X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fi915%2Fi915_prim_vbuf.c;h=79db3b650ebebf600cad591bc8cc88b132feb84a;hb=11ee41fe7f72f7136b531f0c51f820e90a610a79;hp=8a3e466c8453e1ad6f31d62764c3adae0ddd07ca;hpb=f00da2a3ff59c1a7104ac25a1c6eba5a6050ad68;p=mesa.git diff --git a/src/gallium/drivers/i915/i915_prim_vbuf.c b/src/gallium/drivers/i915/i915_prim_vbuf.c index 8a3e466c845..79db3b650eb 100644 --- a/src/gallium/drivers/i915/i915_prim_vbuf.c +++ b/src/gallium/drivers/i915/i915_prim_vbuf.c @@ -41,7 +41,7 @@ #include "draw/draw_context.h" #include "draw/draw_vbuf.h" #include "util/u_debug.h" -#include "pipe/p_inlines.h" +#include "util/u_inlines.h" #include "util/u_math.h" #include "util/u_memory.h" #include "util/u_fifo.h" @@ -52,6 +52,8 @@ #include "i915_state.h" +#define VBUF_MAP_BUFFER + /** * Primitive renderer for i915. */ @@ -73,17 +75,21 @@ struct i915_vbuf_render { unsigned fallback; /* Stuff for the vbo */ - struct intel_buffer *vbo; - size_t vbo_size; - size_t vbo_offset; + struct i915_winsys_buffer *vbo; + size_t vbo_size; /**< current size of allocated buffer */ + size_t vbo_alloc_size; /**< minimum buffer size to allocate */ + size_t vbo_hw_offset; /**< offset that we program the hardware with */ + size_t vbo_sw_offset; /**< offset that we work with */ + size_t vbo_index; /**< index offset to be added to all indices */ void *vbo_ptr; size_t vbo_max_used; + size_t vbo_max_index; /**< index offset to be added to all indices */ - /* stuff for the pool */ - struct util_fifo *pool_fifo; - unsigned pool_used; - unsigned pool_buffer_size; - boolean pool_not_used; +#ifndef VBUF_MAP_BUFFER + size_t map_used_start; + size_t map_used_end; + size_t map_size; +#endif }; @@ -97,6 +103,35 @@ i915_vbuf_render(struct vbuf_render *render) return (struct i915_vbuf_render *)render; } +/** + * If vbo state differs between renderer and context + * push state to the context. This function pushes + * hw_offset to i915->vbo_offset and vbo to i915->vbo. + * + * Side effects: + * May updates context vbo_offset and vbo fields. + */ +static void +i915_vbuf_update_vbo_state(struct vbuf_render *render) +{ + struct i915_vbuf_render *i915_render = i915_vbuf_render(render); + struct i915_context *i915 = i915_render->i915; + + if (i915->vbo != i915_render->vbo || + i915->vbo_offset != i915_render->vbo_hw_offset) { + i915->vbo = i915_render->vbo; + i915->vbo_offset = i915_render->vbo_hw_offset; + i915->dirty |= I915_NEW_VBO; + } +} + +/** + * Callback exported to the draw module. + * Returns the current vertex_info. + * + * Side effects: + * If state is dirty update derived state. + */ static const struct vertex_info * i915_vbuf_render_get_vertex_info(struct vbuf_render *render) { @@ -111,12 +146,18 @@ i915_vbuf_render_get_vertex_info(struct vbuf_render *render) return &i915->current.vertex_info; } +/** + * Reserve space in the vbo for vertices. + * + * Side effects: + * None. + */ static boolean i915_vbuf_render_reserve(struct i915_vbuf_render *i915_render, size_t size) { struct i915_context *i915 = i915_render->i915; - if (i915_render->vbo_size < size + i915_render->vbo_offset) + if (i915_render->vbo_size < size + i915_render->vbo_sw_offset) return FALSE; if (i915->vbo_flushed) @@ -125,65 +166,93 @@ i915_vbuf_render_reserve(struct i915_vbuf_render *i915_render, size_t size) return TRUE; } +/** + * Allocate a new vbo buffer should there not be enough space for + * the requested number of vertices by the draw module. + * + * Side effects: + * Updates hw_offset, sw_offset, index and allocates a new buffer. + * Will set i915->vbo to null on buffer allocation. + */ static void i915_vbuf_render_new_buf(struct i915_vbuf_render *i915_render, size_t size) { struct i915_context *i915 = i915_render->i915; - struct intel_winsys *iws = i915->iws; + struct i915_winsys *iws = i915->iws; if (i915_render->vbo) { - if (i915_render->pool_not_used) - iws->buffer_destroy(iws, i915_render->vbo); - else - u_fifo_add(i915_render->pool_fifo, i915_render->vbo); + iws->buffer_unmap(iws, i915_render->vbo); + iws->buffer_destroy(iws, i915_render->vbo); + /* + * XXX If buffers where referenced then this should be done in + * update_vbo_state but since they arn't and malloc likes to reuse + * memory we need to set it to null + */ + i915->vbo = NULL; i915_render->vbo = NULL; } i915->vbo_flushed = 0; - i915_render->vbo_size = MAX2(size, i915_render->pool_buffer_size); - i915_render->vbo_offset = 0; - - if (i915_render->vbo_size != i915_render->pool_buffer_size) { - i915_render->pool_not_used = TRUE; - i915_render->vbo = iws->buffer_create(iws, i915_render->vbo_size, 64, - INTEL_NEW_VERTEX); - } else { - i915_render->pool_not_used = FALSE; + i915_render->vbo_size = MAX2(size, i915_render->vbo_alloc_size); + i915_render->vbo_hw_offset = 0; + i915_render->vbo_sw_offset = 0; + i915_render->vbo_index = 0; - if (i915_render->pool_used >= 2) { - FLUSH_BATCH(NULL); - i915->vbo_flushed = 0; - i915_render->pool_used = 0; - } - u_fifo_pop(i915_render->pool_fifo, (void**)&i915_render->vbo); +#ifndef VBUF_MAP_BUFFER + if (i915_render->vbo_size > i915_render->map_size) { + i915_render->map_size = i915_render->vbo_size; + FREE(i915_render->vbo_ptr); + i915_render->vbo_ptr = MALLOC(i915_render->map_size); } +#endif + + i915_render->vbo = iws->buffer_create(iws, i915_render->vbo_size, + I915_NEW_VERTEX); + i915_render->vbo_ptr = iws->buffer_map(iws, i915_render->vbo, TRUE); } +/** + * Callback exported to the draw module. + * + * Side effects: + * Updates hw_offset, sw_offset, index and may allocate + * a new buffer. Also updates may update the vbo state + * on the i915 context. + */ static boolean i915_vbuf_render_allocate_vertices(struct vbuf_render *render, ushort vertex_size, ushort nr_vertices) { struct i915_vbuf_render *i915_render = i915_vbuf_render(render); - struct i915_context *i915 = i915_render->i915; size_t size = (size_t)vertex_size * (size_t)nr_vertices; + size_t offset; - /* FIXME: handle failure */ - assert(!i915->vbo); - - if (!i915_vbuf_render_reserve(i915_render, size)) { - - if (i915->vbo_flushed) - i915_render->pool_used = 0; + /* + * Align sw_offset with first multiple of vertex size from hw_offset. + * Set index to be the multiples from from hw_offset to sw_offset. + * i915_vbuf_render_new_buf will reset index, sw_offset, hw_offset + * when it allocates a new buffer this is correct. + */ + { + offset = i915_render->vbo_sw_offset - i915_render->vbo_hw_offset; + offset = util_align_npot(offset, vertex_size); + i915_render->vbo_sw_offset = i915_render->vbo_hw_offset + offset; + i915_render->vbo_index = offset / vertex_size; + } + if (!i915_vbuf_render_reserve(i915_render, size)) i915_vbuf_render_new_buf(i915_render, size); - } + + /* + * If a new buffer has been alocated sw_offset, + * hw_offset & index will be reset by new_buf + */ i915_render->vertex_size = vertex_size; - i915->vbo = i915_render->vbo; - i915->vbo_offset = i915_render->vbo_offset; - i915->dirty |= I915_NEW_VBO; + + i915_vbuf_update_vbo_state(render); if (!i915_render->vbo) return FALSE; @@ -195,14 +264,15 @@ i915_vbuf_render_map_vertices(struct vbuf_render *render) { struct i915_vbuf_render *i915_render = i915_vbuf_render(render); struct i915_context *i915 = i915_render->i915; - struct intel_winsys *iws = i915->iws; if (i915->vbo_flushed) debug_printf("%s bad vbo flush occured stalling on hw\n", __FUNCTION__); - i915_render->vbo_ptr = iws->buffer_map(iws, i915_render->vbo, TRUE); - - return (unsigned char *)i915_render->vbo_ptr + i915->vbo_offset; +#ifdef VBUF_MAP_BUFFER + return (unsigned char *)i915_render->vbo_ptr + i915_render->vbo_sw_offset; +#else + return (unsigned char *)i915_render->vbo_ptr; +#endif } static void @@ -212,10 +282,44 @@ i915_vbuf_render_unmap_vertices(struct vbuf_render *render, { struct i915_vbuf_render *i915_render = i915_vbuf_render(render); struct i915_context *i915 = i915_render->i915; - struct intel_winsys *iws = i915->iws; + struct i915_winsys *iws = i915->iws; + i915_render->vbo_max_index = max_index; i915_render->vbo_max_used = MAX2(i915_render->vbo_max_used, i915_render->vertex_size * (max_index + 1)); - iws->buffer_unmap(iws, i915_render->vbo); +#ifdef VBUF_MAP_BUFFER + (void)iws; +#else + i915_render->map_used_start = i915_render->vertex_size * min_index; + i915_render->map_used_end = i915_render->vertex_size * (max_index + 1); + iws->buffer_write(iws, i915_render->vbo, + i915_render->map_used_start + i915_render->vbo_sw_offset, + i915_render->map_used_end - i915_render->map_used_start, + (unsigned char *)i915_render->vbo_ptr + i915_render->map_used_start); + +#endif +} + +/** + * Ensure that the given max_index given is not larger ushort max. + * If it is larger then ushort max it advanced the hw_offset to the + * same position in the vbo as sw_offset and set index to zero. + * + * Side effects: + * On failure update hw_offset and index. + */ +static void +i915_vbuf_ensure_index_bounds(struct vbuf_render *render, + unsigned max_index) +{ + struct i915_vbuf_render *i915_render = i915_vbuf_render(render); + + if (max_index + i915_render->vbo_index < ((1 << 17) - 1)) + return; + + i915_render->vbo_hw_offset = i915_render->vbo_sw_offset; + i915_render->vbo_index = 0; + + i915_vbuf_update_vbo_state(render); } static boolean @@ -283,7 +387,9 @@ draw_arrays_generate_indices(struct vbuf_render *render, struct i915_vbuf_render *i915_render = i915_vbuf_render(render); struct i915_context *i915 = i915_render->i915; unsigned i; - unsigned end = start + nr; + unsigned end = start + nr + i915_render->vbo_index; + start += i915_render->vbo_index; + switch(type) { case 0: for (i = start; i+1 < end; i += 2) @@ -347,30 +453,32 @@ draw_arrays_fallback(struct vbuf_render *render, struct i915_context *i915 = i915_render->i915; unsigned nr_indices; + nr_indices = draw_arrays_calc_nr_indices(nr, i915_render->fallback); + if (!nr_indices) + return; + + i915_vbuf_ensure_index_bounds(render, start + nr_indices); + if (i915->dirty) i915_update_derived(i915); if (i915->hardware_dirty) i915_emit_hardware_state(i915); - nr_indices = draw_arrays_calc_nr_indices(nr, i915_render->fallback); - if (!nr_indices) - return; - - if (!BEGIN_BATCH(1 + (nr_indices + 1)/2, 1)) { + if (!BEGIN_BATCH(1 + (nr_indices + 1)/2)) { FLUSH_BATCH(NULL); /* Make sure state is re-emitted after a flush: */ - i915_update_derived(i915); i915_emit_hardware_state(i915); i915->vbo_flushed = 1; - if (!BEGIN_BATCH(1 + (nr_indices + 1)/2, 1)) { + if (!BEGIN_BATCH(1 + (nr_indices + 1)/2)) { assert(0); goto out; } } + OUT_BATCH(_3DPRIMITIVE | PRIM_INDIRECT | i915_render->hwprim | @@ -396,22 +504,24 @@ i915_vbuf_render_draw_arrays(struct vbuf_render *render, return; } + i915_vbuf_ensure_index_bounds(render, start + nr); + start += i915_render->vbo_index; + if (i915->dirty) i915_update_derived(i915); if (i915->hardware_dirty) i915_emit_hardware_state(i915); - if (!BEGIN_BATCH(2, 0)) { + if (!BEGIN_BATCH(2)) { FLUSH_BATCH(NULL); /* Make sure state is re-emitted after a flush: */ - i915_update_derived(i915); i915_emit_hardware_state(i915); i915->vbo_flushed = 1; - if (!BEGIN_BATCH(2, 0)) { + if (!BEGIN_BATCH(2)) { assert(0); goto out; } @@ -441,35 +551,36 @@ draw_generate_indices(struct vbuf_render *render, struct i915_vbuf_render *i915_render = i915_vbuf_render(render); struct i915_context *i915 = i915_render->i915; unsigned i; + unsigned o = i915_render->vbo_index; switch(type) { case 0: for (i = 0; i + 1 < nr_indices; i += 2) { - OUT_BATCH(indices[i] | indices[i+1] << 16); + OUT_BATCH((o+indices[i]) | (o+indices[i+1]) << 16); } if (i < nr_indices) { - OUT_BATCH(indices[i]); + OUT_BATCH((o+indices[i])); } break; case PIPE_PRIM_LINE_LOOP: if (nr_indices >= 2) { for (i = 1; i < nr_indices; i++) - OUT_BATCH(indices[i-1] | indices[i] << 16); - OUT_BATCH(indices[i-1] | indices[0] << 16); + OUT_BATCH((o+indices[i-1]) | (o+indices[i]) << 16); + OUT_BATCH((o+indices[i-1]) | (o+indices[0]) << 16); } break; case PIPE_PRIM_QUADS: for (i = 0; i + 3 < nr_indices; i += 4) { - OUT_BATCH(indices[i+0] | indices[i+1] << 16); - OUT_BATCH(indices[i+3] | indices[i+1] << 16); - OUT_BATCH(indices[i+2] | indices[i+3] << 16); + OUT_BATCH((o+indices[i+0]) | (o+indices[i+1]) << 16); + OUT_BATCH((o+indices[i+3]) | (o+indices[i+1]) << 16); + OUT_BATCH((o+indices[i+2]) | (o+indices[i+3]) << 16); } break; case PIPE_PRIM_QUAD_STRIP: for (i = 0; i + 3 < nr_indices; i += 2) { - OUT_BATCH(indices[i+0] | indices[i+1] << 16); - OUT_BATCH(indices[i+3] | indices[i+2] << 16); - OUT_BATCH(indices[i+0] | indices[i+3] << 16); + OUT_BATCH((o+indices[i+0]) | (o+indices[i+1]) << 16); + OUT_BATCH((o+indices[i+3]) | (o+indices[i+2]) << 16); + OUT_BATCH((o+indices[i+0]) | (o+indices[i+3]) << 16); } break; default: @@ -500,9 +611,9 @@ draw_calc_nr_indices(uint nr_indices, unsigned type) } static void -i915_vbuf_render_draw(struct vbuf_render *render, - const ushort *indices, - uint nr_indices) +i915_vbuf_render_draw_elements(struct vbuf_render *render, + const ushort *indices, + uint nr_indices) { struct i915_vbuf_render *i915_render = i915_vbuf_render(render); struct i915_context *i915 = i915_render->i915; @@ -514,22 +625,23 @@ i915_vbuf_render_draw(struct vbuf_render *render, if (!nr_indices) return; + i915_vbuf_ensure_index_bounds(render, i915_render->vbo_max_index); + if (i915->dirty) i915_update_derived(i915); if (i915->hardware_dirty) i915_emit_hardware_state(i915); - if (!BEGIN_BATCH(1 + (nr_indices + 1)/2, 1)) { + if (!BEGIN_BATCH(1 + (nr_indices + 1)/2)) { FLUSH_BATCH(NULL); /* Make sure state is re-emitted after a flush: */ - i915_update_derived(i915); i915_emit_hardware_state(i915); i915->vbo_flushed = 1; - if (!BEGIN_BATCH(1 + (nr_indices + 1)/2, 1)) { + if (!BEGIN_BATCH(1 + (nr_indices + 1)/2)) { assert(0); goto out; } @@ -553,20 +665,30 @@ static void i915_vbuf_render_release_vertices(struct vbuf_render *render) { struct i915_vbuf_render *i915_render = i915_vbuf_render(render); - struct i915_context *i915 = i915_render->i915; - - assert(i915->vbo); - i915_render->vbo_offset += i915_render->vbo_max_used; + i915_render->vbo_sw_offset += i915_render->vbo_max_used; i915_render->vbo_max_used = 0; - i915->vbo = NULL; - i915->dirty |= I915_NEW_VBO; + + /* + * Micro optimization, by calling update here we the offset change + * will be picked up on the next pipe_context::draw_*. + */ + i915_vbuf_update_vbo_state(render); } static void i915_vbuf_render_destroy(struct vbuf_render *render) { struct i915_vbuf_render *i915_render = i915_vbuf_render(render); + struct i915_context *i915 = i915_render->i915; + struct i915_winsys *iws = i915->iws; + + if (i915_render->vbo) { + i915->vbo = NULL; + iws->buffer_unmap(iws, i915_render->vbo); + iws->buffer_destroy(iws, i915_render->vbo); + } + FREE(i915_render); } @@ -577,13 +699,13 @@ static struct vbuf_render * i915_vbuf_render_create(struct i915_context *i915) { struct i915_vbuf_render *i915_render = CALLOC_STRUCT(i915_vbuf_render); - struct intel_winsys *iws = i915->iws; + struct i915_winsys *iws = i915->iws; int i; i915_render->i915 = i915; - - i915_render->base.max_vertex_buffer_bytes = 128*1024; - + + i915_render->base.max_vertex_buffer_bytes = 16*4096; + /* NOTE: it must be such that state and vertices indices fit in a single * batch buffer. */ @@ -594,28 +716,35 @@ i915_vbuf_render_create(struct i915_context *i915) i915_render->base.map_vertices = i915_vbuf_render_map_vertices; i915_render->base.unmap_vertices = i915_vbuf_render_unmap_vertices; i915_render->base.set_primitive = i915_vbuf_render_set_primitive; - i915_render->base.draw = i915_vbuf_render_draw; + i915_render->base.draw_elements = i915_vbuf_render_draw_elements; i915_render->base.draw_arrays = i915_vbuf_render_draw_arrays; i915_render->base.release_vertices = i915_vbuf_render_release_vertices; i915_render->base.destroy = i915_vbuf_render_destroy; +#ifndef VBUF_MAP_BUFFER + i915_render->map_size = 0; + i915_render->map_used_start = 0; + i915_render->map_used_end = 0; +#endif i915_render->vbo = NULL; + i915_render->vbo_ptr = NULL; i915_render->vbo_size = 0; - i915_render->vbo_offset = 0; + i915_render->vbo_hw_offset = 0; + i915_render->vbo_sw_offset = 0; + i915_render->vbo_alloc_size = i915_render->base.max_vertex_buffer_bytes * 4; +#ifdef VBUF_USE_POOL i915_render->pool_used = FALSE; - i915_render->pool_buffer_size = 128 * 4096; + i915_render->pool_buffer_size = i915_render->vbo_alloc_size; i915_render->pool_fifo = u_fifo_create(6); for (i = 0; i < 6; i++) u_fifo_add(i915_render->pool_fifo, - iws->buffer_create(iws, i915_render->pool_buffer_size, 64, - INTEL_NEW_VERTEX)); - -#if 0 - /* TODO JB: is this realy needed? */ - i915_render->vbo_ptr = iws->buffer_map(iws, i915_render->vbo, TRUE); - iws->buffer_unmap(iws, i915_render->vbo); + iws->buffer_create(iws, i915_render->pool_buffer_size, + I915_NEW_VERTEX)); +#else + (void)i; + (void)iws; #endif return &i915_render->base;