, + Offset(struct aos_machine, fpu_rnd_nearest))); + } #endif } @@ -1590,6 +1596,7 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient, Offset( struct draw_vs_varient_aos_sse, machine ))); save_fpu_state( &cp ); + set_fpu_round_nearest( &cp ); /* Note address for loop jump */ -- cgit v1.2.3 From 0a7a0d79f64de9794878c42bc5b79a04772d7ed8 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Wed, 21 May 2008 20:34:52 +0100 Subject: draw: fix x87_ex2 and partially fix lit insn --- src/gallium/auxiliary/draw/draw_vs_aos.c | 112 +++++++++++++++---------------- 1 file changed, 56 insertions(+), 56 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index e736990acc..a365d456d1 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -735,23 +735,26 @@ static void set_fpu_round_nearest( struct aos_compilation *cp ) } -static void emit_x87_ex2( struct aos_compilation *cp ) +static void x87_emit_ex2( struct aos_compilation *cp ) { struct x86_reg st0 = x86_make_reg(file_x87, 0); struct x86_reg st1 = x86_make_reg(file_x87, 1); - struct x86_reg st3 = x86_make_reg(file_x87, 3); + int stack = cp->func->x87_stack; set_fpu_round_neg_inf( cp ); - x87_fld(cp->func, st0); /* a a */ - x87_fprndint( cp->func ); /* int(a) a */ - x87_fld(cp->func, st0); /* int(a) int(a) a */ - x87_fstp(cp->func, st3); /* int(a) a int(a)*/ - x87_fsubp(cp->func, st1); /* frac(a) int(a) */ - x87_f2xm1(cp->func); /* (2^frac(a))-1 int(a)*/ - x87_fld1(cp->func); /* 1 (2^frac(a))-1 int(a)*/ - x87_faddp(cp->func, st1); /* 2^frac(a) int(a) */ - x87_fscale(cp->func); /* 2^a */ + x87_fld(cp->func, st0); /* a a */ + x87_fld(cp->func, st0); /* a a a */ + x87_fprndint( cp->func ); /* flr(a) a a*/ + x87_fsubp(cp->func, st1); /* frac(a) a */ + x87_f2xm1(cp->func); /* (2^frac(a))-1 a */ + x87_fld1(cp->func); /* 1 (2^frac(a))-1 a */ + x87_faddp(cp->func, st1); /* 2^frac(a) a */ + x87_fscale(cp->func); /* 2^a a */ + x87_fstp(cp->func, st1); + + assert( stack == cp->func->x87_stack); + } @@ -907,9 +910,7 @@ static boolean emit_LG2( struct aos_compilation *cp, const struct tgsi_full_inst static boolean emit_EX2( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) { x87_fld_src(cp, &op->FullSrcRegisters[0], 0); - - emit_x87_ex2(cp); - + x87_emit_ex2(cp); x87_fstp_dest4(cp, &op->FullDstRegisters[0]); return TRUE; } @@ -1084,63 +1085,62 @@ static boolean emit_FRC( struct aos_compilation *cp, const struct tgsi_full_inst static boolean emit_LIT( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) { struct x86_reg dst = get_dst_ptr(cp, &op->FullDstRegisters[0]); - struct x86_reg st1 = x86_make_reg(file_x87, 1); - unsigned fixup1, fixup2; unsigned writemask = op->FullDstRegisters[0].DstRegister.WriteMask; - /* Load the interesting parts of arg0: - */ - x87_fld_src(cp, &op->FullSrcRegisters[0], 3); - x87_fld_src(cp, &op->FullSrcRegisters[0], 1); - x87_fld_src(cp, &op->FullSrcRegisters[0], 0); - - if (writemask & TGSI_WRITEMASK_XW) { - x87_fld1(cp->func); - x87_fst_or_nop(cp->func, writemask, 0, dst); - x87_fstp_or_pop(cp->func, writemask, 3, dst); - } if (writemask & TGSI_WRITEMASK_YZ) { + struct x86_reg st1 = x86_make_reg(file_x87, 1); + struct x86_reg st2 = x86_make_reg(file_x87, 2); + + - /* Pre-zero destinations, may be overwritten later... fixme. - */ - x87_fldz(cp->func); - x87_fst_or_nop(cp->func, writemask, 1, dst); - x87_fstp_or_pop(cp->func, writemask, 2, dst); + /* a1' = a1 <= 0 ? 1 : a1; + */ + x87_fldz(cp->func); /* 0 */ + x87_fld1(cp->func); /* 1 0 */ + x87_fld_src(cp, &op->FullSrcRegisters[0], 1); /* a1 1 0 */ + x87_fcomi(cp->func, st2); /* a1 1 0 */ + x87_fcmovb(cp->func, st1); /* a1' 1 0 */ + x87_fstp(cp->func, st1); /* a1' 0 */ + x87_fstp(cp->func, st1); /* a1' */ + + x87_fld_src(cp, &op->FullSrcRegisters[0], 3); /* a3 a1' */ + x87_fxch(cp->func, st1); /* a1' a3 */ + - /* Check arg0[0]: + /* Compute pow(a1, a3) */ - x87_fldz(cp->func); /* 0 a0 a1 a3 */ - x87_fucomp(cp->func, st1); /* a0 a1 a3 */ - x87_fnstsw(cp->func, cp->tmp_EAX); - x86_sahf(cp->func); - fixup1 = x86_jcc_forward(cp->func, cc_AE); - - x87_fstp_or_pop(cp->func, writemask, 1, dst); /* a1 a3 */ + x87_fyl2x(cp->func); /* a3*log2(a1) */ + x87_emit_ex2( cp ); /* 2^(a3*log2(a1)) */ - /* Check arg0[1]: - */ - x87_fldz(cp->func); /* 0 a1 a3 */ - x87_fucomp(cp->func, st1); /* a1 a3 */ - x87_fnstsw(cp->func, cp->tmp_EAX); - x86_sahf(cp->func); - fixup2 = x86_jcc_forward(cp->func, cc_AE); - /* Compute pow(a1, a3) + /* a0' = max2(a0, 0): */ - x87_fyl2x(cp->func); /* a3*log2(a1) */ + x87_fldz(cp->func); /* 0 r2 */ + x87_fld_src(cp, &op->FullSrcRegisters[0], 0); /* a0 0 r2 */ + x87_fcomi(cp->func, st1); + x87_fcmovb(cp->func, st1); /* a0' 0 r2 */ + x87_fstp(cp->func, st1); /* a0' r2 */ - emit_x87_ex2( cp ); /* 2^(a3*log2(a1)) */ + x87_fxch(cp->func, st1); /* a0' r2 */ + x87_fst_or_nop(cp->func, writemask, 1, dst); /* result[1] = a0' */ + + x87_fldz(cp->func); /* 0 a0' r2 */ + x87_fcomi(cp->func, st1); /* 0 a0' r2 */ + x87_fcmovnbe(cp->func, st2); /* r2' a0' r2 */ x87_fstp_or_pop(cp->func, writemask, 2, dst); - - /* Land jumps: - */ - x86_fixup_fwd_jump(cp->func, fixup1); - x86_fixup_fwd_jump(cp->func, fixup2); + x87_fpop(cp->func); + x87_fpop(cp->func); + } + + if (writemask & TGSI_WRITEMASK_XW) { + x87_fld1(cp->func); + x87_fst_or_nop(cp->func, writemask, 0, dst); + x87_fstp_or_pop(cp->func, writemask, 3, dst); } return TRUE; @@ -1222,7 +1222,7 @@ static boolean emit_POW( struct aos_compilation *cp, const struct tgsi_full_inst x87_fld_src(cp, &op->FullSrcRegisters[0], 0); /* a0.x a1.x */ x87_fyl2x(cp->func); /* a1*log2(a0) */ - emit_x87_ex2( cp ); /* 2^(a1*log2(a0)) */ + x87_emit_ex2( cp ); /* 2^(a1*log2(a0)) */ x87_fstp_dest4(cp, &op->FullDstRegisters[0]); return TRUE; -- cgit v1.2.3 From 083f3f5c32a28d2993a8a5a8b4f5ef81224a5ec3 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Wed, 21 May 2008 20:38:23 +0100 Subject: draw: avoid a pointless mov in many sse opcodes --- src/gallium/auxiliary/draw/draw_vs_aos.c | 94 +++++++++++++++++++++----------- 1 file changed, 63 insertions(+), 31 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index a365d456d1..97de43c232 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -174,14 +174,44 @@ static void spill( struct aos_compilation *cp, unsigned idx ) } } +static boolean is_xmm_tmp( struct aos_compilation *cp, + struct x86_reg reg ) +{ + return (reg.file == file_XMM && + cp->xmm[reg.idx].file == TGSI_FILE_NULL); +} + +static struct x86_reg get_xmm_tmp( struct aos_compilation *cp, + struct x86_reg reg ) +{ + if (!is_xmm_tmp(cp, reg)) { + struct x86_reg tmp = aos_get_xmm_reg(cp); + sse_movups(cp->func, tmp, reg); + reg = tmp; + } + + return reg; +} + + struct x86_reg aos_get_xmm_reg( struct aos_compilation *cp ) { unsigned i; unsigned oldest = 0; + boolean found = FALSE; for (i = 0; i < 8; i++) - if (cp->xmm[i].last_used < cp->xmm[oldest].last_used) + if (cp->xmm[i].last_used != cp->insn_counter && + cp->xmm[i].file == TGSI_FILE_NULL) { oldest = i; + found = TRUE; + } + + if (!found) { + for (i = 0; i < 8; i++) + if (cp->xmm[i].last_used < cp->xmm[oldest].last_used) + oldest = i; + } /* Need to write out the old value? */ @@ -237,15 +267,24 @@ void aos_adopt_xmm_reg( struct aos_compilation *cp, unsigned idx, unsigned dirty ) { + unsigned i; + if (reg.file != file_XMM) { assert(0); return; } - invalidate_xmm(cp, file, idx); + for (i = 0; i < 8; i++) { + if (cp->xmm[i].file == file && + cp->xmm[i].idx == idx) { + aos_release_xmm_reg(cp, i); + } + } + cp->xmm[reg.idx].file = file; cp->xmm[reg.idx].idx = idx; cp->xmm[reg.idx].dirty = dirty; + cp->xmm[reg.idx].last_used = cp->insn_counter; } @@ -659,6 +698,7 @@ static void x87_fst_or_nop( struct x86_function *func, unsigned channel, struct x86_reg ptr ) { + assert(ptr.file == file_REG32); if (writemask & (1<FullSrcRegisters[0]); - struct x86_reg dst = aos_get_xmm_reg(cp); struct x86_reg neg = aos_get_internal(cp, IMM_NEGS); + struct x86_reg dst = get_xmm_tmp(cp, arg0); - sse_movups(cp->func, dst, arg0); sse_mulps(cp->func, dst, neg); sse_maxps(cp->func, dst, arg0); @@ -782,9 +822,8 @@ static boolean emit_ADD( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); - struct x86_reg dst = aos_get_xmm_reg(cp); + struct x86_reg dst = get_xmm_tmp(cp, arg0); - sse_movups(cp->func, dst, arg0); sse_addps(cp->func, dst, arg1); store_dest(cp, &op->FullDstRegisters[0], dst); @@ -806,10 +845,9 @@ static boolean emit_DP3( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); - struct x86_reg dst = aos_get_xmm_reg(cp); struct x86_reg tmp = aos_get_xmm_reg(cp); + struct x86_reg dst = get_xmm_tmp(cp, arg0); - sse_movups(cp->func, dst, arg0); sse_mulps(cp->func, dst, arg1); /* Now the hard bit: sum the first 3 values: @@ -831,10 +869,9 @@ static boolean emit_DP4( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); - struct x86_reg dst = aos_get_xmm_reg(cp); struct x86_reg tmp = aos_get_xmm_reg(cp); + struct x86_reg dst = get_xmm_tmp(cp, arg0); - sse_movups(cp->func, dst, arg0); sse_mulps(cp->func, dst, arg1); /* Now the hard bit: sum the values: @@ -854,10 +891,9 @@ static boolean emit_DPH( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); - struct x86_reg dst = aos_get_xmm_reg(cp); struct x86_reg tmp = aos_get_xmm_reg(cp); + struct x86_reg dst = get_xmm_tmp(cp, arg0); - sse_movups(cp->func, dst, arg0); sse_mulps(cp->func, dst, arg1); /* Now the hard bit: sum the values (from DP3): @@ -1152,9 +1188,8 @@ static boolean emit_MAX( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); - struct x86_reg dst = aos_get_xmm_reg(cp); + struct x86_reg dst = get_xmm_tmp(cp, arg0); - sse_movups(cp->func, dst, arg0); sse_maxps(cp->func, dst, arg1); store_dest(cp, &op->FullDstRegisters[0], dst); @@ -1166,9 +1201,8 @@ static boolean emit_MIN( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); - struct x86_reg dst = aos_get_xmm_reg(cp); + struct x86_reg dst = get_xmm_tmp(cp, arg0); - sse_movups(cp->func, dst, arg0); sse_minps(cp->func, dst, arg1); store_dest(cp, &op->FullDstRegisters[0], dst); @@ -1178,9 +1212,9 @@ static boolean emit_MIN( struct aos_compilation *cp, const struct tgsi_full_inst static boolean emit_MOV( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); - struct x86_reg dst = aos_get_xmm_reg(cp); + struct x86_reg dst = get_xmm_tmp(cp, arg0); - sse_movups(cp->func, dst, arg0); + /* potentially nothing to do */ store_dest(cp, &op->FullDstRegisters[0], dst); return TRUE; @@ -1190,9 +1224,8 @@ static boolean emit_MUL( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); - struct x86_reg dst = aos_get_xmm_reg(cp); + struct x86_reg dst = get_xmm_tmp(cp, arg0); - sse_movups(cp->func, dst, arg0); sse_mulps(cp->func, dst, arg1); store_dest(cp, &op->FullDstRegisters[0], dst); @@ -1205,13 +1238,15 @@ static boolean emit_MAD( struct aos_compilation *cp, const struct tgsi_full_inst struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); struct x86_reg arg2 = fetch_src(cp, &op->FullSrcRegisters[2]); - struct x86_reg dst = aos_get_xmm_reg(cp); - sse_movups(cp->func, dst, arg0); - sse_mulps(cp->func, dst, arg1); - sse_addps(cp->func, dst, arg2); + /* If we can't clobber old contents of arg0, get a temporary & copy + * it there, then clobber it... + */ + arg0 = get_xmm_tmp(cp, arg0); - store_dest(cp, &op->FullDstRegisters[0], dst); + sse_mulps(cp->func, arg0, arg1); + sse_addps(cp->func, arg0, arg2); + store_dest(cp, &op->FullDstRegisters[0], arg0); return TRUE; } @@ -1272,10 +1307,9 @@ static boolean emit_SGE( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); - struct x86_reg dst = aos_get_xmm_reg(cp); struct x86_reg ones = aos_get_internal(cp, IMM_ONES); + struct x86_reg dst = get_xmm_tmp(cp, arg0); - sse_movups(cp->func, dst, arg0); sse_cmpps(cp->func, dst, arg1, cc_NotLessThan); sse_andps(cp->func, dst, ones); @@ -1297,10 +1331,9 @@ static boolean emit_SLT( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); - struct x86_reg dst = aos_get_xmm_reg(cp); struct x86_reg ones = aos_get_internal(cp, IMM_ONES); + struct x86_reg dst = get_xmm_tmp(cp, arg0); - sse_movups(cp->func, dst, arg0); sse_cmpps(cp->func, dst, arg1, cc_LessThan); sse_andps(cp->func, dst, ones); @@ -1312,9 +1345,8 @@ static boolean emit_SUB( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); - struct x86_reg dst = aos_get_xmm_reg(cp); + struct x86_reg dst = get_xmm_tmp(cp, arg0); - sse_movups(cp->func, dst, arg0); sse_subps(cp->func, dst, arg1); store_dest(cp, &op->FullDstRegisters[0], dst); -- cgit v1.2.3 From 5b1bd30f22ffa3955150ec008631d0f4754d340f Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Wed, 21 May 2008 20:41:01 +0100 Subject: draw: when preloading args to x87 stack, need to use reverse order --- src/gallium/auxiliary/draw/draw_vs_aos.c | 18 +++++++++++------- 1 file changed, 11 insertions(+), 7 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index 97de43c232..fde92c7226 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -980,7 +980,7 @@ static boolean emit_EXP( struct aos_compilation *cp, const struct tgsi_full_inst x87_fstp_or_pop(cp->func, writemask, 0, dst); /* flr(a) a 2^flr(a) */ - x87_fsubrp(cp->func, st1); /* frac(a) 2^flr(a) */ + x87_fsubp(cp->func, st1); /* frac(a) 2^flr(a) */ x87_fst_or_nop(cp->func, writemask, 1, dst); /* frac(a) 2^flr(a) */ @@ -1041,9 +1041,9 @@ static boolean emit_FLR( struct aos_compilation *cp, const struct tgsi_full_inst /* Load all sources first to avoid aliasing */ - for (i = 0; i < 4; i++) { + for (i = 3; i >= 0; i--) { if (writemask & (1<FullSrcRegisters[0], i); + x87_fld_src(cp, &op->FullSrcRegisters[0], i); } } @@ -1068,9 +1068,9 @@ static boolean emit_RND( struct aos_compilation *cp, const struct tgsi_full_inst /* Load all sources first to avoid aliasing */ - for (i = 0; i < 4; i++) { + for (i = 3; i >= 0; i--) { if (writemask & (1<FullSrcRegisters[0], i); + x87_fld_src(cp, &op->FullSrcRegisters[0], i); } } @@ -1098,7 +1098,7 @@ static boolean emit_FRC( struct aos_compilation *cp, const struct tgsi_full_inst /* suck all the source values onto the stack before writing out any * dst, which may alias... */ - for (i = 0; i < 4; i++) { + for (i = 3; i >= 0; i--) { if (writemask & (1<FullSrcRegisters[0], i); } @@ -1108,7 +1108,7 @@ static boolean emit_FRC( struct aos_compilation *cp, const struct tgsi_full_inst if (writemask & (1<func, st0); /* a a */ x87_fprndint( cp->func ); /* flr(a) a */ - x87_fsubrp(cp->func, st1); /* frc(a) */ + x87_fsubp(cp->func, st1); /* frc(a) */ x87_fstp(cp->func, x86_make_disp(dst, i*4)); } } @@ -1392,6 +1392,8 @@ static boolean emit_instruction( struct aos_compilation *cp, struct tgsi_full_instruction *inst ) { + x87_assert_stack_empty(cp->func); + switch( inst->Instruction.Opcode ) { case TGSI_OPCODE_MOV: return emit_MOV( cp, inst ); @@ -1657,6 +1659,7 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient, break; } + x87_assert_stack_empty(cp.func); cp.insn_counter++; debug_printf("\n"); } @@ -1712,6 +1715,7 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient, x86_pop(cp.func, cp.count_ESI); x86_pop(cp.func, cp.idx_EBX); + x87_assert_stack_empty(cp.func); x86_ret(cp.func); tgsi_parse_free( &parse ); -- cgit v1.2.3 From 6f407b072453eb2bb7077a952257a099db4da025 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Wed, 21 May 2008 20:50:36 +0100 Subject: rtasm: remove debug --- src/gallium/auxiliary/rtasm/rtasm_x86sse.c | 2 -- 1 file changed, 2 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c index a2e8af343b..d78676b8f3 100644 --- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c +++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c @@ -990,14 +990,12 @@ static void note_x87_pop( struct x86_function *p ) { p->x87_stack--; assert(p->x87_stack >= 0); - debug_printf("\nstack: %d\n", p->x87_stack); } static void note_x87_push( struct x86_function *p ) { p->x87_stack++; assert(p->x87_stack <= 7); - debug_printf("\nstack: %d\n", p->x87_stack); } void x87_assert_stack_empty( struct x86_function *p ) -- cgit v1.2.3 From a5c3b499fa40f46298389900e74f1db04f99166a Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 22 May 2008 13:37:48 +0100 Subject: draw: fse works with elts, remove assert --- src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c | 10 ++-------- 1 file changed, 2 insertions(+), 8 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c index 984fbb6767..7fefd391a6 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c @@ -78,11 +78,6 @@ static void fse_prepare( struct draw_pt_middle_end *middle, boolean need_psize = 0; - if (draw->pt.user.elts) { - assert(0); - return ; - } - if (!draw->render->set_primitive( draw->render, prim )) { assert(0); @@ -250,9 +245,8 @@ static void fse_run_linear( struct draw_pt_middle_end *middle, } /* Single routine to fetch vertices, run shader and emit HW verts. - * Clipping and viewport transformation are done elsewhere -- - * either by the API or on hardware, or for some other reason not - * required... + * Clipping is done elsewhere -- either by the API or on hardware, + * or for some other reason not required... */ fse->active->run_linear( fse->active, start, count, -- cgit v1.2.3 From c684ffa02d8d43ee04b99ee63ccd1adb66e81c1a Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 22 May 2008 13:41:49 +0100 Subject: draw: clean up internal immediates in aos sse --- src/gallium/auxiliary/draw/draw_vs_aos.c | 64 +++++++++++++++++++++-------- src/gallium/auxiliary/draw/draw_vs_aos.h | 5 ++- src/gallium/auxiliary/draw/draw_vs_aos_io.c | 10 ++--- 3 files changed, 55 insertions(+), 24 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index fde92c7226..0b8600696a 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -44,12 +44,6 @@ #ifdef PIPE_ARCH_X86 -#define DISASSEM 0 - - - - - static INLINE boolean eq( struct x86_reg a, struct x86_reg b ) { @@ -92,13 +86,6 @@ static struct x86_reg get_reg_ptr(struct aos_compilation *cp, } -struct x86_reg aos_get_internal( struct aos_compilation *cp, - unsigned imm ) -{ - return get_reg_ptr( cp, - AOS_FILE_INTERNAL, - imm ); -} #define X87_CW_EXCEPTION_INV_OP (1<<0) #define X87_CW_EXCEPTION_DENORM_OP (1<<1) @@ -123,6 +110,9 @@ static void init_internals( struct aos_machine *machine ) float inv = 1.0f/255.0f; float f255 = 255.0f; + ASSIGN_4V(machine->internal[IMM_SWZ], 1.0f, -1.0f, 0.0f, 1.0f); + *(unsigned *)&machine->internal[IMM_SWZ][3] = 0xffffffff; + ASSIGN_4V(machine->internal[IMM_ONES], 1.0f, 1.0f, 1.0f, 1.0f); ASSIGN_4V(machine->internal[IMM_NEGS], -1.0f, -1.0f, -1.0f, -1.0f); ASSIGN_4V(machine->internal[IMM_IDENTITY], 0.0f, 0.0f, 0.0f, 1.0f); @@ -337,6 +327,39 @@ struct x86_reg aos_get_shader_reg( struct aos_compilation *cp, +static struct x86_reg aos_get_shader_reg_xmm( struct aos_compilation *cp, + unsigned file, + unsigned idx ) +{ + struct x86_reg reg = aos_get_shader_reg( cp, file, idx ); + + if (reg.file != file_XMM) { + struct x86_reg tmp = aos_get_xmm_reg(cp); + sse_movups(cp->func, tmp, reg); + aos_adopt_xmm_reg( cp, tmp, file, idx, FALSE ); + reg = tmp; + } + + return reg; +} + + + +struct x86_reg aos_get_internal_xmm( struct aos_compilation *cp, + unsigned imm ) +{ + return aos_get_shader_reg_xmm( cp, AOS_FILE_INTERNAL, imm ); +} + + +struct x86_reg aos_get_internal( struct aos_compilation *cp, + unsigned imm ) +{ + return aos_get_shader_reg( cp, AOS_FILE_INTERNAL, imm ); +} + + + /* Emulate pshufd insn in regular SSE, if necessary: @@ -461,15 +484,15 @@ static struct x86_reg fetch_src( struct aos_compilation *cp, arg0 = dst; } - if (negs) { - struct x86_reg imm_negs = aos_get_internal(cp, IMM_NEGS); + if (negs && negs != 0xf) { + struct x86_reg imm_swz = aos_get_internal_xmm(cp, IMM_SWZ); struct x86_reg tmp = aos_get_xmm_reg(cp); /* Load 1,-1,0,0 * Use neg as arg to pshufd * Multiply */ - emit_pshufd(cp, tmp, imm_negs, + emit_pshufd(cp, tmp, imm_swz, SHUF((negs & 1) ? 1 : 0, (negs & 2) ? 1 : 0, (negs & 4) ? 1 : 0, @@ -479,12 +502,17 @@ static struct x86_reg fetch_src( struct aos_compilation *cp, aos_release_xmm_reg(cp, tmp.idx); arg0 = dst; } + else if (negs) { + struct x86_reg imm_negs = aos_get_internal_xmm(cp, IMM_NEGS); + sse_mulps(cp->func, dst, imm_negs); + arg0 = dst; + } + if (abs && abs != 0xf) { ERROR(cp, "unsupported partial abs"); } - - if (abs) { + else if (abs) { struct x86_reg neg = aos_get_internal(cp, IMM_NEGS); struct x86_reg tmp = aos_get_xmm_reg(cp); diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h index c2afd4e9a0..efdc9a38f4 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.h +++ b/src/gallium/auxiliary/draw/draw_vs_aos.h @@ -142,13 +142,16 @@ boolean aos_emit_outputs( struct aos_compilation *cp ); #define IMM_ONES 0 /* 1, 1,1,1 */ -#define IMM_NEGS 1 /* 1,-1,0,0 */ +#define IMM_SWZ 1 /* 1,-1,0, 0xffffffff */ #define IMM_IDENTITY 2 /* 0, 0,0,1 */ #define IMM_INV_255 3 /* 1/255, 1/255, 1/255, 1/255 */ #define IMM_255 4 /* 255, 255, 255, 255 */ +#define IMM_NEGS 5 /* -1,-1,-1,-1 */ struct x86_reg aos_get_internal( struct aos_compilation *cp, unsigned imm ); +struct x86_reg aos_get_internal_xmm( struct aos_compilation *cp, + unsigned imm ); #define ERROR(cp, msg) \ diff --git a/src/gallium/auxiliary/draw/draw_vs_aos_io.c b/src/gallium/auxiliary/draw/draw_vs_aos_io.c index 72b2b3d11d..0dda9df97d 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos_io.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos_io.c @@ -54,7 +54,7 @@ static void emit_load_R32G32B32( struct aos_compilation *cp, struct x86_reg src_ptr ) { sse_movss(cp->func, data, x86_make_disp(src_ptr, 8)); - sse_shufps(cp->func, data, aos_get_internal( cp, IMM_IDENTITY ), SHUF(X,Y,Z,W) ); + sse_shufps(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY ), SHUF(X,Y,Z,W) ); sse_shufps(cp->func, data, data, SHUF(Y,Z,X,W) ); sse_movlps(cp->func, data, src_ptr); } @@ -63,7 +63,7 @@ static void emit_load_R32G32( struct aos_compilation *cp, struct x86_reg data, struct x86_reg src_ptr ) { - sse_movups(cp->func, data, aos_get_internal( cp, IMM_IDENTITY ) ); + sse_movups(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY ) ); sse_movlps(cp->func, data, src_ptr); } @@ -73,7 +73,7 @@ static void emit_load_R32( struct aos_compilation *cp, struct x86_reg src_ptr ) { sse_movss(cp->func, data, src_ptr); - sse_orps(cp->func, data, aos_get_internal( cp, IMM_IDENTITY ) ); + sse_orps(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY ) ); } @@ -82,8 +82,8 @@ static void emit_load_R8G8B8A8_UNORM( struct aos_compilation *cp, struct x86_reg src_ptr ) { sse_movss(cp->func, data, src_ptr); - sse2_punpcklbw(cp->func, data, aos_get_internal( cp, IMM_IDENTITY )); - sse2_punpcklbw(cp->func, data, aos_get_internal( cp, IMM_IDENTITY )); + sse2_punpcklbw(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY )); + sse2_punpcklbw(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY )); sse2_cvtdq2ps(cp->func, data, data); sse_mulps(cp->func, data, aos_get_internal(cp, IMM_INV_255)); } -- cgit v1.2.3 From 05029c919d46299ca259ee8af880d0a65f95ce7c Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 22 May 2008 13:46:06 +0100 Subject: draw: clean up masked writes in aos sse, make some xmm function names clearer --- src/gallium/auxiliary/draw/draw_vs_aos.c | 236 +++++++++++-------------------- 1 file changed, 82 insertions(+), 154 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index 0b8600696a..708ecadbac 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -171,7 +171,7 @@ static boolean is_xmm_tmp( struct aos_compilation *cp, cp->xmm[reg.idx].file == TGSI_FILE_NULL); } -static struct x86_reg get_xmm_tmp( struct aos_compilation *cp, +static struct x86_reg get_xmm_clone( struct aos_compilation *cp, struct x86_reg reg ) { if (!is_xmm_tmp(cp, reg)) { @@ -380,31 +380,37 @@ static void emit_pshufd( struct aos_compilation *cp, } } - - - -/* Helper for writemask: +/* load masks (pack into negs??) + * pshufd - shuffle according to writemask + * and - result, mask + * nand - dest, mask + * or - dest, result */ -static boolean emit_shuf_copy1( struct aos_compilation *cp, - struct x86_reg dst, - struct x86_reg arg0, - struct x86_reg arg1, - ubyte shuf ) +static boolean mask_write( struct aos_compilation *cp, + struct x86_reg dst, + struct x86_reg result, + unsigned mask ) { + struct x86_reg imm_swz = aos_get_internal_xmm(cp, IMM_SWZ); struct x86_reg tmp = aos_get_xmm_reg(cp); - sse_movups(cp->func, dst, arg1); - emit_pshufd(cp, dst, dst, shuf); - emit_pshufd(cp, tmp, arg0, shuf); - - sse_movss(cp->func, dst, tmp); + + emit_pshufd(cp, tmp, imm_swz, + SHUF((mask & 1) ? 2 : 3, + (mask & 2) ? 2 : 3, + (mask & 4) ? 2 : 3, + (mask & 8) ? 2 : 3)); - emit_pshufd(cp, dst, dst, shuf); + sse_andps(cp->func, dst, tmp); + sse_andnps(cp->func, tmp, result); + sse_orps(cp->func, dst, tmp); aos_release_xmm_reg(cp, tmp.idx); return TRUE; } + + /* Helper for writemask: */ static boolean emit_shuf_copy2( struct aos_compilation *cp, @@ -414,17 +420,18 @@ static boolean emit_shuf_copy2( struct aos_compilation *cp, ubyte shuf ) { struct x86_reg tmp = aos_get_xmm_reg(cp); + emit_pshufd(cp, dst, arg1, shuf); emit_pshufd(cp, tmp, arg0, shuf); - sse_shufps(cp->func, dst, tmp, SHUF(X, Y, Z, W)); - emit_pshufd(cp, dst, dst, shuf); aos_release_xmm_reg(cp, tmp.idx); return TRUE; } + + #define SSE_SWIZZLE_NOOP ((0<<0) | (1<<2) | (2<<4) | (3<<6)) @@ -593,131 +600,58 @@ static void store_dest( struct aos_compilation *cp, const struct tgsi_full_dst_register *reg, struct x86_reg result ) { - if (reg->DstRegister.WriteMask == 0) - { - return; - } - else if (reg->DstRegister.WriteMask == TGSI_WRITEMASK_XYZW) - { - if (result.file == file_XMM) { - aos_adopt_xmm_reg(cp, - result, - reg->DstRegister.File, - reg->DstRegister.Index, - TRUE); - } - else { - struct x86_reg dst = aos_get_xmm_reg(cp); - aos_adopt_xmm_reg(cp, - dst, - reg->DstRegister.File, - reg->DstRegister.Index, - TRUE); - sse_movups(cp->func, dst, result); - } - } - else - { - /* Previous value of the dest register: - */ - struct x86_reg old_dst = aos_get_shader_reg(cp, - reg->DstRegister.File, - reg->DstRegister.Index); - - - /* Alloc an xmm reg to hold the new value of the dest register: - */ - struct x86_reg dst = aos_get_xmm_reg(cp); + struct x86_reg dst; + switch (reg->DstRegister.WriteMask) { + case 0: + return; + + case TGSI_WRITEMASK_XYZW: aos_adopt_xmm_reg(cp, - dst, + get_xmm_clone(cp, result), reg->DstRegister.File, reg->DstRegister.Index, - TRUE ); - - switch (reg->DstRegister.WriteMask) { - case TGSI_WRITEMASK_X: - if (result.file == file_XMM) { - sse_movups(cp->func, dst, old_dst); - sse_movss(cp->func, dst, result); - } - else { - struct x86_reg tmp = aos_get_xmm_reg(cp); - sse_movups(cp->func, dst, old_dst); - sse_movss(cp->func, tmp, result); - sse_movss(cp->func, dst, tmp); - aos_release_xmm_reg(cp, tmp.idx); - } - break; - - case TGSI_WRITEMASK_XY: - sse_movups(cp->func, dst, old_dst); - sse_shufps(cp->func, dst, result, SHUF(X, Y, Z, W)); - break; - - case TGSI_WRITEMASK_ZW: - sse_movups(cp->func, dst, result); - sse_shufps(cp->func, dst, old_dst, SHUF(X, Y, Z, W)); - break; - - case TGSI_WRITEMASK_YZW: - if (old_dst.file == file_XMM) { - sse_movups(cp->func, dst, result); - sse_movss(cp->func, dst, old_dst); - } - else { - struct x86_reg tmp = aos_get_xmm_reg(cp); - sse_movups(cp->func, dst, result); - sse_movss(cp->func, tmp, old_dst); - sse_movss(cp->func, dst, tmp); - aos_release_xmm_reg(cp, tmp.idx); - } - break; - - case TGSI_WRITEMASK_Y: - emit_shuf_copy1(cp, dst, result, old_dst, SHUF(Y,X,Z,W)); - break; - - case TGSI_WRITEMASK_Z: - emit_shuf_copy1(cp, dst, result, old_dst, SHUF(Z,Y,X,W)); - break; - - case TGSI_WRITEMASK_W: - emit_shuf_copy1(cp, dst, result, old_dst, SHUF(W,Y,Z,X)); - break; - - case TGSI_WRITEMASK_XZ: - emit_shuf_copy2(cp, dst, result, old_dst, SHUF(X,Z,Y,W)); - break; + TRUE); + return; + default: + break; + } - case TGSI_WRITEMASK_XW: - emit_shuf_copy2(cp, dst, result, old_dst, SHUF(X,W,Z,Y)); + dst = aos_get_shader_reg_xmm(cp, + reg->DstRegister.File, + reg->DstRegister.Index); - case TGSI_WRITEMASK_YZ: - emit_shuf_copy2(cp, dst, result, old_dst, SHUF(Z,Y,X,W)); - break; + switch (reg->DstRegister.WriteMask) { + case TGSI_WRITEMASK_X: + sse_movss(cp->func, dst, get_xmm_clone(cp, result)); + break; + + case TGSI_WRITEMASK_XY: + sse_shufps(cp->func, dst, get_xmm_clone(cp, result), SHUF(X, Y, Z, W)); + break; - case TGSI_WRITEMASK_YW: - emit_shuf_copy2(cp, dst, result, old_dst, SHUF(W,Y,Z,X)); - break; + case TGSI_WRITEMASK_ZW: + result = get_xmm_clone(cp, result); + sse_shufps(cp->func, result, dst, SHUF(X, Y, Z, W)); + dst = result; + break; - case TGSI_WRITEMASK_XZW: - emit_shuf_copy1(cp, dst, old_dst, result, SHUF(Y,X,Z,W)); - break; + case TGSI_WRITEMASK_YZW: + sse_movss(cp->func, result, dst); + dst = result; + break; - case TGSI_WRITEMASK_XYW: - emit_shuf_copy1(cp, dst, old_dst, result, SHUF(Z,Y,X,W)); - break; + default: + mask_write(cp, dst, result, reg->DstRegister.WriteMask); + break; + } - case TGSI_WRITEMASK_XYZ: - emit_shuf_copy1(cp, dst, old_dst, result, SHUF(W,Y,Z,X)); - break; + aos_adopt_xmm_reg(cp, + dst, + reg->DstRegister.File, + reg->DstRegister.Index, + TRUE); - default: - assert(0); /* not possible */ - break; - } - } } @@ -837,7 +771,7 @@ static boolean emit_ABS( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg neg = aos_get_internal(cp, IMM_NEGS); - struct x86_reg dst = get_xmm_tmp(cp, arg0); + struct x86_reg dst = get_xmm_clone(cp, arg0); sse_mulps(cp->func, dst, neg); sse_maxps(cp->func, dst, arg0); @@ -850,7 +784,7 @@ static boolean emit_ADD( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); - struct x86_reg dst = get_xmm_tmp(cp, arg0); + struct x86_reg dst = get_xmm_clone(cp, arg0); sse_addps(cp->func, dst, arg1); @@ -874,7 +808,7 @@ static boolean emit_DP3( struct aos_compilation *cp, const struct tgsi_full_inst struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); struct x86_reg tmp = aos_get_xmm_reg(cp); - struct x86_reg dst = get_xmm_tmp(cp, arg0); + struct x86_reg dst = get_xmm_clone(cp, arg0); sse_mulps(cp->func, dst, arg1); @@ -898,7 +832,7 @@ static boolean emit_DP4( struct aos_compilation *cp, const struct tgsi_full_inst struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); struct x86_reg tmp = aos_get_xmm_reg(cp); - struct x86_reg dst = get_xmm_tmp(cp, arg0); + struct x86_reg dst = get_xmm_clone(cp, arg0); sse_mulps(cp->func, dst, arg1); @@ -920,7 +854,7 @@ static boolean emit_DPH( struct aos_compilation *cp, const struct tgsi_full_inst struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); struct x86_reg tmp = aos_get_xmm_reg(cp); - struct x86_reg dst = get_xmm_tmp(cp, arg0); + struct x86_reg dst = get_xmm_clone(cp, arg0); sse_mulps(cp->func, dst, arg1); @@ -1216,7 +1150,7 @@ static boolean emit_MAX( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); - struct x86_reg dst = get_xmm_tmp(cp, arg0); + struct x86_reg dst = get_xmm_clone(cp, arg0); sse_maxps(cp->func, dst, arg1); @@ -1229,7 +1163,7 @@ static boolean emit_MIN( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); - struct x86_reg dst = get_xmm_tmp(cp, arg0); + struct x86_reg dst = get_xmm_clone(cp, arg0); sse_minps(cp->func, dst, arg1); @@ -1240,7 +1174,7 @@ static boolean emit_MIN( struct aos_compilation *cp, const struct tgsi_full_inst static boolean emit_MOV( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); - struct x86_reg dst = get_xmm_tmp(cp, arg0); + struct x86_reg dst = get_xmm_clone(cp, arg0); /* potentially nothing to do */ @@ -1252,7 +1186,7 @@ static boolean emit_MUL( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); - struct x86_reg dst = get_xmm_tmp(cp, arg0); + struct x86_reg dst = get_xmm_clone(cp, arg0); sse_mulps(cp->func, dst, arg1); @@ -1270,7 +1204,7 @@ static boolean emit_MAD( struct aos_compilation *cp, const struct tgsi_full_inst /* If we can't clobber old contents of arg0, get a temporary & copy * it there, then clobber it... */ - arg0 = get_xmm_tmp(cp, arg0); + arg0 = get_xmm_clone(cp, arg0); sse_mulps(cp->func, arg0, arg1); sse_addps(cp->func, arg0, arg2); @@ -1336,7 +1270,7 @@ static boolean emit_SGE( struct aos_compilation *cp, const struct tgsi_full_inst struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); struct x86_reg ones = aos_get_internal(cp, IMM_ONES); - struct x86_reg dst = get_xmm_tmp(cp, arg0); + struct x86_reg dst = get_xmm_clone(cp, arg0); sse_cmpps(cp->func, dst, arg1, cc_NotLessThan); sse_andps(cp->func, dst, ones); @@ -1360,7 +1294,7 @@ static boolean emit_SLT( struct aos_compilation *cp, const struct tgsi_full_inst struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); struct x86_reg ones = aos_get_internal(cp, IMM_ONES); - struct x86_reg dst = get_xmm_tmp(cp, arg0); + struct x86_reg dst = get_xmm_clone(cp, arg0); sse_cmpps(cp->func, dst, arg1, cc_LessThan); sse_andps(cp->func, dst, ones); @@ -1373,7 +1307,7 @@ static boolean emit_SUB( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); - struct x86_reg dst = get_xmm_tmp(cp, arg0); + struct x86_reg dst = get_xmm_clone(cp, arg0); sse_subps(cp->func, dst, arg1); @@ -1526,9 +1460,9 @@ emit_instruction( struct aos_compilation *cp, static boolean emit_viewport( struct aos_compilation *cp ) { - struct x86_reg pos = aos_get_shader_reg(cp, - TGSI_FILE_OUTPUT, - 0); + struct x86_reg pos = aos_get_shader_reg_xmm(cp, + TGSI_FILE_OUTPUT, + 0); struct x86_reg scale = x86_make_disp(cp->machine_EDX, Offset(struct aos_machine, scale)); @@ -1536,12 +1470,6 @@ static boolean emit_viewport( struct aos_compilation *cp ) struct x86_reg translate = x86_make_disp(cp->machine_EDX, Offset(struct aos_machine, translate)); - if (pos.file != file_XMM) { - struct x86_reg dst = aos_get_xmm_reg(cp); - sse_movups(cp->func, dst, pos); - pos = dst; - } - sse_mulps(cp->func, pos, scale); sse_addps(cp->func, pos, translate); -- cgit v1.2.3 From 7b25c1a4032960752d8a8e950bdf75740b2de2e8 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 22 May 2008 13:47:08 +0100 Subject: draw: remove FPU_MANIP ifdef --- src/gallium/auxiliary/draw/draw_vs_aos.c | 11 +---------- 1 file changed, 1 insertion(+), 10 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index 708ecadbac..d60940bb7a 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -693,48 +693,39 @@ static void x87_fstp_dest4( struct aos_compilation *cp, x87_fstp_or_pop(cp->func, writemask, 3, ptr); } -#define FPU_MANIP 1 /* Save current x87 state and put it into single precision mode. */ static void save_fpu_state( struct aos_compilation *cp ) { -#if FPU_MANIP x87_fnstcw( cp->func, x86_make_disp(cp->machine_EDX, Offset(struct aos_machine, fpu_restore))); -#endif } static void restore_fpu_state( struct aos_compilation *cp ) { -#if FPU_MANIP x87_fnclex(cp->func); x87_fldcw( cp->func, x86_make_disp(cp->machine_EDX, Offset(struct aos_machine, fpu_restore))); -#endif } static void set_fpu_round_neg_inf( struct aos_compilation *cp ) { -#if FPU_MANIP if (cp->fpucntl != FPU_RND_NEG) { cp->fpucntl = FPU_RND_NEG; x87_fnclex(cp->func); x87_fldcw( cp->func, x86_make_disp(cp->machine_EDX, Offset(struct aos_machine, fpu_rnd_neg_inf))); } -#endif } static void set_fpu_round_nearest( struct aos_compilation *cp ) { -#if FPU_MANIP if (cp->fpucntl != FPU_RND_NEAREST) { cp->fpucntl = FPU_RND_NEAREST; x87_fnclex(cp->func); x87_fldcw( cp->func, x86_make_disp(cp->machine_EDX, Offset(struct aos_machine, fpu_rnd_nearest))); } -#endif } @@ -754,7 +745,7 @@ static void x87_emit_ex2( struct aos_compilation *cp ) x87_fld1(cp->func); /* 1 (2^frac(a))-1 a */ x87_faddp(cp->func, st1); /* 2^frac(a) a */ x87_fscale(cp->func); /* 2^a a */ - x87_fstp(cp->func, st1); + x87_fstp(cp->func, st1); /* 2^a */ assert( stack == cp->func->x87_stack); -- cgit v1.2.3 From 6780a6dede31e7f2eb465e1d7b507b3e64fe6ec9 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 22 May 2008 13:48:07 +0100 Subject: draw: shortcircuit shuffle in aos_sse when possible --- src/gallium/auxiliary/draw/draw_vs_aos.c | 18 +++++++++++++----- 1 file changed, 13 insertions(+), 5 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index d60940bb7a..b8fad231ca 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -809,7 +809,9 @@ static boolean emit_DP3( struct aos_compilation *cp, const struct tgsi_full_inst sse_addss(cp->func, dst, tmp); /* a*x+c*z, b*y, ?, ? */ emit_pshufd(cp, tmp, dst, SHUF(Y,X,W,Z)); sse_addss(cp->func, dst, tmp); - sse_shufps(cp->func, dst, dst, SHUF(X, X, X, X)); + + if (op->FullDstRegisters[0].DstRegister.WriteMask != 0x1) + sse_shufps(cp->func, dst, dst, SHUF(X, X, X, X)); aos_release_xmm_reg(cp, tmp.idx); store_dest(cp, &op->FullDstRegisters[0], dst); @@ -833,7 +835,9 @@ static boolean emit_DP4( struct aos_compilation *cp, const struct tgsi_full_inst sse_addps(cp->func, dst, tmp); /* a*x+c*z, b*y+d*w, a*x+c*z, b*y+d*w */ emit_pshufd(cp, tmp, dst, SHUF(Y,X,W,Z)); sse_addss(cp->func, dst, tmp); - sse_shufps(cp->func, dst, dst, SHUF(X, X, X, X)); + + if (op->FullDstRegisters[0].DstRegister.WriteMask != 0x1) + sse_shufps(cp->func, dst, dst, SHUF(X, X, X, X)); aos_release_xmm_reg(cp, tmp.idx); store_dest(cp, &op->FullDstRegisters[0], dst); @@ -857,7 +861,9 @@ static boolean emit_DPH( struct aos_compilation *cp, const struct tgsi_full_inst sse_addss(cp->func, dst, tmp); emit_pshufd(cp, tmp, arg1, SHUF(W,W,W,W)); sse_addss(cp->func, dst, tmp); - sse_shufps(cp->func, dst, dst, SHUF(X, X, X, X)); + + if (op->FullDstRegisters[0].DstRegister.WriteMask != 0x1) + sse_shufps(cp->func, dst, dst, SHUF(X, X, X, X)); aos_release_xmm_reg(cp, tmp.idx); store_dest(cp, &op->FullDstRegisters[0], dst); @@ -1233,7 +1239,8 @@ static boolean emit_RCP( struct aos_compilation *cp, const struct tgsi_full_inst sse_divss(cp->func, dst, arg0); } - sse_shufps(cp->func, dst, dst, SHUF(X, X, X, X)); + if (op->FullDstRegisters[0].DstRegister.WriteMask != 0x1) + sse_shufps(cp->func, dst, dst, SHUF(X, X, X, X)); store_dest(cp, &op->FullDstRegisters[0], dst); return TRUE; @@ -1249,7 +1256,8 @@ static boolean emit_RSQ( struct aos_compilation *cp, const struct tgsi_full_inst /* Extend precision here... */ - sse_shufps(cp->func, dst, dst, SHUF(X, X, X, X)); + if (op->FullDstRegisters[0].DstRegister.WriteMask != 0x1) + sse_shufps(cp->func, dst, dst, SHUF(X, X, X, X)); store_dest(cp, &op->FullDstRegisters[0], dst); return TRUE; -- cgit v1.2.3 From 65cb09249e750b45ec3fc9a57670fc77250efc5e Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 22 May 2008 13:49:38 +0100 Subject: draw: for debug, do rhw divide in aos_sse viewport calcs --- src/gallium/auxiliary/draw/draw_vs_aos.c | 45 +++++++++++++++++++++++++++++++- 1 file changed, 44 insertions(+), 1 deletion(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index b8fad231ca..40de13a98c 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -1481,6 +1481,46 @@ static boolean emit_viewport( struct aos_compilation *cp ) } +/* This is useful to be able to see the results on softpipe. Doesn't + * do proper clipping, just assumes the backend can do it during + * rasterization -- for debug only... + */ +static boolean emit_rhw_viewport( struct aos_compilation *cp ) +{ + struct x86_reg tmp = aos_get_xmm_reg(cp); + struct x86_reg pos = aos_get_shader_reg_xmm(cp, + TGSI_FILE_OUTPUT, + 0); + + struct x86_reg scale = x86_make_disp(cp->machine_EDX, + Offset(struct aos_machine, scale)); + + struct x86_reg translate = x86_make_disp(cp->machine_EDX, + Offset(struct aos_machine, translate)); + + + + emit_pshufd(cp, tmp, pos, SHUF(W, W, W, W)); + sse2_rcpss(cp->func, tmp, tmp); + sse_shufps(cp->func, tmp, tmp, SHUF(X, X, X, X)); + + sse_mulps(cp->func, pos, scale); + sse_mulps(cp->func, pos, tmp); + sse_addps(cp->func, pos, translate); + + /* Set pos[3] = w + */ + mask_write(cp, pos, tmp, TGSI_WRITEMASK_W); + + aos_adopt_xmm_reg( cp, + pos, + TGSI_FILE_OUTPUT, + 0, + TRUE ); + return TRUE; +} + + static boolean note_immediate( struct aos_compilation *cp, struct tgsi_full_immediate *imm ) { @@ -1623,7 +1663,10 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient, goto fail; if (cp.vaos->base.key.viewport) { - emit_viewport(&cp); + if (0) + emit_viewport(&cp); + else + emit_rhw_viewport(&cp); } /* Emit output... TODO: do this eagerly after the last write to a -- cgit v1.2.3 From 260001430bbd28ea17201f1980ab1ebed93b246f Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 22 May 2008 15:24:02 +0100 Subject: draw: use aligned movs within draw_vs_aos.c --- src/gallium/auxiliary/draw/draw_vs_aos.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index 40de13a98c..039e233fe8 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -159,7 +159,7 @@ static void spill( struct aos_compilation *cp, unsigned idx ) cp->xmm[idx].idx); assert(cp->xmm[idx].dirty); - sse_movups(cp->func, oldval, x86_make_reg(file_XMM, idx)); + sse_movaps(cp->func, oldval, x86_make_reg(file_XMM, idx)); cp->xmm[idx].dirty = 0; } } @@ -176,7 +176,7 @@ static struct x86_reg get_xmm_clone( struct aos_compilation *cp, { if (!is_xmm_tmp(cp, reg)) { struct x86_reg tmp = aos_get_xmm_reg(cp); - sse_movups(cp->func, tmp, reg); + sse_movaps(cp->func, tmp, reg); reg = tmp; } @@ -335,7 +335,7 @@ static struct x86_reg aos_get_shader_reg_xmm( struct aos_compilation *cp, if (reg.file != file_XMM) { struct x86_reg tmp = aos_get_xmm_reg(cp); - sse_movups(cp->func, tmp, reg); + sse_movaps(cp->func, tmp, reg); aos_adopt_xmm_reg( cp, tmp, file, idx, FALSE ); reg = tmp; } @@ -374,7 +374,7 @@ static void emit_pshufd( struct aos_compilation *cp, } else { if (!eq(dst, arg0)) - sse_movups(cp->func, dst, arg0); + sse_movaps(cp->func, dst, arg0); sse_shufps(cp->func, dst, dst, shuf); } @@ -523,7 +523,7 @@ static struct x86_reg fetch_src( struct aos_compilation *cp, struct x86_reg neg = aos_get_internal(cp, IMM_NEGS); struct x86_reg tmp = aos_get_xmm_reg(cp); - sse_movups(cp->func, tmp, arg0); + sse_movaps(cp->func, tmp, arg0); sse_mulps(cp->func, tmp, neg); sse_maxps(cp->func, dst, arg0); -- cgit v1.2.3 From 43df4642f1d2f3d2673a1d5e4f5126f5175fb899 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 22 May 2008 20:21:49 +0100 Subject: draw: tweak x87_emit_ex2 to avoid changing x87 fpu settings --- src/gallium/auxiliary/draw/draw_vs_aos.c | 17 +++++++++-------- 1 file changed, 9 insertions(+), 8 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index 039e233fe8..93bb4f9bc0 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -735,16 +735,17 @@ static void x87_emit_ex2( struct aos_compilation *cp ) struct x86_reg st1 = x86_make_reg(file_x87, 1); int stack = cp->func->x87_stack; - set_fpu_round_neg_inf( cp ); +// set_fpu_round_neg_inf( cp ); x87_fld(cp->func, st0); /* a a */ - x87_fld(cp->func, st0); /* a a a */ - x87_fprndint( cp->func ); /* flr(a) a a*/ - x87_fsubp(cp->func, st1); /* frac(a) a */ - x87_f2xm1(cp->func); /* (2^frac(a))-1 a */ - x87_fld1(cp->func); /* 1 (2^frac(a))-1 a */ - x87_faddp(cp->func, st1); /* 2^frac(a) a */ - x87_fscale(cp->func); /* 2^a a */ + x87_fprndint( cp->func ); /* int(a) a*/ + x87_fsubr(cp->func, st1, st0); /* int(a) frc(a) */ + x87_fxch(cp->func, st1); /* frc(a) int(a) */ + x87_f2xm1(cp->func); /* (2^frc(a))-1 int(a) */ + x87_fld1(cp->func); /* 1 (2^frc(a))-1 int(a) */ + x87_faddp(cp->func, st1); /* 2^frac(a) int(a) */ + x87_fscale(cp->func); /* (2^frac(a)*2^int(int(a))) int(a) */ + /* 2^a int(a) */ x87_fstp(cp->func, st1); /* 2^a */ assert( stack == cp->func->x87_stack); -- cgit v1.2.3 From 7106da136069f865747e03c30ca245bc030b241b Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 22 May 2008 20:22:15 +0100 Subject: draw: correct but slow LIT() in aos varient --- src/gallium/auxiliary/draw/draw_vs_aos.c | 19 +++++++++---------- 1 file changed, 9 insertions(+), 10 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index 93bb4f9bc0..930914f609 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -1093,10 +1093,12 @@ static boolean emit_LIT( struct aos_compilation *cp, const struct tgsi_full_inst - /* a1' = a1 <= 0 ? 1 : a1; + /* a1' = a1 <= 0 ? 1 : a1; + * + * Note: use 1.0 to avoid passing zero to */ - x87_fldz(cp->func); /* 0 */ - x87_fld1(cp->func); /* 1 0 */ + x87_fldz(cp->func); /* 1 0 */ + x87_fldz(cp->func); /* 1 0 */ x87_fld_src(cp, &op->FullSrcRegisters[0], 1); /* a1 1 0 */ x87_fcomi(cp->func, st2); /* a1 1 0 */ x87_fcmovb(cp->func, st1); /* a1' 1 0 */ @@ -1119,17 +1121,14 @@ static boolean emit_LIT( struct aos_compilation *cp, const struct tgsi_full_inst x87_fld_src(cp, &op->FullSrcRegisters[0], 0); /* a0 0 r2 */ x87_fcomi(cp->func, st1); x87_fcmovb(cp->func, st1); /* a0' 0 r2 */ - x87_fstp(cp->func, st1); /* a0' r2 */ - x87_fxch(cp->func, st1); /* a0' r2 */ x87_fst_or_nop(cp->func, writemask, 1, dst); /* result[1] = a0' */ - x87_fldz(cp->func); /* 0 a0' r2 */ - x87_fcomi(cp->func, st1); /* 0 a0' r2 */ - x87_fcmovnbe(cp->func, st2); /* r2' a0' r2 */ + x87_fcomi(cp->func, st1); /* a0' 0 r2 */ + x87_fcmovnbe(cp->func, st2); /* r2' 0' r2 */ - x87_fstp_or_pop(cp->func, writemask, 2, dst); - x87_fpop(cp->func); + x87_fstp_or_pop(cp->func, writemask, 2, dst); /* 0 r2 */ + x87_fpop(cp->func); /* r2 */ x87_fpop(cp->func); } -- cgit v1.2.3 From 3b41d619a1b7cc8c356c32af777486461ddd7926 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Fri, 23 May 2008 09:14:17 +0100 Subject: draw: faster LIT(), incorrect though --- src/gallium/auxiliary/draw/draw_vs_aos.c | 6 ++++++ 1 file changed, 6 insertions(+) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index 930914f609..b0c3ac49d2 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -1098,7 +1098,13 @@ static boolean emit_LIT( struct aos_compilation *cp, const struct tgsi_full_inst * Note: use 1.0 to avoid passing zero to */ x87_fldz(cp->func); /* 1 0 */ +#if 1 + x87_fld1(cp->func); /* 1 0 */ +#else + /* Correct but slow due to fp exceptions generated in fyl2x - fix me. + */ x87_fldz(cp->func); /* 1 0 */ +#endif x87_fld_src(cp, &op->FullSrcRegisters[0], 1); /* a1 1 0 */ x87_fcomi(cp->func, st2); /* a1 1 0 */ x87_fcmovb(cp->func, st1); /* a1' 1 0 */ -- cgit v1.2.3 From 6b3723ee8d084a1abbc971b21c58f7c1e66949a7 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sat, 24 May 2008 13:22:15 +0100 Subject: rtasm: add some helpers for calling out from generated code --- src/gallium/auxiliary/rtasm/rtasm_x86sse.c | 44 ++++++++++++++++++++++++++++-- src/gallium/auxiliary/rtasm/rtasm_x86sse.h | 6 ++++ 2 files changed, 47 insertions(+), 3 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c index d78676b8f3..2415b0156b 100644 --- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c +++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c @@ -218,6 +218,8 @@ static void emit_3ub( struct x86_function *p, unsigned char b0, unsigned char b1 /* Build a modRM byte + possible displacement. No treatment of SIB * indexing. BZZT - no way to encode an absolute address. + * + * This is the "/r" field in the x86 manuals... */ static void emit_modrm( struct x86_function *p, struct x86_reg reg, @@ -256,7 +258,8 @@ static void emit_modrm( struct x86_function *p, } } - +/* Emits the "/0".."/7" specialized versions of the modrm ("/r") bytes. + */ static void emit_modrm_noreg( struct x86_function *p, unsigned op, struct x86_reg regmem ) @@ -365,8 +368,7 @@ void x86_jcc( struct x86_function *p, DUMP_I(cc); if (offset < 0) { - int amt = p->csr - p->store; - assert(amt > -offset); + assert(p->csr - p->store > -offset); } if (offset <= 127 && offset >= -128) { @@ -443,6 +445,16 @@ void x86_mov_reg_imm( struct x86_function *p, struct x86_reg dst, int imm ) emit_1i(p, imm); } +void x86_add_reg_imm8( struct x86_function *p, struct x86_reg dst, ubyte imm ) +{ + DUMP_RI( dst, imm ); + assert(dst.mod == mod_REG); + emit_1ub(p, 0x80); + emit_modrm_noreg(p, 0, dst); + emit_1ub(p, imm); +} + + void x86_push( struct x86_function *p, struct x86_reg reg ) { @@ -459,6 +471,17 @@ void x86_push( struct x86_function *p, p->stack_offset += 4; } +void x86_push_imm32( struct x86_function *p, + int imm32 ) +{ + DUMP_I( imm32 ); + emit_1ub(p, 0x68); + emit_1i(p, imm32); + + p->stack_offset += 4; +} + + void x86_pop( struct x86_function *p, struct x86_reg reg ) { @@ -1558,6 +1581,21 @@ void mmx_movq( struct x86_function *p, */ +void x86_cdecl_caller_push_regs( struct x86_function *p ) +{ + x86_push(p, x86_make_reg(file_REG32, reg_AX)); + x86_push(p, x86_make_reg(file_REG32, reg_CX)); + x86_push(p, x86_make_reg(file_REG32, reg_DX)); +} + +void x86_cdecl_caller_pop_regs( struct x86_function *p ) +{ + x86_pop(p, x86_make_reg(file_REG32, reg_DX)); + x86_pop(p, x86_make_reg(file_REG32, reg_CX)); + x86_pop(p, x86_make_reg(file_REG32, reg_AX)); +} + + /* Retreive a reference to one of the function arguments, taking into * account any push/pop activity: */ diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h index 9f7e31e055..63e812fac9 100644 --- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h +++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h @@ -155,6 +155,7 @@ void x86_call( struct x86_function *p, struct x86_reg reg); * I load the immediate into general purpose register and use it. */ void x86_mov_reg_imm( struct x86_function *p, struct x86_reg dst, int imm ); +void x86_add_reg_imm8( struct x86_function *p, struct x86_reg dst, ubyte imm ); /* Macro for sse_shufps() and sse2_pshufd(): @@ -225,6 +226,7 @@ void x86_imul( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void x86_or( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void x86_pop( struct x86_function *p, struct x86_reg reg ); void x86_push( struct x86_function *p, struct x86_reg reg ); +void x86_push_imm32( struct x86_function *p, int imm ); void x86_ret( struct x86_function *p ); void x86_retw( struct x86_function *p, unsigned short imm ); void x86_sub( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); @@ -232,6 +234,10 @@ void x86_test( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void x86_xor( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void x86_sahf( struct x86_function *p ); + +void x86_cdecl_caller_push_regs( struct x86_function *p ); +void x86_cdecl_caller_pop_regs( struct x86_function *p ); + void x87_assert_stack_empty( struct x86_function *p ); void x87_f2xm1( struct x86_function *p ); -- cgit v1.2.3 From 6172f1295cf812108d8ceba15a83ba87880360d3 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sat, 24 May 2008 13:22:29 +0100 Subject: draw: add a debug-print which can be called from inside generated shaders --- src/gallium/auxiliary/draw/draw_vs_aos.c | 67 ++++++++++++++++++++++++++++++++ src/gallium/auxiliary/draw/draw_vs_aos.h | 1 + 2 files changed, 68 insertions(+) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index b0c3ac49d2..aa119f242e 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -752,7 +752,63 @@ static void x87_emit_ex2( struct aos_compilation *cp ) } +static void PIPE_CDECL print_reg( const char *msg, + const float *reg ) +{ + debug_printf("%s: %f %f %f %f\n", msg, reg[0], reg[1], reg[2], reg[3]); +} + +static void emit_print( struct aos_compilation *cp, + const char *message, /* must point to a static string! */ + unsigned file, + unsigned idx ) +{ + struct x86_reg ecx = x86_make_reg( file_REG32, reg_CX ); + struct x86_reg arg = get_reg_ptr( cp, file, idx ); + unsigned i; + + /* There shouldn't be anything on the x87 stack. Can add this + * capacity later if need be. + */ + assert(cp->func->x87_stack == 0); + + /* For absolute correctness, need to spill/invalidate all XMM regs + * too. We're obviously not concerned about performance on this + * debug path, so here goes: + */ + for (i = 0; i < 8; i++) { + if (cp->xmm[i].dirty) + spill(cp, i); + aos_release_xmm_reg(cp, i); + } + + /* Push caller-save (ie scratch) regs. + */ + x86_cdecl_caller_push_regs( cp->func ); + + + /* Push the arguments: + */ + x86_lea( cp->func, ecx, arg ); + x86_push( cp->func, ecx ); + x86_push_imm32( cp->func, (int)message ); + + /* Call the helper. Could call debug_printf directly, but + * print_reg is a nice place to put a breakpoint if need be. + */ + x86_mov_reg_imm( cp->func, ecx, (int)print_reg ); + x86_call( cp->func, ecx ); + x86_pop( cp->func, ecx ); + x86_pop( cp->func, ecx ); + + /* Pop caller-save regs + */ + x86_cdecl_caller_pop_regs( cp->func ); + + /* Done... + */ +} /** * The traditional instructions. All operate on internal registers @@ -1798,6 +1854,17 @@ static void vaos_set_constants( struct draw_vs_varient *varient, memcpy(vaos->machine->constant, constants, (vaos->base.vs->info.file_max[TGSI_FILE_CONSTANT] + 1) * 4 * sizeof(float)); + +#if 0 + unsigned i; + for (i =0; i < vaos->base.vs->info.file_max[TGSI_FILE_CONSTANT] + 1; i++) + debug_printf("state %d: %f %f %f %f\n", + i, + constants[i][0], + constants[i][1], + constants[i][2], + constants[i][3]); +#endif } diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h index efdc9a38f4..a0680ec63d 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.h +++ b/src/gallium/auxiliary/draw/draw_vs_aos.h @@ -96,6 +96,7 @@ struct aos_compilation { unsigned insn_counter; unsigned num_immediates; + unsigned count; struct { unsigned idx:16; -- cgit v1.2.3 From 86e529ad90411d21bca3d70984b2db202e7a0cd6 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sat, 24 May 2008 16:31:11 +0100 Subject: draw: use lookup tables to avoid calling pow() in LIT opcode --- src/gallium/auxiliary/draw/draw_vs_aos.c | 251 ++++++++++++++++++++++++++++++- src/gallium/auxiliary/draw/draw_vs_aos.h | 27 ++++ 2 files changed, 272 insertions(+), 6 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index aa119f242e..1fbb7088ca 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -105,8 +105,31 @@ static struct x86_reg get_reg_ptr(struct aos_compilation *cp, #define X87_CW_ROUND_MASK (3<<10) #define X87_CW_INFINITY (1<<12) +static void do_populate_lut( struct shine_tab *tab, + float unclamped_exponent ) +{ + const float epsilon = 1.0F / 256.0F; + float exponent = CLAMP(unclamped_exponent, -(128.0F - epsilon), (128.0F - epsilon)); + unsigned i; + + tab->exponent = unclamped_exponent; /* for later comparison */ + + tab->values[0] = 0; + if (exponent == 0) { + for (i = 1; i < 258; i++) { + tab->values[i] = 1.0; + } + } + else { + for (i = 1; i < 258; i++) { + tab->values[i] = powf((float)i * epsilon, exponent); + } + } +} + static void init_internals( struct aos_machine *machine ) { + unsigned i; float inv = 1.0f/255.0f; float f255 = 255.0f; @@ -141,6 +164,9 @@ static void init_internals( struct aos_machine *machine ) (1<<6) | X87_CW_ROUND_DOWN | X87_CW_PRECISION_DOUBLE_EXT); + + for (i = 0; i < MAX_SHINE_TAB; i++) + do_populate_lut( &machine->shine_tab[i], 1.0f ); } @@ -1132,26 +1158,231 @@ static boolean emit_FRC( struct aos_compilation *cp, const struct tgsi_full_inst return TRUE; } +static PIPE_CDECL void do_lit( struct aos_machine *machine, + float *result, + const float *in, + unsigned count ) +{ + if (in[0] > 0) + { + if (in[1] <= 0.0) + { + result[0] = 1.0F; + result[1] = in[0]; + result[2] = 1.0; + result[3] = 1.0F; + } + else + { + const float epsilon = 1.0F / 256.0F; + float exponent = CLAMP(in[3], -(128.0F - epsilon), (128.0F - epsilon)); + result[0] = 1.0F; + result[1] = in[0]; + result[2] = powf(in[1], exponent); + result[3] = 1.0; + } + } + else + { + result[0] = 1.0F; + result[1] = 0.0; + result[2] = 0.0; + result[3] = 1.0F; + } +} + + +static PIPE_CDECL void do_lit_lut( struct aos_machine *machine, + float *result, + const float *in, + unsigned count ) +{ + if (in[0] > 0) + { + if (in[1] <= 0.0) + { + result[0] = 1.0F; + result[1] = in[0]; + result[2] = 1.0; + result[3] = 1.0F; + return; + } + + if (machine->lit_info[count].shine_tab->exponent != in[3]) { + machine->lit_info[count].func = do_lit; + goto no_luck; + } + + if (in[1] <= 1.0) + { + const float *tab = machine->lit_info[count].shine_tab->values; + float f = in[1] * 256; + int k = (int)f; + float frac = f - (float)k; + + result[0] = 1.0F; + result[1] = in[0]; + result[2] = tab[k] + frac*(tab[k+1]-tab[k]); + result[3] = 1.0; + return; + } + + no_luck: + { + const float epsilon = 1.0F / 256.0F; + float exponent = CLAMP(in[3], -(128.0F - epsilon), (128.0F - epsilon)); + result[0] = 1.0F; + result[1] = in[0]; + result[2] = powf(in[1], exponent); + result[3] = 1.0; + } + } + else + { + result[0] = 1.0F; + result[1] = 0.0; + result[2] = 0.0; + result[3] = 1.0F; + } +} + + + +static void PIPE_CDECL populate_lut( struct aos_machine *machine, + float *result, + const float *in, + unsigned count ) +{ + unsigned i, tab; + + /* Search for an existing table for this value. Note that without + * static analysis we don't really know if in[3] will be constant, + * but it usually is... + */ + for (tab = 0; tab < 4; tab++) { + if (machine->shine_tab[tab].exponent == in[3]) { + goto found; + } + } + + for (tab = 0, i = 1; i < 4; i++) { + if (machine->shine_tab[i].last_used < machine->shine_tab[tab].last_used) + tab = i; + } + + if (machine->shine_tab[tab].last_used == machine->now) { + /* No unused tables (this is not a ffvertex program...). Just + * call pow each time: + */ + machine->lit_info[count].func = do_lit; + machine->lit_info[count].func( machine, result, in, count ); + return; + } + else { + do_populate_lut( &machine->shine_tab[tab], in[3] ); + } + + found: + machine->shine_tab[tab].last_used = machine->now; + machine->lit_info[count].shine_tab = &machine->shine_tab[tab]; + machine->lit_info[count].func = do_lit_lut; + machine->lit_info[count].func( machine, result, in, count ); +} + + + static boolean emit_LIT( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) { - struct x86_reg dst = get_dst_ptr(cp, &op->FullDstRegisters[0]); + struct x86_reg ecx = x86_make_reg( file_REG32, reg_CX ); unsigned writemask = op->FullDstRegisters[0].DstRegister.WriteMask; + unsigned lit_count = cp->lit_count++; + struct x86_reg result, arg0; + unsigned i; + +#if 1 + /* For absolute correctness, need to spill/invalidate all XMM regs + * too. + */ + for (i = 0; i < 8; i++) { + if (cp->xmm[i].dirty) + spill(cp, i); + aos_release_xmm_reg(cp, i); + } +#endif + + if (writemask != TGSI_WRITEMASK_XYZW) + result = x86_make_disp(cp->machine_EDX, Offset(struct aos_machine, tmp[0])); + else + result = get_dst_ptr(cp, &op->FullDstRegisters[0]); + + + arg0 = fetch_src( cp, &op->FullSrcRegisters[0] ); + if (arg0.file == file_XMM) { + struct x86_reg tmp = x86_make_disp(cp->machine_EDX, + Offset(struct aos_machine, tmp[1])); + sse_movaps( cp->func, tmp, arg0 ); + arg0 = tmp; + } + + + + /* Push caller-save (ie scratch) regs. + */ + x86_cdecl_caller_push_regs( cp->func ); + + /* Push the arguments: + */ + x86_push_imm32( cp->func, lit_count ); + + x86_lea( cp->func, ecx, arg0 ); + x86_push( cp->func, ecx ); + x86_lea( cp->func, ecx, result ); + x86_push( cp->func, ecx ); + x86_push( cp->func, cp->machine_EDX ); + if (lit_count < MAX_LIT_INFO) { + x86_mov( cp->func, ecx, x86_make_disp( cp->machine_EDX, + Offset(struct aos_machine, lit_info) + + lit_count * sizeof(struct lit_info) + + Offset(struct lit_info, func))); + } + else { + x86_mov_reg_imm( cp->func, ecx, (int)do_lit ); + } + + x86_call( cp->func, ecx ); + + x86_pop( cp->func, ecx ); /* fixme... */ + x86_pop( cp->func, ecx ); + x86_pop( cp->func, ecx ); + x86_pop( cp->func, ecx ); + + x86_cdecl_caller_pop_regs( cp->func ); + + if (writemask != TGSI_WRITEMASK_XYZW) { + store_dest( cp, + &op->FullDstRegisters[0], + get_xmm_clone( cp, result ) ); + } + + return TRUE; +} + + +static boolean emit_inline_LIT( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) +{ + struct x86_reg dst = get_dst_ptr(cp, &op->FullDstRegisters[0]); + unsigned writemask = op->FullDstRegisters[0].DstRegister.WriteMask; if (writemask & TGSI_WRITEMASK_YZ) { struct x86_reg st1 = x86_make_reg(file_x87, 1); struct x86_reg st2 = x86_make_reg(file_x87, 2); - - - /* a1' = a1 <= 0 ? 1 : a1; - * - * Note: use 1.0 to avoid passing zero to */ x87_fldz(cp->func); /* 1 0 */ #if 1 @@ -1865,6 +2096,14 @@ static void vaos_set_constants( struct draw_vs_varient *varient, constants[i][2], constants[i][3]); #endif + + { + unsigned i; + for (i = 0; i < MAX_LIT_INFO; i++) { + vaos->machine->lit_info[i].func = populate_lut; + vaos->machine->now++; + } + } } diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h index a0680ec63d..c08c73d4bc 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.h +++ b/src/gallium/auxiliary/draw/draw_vs_aos.h @@ -59,6 +59,25 @@ struct x86_function; #define FPU_RND_NEG 1 #define FPU_RND_NEAREST 2 +struct aos_machine; +typedef void PIPE_CDECL (*lit_func)( struct aos_machine *, + float *result, + const float *in, + unsigned count ); +struct shine_tab { + float exponent; + float values[258]; + unsigned last_used; +}; + +struct lit_info { + lit_func func; + struct shine_tab *shine_tab; +}; + +#define MAX_SHINE_TAB 4 +#define MAX_LIT_INFO 16 + /* This is the temporary storage used by all the aos_sse vs varients. * Create one per context and reuse by passing a pointer in at * vs_varient creation?? @@ -74,6 +93,13 @@ struct aos_machine { float scale[4]; /* viewport */ float translate[4]; /* viewport */ + float tmp[2][4]; /* scratch space for LIT */ + + struct shine_tab shine_tab[MAX_SHINE_TAB]; + struct lit_info lit_info[MAX_LIT_INFO]; + unsigned now; + + ushort fpu_rnd_nearest; ushort fpu_rnd_neg_inf; ushort fpu_restore; @@ -97,6 +123,7 @@ struct aos_compilation { unsigned insn_counter; unsigned num_immediates; unsigned count; + unsigned lit_count; struct { unsigned idx:16; -- cgit v1.2.3 From a6fca8acb5e8ce0e5e6ce91a524e2bb4c180d3ac Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sun, 25 May 2008 11:20:38 +0100 Subject: draw: fix input vs output typo in emit --- src/gallium/auxiliary/draw/draw_vs_aos_io.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos_io.c b/src/gallium/auxiliary/draw/draw_vs_aos_io.c index 0dda9df97d..f39ebb7a17 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos_io.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos_io.c @@ -282,7 +282,7 @@ boolean aos_emit_outputs( struct aos_compilation *cp ) { unsigned i; - for (i = 0; i < cp->vaos->base.vs->info.num_inputs; i++) { + for (i = 0; i < cp->vaos->base.vs->info.num_outputs; i++) { unsigned format = cp->vaos->base.key.element[i].out.format; unsigned offset = cp->vaos->base.key.element[i].out.offset; -- cgit v1.2.3 From 584a3dcf8e4042cc1a5d48d83ea63d0a3c9706c1 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sun, 25 May 2008 11:25:09 +0100 Subject: draw: add viewport support to generic vs varient code --- src/gallium/auxiliary/draw/draw_vs_varient.c | 43 ++++++++++++++++++++++++++-- 1 file changed, 41 insertions(+), 2 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c index f6f621a748..c15c648527 100644 --- a/src/gallium/auxiliary/draw/draw_vs_varient.c +++ b/src/gallium/auxiliary/draw/draw_vs_varient.c @@ -44,7 +44,7 @@ struct draw_vs_varient_generic { struct draw_vs_varient base; - + struct pipe_viewport_state viewport; struct draw_vertex_shader *shader; struct draw_context *draw; @@ -87,6 +87,30 @@ static void vsvg_set_input( struct draw_vs_varient *varient, } +/* Mainly for debug at this stage: + */ +static void do_viewport( struct draw_vs_varient_generic *vsvg, + unsigned count, + void *output_buffer ) +{ + char *ptr = (char *)output_buffer; + const float *scale = vsvg->viewport.scale; + const float *trans = vsvg->viewport.translate; + unsigned stride = vsvg->base.key.output_stride; + unsigned j; + + for (j = 0; j < count; j++, ptr += stride) { + float *data = (float *)ptr; + float w = 1.0f / data[3]; + + data[0] = data[0] * w * scale[0] + trans[0]; + data[1] = data[1] * w * scale[1] + trans[1]; + data[2] = data[2] * w * scale[2] + trans[2]; + data[3] = w; + } +} + + static void vsvg_run_elts( struct draw_vs_varient *varient, const unsigned *elts, unsigned count, @@ -112,6 +136,12 @@ static void vsvg_run_elts( struct draw_vs_varient *varient, vsvg->base.key.output_stride, vsvg->base.key.output_stride); + if (vsvg->base.key.viewport) + do_viewport( vsvg, + count, + output_buffer ); + + //if (!vsvg->already_in_emit_format) vsvg->emit->set_buffer( vsvg->emit, @@ -152,6 +182,12 @@ static void vsvg_run_linear( struct draw_vs_varient *varient, vsvg->base.key.output_stride, vsvg->base.key.output_stride); + if (vsvg->base.key.viewport) + do_viewport( vsvg, + count, + output_buffer ); + + //if (!vsvg->already_in_emit_format) vsvg->emit->set_buffer( vsvg->emit, 0, @@ -171,6 +207,9 @@ static void vsvg_run_linear( struct draw_vs_varient *varient, static void vsvg_set_viewport( struct draw_vs_varient *varient, const struct pipe_viewport_state *viewport ) { + struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient; + + vsvg->viewport = *viewport; } static void vsvg_destroy( struct draw_vs_varient *varient ) @@ -185,7 +224,7 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs, unsigned i; struct translate_key fetch, emit; - if (key->viewport || key->clip) + if (key->clip) return NULL; struct draw_vs_varient_generic *vsvg = CALLOC_STRUCT( draw_vs_varient_generic ); -- cgit v1.2.3 From dc52622fcf5660a9675ed61c359cf7068aa4861b Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sun, 25 May 2008 15:28:30 +0100 Subject: cso: use memcpy rather than structure asignment for copying Apparently gcc will omit to copy hidden padding bytes under some circumstances, which means later on memcmp() will indicate a difference between structs even though all the visible members are identical. --- src/gallium/auxiliary/cso_cache/cso_context.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/cso_cache/cso_context.c b/src/gallium/auxiliary/cso_cache/cso_context.c index a1a3a9efaf..7236bff592 100644 --- a/src/gallium/auxiliary/cso_cache/cso_context.c +++ b/src/gallium/auxiliary/cso_cache/cso_context.c @@ -292,7 +292,7 @@ enum pipe_error cso_set_blend(struct cso_context *ctx, if (!cso) return PIPE_ERROR_OUT_OF_MEMORY; - cso->state = *templ; + memcpy(&cso->state, templ, sizeof(*templ)); cso->data = ctx->pipe->create_blend_state(ctx->pipe, &cso->state); cso->delete_state = (cso_state_callback)ctx->pipe->delete_blend_state; cso->context = ctx->pipe; @@ -350,7 +350,7 @@ enum pipe_error cso_single_sampler(struct cso_context *ctx, if (!cso) return PIPE_ERROR_OUT_OF_MEMORY; - cso->state = *templ; + memcpy(&cso->state, templ, sizeof(*templ)); cso->data = ctx->pipe->create_sampler_state(ctx->pipe, &cso->state); cso->delete_state = (cso_state_callback)ctx->pipe->delete_sampler_state; cso->context = ctx->pipe; @@ -508,7 +508,7 @@ enum pipe_error cso_set_depth_stencil_alpha(struct cso_context *ctx, if (!cso) return PIPE_ERROR_OUT_OF_MEMORY; - cso->state = *templ; + memcpy(&cso->state, templ, sizeof(*templ)); cso->data = ctx->pipe->create_depth_stencil_alpha_state(ctx->pipe, &cso->state); cso->delete_state = (cso_state_callback)ctx->pipe->delete_depth_stencil_alpha_state; cso->context = ctx->pipe; @@ -564,7 +564,7 @@ enum pipe_error cso_set_rasterizer(struct cso_context *ctx, if (!cso) return PIPE_ERROR_OUT_OF_MEMORY; - cso->state = *templ; + memcpy(&cso->state, templ, sizeof(*templ)); cso->data = ctx->pipe->create_rasterizer_state(ctx->pipe, &cso->state); cso->delete_state = (cso_state_callback)ctx->pipe->delete_rasterizer_state; cso->context = ctx->pipe; @@ -726,7 +726,7 @@ enum pipe_error cso_set_vertex_shader(struct cso_context *ctx, if (!cso) return PIPE_ERROR_OUT_OF_MEMORY; - cso->state = *templ; + memcpy(cso->state, templ, sizeof(*templ)); cso->data = ctx->pipe->create_vs_state(ctx->pipe, &cso->state); cso->delete_state = (cso_state_callback)ctx->pipe->delete_vs_state; cso->context = ctx->pipe; -- cgit v1.2.3 From caadc8d944c558e1fa9f23c3616d726337a19862 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sun, 25 May 2008 15:37:47 +0100 Subject: draw: clean up some of the xmm register manipulation function names --- src/gallium/auxiliary/draw/draw_vs_aos.c | 141 +++++++++++++++++-------------- 1 file changed, 78 insertions(+), 63 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index 1fbb7088ca..17b9442d6b 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -190,17 +190,26 @@ static void spill( struct aos_compilation *cp, unsigned idx ) } } -static boolean is_xmm_tmp( struct aos_compilation *cp, - struct x86_reg reg ) + +static struct x86_reg get_xmm_writable( struct aos_compilation *cp, + struct x86_reg reg ) { - return (reg.file == file_XMM && - cp->xmm[reg.idx].file == TGSI_FILE_NULL); + if (reg.file != file_XMM || + cp->xmm[reg.idx].file != TGSI_FILE_NULL) + { + struct x86_reg tmp = aos_get_xmm_reg(cp); + sse_movaps(cp->func, tmp, reg); + reg = tmp; + } + + return reg; } -static struct x86_reg get_xmm_clone( struct aos_compilation *cp, - struct x86_reg reg ) +static struct x86_reg get_xmm( struct aos_compilation *cp, + struct x86_reg reg ) { - if (!is_xmm_tmp(cp, reg)) { + if (reg.file != file_XMM) + { struct x86_reg tmp = aos_get_xmm_reg(cp); sse_movaps(cp->func, tmp, reg); reg = tmp; @@ -210,6 +219,9 @@ static struct x86_reg get_xmm_clone( struct aos_compilation *cp, } +/* Allocate an empty xmm register, either as a temporary or later to + * "adopt" as a shader reg. + */ struct x86_reg aos_get_xmm_reg( struct aos_compilation *cp ) { unsigned i; @@ -251,32 +263,11 @@ void aos_release_xmm_reg( struct aos_compilation *cp, cp->xmm[idx].last_used = 0; } -static void invalidate_xmm( struct aos_compilation *cp, - unsigned file, unsigned idx ) -{ - unsigned i; - - /* Invalidate any old copy of this register in XMM0-7. - */ - for (i = 0; i < 8; i++) { - if (cp->xmm[i].file == file && cp->xmm[i].idx == idx) { - - if (cp->xmm[i].dirty) - spill(cp, i); - - aos_release_xmm_reg(cp, i); - break; - } - } - for (; i < 8; i++) { - if (cp->xmm[i].file == file && cp->xmm[i].idx == idx) { - assert(0); - } - } -} - + +/* Mark an xmm reg as holding the current copy of a shader reg. + */ void aos_adopt_xmm_reg( struct aos_compilation *cp, struct x86_reg reg, unsigned file, @@ -290,6 +281,9 @@ void aos_adopt_xmm_reg( struct aos_compilation *cp, return; } + /* If any xmm reg thinks it holds this shader reg, break the + * illusion. + */ for (i = 0; i < 8; i++) { if (cp->xmm[i].file == file && cp->xmm[i].idx == idx) { @@ -304,12 +298,24 @@ void aos_adopt_xmm_reg( struct aos_compilation *cp, } - +/* Return a pointer to the in-memory copy of the reg, making sure it is uptodate. + */ static struct x86_reg aos_get_shader_reg_ptr( struct aos_compilation *cp, unsigned file, unsigned idx ) { - invalidate_xmm( cp, file, idx ); + unsigned i; + + /* Ensure the in-memory copy of this reg is up-to-date + */ + for (i = 0; i < 8; i++) { + if (cp->xmm[i].file == file && + cp->xmm[i].idx == idx && + cp->xmm[i].dirty) { + spill(cp, i); + } + } + return get_reg_ptr( cp, file, idx ); } @@ -320,7 +326,26 @@ static struct x86_reg aos_get_shader_reg_ptr( struct aos_compilation *cp, static struct x86_reg get_dst_ptr( struct aos_compilation *cp, const struct tgsi_full_dst_register *dst ) { - return aos_get_shader_reg_ptr( cp, dst->DstRegister.File, dst->DstRegister.Index ); + unsigned file = dst->DstRegister.File; + unsigned idx = dst->DstRegister.Index; + unsigned i; + + + /* Ensure in-memory copy of this reg is up-to-date and invalidate + * any xmm copies. + */ + for (i = 0; i < 8; i++) { + if (cp->xmm[i].file == file && + cp->xmm[i].idx == idx) + { + if (cp->xmm[i].dirty) + spill(cp, i); + + aos_release_xmm_reg(cp, i); + } + } + + return get_reg_ptr( cp, file, idx ); } @@ -358,15 +383,7 @@ static struct x86_reg aos_get_shader_reg_xmm( struct aos_compilation *cp, unsigned idx ) { struct x86_reg reg = aos_get_shader_reg( cp, file, idx ); - - if (reg.file != file_XMM) { - struct x86_reg tmp = aos_get_xmm_reg(cp); - sse_movaps(cp->func, tmp, reg); - aos_adopt_xmm_reg( cp, tmp, file, idx, FALSE ); - reg = tmp; - } - - return reg; + return get_xmm( cp, reg ); } @@ -634,7 +651,7 @@ static void store_dest( struct aos_compilation *cp, case TGSI_WRITEMASK_XYZW: aos_adopt_xmm_reg(cp, - get_xmm_clone(cp, result), + get_xmm_writable(cp, result), reg->DstRegister.File, reg->DstRegister.Index, TRUE); @@ -649,15 +666,15 @@ static void store_dest( struct aos_compilation *cp, switch (reg->DstRegister.WriteMask) { case TGSI_WRITEMASK_X: - sse_movss(cp->func, dst, get_xmm_clone(cp, result)); + sse_movss(cp->func, dst, get_xmm(cp, result)); break; case TGSI_WRITEMASK_XY: - sse_shufps(cp->func, dst, get_xmm_clone(cp, result), SHUF(X, Y, Z, W)); + sse_shufps(cp->func, dst, get_xmm(cp, result), SHUF(X, Y, Z, W)); break; case TGSI_WRITEMASK_ZW: - result = get_xmm_clone(cp, result); + result = get_xmm_writable(cp, result); sse_shufps(cp->func, result, dst, SHUF(X, Y, Z, W)); dst = result; break; @@ -845,7 +862,7 @@ static boolean emit_ABS( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg neg = aos_get_internal(cp, IMM_NEGS); - struct x86_reg dst = get_xmm_clone(cp, arg0); + struct x86_reg dst = get_xmm_writable(cp, arg0); sse_mulps(cp->func, dst, neg); sse_maxps(cp->func, dst, arg0); @@ -858,7 +875,7 @@ static boolean emit_ADD( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); - struct x86_reg dst = get_xmm_clone(cp, arg0); + struct x86_reg dst = get_xmm_writable(cp, arg0); sse_addps(cp->func, dst, arg1); @@ -882,10 +899,9 @@ static boolean emit_DP3( struct aos_compilation *cp, const struct tgsi_full_inst struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); struct x86_reg tmp = aos_get_xmm_reg(cp); - struct x86_reg dst = get_xmm_clone(cp, arg0); + struct x86_reg dst = get_xmm_writable(cp, arg0); sse_mulps(cp->func, dst, arg1); - /* Now the hard bit: sum the first 3 values: */ sse_movhlps(cp->func, tmp, dst); @@ -908,7 +924,7 @@ static boolean emit_DP4( struct aos_compilation *cp, const struct tgsi_full_inst struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); struct x86_reg tmp = aos_get_xmm_reg(cp); - struct x86_reg dst = get_xmm_clone(cp, arg0); + struct x86_reg dst = get_xmm_writable(cp, arg0); sse_mulps(cp->func, dst, arg1); @@ -932,7 +948,7 @@ static boolean emit_DPH( struct aos_compilation *cp, const struct tgsi_full_inst struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); struct x86_reg tmp = aos_get_xmm_reg(cp); - struct x86_reg dst = get_xmm_clone(cp, arg0); + struct x86_reg dst = get_xmm_writable(cp, arg0); sse_mulps(cp->func, dst, arg1); @@ -1366,7 +1382,7 @@ static boolean emit_LIT( struct aos_compilation *cp, const struct tgsi_full_inst if (writemask != TGSI_WRITEMASK_XYZW) { store_dest( cp, &op->FullDstRegisters[0], - get_xmm_clone( cp, result ) ); + get_xmm_writable( cp, result ) ); } return TRUE; @@ -1440,7 +1456,7 @@ static boolean emit_MAX( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); - struct x86_reg dst = get_xmm_clone(cp, arg0); + struct x86_reg dst = get_xmm_writable(cp, arg0); sse_maxps(cp->func, dst, arg1); @@ -1453,7 +1469,7 @@ static boolean emit_MIN( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); - struct x86_reg dst = get_xmm_clone(cp, arg0); + struct x86_reg dst = get_xmm_writable(cp, arg0); sse_minps(cp->func, dst, arg1); @@ -1464,7 +1480,7 @@ static boolean emit_MIN( struct aos_compilation *cp, const struct tgsi_full_inst static boolean emit_MOV( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); - struct x86_reg dst = get_xmm_clone(cp, arg0); + struct x86_reg dst = get_xmm_writable(cp, arg0); /* potentially nothing to do */ @@ -1476,7 +1492,7 @@ static boolean emit_MUL( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); - struct x86_reg dst = get_xmm_clone(cp, arg0); + struct x86_reg dst = get_xmm_writable(cp, arg0); sse_mulps(cp->func, dst, arg1); @@ -1494,7 +1510,7 @@ static boolean emit_MAD( struct aos_compilation *cp, const struct tgsi_full_inst /* If we can't clobber old contents of arg0, get a temporary & copy * it there, then clobber it... */ - arg0 = get_xmm_clone(cp, arg0); + arg0 = get_xmm_writable(cp, arg0); sse_mulps(cp->func, arg0, arg1); sse_addps(cp->func, arg0, arg2); @@ -1562,7 +1578,7 @@ static boolean emit_SGE( struct aos_compilation *cp, const struct tgsi_full_inst struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); struct x86_reg ones = aos_get_internal(cp, IMM_ONES); - struct x86_reg dst = get_xmm_clone(cp, arg0); + struct x86_reg dst = get_xmm_writable(cp, arg0); sse_cmpps(cp->func, dst, arg1, cc_NotLessThan); sse_andps(cp->func, dst, ones); @@ -1586,7 +1602,7 @@ static boolean emit_SLT( struct aos_compilation *cp, const struct tgsi_full_inst struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); struct x86_reg ones = aos_get_internal(cp, IMM_ONES); - struct x86_reg dst = get_xmm_clone(cp, arg0); + struct x86_reg dst = get_xmm_writable(cp, arg0); sse_cmpps(cp->func, dst, arg1, cc_LessThan); sse_andps(cp->func, dst, ones); @@ -1599,7 +1615,7 @@ static boolean emit_SUB( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); - struct x86_reg dst = get_xmm_clone(cp, arg0); + struct x86_reg dst = get_xmm_writable(cp, arg0); sse_subps(cp->func, dst, arg1); @@ -1989,7 +2005,6 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient, /* decr count, loop if not zero */ x86_dec(cp.func, cp.count_ESI); -/* x86_test(cp.func, cp.count_ESI, cp.count_ESI); */ x86_jcc(cp.func, cc_NZ, label); restore_fpu_state(&cp); -- cgit v1.2.3 From ce331e3a5e2a0505e01637861bdd7f5e6cfbd041 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sun, 25 May 2008 15:44:17 +0100 Subject: draw: special case for writing out scalar results --- src/gallium/auxiliary/draw/draw_vs_aos.c | 127 +++++++++++++++++++++++++------ 1 file changed, 102 insertions(+), 25 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index 17b9442d6b..aebc230858 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -697,6 +697,72 @@ static void store_dest( struct aos_compilation *cp, } +static void inject_scalar( struct aos_compilation *cp, + struct x86_reg dst, + struct x86_reg result, + unsigned swizzle ) +{ + sse_shufps(cp->func, dst, dst, swizzle); + sse_movss(cp->func, dst, result); + sse_shufps(cp->func, dst, dst, swizzle); +} + + +static void store_scalar_dest( struct aos_compilation *cp, + const struct tgsi_full_dst_register *reg, + struct x86_reg result ) +{ + unsigned writemask = reg->DstRegister.WriteMask; + struct x86_reg dst; + + if (writemask != TGSI_WRITEMASK_X && + writemask != TGSI_WRITEMASK_Y && + writemask != TGSI_WRITEMASK_Z && + writemask != TGSI_WRITEMASK_W && + writemask != 0) + { + result = get_xmm_writable(cp, result); /* already true, right? */ + sse_shufps(cp->func, result, result, SHUF(X,X,X,X)); + store_dest(cp, reg, result); + return; + } + + result = get_xmm(cp, result); + dst = aos_get_shader_reg_xmm(cp, + reg->DstRegister.File, + reg->DstRegister.Index); + + + + switch (reg->DstRegister.WriteMask) { + case TGSI_WRITEMASK_X: + sse_movss(cp->func, dst, result); + break; + + case TGSI_WRITEMASK_Y: + inject_scalar(cp, dst, result, SHUF(Y, X, Z, W)); + break; + + case TGSI_WRITEMASK_Z: + inject_scalar(cp, dst, result, SHUF(Z, Y, X, W)); + break; + + case TGSI_WRITEMASK_W: + inject_scalar(cp, dst, result, SHUF(W, Y, Z, X)); + break; + + default: + break; + } + + aos_adopt_xmm_reg(cp, + dst, + reg->DstRegister.File, + reg->DstRegister.Index, + TRUE); +} + + static void x87_fst_or_nop( struct x86_function *func, unsigned writemask, @@ -909,11 +975,8 @@ static boolean emit_DP3( struct aos_compilation *cp, const struct tgsi_full_inst emit_pshufd(cp, tmp, dst, SHUF(Y,X,W,Z)); sse_addss(cp->func, dst, tmp); - if (op->FullDstRegisters[0].DstRegister.WriteMask != 0x1) - sse_shufps(cp->func, dst, dst, SHUF(X, X, X, X)); - aos_release_xmm_reg(cp, tmp.idx); - store_dest(cp, &op->FullDstRegisters[0], dst); + store_scalar_dest(cp, &op->FullDstRegisters[0], dst); return TRUE; } @@ -935,11 +998,8 @@ static boolean emit_DP4( struct aos_compilation *cp, const struct tgsi_full_inst emit_pshufd(cp, tmp, dst, SHUF(Y,X,W,Z)); sse_addss(cp->func, dst, tmp); - if (op->FullDstRegisters[0].DstRegister.WriteMask != 0x1) - sse_shufps(cp->func, dst, dst, SHUF(X, X, X, X)); - aos_release_xmm_reg(cp, tmp.idx); - store_dest(cp, &op->FullDstRegisters[0], dst); + store_scalar_dest(cp, &op->FullDstRegisters[0], dst); return TRUE; } @@ -961,11 +1021,8 @@ static boolean emit_DPH( struct aos_compilation *cp, const struct tgsi_full_inst emit_pshufd(cp, tmp, arg1, SHUF(W,W,W,W)); sse_addss(cp->func, dst, tmp); - if (op->FullDstRegisters[0].DstRegister.WriteMask != 0x1) - sse_shufps(cp->func, dst, dst, SHUF(X, X, X, X)); - aos_release_xmm_reg(cp, tmp.idx); - store_dest(cp, &op->FullDstRegisters[0], dst); + store_scalar_dest(cp, &op->FullDstRegisters[0], dst); return TRUE; } @@ -1518,7 +1575,9 @@ static boolean emit_MAD( struct aos_compilation *cp, const struct tgsi_full_inst return TRUE; } - +/* Really not sufficient -- need to check for conditions that could + * generate inf/nan values, which will slow things down hugely. + */ static boolean emit_POW( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) { x87_fld_src(cp, &op->FullSrcRegisters[1], 0); /* a1.x */ @@ -1548,27 +1607,45 @@ static boolean emit_RCP( struct aos_compilation *cp, const struct tgsi_full_inst sse_divss(cp->func, dst, arg0); } - if (op->FullDstRegisters[0].DstRegister.WriteMask != 0x1) - sse_shufps(cp->func, dst, dst, SHUF(X, X, X, X)); - - store_dest(cp, &op->FullDstRegisters[0], dst); + store_scalar_dest(cp, &op->FullDstRegisters[0], dst); return TRUE; } + +/* Although rsqrtps() and rcpps() are low precision on some/all SSE + * implementations, it is possible to improve its precision at + * fairly low cost, using a newton/raphson step, as below: + * + * x1 = 2 * rcpps(a) - a * rcpps(a) * rcpps(a) + * x1 = 0.5 * rsqrtps(a) * [3.0 - (a * rsqrtps(a))* rsqrtps(a)] + * + * See: http://softwarecommunity.intel.com/articles/eng/1818.htm + */ static boolean emit_RSQ( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg dst = aos_get_xmm_reg(cp); - sse_rsqrtss(cp->func, dst, arg0); - - /* Extend precision here... - */ - - if (op->FullDstRegisters[0].DstRegister.WriteMask != 0x1) - sse_shufps(cp->func, dst, dst, SHUF(X, X, X, X)); + if (1) { + sse_rsqrtss(cp->func, dst, arg0); + } + else { +#if 0 + /* Extend precision here... + */ + sse_movaps( func, dst, get_temp( TGSI_EXEC_TEMP_HALF_I, TGSI_EXEC_TEMP_HALF_C ) ); + sse_movaps( func, tmp0, get_temp( TGSI_EXEC_TEMP_THREE_I, TGSI_EXEC_TEMP_THREE_C ) ); + + sse_rsqrtss( func, tmp1, src ); /* rsqrtss(a) */ + sse_mulss( func, src, tmp1 ); /* a * rsqrtss(a) */ + sse_mulss( func, dst, tmp1 ); /* .5 * rsqrtss(a) */ + sse_mulss( func, src, tmp1 ); /* a * rsqrtss(a) * rsqrtss(a) */ + sse_subss( func, tmp0, src ); /* 3.0 - (a * rsqrtss(a) * rsqrtss(a)) */ + sse_mulss( func, dst, tmp0 ); /* .5 * r * (3.0 - (a * r * r)) */ +#endif + } - store_dest(cp, &op->FullDstRegisters[0], dst); + store_scalar_dest(cp, &op->FullDstRegisters[0], dst); return TRUE; } -- cgit v1.2.3 From 3afb7198e01516dba38bb3248d4c0161e54650fe Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sun, 25 May 2008 15:45:27 +0100 Subject: draw: remove EXP & LOG from vs_aos.c These don't get hit & look like bug magnets to me... --- src/gallium/auxiliary/draw/draw_vs_aos.c | 85 ++------------------------------ 1 file changed, 4 insertions(+), 81 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index aebc230858..34dc09ead7 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -1066,85 +1066,6 @@ static boolean emit_EX2( struct aos_compilation *cp, const struct tgsi_full_inst return TRUE; } -static boolean emit_EXP( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) -{ - struct x86_reg dst = get_dst_ptr(cp, &op->FullDstRegisters[0]); - struct x86_reg st0 = x86_make_reg(file_x87, 0); - struct x86_reg st1 = x86_make_reg(file_x87, 1); - struct x86_reg st3 = x86_make_reg(file_x87, 3); - unsigned writemask = op->FullDstRegisters[0].DstRegister.WriteMask; - - /* CAUTION: dst may alias arg0! - */ - x87_fld_src(cp, &op->FullSrcRegisters[0], 0); /* arg0.x */ - x87_fld(cp->func, st0); /* arg arg */ - - /* by default, fpu is setup to round-to-nearest. We want to - * change this now, and track the state through to the end of the - * generated function so that it isn't repeated unnecessarily. - * Alternately, could subtract .5 to get round to -inf behaviour. - */ - set_fpu_round_neg_inf( cp ); - x87_fprndint( cp->func ); /* flr(a) a */ - x87_fld(cp->func, st0); /* flr(a) flr(a) a */ - x87_fld1(cp->func); /* 1 floor(a) floor(a) a */ - x87_fst_or_nop(cp->func, writemask, 3, dst); /* stack unchanged */ - - x87_fscale(cp->func); /* 2^floor(a) floor(a) a */ - x87_fst(cp->func, st3); /* 2^floor(a) floor(a) a 2^floor(a)*/ - - x87_fstp_or_pop(cp->func, writemask, 0, dst); /* flr(a) a 2^flr(a) */ - - x87_fsubp(cp->func, st1); /* frac(a) 2^flr(a) */ - - x87_fst_or_nop(cp->func, writemask, 1, dst); /* frac(a) 2^flr(a) */ - - x87_f2xm1(cp->func); /* (2^frac(a))-1 2^flr(a)*/ - x87_fld1(cp->func); /* 1 (2^frac(a))-1 2^flr(a)*/ - x87_faddp(cp->func, st1); /* 2^frac(a) 2^flr(a) */ - x87_fmulp(cp->func, st1); /* 2^a */ - - x87_fstp_or_pop(cp->func, writemask, 2, dst); - -/* dst[0] = 2^floor(tmp); */ -/* dst[1] = frac(tmp); */ -/* dst[2] = 2^floor(tmp) * 2^frac(tmp); */ -/* dst[3] = 1.0F; */ - return TRUE; -} - -static boolean emit_LOG( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) -{ - struct x86_reg dst = get_dst_ptr(cp, &op->FullDstRegisters[0]); - struct x86_reg st0 = x86_make_reg(file_x87, 0); - struct x86_reg st1 = x86_make_reg(file_x87, 1); - struct x86_reg st2 = x86_make_reg(file_x87, 2); - unsigned writemask = op->FullDstRegisters[0].DstRegister.WriteMask; - - /* CAUTION: dst may alias arg0! - */ - x87_fld_src(cp, &op->FullSrcRegisters[0], 0); /* arg0.x */ - x87_fabs(cp->func); /* |arg0.x| */ - x87_fxtract(cp->func); /* mantissa(arg0.x), exponent(arg0.x) */ - x87_fst(cp->func, st2); /* mantissa, exponent, mantissa */ - x87_fld1(cp->func); /* 1, mantissa, exponent, mantissa */ - x87_fyl2x(cp->func); /* log2(mantissa), exponent, mantissa */ - x87_fadd(cp->func, st0, st1); /* e+l2(m), e, m */ - - x87_fstp_or_pop(cp->func, writemask, 2, dst); /* e, m */ - - x87_fld1(cp->func); /* 1, e, m */ - x87_fsub(cp->func, st1, st0); /* 1, e-1, m */ - - x87_fstp_or_pop(cp->func, writemask, 3, dst); /* e-1,m */ - x87_fstp_or_pop(cp->func, writemask, 0, dst); /* m */ - - x87_fadd(cp->func, st0, st0); /* 2m */ - - x87_fstp_or_pop( cp->func, writemask, 1, dst ); - - return TRUE; -} static boolean emit_FLR( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) { @@ -1755,10 +1676,12 @@ emit_instruction( struct aos_compilation *cp, return emit_RSQ(cp, inst); case TGSI_OPCODE_EXP: - return emit_EXP(cp, inst); + /*return emit_EXP(cp, inst);*/ + return FALSE; case TGSI_OPCODE_LOG: - return emit_LOG(cp, inst); + /*return emit_LOG(cp, inst);*/ + return FALSE; case TGSI_OPCODE_MUL: return emit_MUL(cp, inst); -- cgit v1.2.3 From 9c7568965c00dcc2e9403a2f94f1cd09dcd783ae Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sun, 25 May 2008 15:47:04 +0100 Subject: draw: slight tweak for XPD opcode --- src/gallium/auxiliary/draw/draw_vs_aos.c | 21 +++++++-------------- 1 file changed, 7 insertions(+), 14 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index 34dc09ead7..37d04e45a6 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -1626,31 +1626,24 @@ static boolean emit_XPD( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]); - struct x86_reg dst = aos_get_xmm_reg(cp); struct x86_reg tmp0 = aos_get_xmm_reg(cp); struct x86_reg tmp1 = aos_get_xmm_reg(cp); - /* Could avoid tmp0, tmp1 if we overwrote arg0, arg1. Need a way - * to invalidate registers. This will come with better analysis - * (liveness analysis) of the incoming program. - */ - emit_pshufd(cp, dst, arg0, SHUF(Y, Z, X, W)); - emit_pshufd(cp, tmp1, arg1, SHUF(Z, X, Y, W)); - sse_mulps(cp->func, dst, tmp1); - emit_pshufd(cp, tmp0, arg0, SHUF(Z, X, Y, W)); emit_pshufd(cp, tmp1, arg1, SHUF(Y, Z, X, W)); - sse_mulps(cp->func, tmp0, tmp1); - sse_subps(cp->func, dst, tmp0); + sse_mulps(cp->func, tmp1, arg0); + emit_pshufd(cp, tmp0, arg0, SHUF(Y, Z, X, W)); + sse_mulps(cp->func, tmp0, arg1); + sse_subps(cp->func, tmp1, tmp0); + sse_shufps(cp->func, tmp1, tmp1, SHUF(Y, Z, X, W)); +/* dst[2] = arg0[0] * arg1[1] - arg0[1] * arg1[0]; */ /* dst[0] = arg0[1] * arg1[2] - arg0[2] * arg1[1]; */ /* dst[1] = arg0[2] * arg1[0] - arg0[0] * arg1[2]; */ -/* dst[2] = arg0[0] * arg1[1] - arg0[1] * arg1[0]; */ /* dst[3] is undef */ aos_release_xmm_reg(cp, tmp0.idx); - aos_release_xmm_reg(cp, tmp1.idx); - store_dest(cp, &op->FullDstRegisters[0], dst); + store_dest(cp, &op->FullDstRegisters[0], tmp1); return TRUE; } -- cgit v1.2.3 From 721fb5597e687fc1446119002ab03cc428104b29 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Mon, 26 May 2008 00:09:02 +0100 Subject: draw: more aos tweaks --- src/gallium/auxiliary/draw/draw_vs_aos.c | 26 ++++++++++++++++++++++++-- 1 file changed, 24 insertions(+), 2 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index 37d04e45a6..916203c66b 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -202,6 +202,7 @@ static struct x86_reg get_xmm_writable( struct aos_compilation *cp, reg = tmp; } + cp->xmm[reg.idx].last_used = cp->insn_counter; return reg; } @@ -215,6 +216,7 @@ static struct x86_reg get_xmm( struct aos_compilation *cp, reg = tmp; } + cp->xmm[reg.idx].last_used = cp->insn_counter; return reg; } @@ -281,6 +283,18 @@ void aos_adopt_xmm_reg( struct aos_compilation *cp, return; } + /* If this xmm reg is already holding this shader reg, just update + * last_used, and don't clobber the dirty flag... + */ + if (cp->xmm[reg.idx].file == file && + cp->xmm[reg.idx].idx == idx) + { + cp->xmm[reg.idx].dirty |= dirty; + cp->xmm[reg.idx].last_used = cp->insn_counter; + return; + } + + /* If any xmm reg thinks it holds this shader reg, break the * illusion. */ @@ -382,8 +396,16 @@ static struct x86_reg aos_get_shader_reg_xmm( struct aos_compilation *cp, unsigned file, unsigned idx ) { - struct x86_reg reg = aos_get_shader_reg( cp, file, idx ); - return get_xmm( cp, reg ); + struct x86_reg reg = get_xmm( cp, + aos_get_shader_reg( cp, file, idx ) ); + + aos_adopt_xmm_reg( cp, + reg, + file, + idx, + FALSE ); + + return reg; } -- cgit v1.2.3 From 351eca365c0ba488000c3826d5093de6170381e4 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Mon, 26 May 2008 11:03:00 +0100 Subject: draw: extend precision in RSQ opcode --- src/gallium/auxiliary/draw/draw_vs_aos.c | 48 ++++++++++++++++++-------------- src/gallium/auxiliary/draw/draw_vs_aos.h | 1 + 2 files changed, 28 insertions(+), 21 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index 916203c66b..1622358ae1 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -140,7 +140,8 @@ static void init_internals( struct aos_machine *machine ) ASSIGN_4V(machine->internal[IMM_NEGS], -1.0f, -1.0f, -1.0f, -1.0f); ASSIGN_4V(machine->internal[IMM_IDENTITY], 0.0f, 0.0f, 0.0f, 1.0f); ASSIGN_4V(machine->internal[IMM_INV_255], inv, inv, inv, inv); - ASSIGN_4V(machine->internal[IMM_255], f255, f255, f255, f255); + ASSIGN_4V(machine->internal[IMM_255], f255, f255, f255, f255); + ASSIGN_4V(machine->internal[IMM_RSQ], -.5f, 1.5f, 0.0f, 0.0f); machine->fpu_rnd_nearest = (X87_CW_EXCEPTION_INV_OP | @@ -1561,35 +1562,40 @@ static boolean emit_RCP( struct aos_compilation *cp, const struct tgsi_full_inst * * x1 = 2 * rcpps(a) - a * rcpps(a) * rcpps(a) * x1 = 0.5 * rsqrtps(a) * [3.0 - (a * rsqrtps(a))* rsqrtps(a)] + * or: + * x1 = rsqrtps(a) * [1.5 - .5 * a * rsqrtps(a) * rsqrtps(a)] + * * * See: http://softwarecommunity.intel.com/articles/eng/1818.htm */ static boolean emit_RSQ( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) { - struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); - struct x86_reg dst = aos_get_xmm_reg(cp); - if (1) { - sse_rsqrtss(cp->func, dst, arg0); + if (0) { + struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); + struct x86_reg r = aos_get_xmm_reg(cp); + sse_rsqrtss(cp->func, r, arg0); + store_scalar_dest(cp, &op->FullDstRegisters[0], r); + return TRUE; } else { -#if 0 - /* Extend precision here... - */ - sse_movaps( func, dst, get_temp( TGSI_EXEC_TEMP_HALF_I, TGSI_EXEC_TEMP_HALF_C ) ); - sse_movaps( func, tmp0, get_temp( TGSI_EXEC_TEMP_THREE_I, TGSI_EXEC_TEMP_THREE_C ) ); - - sse_rsqrtss( func, tmp1, src ); /* rsqrtss(a) */ - sse_mulss( func, src, tmp1 ); /* a * rsqrtss(a) */ - sse_mulss( func, dst, tmp1 ); /* .5 * rsqrtss(a) */ - sse_mulss( func, src, tmp1 ); /* a * rsqrtss(a) * rsqrtss(a) */ - sse_subss( func, tmp0, src ); /* 3.0 - (a * rsqrtss(a) * rsqrtss(a)) */ - sse_mulss( func, dst, tmp0 ); /* .5 * r * (3.0 - (a * r * r)) */ -#endif - } + struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); + struct x86_reg r = aos_get_xmm_reg(cp); - store_scalar_dest(cp, &op->FullDstRegisters[0], dst); - return TRUE; + struct x86_reg neg_half = get_reg_ptr( cp, AOS_FILE_INTERNAL, IMM_RSQ ); + struct x86_reg one_point_five = x86_make_disp( neg_half, 4 ); + struct x86_reg src = get_xmm_writable( cp, arg0 ); + + sse_rsqrtss( cp->func, r, src ); /* rsqrtss(a) */ + sse_mulss( cp->func, src, neg_half ); /* -.5 * a */ + sse_mulss( cp->func, src, r ); /* -.5 * a * r */ + sse_mulss( cp->func, src, r ); /* -.5 * a * r * r */ + sse_addss( cp->func, src, one_point_five ); /* 1.5 - .5 * a * r * r */ + sse_mulss( cp->func, r, src ); /* r * (1.5 - .5 * a * r * r) */ + + store_scalar_dest(cp, &op->FullDstRegisters[0], r); + return TRUE; + } } diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h index c08c73d4bc..fffe2e4658 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.h +++ b/src/gallium/auxiliary/draw/draw_vs_aos.h @@ -175,6 +175,7 @@ boolean aos_emit_outputs( struct aos_compilation *cp ); #define IMM_INV_255 3 /* 1/255, 1/255, 1/255, 1/255 */ #define IMM_255 4 /* 255, 255, 255, 255 */ #define IMM_NEGS 5 /* -1,-1,-1,-1 */ +#define IMM_RSQ 6 /* -.5,1.5,_,_ */ struct x86_reg aos_get_internal( struct aos_compilation *cp, unsigned imm ); -- cgit v1.2.3 From 529762d5df6d9427f9fa0927e38b9886e412a6bc Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Mon, 26 May 2008 18:29:47 +0100 Subject: draw: add missing break --- src/gallium/auxiliary/draw/draw_pt_emit.c | 1 + 1 file changed, 1 insertion(+) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt_emit.c b/src/gallium/auxiliary/draw/draw_pt_emit.c index 22a83ec78f..e21af3d287 100644 --- a/src/gallium/auxiliary/draw/draw_pt_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_emit.c @@ -100,6 +100,7 @@ void draw_pt_emit_prepare( struct pt_emit *emit, case EMIT_4UB: output_format = PIPE_FORMAT_B8G8R8A8_UNORM; emit_sz = 4 * sizeof(ubyte); + break; default: assert(0); output_format = PIPE_FORMAT_NONE; -- cgit v1.2.3 From d80c24a81a9a46c132fe877dde6919a57cacf8c0 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Mon, 26 May 2008 18:37:34 +0100 Subject: draw: defensively reset render primitive, which can get clobbered by clipping --- src/gallium/auxiliary/draw/draw_pt_emit.c | 17 ++++++++++++++++- 1 file changed, 16 insertions(+), 1 deletion(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt_emit.c b/src/gallium/auxiliary/draw/draw_pt_emit.c index e21af3d287..01396a749e 100644 --- a/src/gallium/auxiliary/draw/draw_pt_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_emit.c @@ -40,6 +40,7 @@ struct pt_emit { struct translate *translate; struct translate_cache *cache; + unsigned prim; }; void draw_pt_emit_prepare( struct pt_emit *emit, @@ -51,8 +52,14 @@ void draw_pt_emit_prepare( struct pt_emit *emit, struct translate_key hw_key; unsigned i; boolean ok; + - ok = draw->render->set_primitive(draw->render, prim); + /* XXX: may need to defensively reset this later on as clipping can + * clobber this state in the render backend. + */ + emit->prim = prim; + + ok = draw->render->set_primitive(draw->render, emit->prim); if (!ok) { assert(0); return; @@ -145,6 +152,14 @@ void draw_pt_emit( struct pt_emit *emit, */ draw_do_flush( draw, DRAW_FLUSH_BACKEND ); + /* XXX: and work out some way to coordinate the render primitive + * between vbuf.c and here... + */ + if (!draw->render->set_primitive(draw->render, emit->prim)) { + assert(0); + return; + } + hw_verts = render->allocate_vertices(render, (ushort)translate->key.output_stride, (ushort)vertex_count); -- cgit v1.2.3 From 93bfc94c351a2eafd43ac7a20b362d969f98d86a Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Mon, 26 May 2008 18:54:35 +0100 Subject: draw: defensively reset render primitive some more --- src/gallium/auxiliary/draw/draw_pt_emit.c | 8 ++++++++ 1 file changed, 8 insertions(+) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt_emit.c b/src/gallium/auxiliary/draw/draw_pt_emit.c index 01396a749e..d0c9577ee6 100644 --- a/src/gallium/auxiliary/draw/draw_pt_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_emit.c @@ -213,6 +213,14 @@ void draw_pt_emit_linear(struct pt_emit *emit, */ draw_do_flush( draw, DRAW_FLUSH_BACKEND ); + /* XXX: and work out some way to coordinate the render primitive + * between vbuf.c and here... + */ + if (!draw->render->set_primitive(draw->render, emit->prim)) { + assert(0); + return; + } + hw_verts = render->allocate_vertices(render, (ushort)translate->key.output_stride, (ushort)count); -- cgit v1.2.3 From aa16e3a2750993afdba16c24237bb6d8d8e4d91a Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Mon, 26 May 2008 19:10:44 +0100 Subject: draw: defensively flush pipeline backend when setting primitive --- src/gallium/auxiliary/draw/draw_pt_emit.c | 4 ++++ 1 file changed, 4 insertions(+) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt_emit.c b/src/gallium/auxiliary/draw/draw_pt_emit.c index d0c9577ee6..e37256b1ff 100644 --- a/src/gallium/auxiliary/draw/draw_pt_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_emit.c @@ -53,6 +53,10 @@ void draw_pt_emit_prepare( struct pt_emit *emit, unsigned i; boolean ok; + /* XXX: need to flush to get prim_vbuf.c to release its allocation?? + */ + draw_do_flush( draw, DRAW_FLUSH_BACKEND ); + /* XXX: may need to defensively reset this later on as clipping can * clobber this state in the render backend. -- cgit v1.2.3 From 4e2567f0ab6afd701bea4c35e388663e90f5cb6c Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 27 May 2008 10:42:58 +0100 Subject: draw: some possible fixes for spilling --- src/gallium/auxiliary/draw/draw_vs_aos.c | 30 ++++++++++++++++++------------ 1 file changed, 18 insertions(+), 12 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index 1622358ae1..99630e4f75 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -253,6 +253,7 @@ struct x86_reg aos_get_xmm_reg( struct aos_compilation *cp ) cp->xmm[oldest].file = TGSI_FILE_NULL; cp->xmm[oldest].idx = 0; + cp->xmm[oldest].dirty = 0; cp->xmm[oldest].last_used = cp->insn_counter; return x86_make_reg(file_XMM, oldest); } @@ -284,24 +285,18 @@ void aos_adopt_xmm_reg( struct aos_compilation *cp, return; } - /* If this xmm reg is already holding this shader reg, just update - * last_used, and don't clobber the dirty flag... - */ - if (cp->xmm[reg.idx].file == file && - cp->xmm[reg.idx].idx == idx) - { - cp->xmm[reg.idx].dirty |= dirty; - cp->xmm[reg.idx].last_used = cp->insn_counter; - return; - } - /* If any xmm reg thinks it holds this shader reg, break the * illusion. */ for (i = 0; i < 8; i++) { if (cp->xmm[i].file == file && - cp->xmm[i].idx == idx) { + cp->xmm[i].idx == idx) + { + /* If an xmm reg is already holding this shader reg, take into account its + * dirty flag... + */ + dirty |= cp->xmm[i].dirty; aos_release_xmm_reg(cp, i); } } @@ -1989,6 +1984,17 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient, debug_printf("\n"); } + + { + unsigned i; + for (i = 0; i < 8; i++) { + if (cp.xmm[i].file != TGSI_FILE_OUTPUT) { + cp.xmm[i].file = TGSI_FILE_NULL; + cp.xmm[i].dirty = 0; + } + } + } + if (cp.error) goto fail; -- cgit v1.2.3 From 5dc44184fa9f07465b7ff2be94394c55392ce5e9 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 27 May 2008 11:10:50 +0100 Subject: draw: fix writemask/shufps confusion --- src/gallium/auxiliary/draw/draw_vs_aos.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index 99630e4f75..434bd2a9ab 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -687,17 +687,18 @@ static void store_dest( struct aos_compilation *cp, sse_movss(cp->func, dst, get_xmm(cp, result)); break; - case TGSI_WRITEMASK_XY: + case TGSI_WRITEMASK_ZW: sse_shufps(cp->func, dst, get_xmm(cp, result), SHUF(X, Y, Z, W)); break; - case TGSI_WRITEMASK_ZW: + case TGSI_WRITEMASK_XY: result = get_xmm_writable(cp, result); sse_shufps(cp->func, result, dst, SHUF(X, Y, Z, W)); dst = result; break; case TGSI_WRITEMASK_YZW: + result = get_xmm_writable(cp, result); sse_movss(cp->func, result, dst); dst = result; break; @@ -891,7 +892,7 @@ static void emit_print( struct aos_compilation *cp, unsigned idx ) { struct x86_reg ecx = x86_make_reg( file_REG32, reg_CX ); - struct x86_reg arg = get_reg_ptr( cp, file, idx ); + struct x86_reg arg = aos_get_shader_reg_ptr( cp, file, idx ); unsigned i; /* There shouldn't be anything on the x87 stack. Can add this -- cgit v1.2.3 From adaaa29218f1babad874f50681ca971fdd3b8a40 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 27 May 2008 11:12:42 +0100 Subject: tgsi: export utils for dumping individual instructions --- src/gallium/auxiliary/tgsi/util/tgsi_dump.c | 24 ++++++++++++------------ src/gallium/auxiliary/tgsi/util/tgsi_dump.h | 18 ++++++++++++++++++ 2 files changed, 30 insertions(+), 12 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/tgsi/util/tgsi_dump.c b/src/gallium/auxiliary/tgsi/util/tgsi_dump.c index 4c65ffd780..648afa2a51 100644 --- a/src/gallium/auxiliary/tgsi/util/tgsi_dump.c +++ b/src/gallium/auxiliary/tgsi/util/tgsi_dump.c @@ -539,9 +539,9 @@ static const char *TGSI_MODULATES[] = "MODULATE_EIGHTH" }; -static void -dump_declaration_short( - struct tgsi_full_declaration *decl ) +void +tgsi_dump_declaration( + const struct tgsi_full_declaration *decl ) { TXT( "\nDCL " ); ENM( decl->Declaration.File, TGSI_FILES_SHORT ); @@ -672,9 +672,9 @@ dump_declaration_verbose( } } -static void -dump_immediate_short( - struct tgsi_full_immediate *imm ) +void +tgsi_dump_immediate( + const struct tgsi_full_immediate *imm ) { unsigned i; @@ -727,9 +727,9 @@ dump_immediate_verbose( } } -static void -dump_instruction_short( - struct tgsi_full_instruction *inst, +void +tgsi_dump_instruction( + const struct tgsi_full_instruction *inst, unsigned instno ) { unsigned i; @@ -1281,17 +1281,17 @@ tgsi_dump( switch( parse.FullToken.Token.Type ) { case TGSI_TOKEN_TYPE_DECLARATION: - dump_declaration_short( + tgsi_dump_declaration( &parse.FullToken.FullDeclaration ); break; case TGSI_TOKEN_TYPE_IMMEDIATE: - dump_immediate_short( + tgsi_dump_immediate( &parse.FullToken.FullImmediate ); break; case TGSI_TOKEN_TYPE_INSTRUCTION: - dump_instruction_short( + tgsi_dump_instruction( &parse.FullToken.FullInstruction, instno ); instno++; diff --git a/src/gallium/auxiliary/tgsi/util/tgsi_dump.h b/src/gallium/auxiliary/tgsi/util/tgsi_dump.h index beb0155d56..ca83bdef20 100644 --- a/src/gallium/auxiliary/tgsi/util/tgsi_dump.h +++ b/src/gallium/auxiliary/tgsi/util/tgsi_dump.h @@ -14,6 +14,24 @@ tgsi_dump( const struct tgsi_token *tokens, unsigned flags ); +struct tgsi_full_immediate; +struct tgsi_full_instruction; +struct tgsi_full_declaration; + +void +tgsi_dump_immediate( + const struct tgsi_full_immediate *imm ); + +void +tgsi_dump_instruction( + const struct tgsi_full_instruction *inst, + unsigned instno ); + +void +tgsi_dump_declaration( + const struct tgsi_full_declaration *decl ); + + #if defined __cplusplus } #endif -- cgit v1.2.3 From f7946bc7c0435ab2926cd729dfd8312222a3aa2a Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 27 May 2008 11:15:31 +0100 Subject: draw: dump individual instructions as they are processed --- src/gallium/auxiliary/draw/draw_vs_aos.c | 26 ++++++++++++++++++++++++-- 1 file changed, 24 insertions(+), 2 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index 434bd2a9ab..d3989fe107 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -42,7 +42,20 @@ #include "rtasm/rtasm_x86sse.h" #ifdef PIPE_ARCH_X86 +#define DISASSEM 0 +static const char *files[] = +{ + "NULL", + "CONST", + "IN", + "OUT", + "TEMP", + "SAMP", + "ADDR", + "IMM", + "INTERNAL", +}; static INLINE boolean eq( struct x86_reg a, struct x86_reg b ) @@ -184,7 +197,11 @@ static void spill( struct aos_compilation *cp, unsigned idx ) struct x86_reg oldval = get_reg_ptr(cp, cp->xmm[idx].file, cp->xmm[idx].idx); - + + if (0) debug_printf("\nspill %s[%d]", + files[cp->xmm[idx].file], + cp->xmm[idx].idx); + assert(cp->xmm[idx].dirty); sse_movaps(cp->func, oldval, x86_make_reg(file_XMM, idx)); cp->xmm[idx].dirty = 0; @@ -1975,6 +1992,9 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient, break; case TGSI_TOKEN_TYPE_INSTRUCTION: + if (DISASSEM) + tgsi_dump_instruction( &parse.FullToken.FullInstruction, cp.insn_counter ); + if (!emit_instruction( &cp, &parse.FullToken.FullInstruction )) goto fail; break; @@ -1982,7 +2002,9 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient, x87_assert_stack_empty(cp.func); cp.insn_counter++; - debug_printf("\n"); + + if (DISASSEM) + debug_printf("\n"); } -- cgit v1.2.3 From 50c1d329b95ad78e03ca4d537daee4d11f308c7a Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 27 May 2008 11:58:55 +0100 Subject: draw: restore extras path in draw_pt_vcache.c, keep pipeline flags out of non-pipeline elts --- src/gallium/auxiliary/draw/draw_pt_vcache.c | 117 +++++++++++++++------------- 1 file changed, 65 insertions(+), 52 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt_vcache.c b/src/gallium/auxiliary/draw/draw_pt_vcache.c index 6c17edba34..96e02fbf3a 100644 --- a/src/gallium/auxiliary/draw/draw_pt_vcache.c +++ b/src/gallium/auxiliary/draw/draw_pt_vcache.c @@ -104,10 +104,22 @@ static INLINE void vcache_elt( struct vcache_frontend *vcache, static void vcache_triangle( struct vcache_frontend *vcache, - ushort flags, unsigned i0, unsigned i1, unsigned i2 ) +{ + vcache_elt(vcache, i0, 0); + vcache_elt(vcache, i1, 0); + vcache_elt(vcache, i2, 0); + vcache_check_flush(vcache); +} + + +static void vcache_triangle_flags( struct vcache_frontend *vcache, + ushort flags, + unsigned i0, + unsigned i1, + unsigned i2 ) { vcache_elt(vcache, i0, flags); vcache_elt(vcache, i1, 0); @@ -116,9 +128,19 @@ static void vcache_triangle( struct vcache_frontend *vcache, } static void vcache_line( struct vcache_frontend *vcache, - ushort flags, unsigned i0, unsigned i1 ) +{ + vcache_elt(vcache, i0, 0); + vcache_elt(vcache, i1, 0); + vcache_check_flush(vcache); +} + + +static void vcache_line_flags( struct vcache_frontend *vcache, + ushort flags, + unsigned i0, + unsigned i1 ) { vcache_elt(vcache, i0, flags); vcache_elt(vcache, i1, 0); @@ -139,63 +161,46 @@ static void vcache_quad( struct vcache_frontend *vcache, unsigned i2, unsigned i3 ) { - vcache_triangle( vcache, - ( DRAW_PIPE_RESET_STIPPLE | - DRAW_PIPE_EDGE_FLAG_0 | - DRAW_PIPE_EDGE_FLAG_2 ), - i0, i1, i3 ); - - vcache_triangle( vcache, - ( DRAW_PIPE_EDGE_FLAG_0 | - DRAW_PIPE_EDGE_FLAG_1 ), - i1, i2, i3 ); + vcache_triangle( vcache, i0, i1, i3 ); + vcache_triangle( vcache, i1, i2, i3 ); +} + +static void vcache_ef_quad( struct vcache_frontend *vcache, + unsigned i0, + unsigned i1, + unsigned i2, + unsigned i3 ) +{ + vcache_triangle_flags( vcache, + ( DRAW_PIPE_RESET_STIPPLE | + DRAW_PIPE_EDGE_FLAG_0 | + DRAW_PIPE_EDGE_FLAG_2 ), + i0, i1, i3 ); + + vcache_triangle_flags( vcache, + ( DRAW_PIPE_EDGE_FLAG_0 | + DRAW_PIPE_EDGE_FLAG_1 ), + i1, i2, i3 ); } /* At least for now, we're back to using a template include file for * this. The two paths aren't too different though - it may be * possible to reunify them. */ -#define TRIANGLE(flags,i0,i1,i2) \ - vcache_triangle(vcache, \ - flags, \ - get_elt(elts,i0), \ - get_elt(elts,i1), \ - get_elt(elts,i2)) - -#define QUAD(i0,i1,i2,i3) \ - vcache_quad(vcache, \ - get_elt(elts,i0), \ - get_elt(elts,i1), \ - get_elt(elts,i2), \ - get_elt(elts,i3)) - -#define LINE(flags,i0,i1) \ - vcache_line(vcache, \ - flags, \ - get_elt(elts,i0), \ - get_elt(elts,i1)) - -#define POINT(i0) \ - vcache_point(vcache, \ - get_elt(elts,i0)) - +#define TRIANGLE(vc,flags,i0,i1,i2) vcache_triangle_flags(vc,flags,i0,i1,i2) +#define QUAD(vc,i0,i1,i2,i3) vcache_ef_quad(vc,i0,i1,i2,i3) +#define LINE(vc,flags,i0,i1) vcache_line_flags(vc,flags,i0,i1) +#define POINT(vc,i0) vcache_point(vc,i0) +#define FUNC vcache_run_extras +#include "draw_pt_vcache_tmp.h" + +#define TRIANGLE(vc,flags,i0,i1,i2) vcache_triangle(vc,i0,i1,i2) +#define QUAD(vc,i0,i1,i2,i3) vcache_quad(vc,i0,i1,i2,i3) +#define LINE(vc,flags,i0,i1) vcache_line(vc,i0,i1) +#define POINT(vc,i0) vcache_point(vc,i0) #define FUNC vcache_run -#define ARGS \ - struct draw_pt_front_end *frontend, \ - pt_elt_func get_elt, \ - const void *elts - -#define LOCAL_VARS \ - struct vcache_frontend *vcache = (struct vcache_frontend *)frontend; \ - struct draw_context *draw = vcache->draw; \ - boolean flatfirst = (draw->rasterizer->flatshade && \ - draw->rasterizer->flatshade_first); \ - unsigned prim = vcache->input_prim; \ - unsigned i, flags; +#include "draw_pt_vcache_tmp.h" -#define FLUSH vcache_flush( vcache ) - -#include "draw_pt_decompose.h" @@ -208,7 +213,15 @@ static void vcache_prepare( struct draw_pt_front_end *frontend, { struct vcache_frontend *vcache = (struct vcache_frontend *)frontend; - vcache->base.run = vcache_run; + if (opt & PT_PIPELINE) + { + vcache->base.run = vcache_run_extras; + } + else + { + vcache->base.run = vcache_run; + } + vcache->input_prim = prim; vcache->output_prim = draw_pt_reduced_prim(prim); -- cgit v1.2.3 From a08c574bfcf72c7f7ffbeb35c10347b491ef87fb Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 27 May 2008 12:26:23 +0100 Subject: draw: hook up viewport / rhw emit to varient key state --- .../auxiliary/draw/draw_pt_fetch_shade_emit.c | 4 +- src/gallium/auxiliary/draw/draw_vs_aos.c | 16 +++---- src/gallium/auxiliary/draw/draw_vs_varient.c | 52 ++++++++++++++++++---- 3 files changed, 53 insertions(+), 19 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c index 7fefd391a6..2f2e7195b3 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c @@ -94,8 +94,8 @@ static void fse_prepare( struct draw_pt_middle_end *middle, fse->key.nr_elements = MAX2(num_vs_outputs, /* outputs - translate to hw format */ num_vs_inputs); /* inputs - fetch from api format */ - fse->key.viewport = 1; - fse->key.clip = 0; + fse->key.viewport = !draw->identity_viewport; + fse->key.clip = !draw->bypass_clipping; fse->key.pad = 0; memset(fse->key.element, 0, diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index d3989fe107..c63553216c 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -2021,11 +2021,14 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient, if (cp.error) goto fail; - if (cp.vaos->base.key.viewport) { - if (0) - emit_viewport(&cp); - else - emit_rhw_viewport(&cp); + if (cp.vaos->base.key.clip) { + /* not really handling clipping, just do the rhw so we can + * see the results... + */ + emit_rhw_viewport(&cp); + } + else if (cp.vaos->base.key.viewport) { + emit_viewport(&cp); } /* Emit output... TODO: do this eagerly after the last write to a @@ -2188,9 +2191,6 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs, { struct draw_vs_varient_aos_sse *vaos = CALLOC_STRUCT(draw_vs_varient_aos_sse); - if (key->clip) - return NULL; - if (!vaos) goto fail; diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c index c15c648527..d4deabfff3 100644 --- a/src/gallium/auxiliary/draw/draw_vs_varient.c +++ b/src/gallium/auxiliary/draw/draw_vs_varient.c @@ -89,9 +89,9 @@ static void vsvg_set_input( struct draw_vs_varient *varient, /* Mainly for debug at this stage: */ -static void do_viewport( struct draw_vs_varient_generic *vsvg, - unsigned count, - void *output_buffer ) +static void do_rhw_viewport( struct draw_vs_varient_generic *vsvg, + unsigned count, + void *output_buffer ) { char *ptr = (char *)output_buffer; const float *scale = vsvg->viewport.scale; @@ -109,6 +109,25 @@ static void do_viewport( struct draw_vs_varient_generic *vsvg, data[3] = w; } } + +static void do_viewport( struct draw_vs_varient_generic *vsvg, + unsigned count, + void *output_buffer ) +{ + char *ptr = (char *)output_buffer; + const float *scale = vsvg->viewport.scale; + const float *trans = vsvg->viewport.translate; + unsigned stride = vsvg->base.key.output_stride; + unsigned j; + + for (j = 0; j < count; j++, ptr += stride) { + float *data = (float *)ptr; + + data[0] = data[0] * scale[0] + trans[0]; + data[1] = data[1] * scale[1] + trans[1]; + data[2] = data[2] * scale[2] + trans[2]; + } +} static void vsvg_run_elts( struct draw_vs_varient *varient, @@ -136,10 +155,20 @@ static void vsvg_run_elts( struct draw_vs_varient *varient, vsvg->base.key.output_stride, vsvg->base.key.output_stride); - if (vsvg->base.key.viewport) + + if (vsvg->base.key.clip) { + /* not really handling clipping, just do the rhw so we can + * see the results... + */ + do_rhw_viewport( vsvg, + count, + output_buffer ); + } + else if (vsvg->base.key.viewport) { do_viewport( vsvg, count, output_buffer ); + } //if (!vsvg->already_in_emit_format) @@ -182,11 +211,19 @@ static void vsvg_run_linear( struct draw_vs_varient *varient, vsvg->base.key.output_stride, vsvg->base.key.output_stride); - if (vsvg->base.key.viewport) + if (vsvg->base.key.clip) { + /* not really handling clipping, just do the rhw so we can + * see the results... + */ + do_rhw_viewport( vsvg, + count, + output_buffer ); + } + else if (vsvg->base.key.viewport) { do_viewport( vsvg, count, output_buffer ); - + } //if (!vsvg->already_in_emit_format) vsvg->emit->set_buffer( vsvg->emit, @@ -224,9 +261,6 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs, unsigned i; struct translate_key fetch, emit; - if (key->clip) - return NULL; - struct draw_vs_varient_generic *vsvg = CALLOC_STRUCT( draw_vs_varient_generic ); if (vsvg == NULL) return NULL; -- cgit v1.2.3 From 9752ebd99e16646fed3c14712fc0af2c34c9e48f Mon Sep 17 00:00:00 2001 From: Jakob Bornecrantz Date: Tue, 27 May 2008 12:59:46 +0200 Subject: draw: Fix for EMIT_4UB case --- src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c index 2f2e7195b3..c6249b4b41 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c @@ -143,7 +143,10 @@ static void fse_prepare( struct draw_pt_middle_end *middle, output_format = PIPE_FORMAT_R32_FLOAT; emit_sz = 1 * sizeof(float); vs_output = num_vs_outputs + 1; - + break; + case EMIT_4UB: + output_format = PIPE_FORMAT_B8G8R8A8_UNORM; + emit_sz = 4 * sizeof(ubyte); break; default: assert(0); -- cgit v1.2.3 From 9f9f6c21be105de41a58128605b911e679efe8f0 Mon Sep 17 00:00:00 2001 From: Jakob Bornecrantz Date: Tue, 27 May 2008 13:15:52 +0200 Subject: draw: Fixed typo in draw_pt_varray and added comments --- src/gallium/auxiliary/draw/draw_pt_varray.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt_varray.c b/src/gallium/auxiliary/draw/draw_pt_varray.c index 06fd866ccd..260f28f284 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray.c +++ b/src/gallium/auxiliary/draw/draw_pt_varray.c @@ -158,14 +158,14 @@ static INLINE void varray_point( struct varray_frontend *varray, static unsigned decompose_prim[PIPE_PRIM_POLYGON + 1] = { PIPE_PRIM_POINTS, PIPE_PRIM_LINES, + PIPE_PRIM_LINES, /* decomposed LINELOOP */ PIPE_PRIM_LINE_STRIP, - PIPE_PRIM_LINES, /* decomposed */ PIPE_PRIM_TRIANGLES, PIPE_PRIM_TRIANGLE_STRIP, - PIPE_PRIM_TRIANGLES, /* decomposed */ + PIPE_PRIM_TRIANGLES, /* decomposed TRI_FAN */ PIPE_PRIM_QUADS, PIPE_PRIM_QUAD_STRIP, - PIPE_PRIM_TRIANGLES /* decomposed */ + PIPE_PRIM_TRIANGLES /* decomposed POLYGON */ }; -- cgit v1.2.3 From f64c44ad3e55467ce964871502445cf5a0fb46d6 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 27 May 2008 14:17:57 +0100 Subject: draw: remove dead file --- src/gallium/auxiliary/draw/draw_pt_middle_fse.c | 705 ------------------------ 1 file changed, 705 deletions(-) delete mode 100644 src/gallium/auxiliary/draw/draw_pt_middle_fse.c (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt_middle_fse.c b/src/gallium/auxiliary/draw/draw_pt_middle_fse.c deleted file mode 100644 index 643ea151c1..0000000000 --- a/src/gallium/auxiliary/draw/draw_pt_middle_fse.c +++ /dev/null @@ -1,705 +0,0 @@ -/************************************************************************** - * - * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. - * All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the - * "Software"), to deal in the Software without restriction, including - * without limitation the rights to use, copy, modify, merge, publish, - * distribute, sub license, and/or sell copies of the Software, and to - * permit persons to whom the Software is furnished to do so, subject to - * the following conditions: - * - * The above copyright notice and this permission notice (including the - * next paragraph) shall be included in all copies or substantial portions - * of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF - * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. - * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR - * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, - * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE - * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - * - **************************************************************************/ - - /* - * Authors: - * Keith Whitwell - */ - - -#include "pipe/p_util.h" -#include "draw/draw_context.h" -#include "draw/draw_private.h" -#include "draw/draw_vbuf.h" -#include "draw/draw_vertex.h" -#include "draw/draw_pt.h" -#include "draw/draw_vs.h" - -#include "translate/translate.h" - -struct fetch_shade_emit; - -struct fse_shader { - struct translate_key key; - - void (*run_linear)( const struct fetch_shade_emit *fse, - unsigned start, - unsigned count, - char *buffer ); - - void (*run_elts)( const struct fetch_shade_emit *fse, - const unsigned *fetch_elts, - unsigned fetch_count, - char *buffer ); - -}; - -/* Prototype fetch, shade, emit-hw-verts all in one go. - */ -struct fetch_shade_emit { - struct draw_pt_middle_end base; - struct draw_context *draw; - - struct translate_key key; - - /* Temporaries: - */ - const float *constants; - unsigned pitch[PIPE_MAX_ATTRIBS]; - const ubyte *src[PIPE_MAX_ATTRIBS]; - unsigned prim; - - /* Points to one of the three hardwired example shaders, below: - */ - struct fse_shader *active; - - /* Temporary: A list of hard-wired shaders. Of course the plan - * would be to generate these for a given (vertex-shader, - * translate-key) pair... - */ - struct fse_shader shader[10]; - int nr_shaders; -}; - - - -/* Not quite passthrough yet -- we're still running the 'shader' here, - * inlined into the vertex fetch function. - */ -static void shader0_run_linear( const struct fetch_shade_emit *fse, - unsigned start, - unsigned count, - char *buffer ) -{ - unsigned i; - - const float *m = fse->constants; - const ubyte *xyz = fse->src[0] + start * fse->pitch[0]; - const ubyte *rgb = fse->src[1] + start * fse->pitch[1]; - const ubyte *st = fse->src[2] + start * fse->pitch[2]; - - float *out = (float *)buffer; - - /* loop over vertex attributes (vertex shader inputs) - */ - for (i = 0; i < count; i++) { - { - const float *in = (const float *)xyz; - const float ix = in[0], iy = in[1], iz = in[2]; - - out[0] = m[0] * ix + m[4] * iy + m[8] * iz + m[12]; - out[1] = m[1] * ix + m[5] * iy + m[9] * iz + m[13]; - out[2] = m[2] * ix + m[6] * iy + m[10] * iz + m[14]; - out[3] = m[3] * ix + m[7] * iy + m[11] * iz + m[15]; - xyz += fse->pitch[0]; - } - - { - const float *in = (const float *)rgb; - out[4] = in[0]; - out[5] = in[1]; - out[6] = in[2]; - out[7] = 1.0f; - rgb += fse->pitch[1]; - } - - { - const float *in = (const float *)st; - out[8] = in[0]; - out[9] = in[1]; - out[10] = 0.0f; - out[11] = 1.0f; - st += fse->pitch[2]; - } - - out += 12; - } -} - - - -static void shader1_run_linear( const struct fetch_shade_emit *fse, - unsigned start, - unsigned count, - char *buffer ) -{ - unsigned i; - const float *m = (const float *)fse->constants; - const ubyte *xyz = fse->src[0] + start * fse->pitch[0]; - const ubyte *rgb = fse->src[1] + start * fse->pitch[1]; - float *out = (float *)buffer; - -// debug_printf("rgb %f %f %f\n", rgb[0], rgb[1], rgb[2]); - - - for (i = 0; i < count; i++) { - { - const float *in = (const float *)xyz; - const float ix = in[0], iy = in[1], iz = in[2]; - - out[0] = m[0] * ix + m[4] * iy + m[8] * iz + m[12]; - out[1] = m[1] * ix + m[5] * iy + m[9] * iz + m[13]; - out[2] = m[2] * ix + m[6] * iy + m[10] * iz + m[14]; - out[3] = m[3] * ix + m[7] * iy + m[11] * iz + m[15]; - xyz += fse->pitch[0]; - } - - { - const float *in = (const float *)rgb; - out[4] = in[0]; - out[5] = in[1]; - out[6] = in[2]; - out[7] = 1.0f; - rgb += fse->pitch[1]; - } - - out += 8; - } -} - - - - -static void shader2_run_linear( const struct fetch_shade_emit *fse, - unsigned start, - unsigned count, - char *buffer ) -{ - unsigned i; - const float *m = (const float *)fse->constants; - const ubyte *xyz = fse->src[0] + start * fse->pitch[0]; - const ubyte *rgb = fse->src[1] + start * fse->pitch[1]; - const float psiz = 1.0; - float *out = (float *)buffer; - - - assert(fse->pitch[1] == 0); - - for (i = 0; i < count; i++) { - { - const float *in = (const float *)xyz; - const float ix = in[0], iy = in[1], iz = in[2]; - - out[0] = m[0] * ix + m[4] * iy + m[8] * iz + m[12]; - out[1] = m[1] * ix + m[5] * iy + m[9] * iz + m[13]; - out[2] = m[2] * ix + m[6] * iy + m[10] * iz + m[14]; - out[3] = m[3] * ix + m[7] * iy + m[11] * iz + m[15]; - xyz += fse->pitch[0]; - } - - { - const float *in = (const float *)rgb; - out[4] = in[0]; - out[5] = in[1]; - out[6] = in[2]; - out[7] = 1.0f; - rgb += fse->pitch[1]; - } - - { - out[8] = psiz; - } - - out += 9; - } -} - - - - -static void shader0_run_elts( const struct fetch_shade_emit *fse, - const unsigned *elts, - unsigned count, - char *buffer ) -{ - unsigned i; - const float *m = fse->constants; - float *out = (float *)buffer; - - - /* loop over vertex attributes (vertex shader inputs) - */ - for (i = 0; i < count; i++) { - unsigned elt = elts[i]; - { - const ubyte *xyz = fse->src[0] + elt * fse->pitch[0]; - const float *in = (const float *)xyz; - const float ix = in[0], iy = in[1], iz = in[2]; - - out[0] = m[0] * ix + m[4] * iy + m[8] * iz + m[12]; - out[1] = m[1] * ix + m[5] * iy + m[9] * iz + m[13]; - out[2] = m[2] * ix + m[6] * iy + m[10] * iz + m[14]; - out[3] = m[3] * ix + m[7] * iy + m[11] * iz + m[15]; - } - - { - const ubyte *rgb = fse->src[1] + elt * fse->pitch[1]; - const float *in = (const float *)rgb; - out[4] = in[0]; - out[5] = in[1]; - out[6] = in[2]; - out[7] = 1.0f; - } - - { - const ubyte *st = fse->src[2] + elt * fse->pitch[2]; - const float *in = (const float *)st; - out[8] = in[0]; - out[9] = in[1]; - out[10] = 0.0f; - out[11] = 1.0f; - } - - out += 12; - } -} - - - -static void shader1_run_elts( const struct fetch_shade_emit *fse, - const unsigned *elts, - unsigned count, - char *buffer ) -{ - unsigned i; - const float *m = (const float *)fse->constants; - float *out = (float *)buffer; - - for (i = 0; i < count; i++) { - unsigned elt = elts[i]; - - { - const ubyte *xyz = fse->src[0] + elt * fse->pitch[0]; - const float *in = (const float *)xyz; - const float ix = in[0], iy = in[1], iz = in[2]; - - out[0] = m[0] * ix + m[4] * iy + m[8] * iz + m[12]; - out[1] = m[1] * ix + m[5] * iy + m[9] * iz + m[13]; - out[2] = m[2] * ix + m[6] * iy + m[10] * iz + m[14]; - out[3] = m[3] * ix + m[7] * iy + m[11] * iz + m[15]; - xyz += fse->pitch[0]; - } - - { - const ubyte *rgb = fse->src[1] + elt * fse->pitch[1]; - const float *in = (const float *)rgb; - out[4] = in[0]; - out[5] = in[1]; - out[6] = in[2]; - out[7] = 1.0f; - rgb += fse->pitch[1]; - } - - out += 8; - } -} - - - - -static void shader2_run_elts( const struct fetch_shade_emit *fse, - const unsigned *elts, - unsigned count, - char *buffer ) -{ - unsigned i; - const float *m = (const float *)fse->constants; - const float psiz = 1.0; - float *out = (float *)buffer; - - for (i = 0; i < count; i++) { - unsigned elt = elts[i]; - { - const ubyte *xyz = fse->src[0] + elt * fse->pitch[0]; - const float *in = (const float *)xyz; - const float ix = in[0], iy = in[1], iz = in[2]; - - out[0] = m[0] * ix + m[4] * iy + m[8] * iz + m[12]; - out[1] = m[1] * ix + m[5] * iy + m[9] * iz + m[13]; - out[2] = m[2] * ix + m[6] * iy + m[10] * iz + m[14]; - out[3] = m[3] * ix + m[7] * iy + m[11] * iz + m[15]; - } - - { - const ubyte *rgb = fse->src[1] + elt * fse->pitch[1]; - out[4] = rgb[0]; - out[5] = rgb[1]; - out[6] = rgb[2]; - out[7] = 1.0f; - } - - { - out[8] = psiz; - } - - out += 9; - } -} - - - -static void fse_prepare( struct draw_pt_middle_end *middle, - unsigned prim, - unsigned opt ) -{ - struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle; - struct draw_context *draw = fse->draw; - unsigned num_vs_inputs = draw->vs.vertex_shader->info.num_inputs; - unsigned num_vs_outputs = draw->vs.vertex_shader->info.num_outputs; - const struct vertex_info *vinfo; - unsigned i; - boolean need_psize = 0; - - - if (draw->pt.user.elts) { - assert(0); - return ; - } - - if (!draw->render->set_primitive( draw->render, - prim )) { - assert(0); - return; - } - - /* Must do this after set_primitive() above: - */ - vinfo = draw->render->get_vertex_info(draw->render); - - - - fse->key.nr_elements = MAX2(num_vs_outputs, /* outputs - translate to hw format */ - num_vs_inputs); /* inputs - fetch from api format */ - - fse->key.output_stride = vinfo->size * 4; - memset(fse->key.element, 0, - fse->key.nr_elements * sizeof(fse->key.element[0])); - - for (i = 0; i < num_vs_inputs; i++) { - const struct pipe_vertex_element *src = &draw->pt.vertex_element[i]; - fse->key.element[i].input_format = src->src_format; - - /* Consider ignoring these at this point, ie make generated - * programs independent of this state: - */ - fse->key.element[i].input_buffer = 0; //src->vertex_buffer_index; - fse->key.element[i].input_offset = 0; //src->src_offset; - } - - - { - unsigned dst_offset = 0; - - for (i = 0; i < vinfo->num_attribs; i++) { - unsigned emit_sz = 0; - unsigned output_format = PIPE_FORMAT_NONE; - unsigned vs_output = vinfo->src_index[i]; - - switch (vinfo->emit[i]) { - case EMIT_4F: - output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; - emit_sz = 4 * sizeof(float); - break; - case EMIT_3F: - output_format = PIPE_FORMAT_R32G32B32_FLOAT; - emit_sz = 3 * sizeof(float); - break; - case EMIT_2F: - output_format = PIPE_FORMAT_R32G32_FLOAT; - emit_sz = 2 * sizeof(float); - break; - case EMIT_1F: - output_format = PIPE_FORMAT_R32_FLOAT; - emit_sz = 1 * sizeof(float); - break; - case EMIT_1F_PSIZE: - need_psize = 1; - output_format = PIPE_FORMAT_R32_FLOAT; - emit_sz = 1 * sizeof(float); - vs_output = num_vs_outputs + 1; - - break; - default: - assert(0); - break; - } - - /* The elements in the key correspond to vertex shader output - * numbers, not to positions in the hw vertex description -- - * that's handled by the output_offset field. - */ - fse->key.element[vs_output].output_format = output_format; - fse->key.element[vs_output].output_offset = dst_offset; - - dst_offset += emit_sz; - assert(fse->key.output_stride >= dst_offset); - } - } - - /* To make psize work, really need to tell the vertex shader to - * copy that value from input->output. For 'translate' this was - * implicit for all elements. - */ -#if 0 - if (need_psize) { - unsigned input = num_vs_inputs + 1; - const struct pipe_vertex_element *src = &draw->pt.vertex_element[i]; - fse->key.element[i].input_format = PIPE_FORMAT_R32_FLOAT; - fse->key.element[i].input_buffer = 0; //nr_buffers + 1; - fse->key.element[i].input_offset = 0; - - fse->key.nr_elements += 1; - - } -#endif - - fse->constants = draw->pt.user.constants; - - /* Would normally look up a vertex shader and peruse its list of - * varients somehow. We omitted that step and put all the - * hardcoded "shaders" into an array. We're just making the - * assumption that this happens to be a matching shader... ie - * you're running isosurf, aren't you? - */ - fse->active = NULL; - for (i = 0; i < fse->nr_shaders; i++) { - if (translate_key_compare( &fse->key, &fse->shader[i].key) == 0) - fse->active = &fse->shader[i]; - } - - if (!fse->active) { - assert(0); - return ; - } - - /* Now set buffer pointers: - */ - for (i = 0; i < num_vs_inputs; i++) { - unsigned buf = draw->pt.vertex_element[i].vertex_buffer_index; - - fse->src[i] = ((const ubyte *) draw->pt.user.vbuffer[buf] + - draw->pt.vertex_buffer[buf].buffer_offset + - draw->pt.vertex_element[i].src_offset); - - fse->pitch[i] = draw->pt.vertex_buffer[buf].pitch; - - } - - - //return TRUE; -} - - - - - - - -static void fse_run_linear( struct draw_pt_middle_end *middle, - unsigned start, - unsigned count ) -{ - struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle; - struct draw_context *draw = fse->draw; - - char *hw_verts; - - /* XXX: need to flush to get prim_vbuf.c to release its allocation?? - */ - draw_do_flush( draw, DRAW_FLUSH_BACKEND ); - - hw_verts = draw->render->allocate_vertices( draw->render, - (ushort)fse->key.output_stride, - (ushort)count ); - - if (!hw_verts) { - assert(0); - return; - } - - /* Single routine to fetch vertices, run shader and emit HW verts. - * Clipping and viewport transformation are done elsewhere -- - * either by the API or on hardware, or for some other reason not - * required... - */ - fse->active->run_linear( fse, - start, count, - hw_verts ); - - /* Draw arrays path to avoid re-emitting index list again and - * again. - */ - draw->render->draw_arrays( draw->render, - 0, - count ); - - - draw->render->release_vertices( draw->render, - hw_verts, - fse->key.output_stride, - count ); -} - - -static void -fse_run(struct draw_pt_middle_end *middle, - const unsigned *fetch_elts, - unsigned fetch_count, - const ushort *draw_elts, - unsigned draw_count ) -{ - struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle; - struct draw_context *draw = fse->draw; - void *hw_verts; - - /* XXX: need to flush to get prim_vbuf.c to release its allocation?? - */ - draw_do_flush( draw, DRAW_FLUSH_BACKEND ); - - hw_verts = draw->render->allocate_vertices( draw->render, - (ushort)fse->key.output_stride, - (ushort)fetch_count ); - if (!hw_verts) { - assert(0); - return; - } - - - /* Single routine to fetch vertices, run shader and emit HW verts. - */ - fse->active->run_elts( fse, - fetch_elts, - fetch_count, - hw_verts ); - - draw->render->draw( draw->render, - draw_elts, - draw_count ); - - draw->render->release_vertices( draw->render, - hw_verts, - fse->key.output_stride, - fetch_count ); - -} - - -static void fse_finish( struct draw_pt_middle_end *middle ) -{ -} - - -static void -fse_destroy( struct draw_pt_middle_end *middle ) -{ - FREE(middle); -} - -struct draw_pt_middle_end *draw_pt_middle_fse( struct draw_context *draw ) -{ - struct fetch_shade_emit *fse = CALLOC_STRUCT(fetch_shade_emit); - if (!fse) - return NULL; - - fse->base.prepare = fse_prepare; - fse->base.run = fse_run; - fse->base.run_linear = fse_run_linear; - fse->base.finish = fse_finish; - fse->base.destroy = fse_destroy; - fse->draw = draw; - - fse->shader[0].run_linear = shader0_run_linear; - fse->shader[0].run_elts = shader0_run_elts; - fse->shader[0].key.nr_elements = 3; - fse->shader[0].key.output_stride = 12 * sizeof(float); - - fse->shader[0].key.element[0].input_format = PIPE_FORMAT_R32G32B32_FLOAT; - fse->shader[0].key.element[0].input_buffer = 0; - fse->shader[0].key.element[0].input_offset = 0; - fse->shader[0].key.element[0].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; - fse->shader[0].key.element[0].output_offset = 0; - - fse->shader[0].key.element[1].input_format = PIPE_FORMAT_R32G32B32_FLOAT; - fse->shader[0].key.element[1].input_buffer = 0; - fse->shader[0].key.element[1].input_offset = 0; - fse->shader[0].key.element[1].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; - fse->shader[0].key.element[1].output_offset = 16; - - fse->shader[0].key.element[1].input_format = PIPE_FORMAT_R32G32_FLOAT; - fse->shader[0].key.element[1].input_buffer = 0; - fse->shader[0].key.element[1].input_offset = 0; - fse->shader[0].key.element[1].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; - fse->shader[0].key.element[1].output_offset = 32; - - fse->shader[1].run_linear = shader1_run_linear; - fse->shader[1].run_elts = shader1_run_elts; - fse->shader[1].key.nr_elements = 2; - fse->shader[1].key.output_stride = 8 * sizeof(float); - - fse->shader[1].key.element[0].input_format = PIPE_FORMAT_R32G32B32_FLOAT; - fse->shader[1].key.element[0].input_buffer = 0; - fse->shader[1].key.element[0].input_offset = 0; - fse->shader[1].key.element[0].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; - fse->shader[1].key.element[0].output_offset = 0; - - fse->shader[1].key.element[1].input_format = PIPE_FORMAT_R32G32B32_FLOAT; - fse->shader[1].key.element[1].input_buffer = 0; - fse->shader[1].key.element[1].input_offset = 0; - fse->shader[1].key.element[1].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; - fse->shader[1].key.element[1].output_offset = 16; - - fse->shader[2].run_linear = shader2_run_linear; - fse->shader[2].run_elts = shader2_run_elts; - fse->shader[2].key.nr_elements = 3; - fse->shader[2].key.output_stride = 9 * sizeof(float); - - fse->shader[2].key.element[0].input_format = PIPE_FORMAT_R32G32B32_FLOAT; - fse->shader[2].key.element[0].input_buffer = 0; - fse->shader[2].key.element[0].input_offset = 0; - fse->shader[2].key.element[0].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; - fse->shader[2].key.element[0].output_offset = 0; - - fse->shader[2].key.element[1].input_format = PIPE_FORMAT_R32G32B32_FLOAT; - fse->shader[2].key.element[1].input_buffer = 0; - fse->shader[2].key.element[1].input_offset = 0; - fse->shader[2].key.element[1].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; - fse->shader[2].key.element[1].output_offset = 16; - - /* psize is special - * -- effectively add it here as another input!?! - * -- who knows how to add it as a buffer? - */ - fse->shader[2].key.element[2].input_format = PIPE_FORMAT_R32_FLOAT; - fse->shader[2].key.element[2].input_buffer = 0; - fse->shader[2].key.element[2].input_offset = 0; - fse->shader[2].key.element[2].output_format = PIPE_FORMAT_R32_FLOAT; - fse->shader[2].key.element[2].output_offset = 32; - - fse->nr_shaders = 3; - - return &fse->base; -} -- cgit v1.2.3 From f8762ba5234fd1b44e11e76bb5f58d2305c90572 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 27 May 2008 14:42:15 +0100 Subject: draw: explicitly list nr_inputs, outputs in varient key --- src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c | 14 +++++++++----- src/gallium/auxiliary/draw/draw_vs.h | 6 ++++-- src/gallium/auxiliary/draw/draw_vs_aos.c | 2 +- src/gallium/auxiliary/draw/draw_vs_aos_io.c | 4 ++-- src/gallium/auxiliary/draw/draw_vs_varient.c | 10 +++++----- 5 files changed, 21 insertions(+), 15 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c index c6249b4b41..581026dcb0 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c @@ -72,7 +72,6 @@ static void fse_prepare( struct draw_pt_middle_end *middle, struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle; struct draw_context *draw = fse->draw; unsigned num_vs_inputs = draw->vs.vertex_shader->info.num_inputs; - unsigned num_vs_outputs = draw->vs.vertex_shader->info.num_outputs; const struct vertex_info *vinfo; unsigned i; boolean need_psize = 0; @@ -91,8 +90,11 @@ static void fse_prepare( struct draw_pt_middle_end *middle, fse->key.output_stride = vinfo->size * 4; - fse->key.nr_elements = MAX2(num_vs_outputs, /* outputs - translate to hw format */ - num_vs_inputs); /* inputs - fetch from api format */ + fse->key.nr_outputs = vinfo->num_attribs; + fse->key.nr_inputs = num_vs_inputs; + + fse->key.nr_elements = MAX2(fse->key.nr_outputs, /* outputs - translate to hw format */ + fse->key.nr_inputs); /* inputs - fetch from api format */ fse->key.viewport = !draw->identity_viewport; fse->key.clip = !draw->bypass_clipping; @@ -142,7 +144,7 @@ static void fse_prepare( struct draw_pt_middle_end *middle, need_psize = 1; output_format = PIPE_FORMAT_R32_FLOAT; emit_sz = 1 * sizeof(float); - vs_output = num_vs_outputs + 1; + vs_output = vinfo->num_attribs + 1; break; case EMIT_4UB: output_format = PIPE_FORMAT_B8G8R8A8_UNORM; @@ -177,7 +179,9 @@ static void fse_prepare( struct draw_pt_middle_end *middle, fse->key.element[i].input_buffer = 0; //nr_buffers + 1; fse->key.element[i].input_offset = 0; - fse->key.nr_elements += 1; + fse->key.nr_inputs += 1; + fse->key.nr_elements = MAX2(fse->key.nr_inputs, + fse->key.nr_outputs); } #endif diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h index ff3e19b2a8..17902ab86a 100644 --- a/src/gallium/auxiliary/draw/draw_vs.h +++ b/src/gallium/auxiliary/draw/draw_vs.h @@ -58,10 +58,12 @@ struct draw_vs_element { struct draw_vs_varient_key { unsigned output_stride; - unsigned nr_elements:16; + unsigned nr_elements:8; /* max2(nr_inputs, nr_outputs) */ + unsigned nr_inputs:8; + unsigned nr_outputs:8; unsigned viewport:1; unsigned clip:1; - unsigned pad:14; + unsigned pad:5; struct draw_vs_element element[PIPE_MAX_ATTRIBS]; }; diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index c63553216c..e2e96470f7 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -2095,7 +2095,7 @@ static void vaos_set_buffer( struct draw_vs_varient *varient, struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient; unsigned i; - for (i = 0; i < vaos->base.vs->info.num_inputs; i++) { + for (i = 0; i < vaos->base.key.nr_inputs; i++) { if (vaos->base.key.element[i].in.buffer == buf) { vaos->machine->attrib[i].input_ptr = ((char *)ptr + vaos->base.key.element[i].in.offset); diff --git a/src/gallium/auxiliary/draw/draw_vs_aos_io.c b/src/gallium/auxiliary/draw/draw_vs_aos_io.c index f39ebb7a17..ef265d61cf 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos_io.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos_io.c @@ -180,7 +180,7 @@ boolean aos_fetch_inputs( struct aos_compilation *cp, boolean linear ) { unsigned i; - for (i = 0; i < cp->vaos->base.vs->info.num_inputs; i++) { + for (i = 0; i < cp->vaos->base.key.nr_inputs; i++) { if (!load_input( cp, i, linear )) return FALSE; cp->insn_counter++; @@ -282,7 +282,7 @@ boolean aos_emit_outputs( struct aos_compilation *cp ) { unsigned i; - for (i = 0; i < cp->vaos->base.vs->info.num_outputs; i++) { + for (i = 0; i < cp->vaos->base.key.nr_outputs; i++) { unsigned format = cp->vaos->base.key.element[i].out.format; unsigned offset = cp->vaos->base.key.element[i].out.offset; diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c index d4deabfff3..dab46e8eed 100644 --- a/src/gallium/auxiliary/draw/draw_vs_varient.c +++ b/src/gallium/auxiliary/draw/draw_vs_varient.c @@ -276,11 +276,11 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs, - /* OK, have to build a new one: + /* Build free-standing fetch and emit functions: */ - fetch.nr_elements = vs->info.num_inputs; + fetch.nr_elements = key->nr_inputs; fetch.output_stride = 0; - for (i = 0; i < vs->info.num_inputs; i++) { + for (i = 0; i < key->nr_inputs; i++) { fetch.element[i].input_format = key->element[i].in.format; fetch.element[i].input_buffer = key->element[i].in.buffer; fetch.element[i].input_offset = key->element[i].in.offset; @@ -290,9 +290,9 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs, } - emit.nr_elements = vs->info.num_outputs; + emit.nr_elements = key->nr_outputs; emit.output_stride = key->output_stride; - for (i = 0; i < vs->info.num_outputs; i++) { + for (i = 0; i < key->nr_outputs; i++) { emit.element[i].input_format = PIPE_FORMAT_R32G32B32A32_FLOAT; emit.element[i].input_buffer = 0; emit.element[i].input_offset = i * 4 * sizeof(float); -- cgit v1.2.3 From 2ec419d40dba43305c28fca9658ea00541f67821 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 27 May 2008 17:45:54 +0100 Subject: draw: fix ABS aliasing bug --- src/gallium/auxiliary/draw/draw_vs_aos.c | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index e2e96470f7..1c63677e6e 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -964,12 +964,13 @@ static boolean emit_ABS( struct aos_compilation *cp, const struct tgsi_full_inst { struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]); struct x86_reg neg = aos_get_internal(cp, IMM_NEGS); - struct x86_reg dst = get_xmm_writable(cp, arg0); + struct x86_reg tmp = aos_get_xmm_reg(cp); - sse_mulps(cp->func, dst, neg); - sse_maxps(cp->func, dst, arg0); + sse_movaps(cp->func, tmp, arg0); + sse_mulps(cp->func, tmp, neg); + sse_maxps(cp->func, tmp, arg0); - store_dest(cp, &op->FullDstRegisters[0], dst); + store_dest(cp, &op->FullDstRegisters[0], tmp); return TRUE; } -- cgit v1.2.3 From 660fee8351542dadc0d5550164e753f7c2d67261 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 27 May 2008 17:49:13 +0100 Subject: draw: ensure vs outputs mapped correctly to vinfo attribs --- src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c | 1 + src/gallium/auxiliary/draw/draw_vs.h | 19 ++++++++++--------- src/gallium/auxiliary/draw/draw_vs_aos_io.c | 3 ++- 3 files changed, 13 insertions(+), 10 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c index 581026dcb0..43d7095f76 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c @@ -160,6 +160,7 @@ static void fse_prepare( struct draw_pt_middle_end *middle, * that's handled by the output_offset field. */ fse->key.element[vs_output].out.format = output_format; + fse->key.element[vs_output].out.vs_output = vs_output; fse->key.element[vs_output].out.offset = dst_offset; dst_offset += emit_sz; diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h index 17902ab86a..01171bc23d 100644 --- a/src/gallium/auxiliary/draw/draw_vs.h +++ b/src/gallium/auxiliary/draw/draw_vs.h @@ -38,22 +38,23 @@ struct draw_context; struct pipe_shader_state; -struct draw_vs_input +struct draw_varient_input { enum pipe_format format; unsigned buffer; unsigned offset; }; -struct draw_vs_output +struct draw_varient_output { - enum pipe_format format; - unsigned offset; + enum pipe_format format; /* output format */ + unsigned vs_output:8; /* which vertex shader output is this? */ + unsigned offset:24; /* offset into output vertex */ }; -struct draw_vs_element { - struct draw_vs_input in; - struct draw_vs_output out; +struct draw_varient_element { + struct draw_varient_input in; + struct draw_varient_output out; }; struct draw_vs_varient_key { @@ -64,7 +65,7 @@ struct draw_vs_varient_key { unsigned viewport:1; unsigned clip:1; unsigned pad:5; - struct draw_vs_element element[PIPE_MAX_ATTRIBS]; + struct draw_varient_element element[PIPE_MAX_ATTRIBS]; }; struct draw_vs_varient; @@ -201,7 +202,7 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs, static INLINE int draw_vs_varient_keysize( const struct draw_vs_varient_key *key ) { - return 2 * sizeof(int) + key->nr_elements * sizeof(struct draw_vs_element); + return 2 * sizeof(int) + key->nr_elements * sizeof(struct draw_varient_element); } static INLINE int draw_vs_varient_key_compare( const struct draw_vs_varient_key *a, diff --git a/src/gallium/auxiliary/draw/draw_vs_aos_io.c b/src/gallium/auxiliary/draw/draw_vs_aos_io.c index ef265d61cf..cebfaf6474 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos_io.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos_io.c @@ -285,10 +285,11 @@ boolean aos_emit_outputs( struct aos_compilation *cp ) for (i = 0; i < cp->vaos->base.key.nr_outputs; i++) { unsigned format = cp->vaos->base.key.element[i].out.format; unsigned offset = cp->vaos->base.key.element[i].out.offset; + unsigned vs_output = cp->vaos->base.key.element[i].out.vs_output; struct x86_reg data = aos_get_shader_reg( cp, TGSI_FILE_OUTPUT, - i ); + vs_output ); if (data.file != file_XMM) { struct x86_reg tmp = aos_get_xmm_reg( cp ); -- cgit v1.2.3 From fd20d1c7e8bbe2f40d73679b1514023772cfd8f6 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 27 May 2008 17:51:48 +0100 Subject: draw: add disabled debug code --- src/gallium/auxiliary/draw/draw_pt_emit.c | 15 +++++++++++- .../auxiliary/draw/draw_pt_fetch_shade_emit.c | 28 +++++++++++++++++++++- 2 files changed, 41 insertions(+), 2 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt_emit.c b/src/gallium/auxiliary/draw/draw_pt_emit.c index e37256b1ff..cf87cde996 100644 --- a/src/gallium/auxiliary/draw/draw_pt_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_emit.c @@ -41,6 +41,8 @@ struct pt_emit { struct translate_cache *cache; unsigned prim; + + const struct vertex_info *vinfo; }; void draw_pt_emit_prepare( struct pt_emit *emit, @@ -71,7 +73,7 @@ void draw_pt_emit_prepare( struct pt_emit *emit, /* Must do this after set_primitive() above: */ - vinfo = draw->render->get_vertex_info(draw->render); + emit->vinfo = vinfo = draw->render->get_vertex_info(draw->render); /* Translate from pipeline vertices to hw vertices. @@ -245,6 +247,17 @@ void draw_pt_emit_linear(struct pt_emit *emit, vertex_count, hw_verts); + if (0) { + unsigned i; + for (i = 0; i < vertex_count; i++) { + debug_printf("\n\n%s vertex %d:\n", __FUNCTION__, i); + draw_dump_emitted_vertex( emit->vinfo, + (const uint8_t *)hw_verts + + translate->key.output_stride * i ); + } + } + + render->draw_arrays(render, start, count); render->release_vertices(render, diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c index 43d7095f76..85d0bdfcab 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c @@ -60,6 +60,9 @@ struct fetch_shade_emit { struct draw_vs_varient_key key; struct draw_vs_varient *active; + + + const struct vertex_info *vinfo; }; @@ -85,7 +88,7 @@ static void fse_prepare( struct draw_pt_middle_end *middle, /* Must do this after set_primitive() above: */ - vinfo = draw->render->get_vertex_info(draw->render); + fse->vinfo = vinfo = draw->render->get_vertex_info(draw->render); @@ -267,6 +270,18 @@ static void fse_run_linear( struct draw_pt_middle_end *middle, 0, count ); + if (0) { + unsigned i; + for (i = 0; i < count; i++) { + debug_printf("\n\n%s vertex %d: (stride %d, offset %d)\n", __FUNCTION__, i, + fse->key.output_stride, + fse->key.output_stride * i); + + draw_dump_emitted_vertex( fse->vinfo, + (const uint8_t *)hw_verts + fse->key.output_stride * i ); + } + } + draw->render->release_vertices( draw->render, hw_verts, @@ -311,6 +326,17 @@ fse_run(struct draw_pt_middle_end *middle, draw_elts, draw_count ); + if (0) { + unsigned i; + for (i = 0; i < fetch_count; i++) { + debug_printf("\n\n%s vertex %d:\n", __FUNCTION__, i); + draw_dump_emitted_vertex( fse->vinfo, + (const uint8_t *)hw_verts + + fse->key.output_stride * i ); + } + } + + draw->render->release_vertices( draw->render, hw_verts, fse->key.output_stride, -- cgit v1.2.3 From 7b85ea19de09d4e7e077ca147528e90e52683690 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 27 May 2008 19:01:57 +0100 Subject: draw: support psize in vs_varient paths Preserve the vinfo "EMIT_*" format descriptors in the varient key, and deal with PSIZE directly in each implementation. --- .../auxiliary/draw/draw_pt_fetch_shade_emit.c | 35 ++------------------ src/gallium/auxiliary/draw/draw_vertex.h | 21 ++++++++++++ src/gallium/auxiliary/draw/draw_vs_aos.c | 3 ++ src/gallium/auxiliary/draw/draw_vs_aos.h | 1 + src/gallium/auxiliary/draw/draw_vs_aos_io.c | 37 ++++++++++++++-------- src/gallium/auxiliary/draw/draw_vs_varient.c | 24 +++++++++++--- 6 files changed, 71 insertions(+), 50 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c index 85d0bdfcab..729c7db999 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c @@ -77,7 +77,6 @@ static void fse_prepare( struct draw_pt_middle_end *middle, unsigned num_vs_inputs = draw->vs.vertex_shader->info.num_inputs; const struct vertex_info *vinfo; unsigned i; - boolean need_psize = 0; if (!draw->render->set_primitive( draw->render, @@ -123,34 +122,24 @@ static void fse_prepare( struct draw_pt_middle_end *middle, for (i = 0; i < vinfo->num_attribs; i++) { unsigned emit_sz = 0; - unsigned output_format = PIPE_FORMAT_NONE; - unsigned vs_output = vinfo->src_index[i]; switch (vinfo->emit[i]) { case EMIT_4F: - output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; emit_sz = 4 * sizeof(float); break; case EMIT_3F: - output_format = PIPE_FORMAT_R32G32B32_FLOAT; emit_sz = 3 * sizeof(float); break; case EMIT_2F: - output_format = PIPE_FORMAT_R32G32_FLOAT; emit_sz = 2 * sizeof(float); break; case EMIT_1F: - output_format = PIPE_FORMAT_R32_FLOAT; emit_sz = 1 * sizeof(float); break; case EMIT_1F_PSIZE: - need_psize = 1; - output_format = PIPE_FORMAT_R32_FLOAT; emit_sz = 1 * sizeof(float); - vs_output = vinfo->num_attribs + 1; break; case EMIT_4UB: - output_format = PIPE_FORMAT_B8G8R8A8_UNORM; emit_sz = 4 * sizeof(ubyte); break; default: @@ -162,33 +151,15 @@ static void fse_prepare( struct draw_pt_middle_end *middle, * numbers, not to positions in the hw vertex description -- * that's handled by the output_offset field. */ - fse->key.element[vs_output].out.format = output_format; - fse->key.element[vs_output].out.vs_output = vs_output; - fse->key.element[vs_output].out.offset = dst_offset; + fse->key.element[i].out.format = vinfo->emit[i]; + fse->key.element[i].out.vs_output = vinfo->src_index[i]; + fse->key.element[i].out.offset = dst_offset; dst_offset += emit_sz; assert(fse->key.output_stride >= dst_offset); } } - /* To make psize work, really need to tell the vertex shader to - * copy that value from input->output. For 'translate' this was - * implicit for all elements. - */ -#if 0 - if (need_psize) { - unsigned input = num_vs_inputs + 1; - const struct pipe_vertex_element *src = &draw->pt.vertex_element[i]; - fse->key.element[i].input_format = PIPE_FORMAT_R32_FLOAT; - fse->key.element[i].input_buffer = 0; //nr_buffers + 1; - fse->key.element[i].input_offset = 0; - - fse->key.nr_inputs += 1; - fse->key.nr_elements = MAX2(fse->key.nr_inputs, - fse->key.nr_outputs); - - } -#endif /* Would normally look up a vertex shader and peruse its list of * varients somehow. We omitted that step and put all the diff --git a/src/gallium/auxiliary/draw/draw_vertex.h b/src/gallium/auxiliary/draw/draw_vertex.h index 6d8bac5138..16c65c4317 100644 --- a/src/gallium/auxiliary/draw/draw_vertex.h +++ b/src/gallium/auxiliary/draw/draw_vertex.h @@ -109,4 +109,25 @@ extern void draw_compute_vertex_size(struct vertex_info *vinfo); void draw_dump_emitted_vertex(const struct vertex_info *vinfo, const uint8_t *data); + +static INLINE unsigned draw_translate_vinfo_format(unsigned format ) +{ + switch (format) { + case EMIT_1F: + case EMIT_1F_PSIZE: + return PIPE_FORMAT_R32_FLOAT; + case EMIT_2F: + return PIPE_FORMAT_R32G32_FLOAT; + case EMIT_3F: + return PIPE_FORMAT_R32G32B32_FLOAT; + case EMIT_4F: + return PIPE_FORMAT_R32G32B32A32_FLOAT; + case EMIT_4UB: + return PIPE_FORMAT_R8G8B8A8_UNORM; + default: + return PIPE_FORMAT_NONE; + } +} + + #endif /* DRAW_VERTEX_H */ diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index 1c63677e6e..d3770b2c53 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -2126,6 +2126,7 @@ static void vaos_run_elts( struct draw_vs_varient *varient, { struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient; + vaos->machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size; vaos->gen_run_elts( varient, elts, count, @@ -2139,6 +2140,7 @@ static void vaos_run_linear( struct draw_vs_varient *varient, { struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient; + vaos->machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size; vaos->gen_run_linear( varient, start, count, @@ -2204,6 +2206,7 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs, vaos->base.run_linear = vaos_run_linear; vaos->base.run_elts = vaos_run_elts; + vaos->draw = vs->draw; vaos->machine = align_malloc( sizeof(struct aos_machine), 16 ); if (!vaos->machine) goto fail; diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h index fffe2e4658..b47413ff43 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.h +++ b/src/gallium/auxiliary/draw/draw_vs_aos.h @@ -176,6 +176,7 @@ boolean aos_emit_outputs( struct aos_compilation *cp ); #define IMM_255 4 /* 255, 255, 255, 255 */ #define IMM_NEGS 5 /* -1,-1,-1,-1 */ #define IMM_RSQ 6 /* -.5,1.5,_,_ */ +#define IMM_PSIZE 7 /* not really an immediate - updated each run */ struct x86_reg aos_get_internal( struct aos_compilation *cp, unsigned imm ); diff --git a/src/gallium/auxiliary/draw/draw_vs_aos_io.c b/src/gallium/auxiliary/draw/draw_vs_aos_io.c index cebfaf6474..836110f382 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos_io.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos_io.c @@ -33,6 +33,7 @@ #include "tgsi/exec/tgsi_exec.h" #include "draw_vs.h" #include "draw_vs_aos.h" +#include "draw_vertex.h" #include "rtasm/rtasm_x86sse.h" @@ -249,24 +250,27 @@ static boolean emit_output( struct aos_compilation *cp, unsigned format ) { switch (format) { - case PIPE_FORMAT_R32_FLOAT: + case EMIT_1F: + case EMIT_1F_PSIZE: emit_store_R32(cp, ptr, dataXMM); break; - case PIPE_FORMAT_R32G32_FLOAT: + case EMIT_2F: emit_store_R32G32(cp, ptr, dataXMM); break; - case PIPE_FORMAT_R32G32B32_FLOAT: + case EMIT_3F: emit_store_R32G32B32(cp, ptr, dataXMM); break; - case PIPE_FORMAT_R32G32B32A32_FLOAT: + case EMIT_4F: emit_store_R32G32B32A32(cp, ptr, dataXMM); break; - case PIPE_FORMAT_B8G8R8A8_UNORM: - emit_swizzle(cp, dataXMM, dataXMM, SHUF(Z,Y,X,W)); - emit_store_R8G8B8A8_UNORM(cp, ptr, dataXMM); - break; - case PIPE_FORMAT_R8G8B8A8_UNORM: - emit_store_R8G8B8A8_UNORM(cp, ptr, dataXMM); + case EMIT_4UB: + if (1) { + emit_swizzle(cp, dataXMM, dataXMM, SHUF(Z,Y,X,W)); + emit_store_R8G8B8A8_UNORM(cp, ptr, dataXMM); + } + else { + emit_store_R8G8B8A8_UNORM(cp, ptr, dataXMM); + } break; default: ERROR(cp, "unhandled output format"); @@ -287,9 +291,16 @@ boolean aos_emit_outputs( struct aos_compilation *cp ) unsigned offset = cp->vaos->base.key.element[i].out.offset; unsigned vs_output = cp->vaos->base.key.element[i].out.vs_output; - struct x86_reg data = aos_get_shader_reg( cp, - TGSI_FILE_OUTPUT, - vs_output ); + struct x86_reg data; + + if (format == EMIT_1F_PSIZE) { + data = aos_get_internal_xmm( cp, IMM_PSIZE ); + } + else { + data = aos_get_shader_reg( cp, + TGSI_FILE_OUTPUT, + vs_output ); + } if (data.file != file_XMM) { struct x86_reg tmp = aos_get_xmm_reg( cp ); diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c index dab46e8eed..119a3a04b5 100644 --- a/src/gallium/auxiliary/draw/draw_vs_varient.c +++ b/src/gallium/auxiliary/draw/draw_vs_varient.c @@ -231,6 +231,10 @@ static void vsvg_run_linear( struct draw_vs_varient *varient, output_buffer, vsvg->base.key.output_stride ); + vsvg->emit->set_buffer( vsvg->emit, + 1, + &vsvg->draw->rasterizer->point_size, + 0); vsvg->emit->run( vsvg->emit, 0, count, @@ -293,11 +297,21 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs, emit.nr_elements = key->nr_outputs; emit.output_stride = key->output_stride; for (i = 0; i < key->nr_outputs; i++) { - emit.element[i].input_format = PIPE_FORMAT_R32G32B32A32_FLOAT; - emit.element[i].input_buffer = 0; - emit.element[i].input_offset = i * 4 * sizeof(float); - emit.element[i].output_format = key->element[i].out.format; - emit.element[i].output_offset = key->element[i].out.offset; + if (key->element[i].out.format != EMIT_1F_PSIZE) + { + emit.element[i].input_format = PIPE_FORMAT_R32G32B32A32_FLOAT; + emit.element[i].input_buffer = 0; + emit.element[i].input_offset = key->element[i].out.vs_output * 4 * sizeof(float); + emit.element[i].output_format = draw_translate_vinfo_format(key->element[i].out.format); + emit.element[i].output_offset = key->element[i].out.offset; + } + else { + emit.element[i].input_format = PIPE_FORMAT_R32_FLOAT; + emit.element[i].input_buffer = 1; + emit.element[i].input_offset = 0; + emit.element[i].output_format = PIPE_FORMAT_R32_FLOAT; + emit.element[i].output_offset = key->element[i].out.offset; + } } vsvg->fetch = draw_vs_get_fetch( vs->draw, &fetch ); -- cgit v1.2.3 From e38bb10824fc3dc194d7cc6987f3f4957784310e Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 27 May 2008 23:21:50 +0100 Subject: draw: reset stipple counter whenever non-line prim encountered --- src/gallium/auxiliary/draw/draw_pipe_stipple.c | 20 ++++++++++++++++++-- 1 file changed, 18 insertions(+), 2 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/draw/draw_pipe_stipple.c b/src/gallium/auxiliary/draw/draw_pipe_stipple.c index 4673d5dcba..9522b79582 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_stipple.c +++ b/src/gallium/auxiliary/draw/draw_pipe_stipple.c @@ -175,6 +175,22 @@ reset_stipple_counter(struct draw_stage *stage) stage->next->reset_stipple_counter( stage->next ); } +static void +stipple_reset_point(struct draw_stage *stage, struct prim_header *header) +{ + struct stipple_stage *stipple = stipple_stage(stage); + stipple->counter = 0; + stage->next->point(stage->next, header); +} + +static void +stipple_reset_tri(struct draw_stage *stage, struct prim_header *header) +{ + struct stipple_stage *stipple = stipple_stage(stage); + stipple->counter = 0; + stage->next->tri(stage->next, header); +} + static void stipple_first_line(struct draw_stage *stage, @@ -220,9 +236,9 @@ struct draw_stage *draw_stipple_stage( struct draw_context *draw ) stipple->stage.draw = draw; stipple->stage.next = NULL; - stipple->stage.point = draw_pipe_passthrough_point; + stipple->stage.point = stipple_reset_point; stipple->stage.line = stipple_first_line; - stipple->stage.tri = draw_pipe_passthrough_tri; + stipple->stage.tri = stipple_reset_tri; stipple->stage.reset_stipple_counter = reset_stipple_counter; stipple->stage.flush = stipple_flush; stipple->stage.destroy = stipple_destroy; -- cgit v1.2.3