* 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
+ * IN NO EVENT SHALL VMWARE 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 "util/u_math.h"
#include "util/u_memory.h"
+#include "util/u_prim.h"
#include "draw/draw_context.h"
#include "draw/draw_gs.h"
+#include "draw/draw_tess.h"
#include "draw/draw_vbuf.h"
#include "draw/draw_vertex.h"
#include "draw/draw_pt.h"
+#include "draw/draw_prim_assembler.h"
#include "draw/draw_vs.h"
#include "draw/draw_llvm.h"
+#include "gallivm/lp_bld_init.h"
+#include "gallivm/lp_bld_debug.h"
struct llvm_middle_end {
};
+/** cast wrapper */
+static inline struct llvm_middle_end *
+llvm_middle_end(struct draw_pt_middle_end *middle)
+{
+ return (struct llvm_middle_end *) middle;
+}
+
+
static void
-llvm_middle_end_prepare( struct draw_pt_middle_end *middle,
- unsigned in_prim,
- unsigned opt,
- unsigned *max_vertices )
+llvm_middle_end_prepare_gs(struct llvm_middle_end *fpme)
{
- struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
struct draw_context *draw = fpme->draw;
- struct llvm_vertex_shader *shader =
- llvm_vertex_shader(draw->vs.vertex_shader);
- struct draw_llvm_variant_key key;
- struct draw_llvm_variant *variant = NULL;
- struct draw_llvm_variant_list_item *li;
+ struct draw_llvm *llvm = fpme->llvm;
+ struct draw_geometry_shader *gs = draw->gs.geometry_shader;
+ struct draw_gs_llvm_variant_key *key;
+ struct draw_gs_llvm_variant *variant = NULL;
+ struct draw_gs_llvm_variant_list_item *li;
+ struct llvm_geometry_shader *shader = llvm_geometry_shader(gs);
+ char store[DRAW_GS_LLVM_MAX_VARIANT_KEY_SIZE];
unsigned i;
- unsigned instance_id_index = ~0;
-
- unsigned out_prim = (draw->gs.geometry_shader ?
- draw->gs.geometry_shader->output_primitive :
- in_prim);
-
- /* Add one to num_outputs because the pipeline occasionally tags on
- * an additional texcoord, eg for AA lines.
- */
- unsigned nr = MAX2( shader->base.info.num_inputs,
- shader->base.info.num_outputs + 1 );
+ key = draw_gs_llvm_make_variant_key(llvm, store);
- /* Scan for instanceID system value.
- */
- for (i = 0; i < shader->base.info.num_inputs; i++) {
- if (shader->base.info.input_semantic_name[i] == TGSI_SEMANTIC_INSTANCEID) {
- instance_id_index = i;
+ /* Search shader's list of variants for the key */
+ li = first_elem(&shader->variants);
+ while (!at_end(&shader->variants, li)) {
+ if (memcmp(&li->base->key, key, shader->variant_key_size) == 0) {
+ variant = li->base;
break;
}
+ li = next_elem(li);
}
- fpme->input_prim = in_prim;
- fpme->opt = opt;
+ if (variant) {
+ /* found the variant, move to head of global list (for LRU) */
+ move_to_head(&llvm->gs_variants_list, &variant->list_item_global);
+ }
+ else {
+ /* Need to create new variant */
+
+ /* First check if we've created too many variants. If so, free
+ * 3.125% of the LRU to avoid using too much memory.
+ */
+ if (llvm->nr_gs_variants >= DRAW_MAX_SHADER_VARIANTS) {
+ if (gallivm_debug & GALLIVM_DEBUG_PERF) {
+ debug_printf("Evicting GS: %u gs variants,\t%u total variants\n",
+ shader->variants_cached, llvm->nr_gs_variants);
+ }
- /* Always leave room for the vertex header whether we need it or
- * not. It's hard to get rid of it in particular because of the
- * viewport code in draw_pt_post_vs.c.
- */
- fpme->vertex_size = sizeof(struct vertex_header) + nr * 4 * sizeof(float);
+ /*
+ * XXX: should we flush here ?
+ */
+ for (i = 0; i < DRAW_MAX_SHADER_VARIANTS / 32; i++) {
+ struct draw_gs_llvm_variant_list_item *item;
+ if (is_empty_list(&llvm->gs_variants_list)) {
+ break;
+ }
+ item = last_elem(&llvm->gs_variants_list);
+ assert(item);
+ assert(item->base);
+ draw_gs_llvm_destroy_variant(item->base);
+ }
+ }
+ variant = draw_gs_llvm_create_variant(llvm, gs->info.num_outputs, key);
- /* XXX: it's not really gl rasterization rules we care about here,
- * but gl vs dx9 clip spaces.
- */
- draw_pt_post_vs_prepare( fpme->post_vs,
- (boolean)draw->bypass_clipping,
- (boolean)(draw->identity_viewport),
- (boolean)draw->rasterizer->gl_rasterization_rules,
- (draw->vs.edgeflag_output ? TRUE : FALSE) );
+ if (variant) {
+ insert_at_head(&shader->variants, &variant->list_item_local);
+ insert_at_head(&llvm->gs_variants_list,
+ &variant->list_item_global);
+ llvm->nr_gs_variants++;
+ shader->variants_cached++;
+ }
+ }
- draw_pt_so_emit_prepare( fpme->so_emit );
+ gs->current_variant = variant;
+}
- if (!(opt & PT_PIPELINE)) {
- draw_pt_emit_prepare( fpme->emit,
- out_prim,
- max_vertices );
+static void
+llvm_middle_end_prepare_tcs(struct llvm_middle_end *fpme)
+{
+ struct draw_context *draw = fpme->draw;
+ struct draw_llvm *llvm = fpme->llvm;
+ struct draw_tess_ctrl_shader *tcs = draw->tcs.tess_ctrl_shader;
+ struct draw_tcs_llvm_variant_key *key;
+ struct draw_tcs_llvm_variant *variant = NULL;
+ struct draw_tcs_llvm_variant_list_item *li;
+ struct llvm_tess_ctrl_shader *shader = llvm_tess_ctrl_shader(tcs);
+ char store[DRAW_TCS_LLVM_MAX_VARIANT_KEY_SIZE];
+ unsigned i;
+
+ key = draw_tcs_llvm_make_variant_key(llvm, store);
- *max_vertices = MAX2( *max_vertices,
- DRAW_PIPE_MAX_VERTICES );
+ /* Search shader's list of variants for the key */
+ li = first_elem(&shader->variants);
+ while (!at_end(&shader->variants, li)) {
+ if (memcmp(&li->base->key, key, shader->variant_key_size) == 0) {
+ variant = li->base;
+ break;
+ }
+ li = next_elem(li);
+ }
+
+ if (variant) {
+ /* found the variant, move to head of global list (for LRU) */
+ move_to_head(&llvm->tcs_variants_list, &variant->list_item_global);
}
else {
- *max_vertices = DRAW_PIPE_MAX_VERTICES;
+ /* Need to create new variant */
+
+ /* First check if we've created too many variants. If so, free
+ * 3.125% of the LRU to avoid using too much memory.
+ */
+ if (llvm->nr_tcs_variants >= DRAW_MAX_SHADER_VARIANTS) {
+ if (gallivm_debug & GALLIVM_DEBUG_PERF) {
+ debug_printf("Evicting TCS: %u tcs variants,\t%u total variants\n",
+ shader->variants_cached, llvm->nr_tcs_variants);
+ }
+
+ /*
+ * XXX: should we flush here ?
+ */
+ for (i = 0; i < DRAW_MAX_SHADER_VARIANTS / 32; i++) {
+ struct draw_tcs_llvm_variant_list_item *item;
+ if (is_empty_list(&llvm->tcs_variants_list)) {
+ break;
+ }
+ item = last_elem(&llvm->tcs_variants_list);
+ assert(item);
+ assert(item->base);
+ draw_tcs_llvm_destroy_variant(item->base);
+ }
+ }
+
+ variant = draw_tcs_llvm_create_variant(llvm, 0, key);
+
+ if (variant) {
+ insert_at_head(&shader->variants, &variant->list_item_local);
+ insert_at_head(&llvm->tcs_variants_list,
+ &variant->list_item_global);
+ llvm->nr_tcs_variants++;
+ shader->variants_cached++;
+ }
}
- /* return even number */
- *max_vertices = *max_vertices & ~1;
+ tcs->current_variant = variant;
+}
- draw_llvm_make_variant_key(fpme->llvm, &key);
+static void
+llvm_middle_end_prepare_tes(struct llvm_middle_end *fpme)
+{
+ struct draw_context *draw = fpme->draw;
+ struct draw_llvm *llvm = fpme->llvm;
+ struct draw_tess_eval_shader *tes = draw->tes.tess_eval_shader;
+ struct draw_tes_llvm_variant_key *key;
+ struct draw_tes_llvm_variant *variant = NULL;
+ struct draw_tes_llvm_variant_list_item *li;
+ struct llvm_tess_eval_shader *shader = llvm_tess_eval_shader(tes);
+ char store[DRAW_TES_LLVM_MAX_VARIANT_KEY_SIZE];
+ unsigned i;
+
+ key = draw_tes_llvm_make_variant_key(llvm, store);
+ /* Search shader's list of variants for the key */
li = first_elem(&shader->variants);
- while(!at_end(&shader->variants, li)) {
- if(memcmp(&li->base->key, &key, sizeof key) == 0) {
+ while (!at_end(&shader->variants, li)) {
+ if (memcmp(&li->base->key, key, shader->variant_key_size) == 0) {
variant = li->base;
break;
}
}
if (variant) {
- move_to_head(&fpme->llvm->vs_variants_list, &variant->list_item_global);
+ /* found the variant, move to head of global list (for LRU) */
+ move_to_head(&llvm->tes_variants_list, &variant->list_item_global);
}
else {
- unsigned i;
- if (fpme->llvm->nr_variants >= DRAW_MAX_SHADER_VARIANTS) {
+ /* Need to create new variant */
+
+ /* First check if we've created too many variants. If so, free
+ * 3.125% of the LRU to avoid using too much memory.
+ */
+ if (llvm->nr_tes_variants >= DRAW_MAX_SHADER_VARIANTS) {
+ if (gallivm_debug & GALLIVM_DEBUG_PERF) {
+ debug_printf("Evicting TES: %u tes variants,\t%u total variants\n",
+ shader->variants_cached, llvm->nr_tes_variants);
+ }
+
/*
* XXX: should we flush here ?
*/
- for (i = 0; i < DRAW_MAX_SHADER_VARIANTS / 4; i++) {
- struct draw_llvm_variant_list_item *item =
- last_elem(&fpme->llvm->vs_variants_list);
- draw_llvm_destroy_variant(item->base);
+ for (i = 0; i < DRAW_MAX_SHADER_VARIANTS / 32; i++) {
+ struct draw_tes_llvm_variant_list_item *item;
+ if (is_empty_list(&llvm->tes_variants_list)) {
+ break;
+ }
+ item = last_elem(&llvm->tes_variants_list);
+ assert(item);
+ assert(item->base);
+ draw_tes_llvm_destroy_variant(item->base);
}
}
- variant = draw_llvm_create_variant(fpme->llvm, nr);
+ variant = draw_tes_llvm_create_variant(llvm, tes->info.num_outputs, key);
if (variant) {
insert_at_head(&shader->variants, &variant->list_item_local);
- insert_at_head(&fpme->llvm->vs_variants_list, &variant->list_item_global);
- fpme->llvm->nr_variants++;
+ insert_at_head(&llvm->tes_variants_list,
+ &variant->list_item_global);
+ llvm->nr_tes_variants++;
shader->variants_cached++;
}
}
- fpme->current_variant = variant;
+ tes->current_variant = variant;
+}
+
+/**
+ * Prepare/validate middle part of the vertex pipeline.
+ * NOTE: if you change this function, also look at the non-LLVM
+ * function fetch_pipeline_prepare() for similar changes.
+ */
+static void
+llvm_middle_end_prepare( struct draw_pt_middle_end *middle,
+ unsigned in_prim,
+ unsigned opt,
+ unsigned *max_vertices )
+{
+ struct llvm_middle_end *fpme = llvm_middle_end(middle);
+ struct draw_context *draw = fpme->draw;
+ struct draw_llvm *llvm = fpme->llvm;
+ struct draw_vertex_shader *vs = draw->vs.vertex_shader;
+ struct draw_geometry_shader *gs = draw->gs.geometry_shader;
+ struct draw_tess_ctrl_shader *tcs = draw->tcs.tess_ctrl_shader;
+ struct draw_tess_eval_shader *tes = draw->tes.tess_eval_shader;
+ const unsigned out_prim = gs ? gs->output_primitive : tes ? get_tes_output_prim(tes) :
+ u_assembled_prim(in_prim);
+ unsigned point_clip = draw->rasterizer->fill_front == PIPE_POLYGON_MODE_POINT ||
+ out_prim == PIPE_PRIM_POINTS;
+ unsigned nr;
+
+ fpme->input_prim = in_prim;
+ fpme->opt = opt;
+
+ draw_pt_post_vs_prepare( fpme->post_vs,
+ draw->clip_xy,
+ draw->clip_z,
+ draw->clip_user,
+ point_clip ? draw->guard_band_points_xy :
+ draw->guard_band_xy,
+ draw->bypass_viewport,
+ draw->rasterizer->clip_halfz,
+ (draw->vs.edgeflag_output ? TRUE : FALSE) );
+
+ draw_pt_so_emit_prepare( fpme->so_emit, gs == NULL );
+
+ if (!(opt & PT_PIPELINE)) {
+ draw_pt_emit_prepare( fpme->emit, out_prim,
+ max_vertices );
+
+ *max_vertices = MAX2( *max_vertices, 4096 );
+ }
+ else {
+ /* limit max fetches by limiting max_vertices */
+ *max_vertices = 4096;
+ }
+
+ /* Get the number of float[4] attributes per vertex.
+ * Note: this must be done after draw_pt_emit_prepare() since that
+ * can effect the vertex size.
+ */
+ nr = MAX2(vs->info.num_inputs, draw_total_vs_outputs(draw));
- /*XXX we only support one constant buffer */
- fpme->llvm->jit_context.vs_constants =
- draw->pt.user.vs_constants[0];
- fpme->llvm->jit_context.gs_constants =
- draw->pt.user.gs_constants[0];
+ /* Always leave room for the vertex header whether we need it or
+ * not. It's hard to get rid of it in particular because of the
+ * viewport code in draw_pt_post_vs.c.
+ */
+ fpme->vertex_size = sizeof(struct vertex_header) + nr * 4 * sizeof(float);
+
+ /* return even number */
+ *max_vertices = *max_vertices & ~1;
+
+ /* Find/create the vertex shader variant */
+ {
+ struct draw_llvm_variant_key *key;
+ struct draw_llvm_variant *variant = NULL;
+ struct draw_llvm_variant_list_item *li;
+ struct llvm_vertex_shader *shader = llvm_vertex_shader(vs);
+ char store[DRAW_LLVM_MAX_VARIANT_KEY_SIZE];
+ unsigned i;
+
+ key = draw_llvm_make_variant_key(llvm, store);
+
+ /* Search shader's list of variants for the key */
+ li = first_elem(&shader->variants);
+ while (!at_end(&shader->variants, li)) {
+ if (memcmp(&li->base->key, key, shader->variant_key_size) == 0) {
+ variant = li->base;
+ break;
+ }
+ li = next_elem(li);
+ }
+
+ if (variant) {
+ /* found the variant, move to head of global list (for LRU) */
+ move_to_head(&llvm->vs_variants_list, &variant->list_item_global);
+ }
+ else {
+ /* Need to create new variant */
+
+ /* First check if we've created too many variants. If so, free
+ * 3.125% of the LRU to avoid using too much memory.
+ */
+ if (llvm->nr_variants >= DRAW_MAX_SHADER_VARIANTS) {
+ if (gallivm_debug & GALLIVM_DEBUG_PERF) {
+ debug_printf("Evicting VS: %u vs variants,\t%u total variants\n",
+ shader->variants_cached, llvm->nr_variants);
+ }
+
+ /*
+ * XXX: should we flush here ?
+ */
+ for (i = 0; i < DRAW_MAX_SHADER_VARIANTS / 32; i++) {
+ struct draw_llvm_variant_list_item *item;
+ if (is_empty_list(&llvm->vs_variants_list)) {
+ break;
+ }
+ item = last_elem(&llvm->vs_variants_list);
+ assert(item);
+ assert(item->base);
+ draw_llvm_destroy_variant(item->base);
+ }
+ }
+
+ variant = draw_llvm_create_variant(llvm, nr, key);
+
+ if (variant) {
+ insert_at_head(&shader->variants, &variant->list_item_local);
+ insert_at_head(&llvm->vs_variants_list,
+ &variant->list_item_global);
+ llvm->nr_variants++;
+ shader->variants_cached++;
+ }
+ }
+
+ fpme->current_variant = variant;
+ }
+
+ if (gs) {
+ llvm_middle_end_prepare_gs(fpme);
+ }
+ if (tcs) {
+ llvm_middle_end_prepare_tcs(fpme);
+ }
+ if (tes) {
+ llvm_middle_end_prepare_tes(fpme);
+ }
+}
+
+
+/**
+ * Bind/update constant buffer pointers, clip planes and viewport dims.
+ * These are "light weight" parameters which aren't baked into the
+ * generated code. Updating these items is much cheaper than revalidating
+ * and rebuilding the generated pipeline code.
+ */
+static void
+llvm_middle_end_bind_parameters(struct draw_pt_middle_end *middle)
+{
+ static const float fake_const_buf[4];
+ struct llvm_middle_end *fpme = llvm_middle_end(middle);
+ struct draw_context *draw = fpme->draw;
+ struct draw_llvm *llvm = fpme->llvm;
+ unsigned i;
+
+ for (i = 0; i < ARRAY_SIZE(llvm->jit_context.vs_constants); ++i) {
+ /*
+ * There could be a potential issue with rounding this up, as the
+ * shader expects 16-byte allocations, the fix is likely to move
+ * to LOAD intrinsic in the future and remove the vec4 constraint.
+ */
+ int num_consts =
+ DIV_ROUND_UP(draw->pt.user.vs_constants_size[i], (sizeof(float) * 4));
+ llvm->jit_context.vs_constants[i] = draw->pt.user.vs_constants[i];
+ llvm->jit_context.num_vs_constants[i] = num_consts;
+ if (num_consts == 0) {
+ llvm->jit_context.vs_constants[i] = fake_const_buf;
+ }
+ }
+ for (i = 0; i < ARRAY_SIZE(llvm->jit_context.vs_ssbos); ++i) {
+ int num_ssbos = draw->pt.user.vs_ssbos_size[i];
+ llvm->jit_context.vs_ssbos[i] = draw->pt.user.vs_ssbos[i];
+ llvm->jit_context.num_vs_ssbos[i] = num_ssbos;
+ if (num_ssbos == 0) {
+ llvm->jit_context.vs_ssbos[i] = (const uint32_t *)fake_const_buf;
+ }
+ }
+
+ for (i = 0; i < ARRAY_SIZE(llvm->gs_jit_context.constants); ++i) {
+ int num_consts =
+ DIV_ROUND_UP(draw->pt.user.gs_constants_size[i], (sizeof(float) * 4));
+ llvm->gs_jit_context.constants[i] = draw->pt.user.gs_constants[i];
+ llvm->gs_jit_context.num_constants[i] = num_consts;
+ if (num_consts == 0) {
+ llvm->gs_jit_context.constants[i] = fake_const_buf;
+ }
+ }
+ for (i = 0; i < ARRAY_SIZE(llvm->gs_jit_context.ssbos); ++i) {
+ int num_ssbos = draw->pt.user.gs_ssbos_size[i];
+ llvm->gs_jit_context.ssbos[i] = draw->pt.user.gs_ssbos[i];
+ llvm->gs_jit_context.num_ssbos[i] = num_ssbos;
+ if (num_ssbos == 0) {
+ llvm->gs_jit_context.ssbos[i] = (const uint32_t *)fake_const_buf;
+ }
+ }
+
+ for (i = 0; i < ARRAY_SIZE(llvm->tcs_jit_context.constants); ++i) {
+ int num_consts =
+ DIV_ROUND_UP(draw->pt.user.tcs_constants_size[i], (sizeof(float) * 4));
+ llvm->tcs_jit_context.constants[i] = draw->pt.user.tcs_constants[i];
+ llvm->tcs_jit_context.num_constants[i] = num_consts;
+ if (num_consts == 0) {
+ llvm->tcs_jit_context.constants[i] = fake_const_buf;
+ }
+ }
+ for (i = 0; i < ARRAY_SIZE(llvm->tcs_jit_context.ssbos); ++i) {
+ int num_ssbos = draw->pt.user.tcs_ssbos_size[i];
+ llvm->tcs_jit_context.ssbos[i] = draw->pt.user.tcs_ssbos[i];
+ llvm->tcs_jit_context.num_ssbos[i] = num_ssbos;
+ if (num_ssbos == 0) {
+ llvm->tcs_jit_context.ssbos[i] = (const uint32_t *)fake_const_buf;
+ }
+ }
+
+ for (i = 0; i < ARRAY_SIZE(llvm->tes_jit_context.constants); ++i) {
+ int num_consts =
+ DIV_ROUND_UP(draw->pt.user.tes_constants_size[i], (sizeof(float) * 4));
+ llvm->tes_jit_context.constants[i] = draw->pt.user.tes_constants[i];
+ llvm->tes_jit_context.num_constants[i] = num_consts;
+ if (num_consts == 0) {
+ llvm->tes_jit_context.constants[i] = fake_const_buf;
+ }
+ }
+ for (i = 0; i < ARRAY_SIZE(llvm->tes_jit_context.ssbos); ++i) {
+ int num_ssbos = draw->pt.user.tes_ssbos_size[i];
+ llvm->tes_jit_context.ssbos[i] = draw->pt.user.tes_ssbos[i];
+ llvm->tes_jit_context.num_ssbos[i] = num_ssbos;
+ if (num_ssbos == 0) {
+ llvm->tes_jit_context.ssbos[i] = (const uint32_t *)fake_const_buf;
+ }
+ }
+
+ llvm->jit_context.planes =
+ (float (*)[DRAW_TOTAL_CLIP_PLANES][4]) draw->pt.user.planes[0];
+ llvm->gs_jit_context.planes =
+ (float (*)[DRAW_TOTAL_CLIP_PLANES][4]) draw->pt.user.planes[0];
+
+ llvm->jit_context.viewports = draw->viewports;
+ llvm->gs_jit_context.viewports = draw->viewports;
}
-static void pipeline(struct llvm_middle_end *llvm,
- const struct draw_vertex_info *vert_info,
- const struct draw_prim_info *prim_info)
+static void
+pipeline(struct llvm_middle_end *llvm,
+ const struct draw_vertex_info *vert_info,
+ const struct draw_prim_info *prim_info)
{
if (prim_info->linear)
draw_pipeline_run_linear( llvm->draw,
prim_info );
}
-static void emit(struct pt_emit *emit,
- const struct draw_vertex_info *vert_info,
- const struct draw_prim_info *prim_info)
+
+static void
+emit(struct pt_emit *emit,
+ const struct draw_vertex_info *vert_info,
+ const struct draw_prim_info *prim_info)
{
if (prim_info->linear) {
draw_pt_emit_linear(emit, vert_info, prim_info);
}
}
+
static void
-llvm_pipeline_generic( struct draw_pt_middle_end *middle,
- const struct draw_fetch_info *fetch_info,
- const struct draw_prim_info *prim_info )
+llvm_pipeline_generic(struct draw_pt_middle_end *middle,
+ const struct draw_fetch_info *fetch_info,
+ const struct draw_prim_info *in_prim_info)
{
- struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
+ struct llvm_middle_end *fpme = llvm_middle_end(middle);
struct draw_context *draw = fpme->draw;
struct draw_geometry_shader *gshader = draw->gs.geometry_shader;
- struct draw_prim_info gs_prim_info;
+ struct draw_tess_ctrl_shader *tcs_shader = draw->tcs.tess_ctrl_shader;
+ struct draw_tess_eval_shader *tes_shader = draw->tes.tess_eval_shader;
+ struct draw_prim_info tcs_prim_info;
+ struct draw_prim_info tes_prim_info;
+ struct draw_prim_info gs_prim_info[TGSI_MAX_VERTEX_STREAMS];
struct draw_vertex_info llvm_vert_info;
- struct draw_vertex_info gs_vert_info;
+ struct draw_vertex_info tcs_vert_info;
+ struct draw_vertex_info tes_vert_info;
+ struct draw_vertex_info gs_vert_info[TGSI_MAX_VERTEX_STREAMS];
struct draw_vertex_info *vert_info;
+ struct draw_prim_info ia_prim_info;
+ struct draw_vertex_info ia_vert_info;
+ const struct draw_prim_info *prim_info = in_prim_info;
+ boolean free_prim_info = FALSE;
unsigned opt = fpme->opt;
+ boolean clipped = 0;
+ unsigned start_or_maxelt, vid_base;
+ const unsigned *elts;
+ ushort *tes_elts_out = NULL;
+ assert(fetch_info->count > 0);
llvm_vert_info.count = fetch_info->count;
llvm_vert_info.vertex_size = fpme->vertex_size;
llvm_vert_info.stride = fpme->vertex_size;
- llvm_vert_info.verts =
- (struct vertex_header *)MALLOC(fpme->vertex_size *
- align(fetch_info->count, 4));
+ llvm_vert_info.verts = (struct vertex_header *)
+ MALLOC(fpme->vertex_size *
+ align(fetch_info->count, lp_native_vector_width / 32));
if (!llvm_vert_info.verts) {
assert(0);
return;
}
- if (fetch_info->linear)
- fpme->current_variant->jit_func( &fpme->llvm->jit_context,
- llvm_vert_info.verts,
- (const char **)draw->pt.user.vbuffer,
- fetch_info->start,
- fetch_info->count,
- fpme->vertex_size,
- draw->pt.vertex_buffer,
- draw->instance_id);
- else
- fpme->current_variant->jit_func_elts( &fpme->llvm->jit_context,
- llvm_vert_info.verts,
- (const char **)draw->pt.user.vbuffer,
- fetch_info->elts,
- fetch_info->count,
- fpme->vertex_size,
- draw->pt.vertex_buffer,
- draw->instance_id);
+ if (draw->collect_statistics) {
+ draw->statistics.ia_vertices += prim_info->count;
+ if (prim_info->prim == PIPE_PRIM_PATCHES)
+ draw->statistics.ia_primitives += prim_info->count / draw->pt.vertices_per_patch;
+ else
+ draw->statistics.ia_primitives +=
+ u_decomposed_prims_for_vertices(prim_info->prim, prim_info->count);
+ draw->statistics.vs_invocations += fetch_info->count;
+ }
+
+ if (fetch_info->linear) {
+ start_or_maxelt = fetch_info->start;
+ vid_base = draw->start_index;
+ elts = NULL;
+ }
+ else {
+ start_or_maxelt = draw->pt.user.eltMax;
+ vid_base = draw->pt.user.eltBias;
+ elts = fetch_info->elts;
+ }
+ clipped = fpme->current_variant->jit_func(&fpme->llvm->jit_context,
+ llvm_vert_info.verts,
+ draw->pt.user.vbuffer,
+ fetch_info->count,
+ start_or_maxelt,
+ fpme->vertex_size,
+ draw->pt.vertex_buffer,
+ draw->instance_id,
+ vid_base,
+ draw->start_instance,
+ elts, draw->pt.user.drawid);
/* Finished with fetch and vs:
*/
fetch_info = NULL;
vert_info = &llvm_vert_info;
+ if (opt & PT_SHADE) {
+ struct draw_vertex_shader *vshader = draw->vs.vertex_shader;
+ if (tcs_shader) {
+ draw_tess_ctrl_shader_run(tcs_shader,
+ draw->pt.user.tcs_constants,
+ draw->pt.user.tcs_constants_size,
+ vert_info,
+ prim_info,
+ &vshader->info,
+ &tcs_vert_info,
+ &tcs_prim_info);
+ FREE(vert_info->verts);
+ vert_info = &tcs_vert_info;
+ prim_info = &tcs_prim_info;
+
+ } else if (tes_shader) {
+ unsigned num_prims = prim_info->count / draw->pt.vertices_per_patch;
+ tcs_prim_info = *prim_info;
+ tcs_prim_info.primitive_count = num_prims;
+ prim_info = &tcs_prim_info;
+ }
+
+ if (tes_shader) {
+ draw_tess_eval_shader_run(tes_shader,
+ draw->pt.user.tes_constants,
+ draw->pt.user.tes_constants_size,
+ tcs_shader ? tcs_shader->vertices_out : draw->pt.vertices_per_patch,
+ vert_info,
+ prim_info,
+ tcs_shader ? &tcs_shader->info : &vshader->info,
+ &tes_vert_info,
+ &tes_prim_info, &tes_elts_out);
+
+ FREE(vert_info->verts);
+ vert_info = &tes_vert_info;
+ prim_info = &tes_prim_info;
+ free_prim_info = TRUE;
+
+ /*
+ * pt emit can only handle ushort number of vertices (see
+ * render->allocate_vertices).
+ * vsplit guarantees there's never more than 4096, however GS can
+ * easily blow this up (by a factor of 256 (or even 1024) max).
+ */
+ if (vert_info->count > 65535) {
+ opt |= PT_PIPELINE;
+ }
+ }
+ }
if ((opt & PT_SHADE) && gshader) {
+ struct draw_vertex_shader *vshader = draw->vs.vertex_shader;
draw_geometry_shader_run(gshader,
draw->pt.user.gs_constants,
draw->pt.user.gs_constants_size,
vert_info,
prim_info,
- &gs_vert_info,
- &gs_prim_info);
+ tes_shader ? &tes_shader->info : &vshader->info,
+ gs_vert_info,
+ gs_prim_info);
+
+ FREE(vert_info->verts);
+ if (free_prim_info) {
+ FREE(prim_info->primitive_lengths);
+ FREE(tes_elts_out);
+ }
+ vert_info = &gs_vert_info[0];
+ prim_info = &gs_prim_info[0];
+ free_prim_info = FALSE;
+ /*
+ * pt emit can only handle ushort number of vertices (see
+ * render->allocate_vertices).
+ * vsplit guarantees there's never more than 4096, however GS can
+ * easily blow this up (by a factor of 256 (or even 1024) max).
+ */
+ if (vert_info->count > 65535) {
+ opt |= PT_PIPELINE;
+ }
+ } else {
+ if (draw_prim_assembler_is_required(draw, prim_info, vert_info)) {
+ draw_prim_assembler_run(draw, prim_info, vert_info,
+ &ia_prim_info, &ia_vert_info);
+
+ if (ia_vert_info.count) {
+ FREE(vert_info->verts);
+ vert_info = &ia_vert_info;
+ prim_info = &ia_prim_info;
+ free_prim_info = TRUE;
+ }
+ }
+ }
+ if (prim_info->count == 0) {
+ debug_printf("GS/IA didn't emit any vertices!\n");
FREE(vert_info->verts);
- vert_info = &gs_vert_info;
- prim_info = &gs_prim_info;
+ if (free_prim_info) {
+ FREE(tes_elts_out);
+ FREE(prim_info->primitive_lengths);
+ }
+ return;
}
/* stream output needs to be done before clipping */
- draw_pt_so_emit( fpme->so_emit,
- vert_info,
- prim_info );
+ draw_pt_so_emit( fpme->so_emit, gshader ? gshader->num_vertex_streams : 1, vert_info, prim_info );
- if (draw_pt_post_vs_run( fpme->post_vs, vert_info )) {
- opt |= PT_PIPELINE;
- }
+ draw_stats_clipper_primitives(draw, prim_info);
- /* Do we need to run the pipeline?
+ /*
+ * if there's no position, need to stop now, or the latter stages
+ * will try to access non-existent position output.
*/
- if (opt & PT_PIPELINE) {
- pipeline( fpme,
- vert_info,
- prim_info );
- }
- else {
- emit( fpme->emit,
- vert_info,
- prim_info );
+ if (draw_current_shader_position_output(draw) != -1) {
+ if ((opt & PT_SHADE) && (gshader || tes_shader ||
+ draw->vs.vertex_shader->info.writes_viewport_index)) {
+ clipped = draw_pt_post_vs_run( fpme->post_vs, vert_info, prim_info );
+ }
+ /* "clipped" also includes non-one edgeflag */
+ if (clipped) {
+ opt |= PT_PIPELINE;
+ }
+
+ /* Do we need to run the pipeline? Now will come here if clipped
+ */
+ if (opt & PT_PIPELINE) {
+ pipeline( fpme, vert_info, prim_info );
+ }
+ else {
+ emit( fpme->emit, vert_info, prim_info );
+ }
}
FREE(vert_info->verts);
+ if (free_prim_info) {
+ FREE(tes_elts_out);
+ FREE(prim_info->primitive_lengths);
+ }
}
-static void llvm_middle_end_run( struct draw_pt_middle_end *middle,
- const unsigned *fetch_elts,
- unsigned fetch_count,
- const ushort *draw_elts,
- unsigned draw_count )
+static inline unsigned
+prim_type(unsigned prim, unsigned flags)
+{
+ if (flags & DRAW_LINE_LOOP_AS_STRIP)
+ return PIPE_PRIM_LINE_STRIP;
+ else
+ return prim;
+}
+
+
+static void
+llvm_middle_end_run(struct draw_pt_middle_end *middle,
+ const unsigned *fetch_elts,
+ unsigned fetch_count,
+ const ushort *draw_elts,
+ unsigned draw_count,
+ unsigned prim_flags)
{
- struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
+ struct llvm_middle_end *fpme = llvm_middle_end(middle);
struct draw_fetch_info fetch_info;
struct draw_prim_info prim_info;
prim_info.start = 0;
prim_info.count = draw_count;
prim_info.elts = draw_elts;
- prim_info.prim = fpme->input_prim;
+ prim_info.prim = prim_type(fpme->input_prim, prim_flags);
+ prim_info.flags = prim_flags;
prim_info.primitive_count = 1;
prim_info.primitive_lengths = &draw_count;
}
-static void llvm_middle_end_linear_run( struct draw_pt_middle_end *middle,
- unsigned start,
- unsigned count)
+static void
+llvm_middle_end_linear_run(struct draw_pt_middle_end *middle,
+ unsigned start,
+ unsigned count,
+ unsigned prim_flags)
{
- struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
+ struct llvm_middle_end *fpme = llvm_middle_end(middle);
struct draw_fetch_info fetch_info;
struct draw_prim_info prim_info;
prim_info.start = 0;
prim_info.count = count;
prim_info.elts = NULL;
- prim_info.prim = fpme->input_prim;
+ prim_info.prim = prim_type(fpme->input_prim, prim_flags);
+ prim_info.flags = prim_flags;
prim_info.primitive_count = 1;
prim_info.primitive_lengths = &count;
}
-
static boolean
-llvm_middle_end_linear_run_elts( struct draw_pt_middle_end *middle,
- unsigned start,
- unsigned count,
- const ushort *draw_elts,
- unsigned draw_count )
+llvm_middle_end_linear_run_elts(struct draw_pt_middle_end *middle,
+ unsigned start,
+ unsigned count,
+ const ushort *draw_elts,
+ unsigned draw_count,
+ unsigned prim_flags)
{
- struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
+ struct llvm_middle_end *fpme = llvm_middle_end(middle);
struct draw_fetch_info fetch_info;
struct draw_prim_info prim_info;
prim_info.start = 0;
prim_info.count = draw_count;
prim_info.elts = draw_elts;
- prim_info.prim = fpme->input_prim;
+ prim_info.prim = prim_type(fpme->input_prim, prim_flags);
+ prim_info.flags = prim_flags;
prim_info.primitive_count = 1;
prim_info.primitive_lengths = &draw_count;
}
-
-static void llvm_middle_end_finish( struct draw_pt_middle_end *middle )
+static void
+llvm_middle_end_finish(struct draw_pt_middle_end *middle)
{
/* nothing to do */
}
-static void llvm_middle_end_destroy( struct draw_pt_middle_end *middle )
+
+static void
+llvm_middle_end_destroy(struct draw_pt_middle_end *middle)
{
- struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
+ struct llvm_middle_end *fpme = llvm_middle_end(middle);
if (fpme->fetch)
draw_pt_fetch_destroy( fpme->fetch );
{
struct llvm_middle_end *fpme = 0;
- if (!draw->engine)
+ if (!draw->llvm)
return NULL;
fpme = CALLOC_STRUCT( llvm_middle_end );
goto fail;
fpme->base.prepare = llvm_middle_end_prepare;
+ fpme->base.bind_parameters = llvm_middle_end_bind_parameters;
fpme->base.run = llvm_middle_end_run;
fpme->base.run_linear = llvm_middle_end_linear_run;
fpme->base.run_linear_elts = llvm_middle_end_linear_run_elts;
projects / mesa.git / blobdiff