[Mesa-dev] [PATCH 5/6] gallium/llvm: implement geometry shaders in the llvm paths
Brian Paul
brianp at vmware.com
Wed Mar 27 07:41:05 PDT 2013
On 03/26/2013 10:38 AM, Zack Rusin wrote:
> This commits implements code generation of the geometry shaders in
> the SOA paths. All the code is there but bugs are likely present.
>
> Signed-off-by: Zack Rusin<zackr at vmware.com>
> ---
> src/gallium/auxiliary/draw/draw_context.c | 17 +-
> src/gallium/auxiliary/draw/draw_context.h | 5 +
> src/gallium/auxiliary/draw/draw_gs.c | 297 +++++++++-
> src/gallium/auxiliary/draw/draw_gs.h | 26 +-
> src/gallium/auxiliary/draw/draw_llvm.c | 567 ++++++++++++++++++--
> src/gallium/auxiliary/draw/draw_llvm.h | 161 +++++-
> .../draw/draw_pt_fetch_shade_pipeline_llvm.c | 75 +++
> src/gallium/auxiliary/gallivm/lp_bld_tgsi.h | 27 +-
> src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c | 173 +++++-
> src/gallium/drivers/llvmpipe/lp_state_fs.c | 4 +-
> 10 files changed, 1273 insertions(+), 79 deletions(-)
>
> diff --git a/src/gallium/auxiliary/draw/draw_context.c b/src/gallium/auxiliary/draw/draw_context.c
> index 6b70ac8..d64b82b 100644
> --- a/src/gallium/auxiliary/draw/draw_context.c
> +++ b/src/gallium/auxiliary/draw/draw_context.c
> @@ -46,7 +46,7 @@
> #include "gallivm/lp_bld_limits.h"
> #include "draw_llvm.h"
>
> -static boolean
> +boolean
> draw_get_option_use_llvm(void)
> {
> static boolean first = TRUE;
> @@ -808,16 +808,15 @@ draw_set_mapped_texture(struct draw_context *draw,
> uint32_t img_stride[PIPE_MAX_TEXTURE_LEVELS],
> uint32_t mip_offsets[PIPE_MAX_TEXTURE_LEVELS])
> {
> - if (shader_stage == PIPE_SHADER_VERTEX) {
> #ifdef HAVE_LLVM
> - if (draw->llvm)
> - draw_llvm_set_mapped_texture(draw,
> - sview_idx,
> - width, height, depth, first_level,
> - last_level, base_ptr,
> - row_stride, img_stride, mip_offsets);
> + if (draw->llvm)
> + draw_llvm_set_mapped_texture(draw,
> + shader_stage,
> + sview_idx,
> + width, height, depth, first_level,
> + last_level, base_ptr,
> + row_stride, img_stride, mip_offsets);
> #endif
> - }
> }
>
> /**
> diff --git a/src/gallium/auxiliary/draw/draw_context.h b/src/gallium/auxiliary/draw/draw_context.h
> index 18c8595..369f6c8 100644
> --- a/src/gallium/auxiliary/draw/draw_context.h
> +++ b/src/gallium/auxiliary/draw/draw_context.h
> @@ -282,4 +282,9 @@ draw_get_shader_param(unsigned shader, enum pipe_shader_cap param);
> int
> draw_get_shader_param_no_llvm(unsigned shader, enum pipe_shader_cap param);
>
> +#ifdef HAVE_LLVM
> +boolean
> +draw_get_option_use_llvm(void);
> +#endif
> +
> #endif /* DRAW_CONTEXT_H */
> diff --git a/src/gallium/auxiliary/draw/draw_gs.c b/src/gallium/auxiliary/draw/draw_gs.c
> index 2b50c9c..c1e1f56 100644
> --- a/src/gallium/auxiliary/draw/draw_gs.c
> +++ b/src/gallium/auxiliary/draw/draw_gs.c
> @@ -29,6 +29,9 @@
>
> #include "draw_private.h"
> #include "draw_context.h"
> +#ifdef HAVE_LLVM
> +#include "draw_llvm.h"
> +#endif
>
> #include "tgsi/tgsi_parse.h"
> #include "tgsi/tgsi_exec.h"
> @@ -61,7 +64,7 @@ draw_gs_get_input_index(int semantic, int index,
> static INLINE boolean
> draw_gs_should_flush(struct draw_geometry_shader *shader)
> {
> - return (shader->fetched_prim_count == 4);
> + return (shader->fetched_prim_count == shader->vector_length);
> }
>
> /*#define DEBUG_OUTPUTS 1*/
> @@ -176,7 +179,7 @@ static void tgsi_fetch_gs_input(struct draw_geometry_shader *shader,
> }
>
> static void tgsi_gs_prepare(struct draw_geometry_shader *shader,
> - const void *constants[PIPE_MAX_CONSTANT_BUFFERS],
> + const void *constants[PIPE_MAX_CONSTANT_BUFFERS],
> const unsigned constants_size[PIPE_MAX_CONSTANT_BUFFERS])
> {
> struct tgsi_exec_machine *machine = shader->machine;
> @@ -199,10 +202,148 @@ static unsigned tgsi_gs_run(struct draw_geometry_shader *shader,
> /* run interpreter */
> tgsi_exec_machine_run(machine);
>
> - return
> + return
> machine->Temps[TGSI_EXEC_TEMP_PRIMITIVE_I].xyzw[TGSI_EXEC_TEMP_PRIMITIVE_C].u[0];
> }
>
> +#ifdef HAVE_LLVM
> +
> +static void
> +llvm_fetch_gs_input(struct draw_geometry_shader *shader,
> + unsigned *indices,
> + unsigned num_vertices,
> + unsigned prim_idx)
> +{
> + unsigned slot, vs_slot, i;
> + unsigned input_vertex_stride = shader->input_vertex_stride;
> + const float (*input_ptr)[4];
> + float (*input_data)[6][PIPE_MAX_SHADER_INPUTS][TGSI_NUM_CHANNELS][TGSI_NUM_CHANNELS] =&shader->gs_input->data;
> +
> + input_ptr = shader->input;
> +
> + for (i = 0; i< num_vertices; ++i) {
> + const float (*input)[4];
> +#if DEBUG_INPUTS
> + debug_printf("%d) vertex index = %d (prim idx = %d)\n",
> + i, indices[i], prim_idx);
> +#endif
> + input = (const float (*)[4])(
> + (const char *)input_ptr + (indices[i] * input_vertex_stride));
> + for (slot = 0, vs_slot = 0; slot< shader->info.num_inputs; ++slot) {
> + if (shader->info.input_semantic_name[slot] == TGSI_SEMANTIC_PRIMID) {
> + (*input_data)[i][slot][0][prim_idx] = (float)shader->in_prim_idx;
> + (*input_data)[i][slot][1][prim_idx] = (float)shader->in_prim_idx;
> + (*input_data)[i][slot][2][prim_idx] = (float)shader->in_prim_idx;
> + (*input_data)[i][slot][3][prim_idx] = (float)shader->in_prim_idx;
> + } else {
> + vs_slot = draw_gs_get_input_index(
> + shader->info.input_semantic_name[slot],
> + shader->info.input_semantic_index[slot],
> + shader->input_info);
> +#if DEBUG_INPUTS
> + debug_printf("\tSlot = %d, vs_slot = %d, idx = %d:\n",
> + slot, vs_slot, idx);
> +#endif
> +#if 0
> + assert(!util_is_inf_or_nan(input[vs_slot][0]));
> + assert(!util_is_inf_or_nan(input[vs_slot][1]));
> + assert(!util_is_inf_or_nan(input[vs_slot][2]));
> + assert(!util_is_inf_or_nan(input[vs_slot][3]));
> +#endif
> + (*input_data)[i][slot][0][prim_idx] = input[vs_slot][0];
> + (*input_data)[i][slot][1][prim_idx] = input[vs_slot][1];
> + (*input_data)[i][slot][2][prim_idx] = input[vs_slot][2];
> + (*input_data)[i][slot][3][prim_idx] = input[vs_slot][3];
> +#if DEBUG_INPUTS
> + debug_printf("\t\t%f %f %f %f\n",
> + (*input_data)[i][slot][0][prim_idx],
> + (*input_data)[i][slot][1][prim_idx],
> + (*input_data)[i][slot][2][prim_idx],
> + (*input_data)[i][slot][3][prim_idx]);
> +#endif
> + ++vs_slot;
> + }
> + }
> + }
> +}
> +
> +static void
> +llvm_fetch_gs_outputs(struct draw_geometry_shader *shader,
> + unsigned num_primitives,
> + float (**p_output)[4])
> +{
> + int total_verts = 0;
> + int vertex_count = 0;
> + int total_prims = 0;
> + int max_prims_per_invocation = 0;
> + char *output_ptr = (char*)shader->gs_output;
> + int i, j, prim_idx;
> +
> + for (i = 0; i< shader->vector_length; ++i) {
> + int prims = shader->llvm_emitted_primitives[i];
> + total_prims += prims;
> + max_prims_per_invocation = MAX2(max_prims_per_invocation, prims);
> + }
> + for (int i = 0; i< shader->vector_length; ++i) {
remove int decl.
> + total_verts += shader->llvm_emitted_vertices[i];
> + }
> +
> +
> + output_ptr += shader->emitted_vertices * shader->vertex_size;
> + for (i = 0; i< shader->vector_length - 1; ++i) {
> + int current_verts = shader->llvm_emitted_vertices[i];
> +
> + if (current_verts != shader->max_output_vertices) {
> + memcpy(output_ptr + (vertex_count + current_verts) * shader->vertex_size,
> + output_ptr + (vertex_count + shader->max_output_vertices) * shader->vertex_size,
> + shader->vertex_size * (total_verts - vertex_count - current_verts));
> + }
> + vertex_count += current_verts;
> + }
> +
> + prim_idx = 0;
> + for (i = 0; i< shader->vector_length; ++i) {
> + int num_prims = shader->llvm_emitted_primitives[i];
> + for (j = 0; j< num_prims; ++j) {
> + int prim_length =
> + shader->llvm_prim_lengths[j][i];
> + shader->primitive_lengths[shader->emitted_primitives + prim_idx] =
> + prim_length;
> + ++prim_idx;
> + }
> + }
> +
> + shader->emitted_primitives += total_prims;
> + shader->emitted_vertices += total_verts;
> +}
> +
> +static void
> +llvm_gs_prepare(struct draw_geometry_shader *shader,
> + const void *constants[PIPE_MAX_CONSTANT_BUFFERS],
> + const unsigned constants_size[PIPE_MAX_CONSTANT_BUFFERS])
> +{
> +}
> +
> +static unsigned
> +llvm_gs_run(struct draw_geometry_shader *shader,
> + unsigned input_primitives)
> +{
> + unsigned ret;
> + char *input = (char*)shader->gs_output;
> +
> + input += (shader->emitted_vertices * shader->vertex_size);
> +
> + ret = shader->current_variant->jit_func(
> + shader->jit_context, shader->gs_input->data,
> + (struct vertex_header*)input,
> + input_primitives,
> + shader->draw->instance_id);
> +
> + return ret;
> +}
> +
> +#endif
> +
> static void gs_flush(struct draw_geometry_shader *shader)
> {
> unsigned out_prim_count;
> @@ -213,13 +354,15 @@ static void gs_flush(struct draw_geometry_shader *shader)
> input_primitives<= 4);
>
> out_prim_count = shader->run(shader, input_primitives);
> + shader->fetch_outputs(shader, out_prim_count,
> +&shader->tmp_output);
> +
> #if 0
> debug_printf("PRIM emitted prims = %d (verts=%d), cur prim count = %d\n",
> shader->emitted_primitives, shader->emitted_vertices,
> out_prim_count);
> #endif
> - shader->fetch_outputs(shader, out_prim_count,
> -&shader->tmp_output);
> +
> shader->fetched_prim_count = 0;
> }
>
> @@ -325,8 +468,8 @@ static void gs_tri_adj(struct draw_geometry_shader *shader,
> * Execute geometry shader.
> */
> int draw_geometry_shader_run(struct draw_geometry_shader *shader,
> - const void *constants[PIPE_MAX_CONSTANT_BUFFERS],
> - const unsigned constants_size[PIPE_MAX_CONSTANT_BUFFERS],
> + const void *constants[PIPE_MAX_CONSTANT_BUFFERS],
> + const unsigned constants_size[PIPE_MAX_CONSTANT_BUFFERS],
> const struct draw_vertex_info *input_verts,
> const struct draw_prim_info *input_prim,
> const struct tgsi_shader_info *input_info,
> @@ -338,14 +481,20 @@ int draw_geometry_shader_run(struct draw_geometry_shader *shader,
> unsigned num_outputs = shader->info.num_outputs;
> unsigned vertex_size = sizeof(struct vertex_header) + num_outputs * 4 * sizeof(float);
> unsigned num_input_verts = input_prim->linear ?
> - input_verts->count :
> - input_prim->count;
> + input_verts->count :
> + input_prim->count;
> unsigned num_in_primitives =
> - MAX2(u_gs_prims_for_vertices(input_prim->prim, num_input_verts),
> - u_gs_prims_for_vertices(shader->input_primitive, num_input_verts));
> + align(
> + MAX2(u_gs_prims_for_vertices(input_prim->prim, num_input_verts),
> + u_gs_prims_for_vertices(shader->input_primitive, num_input_verts)),
> + shader->vector_length);
> unsigned max_out_prims = u_gs_prims_for_vertices(shader->output_primitive,
> shader->max_output_vertices)
> - * num_in_primitives;
> + * num_in_primitives;
> +
> + //Assume at least one primitive
> + max_out_prims = MAX2(max_out_prims, 1);
> +
>
> output_verts->vertex_size = vertex_size;
> output_verts->stride = output_verts->vertex_size;
> @@ -379,6 +528,34 @@ int draw_geometry_shader_run(struct draw_geometry_shader *shader,
> FREE(shader->primitive_lengths);
> shader->primitive_lengths = MALLOC(max_out_prims * sizeof(unsigned));
>
> +
> +#ifdef HAVE_LLVM
> + if (draw_get_option_use_llvm()) {
> + shader->gs_output = output_verts->verts;
> + if (max_out_prims> shader->max_out_prims) {
> + int i;
'unsigned i' since shader->max_out_prims is unsigned, to avoid MVSC
warnings.
> + if (shader->llvm_prim_lengths) {
> + for (i = 0; i< shader->max_out_prims; ++i) {
> + align_free(shader->llvm_prim_lengths[i]);
> + }
> + free(shader->llvm_prim_lengths);
> + }
> +
> + shader->llvm_prim_lengths = malloc(max_out_prims * sizeof(unsigned*));
> + for (i = 0; i< max_out_prims; ++i) {
> + int vector_size = shader->vector_length * sizeof(unsigned);
> + shader->llvm_prim_lengths[i] =
> + align_malloc(vector_size, vector_size);
> + }
> +
> + shader->max_out_prims = max_out_prims;
> + }
> + shader->jit_context->prim_lengths = shader->llvm_prim_lengths;
> + shader->jit_context->emitted_vertices = shader->llvm_emitted_vertices;
> + shader->jit_context->emitted_prims = shader->llvm_emitted_primitives;
> + }
> +#endif
> +
> shader->prepare(shader, constants, constants_size);
>
> if (input_prim->linear)
> @@ -458,10 +635,27 @@ struct draw_geometry_shader *
> draw_create_geometry_shader(struct draw_context *draw,
> const struct pipe_shader_state *state)
> {
> +#ifdef HAVE_LLVM
> + struct llvm_geometry_shader *llvm_gs;
> +#endif
> struct draw_geometry_shader *gs;
> unsigned i;
>
> - gs = CALLOC_STRUCT(draw_geometry_shader);
> +#ifdef HAVE_LLVM
> + if (draw_get_option_use_llvm()) {
> + llvm_gs = CALLOC_STRUCT(llvm_geometry_shader);
> +
> + if (llvm_gs == NULL)
> + return NULL;
> +
> + gs =&llvm_gs->base;
> +
> + make_empty_list(&llvm_gs->variants);
> + } else
> +#endif
> + {
> + gs = CALLOC_STRUCT(draw_geometry_shader);
> + }
>
> if (!gs)
> return NULL;
> @@ -480,6 +674,14 @@ draw_create_geometry_shader(struct draw_context *draw,
> gs->input_primitive = PIPE_PRIM_TRIANGLES;
> gs->output_primitive = PIPE_PRIM_TRIANGLE_STRIP;
> gs->max_output_vertices = 32;
> + gs->max_out_prims = 0;
> +
> + if (draw_get_option_use_llvm()) {
> + //gs->vector_length = lp_native_vector_width / 32;
> + gs->vector_length = TGSI_NUM_CHANNELS;
Is this a "to do" item or something unresolved?
> + } else {
> + gs->vector_length = TGSI_NUM_CHANNELS;
> + }
>
> for (i = 0; i< gs->info.num_properties; ++i) {
> if (gs->info.properties[i].name ==
> @@ -501,10 +703,36 @@ draw_create_geometry_shader(struct draw_context *draw,
>
> gs->machine = draw->gs.tgsi.machine;
>
> - gs->fetch_outputs = tgsi_fetch_gs_outputs;
> - gs->fetch_inputs = tgsi_fetch_gs_input;
> - gs->prepare = tgsi_gs_prepare;
> - gs->run = tgsi_gs_run;
> +#ifdef HAVE_LLVM
> + if (draw_get_option_use_llvm()) {
> + int vector_size = gs->vector_length * sizeof(float);
> + gs->gs_input = align_malloc(sizeof(struct draw_gs_inputs), 16);
> + memset(gs->gs_input, 0, sizeof(struct draw_gs_inputs));
> + gs->llvm_prim_lengths = 0;
> +
> + gs->llvm_emitted_primitives = align_malloc(vector_size, vector_size);
> + gs->llvm_emitted_vertices = align_malloc(vector_size, vector_size);
> +
> + gs->fetch_outputs = llvm_fetch_gs_outputs;
> + gs->fetch_inputs = llvm_fetch_gs_input;
> + gs->prepare = llvm_gs_prepare;
> + gs->run = llvm_gs_run;
> +
> + gs->jit_context =&draw->llvm->gs_jit_context;
> +
> +
> + llvm_gs->variant_key_size =
> + draw_gs_llvm_variant_key_size(
> + MAX2(gs->info.file_max[TGSI_FILE_SAMPLER]+1,
> + gs->info.file_max[TGSI_FILE_SAMPLER_VIEW]+1));
> + } else
> +#endif
> + {
> + gs->fetch_outputs = tgsi_fetch_gs_outputs;
> + gs->fetch_inputs = tgsi_fetch_gs_input;
> + gs->prepare = tgsi_gs_prepare;
> + gs->run = tgsi_gs_run;
> + }
>
> return gs;
> }
> @@ -529,7 +757,42 @@ void draw_bind_geometry_shader(struct draw_context *draw,
> void draw_delete_geometry_shader(struct draw_context *draw,
> struct draw_geometry_shader *dgs)
> {
> +#ifdef HAVE_LLVM
> + if (draw_get_option_use_llvm()) {
> + struct llvm_geometry_shader *shader = llvm_geometry_shader(dgs);
> + struct draw_gs_llvm_variant_list_item *li;
> +
> + li = first_elem(&shader->variants);
> + while(!at_end(&shader->variants, li)) {
> + struct draw_gs_llvm_variant_list_item *next = next_elem(li);
> + draw_gs_llvm_destroy_variant(li->base);
> + li = next;
> + }
> +
> + assert(shader->variants_cached == 0);
> +
> + if (dgs->llvm_prim_lengths) {
> + unsigned i;
> + for (i = 0; i< dgs->max_out_prims; ++i) {
> + align_free(dgs->llvm_prim_lengths[i]);
> + }
> + free(dgs->llvm_prim_lengths);
> + }
> + align_free(dgs->llvm_emitted_primitives);
> + align_free(dgs->llvm_emitted_vertices);
> +
> + align_free(dgs->gs_input);
> + }
> +#endif
> +
> FREE(dgs->primitive_lengths);
> FREE((void*) dgs->state.tokens);
> FREE(dgs);
> }
> +
> +
> +void draw_gs_set_current_variant(struct draw_geometry_shader *shader,
> + struct draw_gs_llvm_variant *variant)
> +{
> + shader->current_variant = variant;
> +}
> diff --git a/src/gallium/auxiliary/draw/draw_gs.h b/src/gallium/auxiliary/draw/draw_gs.h
> index 7ab4f04..62bfcab 100644
> --- a/src/gallium/auxiliary/draw/draw_gs.h
> +++ b/src/gallium/auxiliary/draw/draw_gs.h
> @@ -31,11 +31,19 @@
> #include "draw_context.h"
> #include "draw_private.h"
>
> -
> #define MAX_TGSI_PRIMITIVES 4
>
> struct draw_context;
>
> +#ifdef HAVE_LLVM
> +struct draw_gs_jit_context;
> +struct draw_gs_llvm_variant;
> +
> +struct draw_gs_inputs {
> + float data[6][PIPE_MAX_SHADER_INPUTS][TGSI_NUM_CHANNELS][TGSI_NUM_CHANNELS];
Could you add some extra comments explaining all these array
dimensions? it's a bit overwhelming.
> +};
> +#endif
> +
> /**
> * Private version of the compiled geometry shader
> */
> @@ -66,6 +74,19 @@ struct draw_geometry_shader {
> unsigned fetched_prim_count;
> const float (*input)[4];
> const struct tgsi_shader_info *input_info;
> + unsigned vector_length;
> + unsigned max_out_prims;
> +
> +#ifdef HAVE_LLVM
> + struct draw_gs_inputs *gs_input;
> + struct draw_gs_jit_context *jit_context;
> + struct draw_gs_llvm_variant *current_variant;
> + struct vertex_header *gs_output;
> +
> + int **llvm_prim_lengths;
> + int *llvm_emitted_primitives;
> + int *llvm_emitted_vertices;
Should any of those be unsigned?
> +#endif
>
> void (*fetch_inputs)(struct draw_geometry_shader *shader,
> unsigned *indices,
> @@ -102,4 +123,7 @@ void draw_geometry_shader_prepare(struct draw_geometry_shader *shader,
> int draw_gs_max_output_vertices(struct draw_geometry_shader *shader,
> unsigned pipe_prim);
>
> +void draw_gs_set_current_variant(struct draw_geometry_shader *shader,
> + struct draw_gs_llvm_variant *variant);
> +
> #endif
> +static void
> +draw_gs_llvm_generate(struct draw_llvm *llvm,
> + struct draw_gs_llvm_variant *variant)
> +{
> + struct gallivm_state *gallivm = variant->gallivm;
> + LLVMContextRef context = gallivm->context;
> + LLVMTypeRef int32_type = LLVMInt32TypeInContext(context);
> + LLVMTypeRef arg_types[5];
> + LLVMTypeRef func_type;
> + LLVMValueRef variant_func;
> + LLVMValueRef context_ptr;
> + LLVMBasicBlockRef block;
> + LLVMBuilderRef builder;
> + LLVMValueRef io_ptr, input_array, num_prims, mask_val;
> + struct lp_build_sampler_soa *sampler = 0;
> + struct lp_build_context bld;
> + struct lp_bld_tgsi_system_values system_values;
> + struct lp_type gs_type;
> + unsigned i;
> + struct lp_build_tgsi_gs_args gs_args;
> + const struct tgsi_token *tokens = variant->shader->base.state.tokens;
> + LLVMValueRef consts_ptr;
> + LLVMValueRef outputs[PIPE_MAX_SHADER_OUTPUTS][TGSI_NUM_CHANNELS];
> + struct lp_build_mask_context mask;
> +
> + memset(&system_values, 0, sizeof(system_values));
> +
> + assert(variant->vertex_header_ptr_type);
> +
> + arg_types[0] = get_gs_context_ptr_type(variant); /* context */
> + arg_types[1] = variant->input_array_type; /* input */
> + arg_types[2] = variant->vertex_header_ptr_type; /* vertex_header */
> + arg_types[3] = int32_type; /* num_prims */
> + arg_types[4] = int32_type; /* instance_id */
> +
> + func_type = LLVMFunctionType(int32_type, arg_types, Elements(arg_types), 0);
> +
> + variant_func = LLVMAddFunction(gallivm->module, "draw_geometry_shader",
> + func_type);
> + variant->function = variant_func;
> +
> + LLVMSetFunctionCallConv(variant_func, LLVMCCallConv);
> +
> + for (i = 0; i< Elements(arg_types); ++i)
> + if (LLVMGetTypeKind(arg_types[i]) == LLVMPointerTypeKind)
> + LLVMAddAttribute(LLVMGetParam(variant_func, i),
> + LLVMNoAliasAttribute);
> +
> + context_ptr = LLVMGetParam(variant_func, 0);
> + input_array = LLVMGetParam(variant_func, 1);
> + io_ptr = LLVMGetParam(variant_func, 2);
> + num_prims = LLVMGetParam(variant_func, 3);
> + system_values.instance_id = LLVMGetParam(variant_func, 4);
> +
> + lp_build_name(context_ptr, "context");
> + lp_build_name(input_array, "input");
> + lp_build_name(io_ptr, "io");
lp_build_name(num_prims, "num_prims");
> + lp_build_name(system_values.instance_id, "instance_id");
> +
> + variant->context_ptr = context_ptr;
> + variant->io_ptr = io_ptr;
> + variant->num_prims = num_prims;
> +
> + gs_args.input = input_array;
> + gs_args.emit_vertex = draw_gs_llvm_emit_vertex;
> + gs_args.end_primitive = draw_gs_llvm_end_primitive;
> + gs_args.gs_epilogue = draw_gs_llvm_epilogue;
> + gs_args.user_data = variant;
> +
> + /*
> + * Function body
> + */
> +
> + block = LLVMAppendBasicBlockInContext(gallivm->context, variant_func, "entry");
> + builder = gallivm->builder;
> + LLVMPositionBuilderAtEnd(builder, block);
> +
> + lp_build_context_init(&bld, gallivm, lp_type_int(32));
> +
> + memset(&gs_type, 0, sizeof gs_type);
> + gs_type.floating = TRUE; /* floating point values */
> + gs_type.sign = TRUE; /* values are signed */
> + gs_type.norm = FALSE; /* values are not limited to [0,1] or [-1,1] */
> + gs_type.width = 32; /* 32-bit float */
> + gs_type.length = variant->shader->base.vector_length;
> +
> + consts_ptr = draw_gs_jit_context_constants(variant->gallivm, context_ptr);
> +
> + /* code generated texture sampling */
> + sampler = draw_llvm_sampler_soa_create(variant->key.samplers,
> + context_ptr);
> +
> + mask_val = generate_mask_value(variant, gs_type);
> + lp_build_mask_begin(&mask, gallivm, gs_type, mask_val);
> +
> + lp_build_tgsi_soa(variant->gallivm,
> + tokens,
> + gs_type,
> +&mask,
> + consts_ptr,
> +&system_values,
> + NULL /*pos*/,
> + NULL,
> + outputs,
> + sampler,
> +&llvm->draw->gs.geometry_shader->info,
> +&gs_args);
> +
> + lp_build_mask_end(&mask);
> +
> + LLVMBuildRet(builder, lp_build_zero(gallivm, lp_type_uint(32)));
> +
> + gallivm_verify_function(gallivm, variant_func);
> +}
> +
> +
> +struct draw_gs_llvm_variant *
> +draw_gs_llvm_create_variant(struct draw_llvm *llvm,
> + unsigned num_outputs,
> + const struct draw_gs_llvm_variant_key *key)
> +{
> + struct draw_gs_llvm_variant *variant;
> + struct llvm_geometry_shader *shader =
> + llvm_geometry_shader(llvm->draw->gs.geometry_shader);
> + LLVMTypeRef vertex_header;
> +
> + variant = MALLOC(sizeof *variant +
> + shader->variant_key_size -
> + sizeof variant->key);
> + if (variant == NULL)
> + return NULL;
> +
> + variant->llvm = llvm;
> + variant->shader = shader;
> +
> + variant->gallivm = gallivm_create();
> +
> + create_gs_jit_types(variant);
> +
> + memcpy(&variant->key, key, shader->variant_key_size);
> +
> + vertex_header = create_jit_vertex_header(variant->gallivm, num_outputs);
> +
> + variant->vertex_header_ptr_type = LLVMPointerType(vertex_header, 0);
> +
> + draw_gs_llvm_generate(llvm, variant);
> +
> + gallivm_compile_module(variant->gallivm);
> +
> + variant->jit_func = (draw_gs_jit_func)
> + gallivm_jit_function(variant->gallivm, variant->function);
> +
> + variant->list_item_global.base = variant;
> + variant->list_item_local.base = variant;
> + /*variant->no = */shader->variants_created++;
> + variant->list_item_global.base = variant;
> +
> + return variant;
> +}
> +
> +void
> +draw_gs_llvm_destroy_variant(struct draw_gs_llvm_variant *variant)
> +{
> + struct draw_llvm *llvm = variant->llvm;
> +
> + if (variant->function) {
> + gallivm_free_function(variant->gallivm,
> + variant->function, variant->jit_func);
> + }
> +
> + gallivm_destroy(variant->gallivm);
> +
> + remove_from_list(&variant->list_item_local);
> + variant->shader->variants_cached--;
> + remove_from_list(&variant->list_item_global);
> + llvm->nr_gs_variants--;
> + FREE(variant);
> +}
> +
> +struct draw_gs_llvm_variant_key *
> +draw_gs_llvm_make_variant_key(struct draw_llvm *llvm, char *store)
> +{
> + unsigned i;
> + struct draw_gs_llvm_variant_key *key;
> + struct draw_sampler_static_state *draw_sampler;
> +
> + key = (struct draw_gs_llvm_variant_key *)store;
> +
> + /* All variants of this shader will have the same value for
> + * nr_samplers. Not yet trying to compact away holes in the
> + * sampler array.
> + */
> + key->nr_samplers = llvm->draw->gs.geometry_shader->info.file_max[TGSI_FILE_SAMPLER] + 1;
> + if (llvm->draw->gs.geometry_shader->info.file_max[TGSI_FILE_SAMPLER_VIEW] != -1) {
> + key->nr_sampler_views =
> + llvm->draw->gs.geometry_shader->info.file_max[TGSI_FILE_SAMPLER_VIEW] + 1;
> + }
> + else {
> + key->nr_sampler_views = key->nr_samplers;
> + }
> +
> + draw_sampler = key->samplers;
> +
> + memset(draw_sampler, 0, MAX2(key->nr_samplers, key->nr_sampler_views) * sizeof *draw_sampler);
> +
> + for (i = 0 ; i< key->nr_samplers; i++) {
> + lp_sampler_static_sampler_state(&draw_sampler[i].sampler_state,
> + llvm->draw->samplers[PIPE_SHADER_GEOMETRY][i]);
> + }
> + for (i = 0 ; i< key->nr_sampler_views; i++) {
> + lp_sampler_static_texture_state(&draw_sampler[i].texture_state,
> + llvm->draw->sampler_views[PIPE_SHADER_GEOMETRY][i]);
> + }
> +
> + return key;
> +}
> +
> +void
> +draw_gs_llvm_dump_variant_key(struct draw_gs_llvm_variant_key *key)
> +{
> +
Remove blank line.
> + unsigned i;
> + struct draw_sampler_static_state *sampler = key->samplers;
> +
> + for (i = 0 ; i< key->nr_sampler_views; i++) {
> + debug_printf("sampler[%i].src_format = %s\n", i,
> + util_format_name(sampler[i].texture_state.format));
> + }
> +}
More information about the mesa-dev
mailing list