#include "compiler/nir_types.h"
+static int
+type_size(const struct glsl_type *type)
+{
+ return glsl_count_attribute_slots(type, false);
+}
+
static void scan_instruction(struct tgsi_shader_info *info,
nir_instr *instr)
{
} else if (instr->type == nir_instr_type_tex) {
nir_tex_instr *tex = nir_instr_as_tex(instr);
+ if (!tex->texture) {
+ info->samplers_declared |=
+ u_bit_consecutive(tex->sampler_index, 1);
+ }
+
switch (tex->op) {
case nir_texop_tex:
case nir_texop_txb:
nir_function *func;
unsigned i;
- assert(nir->stage == MESA_SHADER_VERTEX ||
- nir->stage == MESA_SHADER_FRAGMENT);
+ assert(nir->info.stage == MESA_SHADER_VERTEX ||
+ nir->info.stage == MESA_SHADER_FRAGMENT);
- info->processor = pipe_shader_type_from_mesa(nir->stage);
+ info->processor = pipe_shader_type_from_mesa(nir->info.stage);
info->num_tokens = 2; /* indicate that the shader is non-empty */
info->num_instructions = 2;
- info->num_inputs = nir->num_inputs;
- info->num_outputs = nir->num_outputs;
+ if (nir->info.stage == MESA_SHADER_GEOMETRY) {
+ info->properties[TGSI_PROPERTY_GS_INPUT_PRIM] = nir->info.gs.input_primitive;
+ info->properties[TGSI_PROPERTY_GS_OUTPUT_PRIM] = nir->info.gs.output_primitive;
+ info->properties[TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES] = nir->info.gs.vertices_out;
+ info->properties[TGSI_PROPERTY_GS_INVOCATIONS] = nir->info.gs.invocations;
+ }
i = 0;
+ uint64_t processed_inputs = 0;
+ unsigned num_inputs = 0;
nir_foreach_variable(variable, &nir->inputs) {
unsigned semantic_name, semantic_index;
unsigned attrib_count = glsl_count_attribute_slots(variable->type,
- nir->stage == MESA_SHADER_VERTEX);
+ nir->info.stage == MESA_SHADER_VERTEX);
assert(attrib_count == 1 && "not implemented");
* tracker has already mapped them to attributes via
* variable->data.driver_location.
*/
- if (nir->stage == MESA_SHADER_VERTEX)
+ if (nir->info.stage == MESA_SHADER_VERTEX)
continue;
/* Fragment shader position is a system value. */
- if (nir->stage == MESA_SHADER_FRAGMENT &&
+ if (nir->info.stage == MESA_SHADER_FRAGMENT &&
variable->data.location == VARYING_SLOT_POS) {
if (variable->data.pixel_center_integer)
info->properties[TGSI_PROPERTY_FS_COORD_PIXEL_CENTER] =
TGSI_FS_COORD_PIXEL_CENTER_INTEGER;
+
+ num_inputs++;
continue;
}
+ i = variable->data.driver_location;
+ if (processed_inputs & ((uint64_t)1 << i))
+ continue;
+
+ processed_inputs |= ((uint64_t)1 << i);
+ num_inputs++;
+
tgsi_get_gl_varying_semantic(variable->data.location, true,
&semantic_name, &semantic_index);
info->colors_read |= 0x0f;
else if (variable->data.location == VARYING_SLOT_COL1)
info->colors_read |= 0xf0;
-
- i++;
}
+ if (nir->info.stage != MESA_SHADER_VERTEX)
+ info->num_inputs = num_inputs;
+ else
+ info->num_inputs = nir->num_inputs;
+
i = 0;
+ uint64_t processed_outputs = 0;
+ unsigned num_outputs = 0;
nir_foreach_variable(variable, &nir->outputs) {
unsigned semantic_name, semantic_index;
- if (nir->stage == MESA_SHADER_FRAGMENT) {
+ if (nir->info.stage == MESA_SHADER_FRAGMENT) {
tgsi_get_gl_frag_result_semantic(variable->data.location,
&semantic_name, &semantic_index);
} else {
&semantic_name, &semantic_index);
}
+ i = variable->data.driver_location;
+ if (processed_outputs & ((uint64_t)1 << i))
+ continue;
+
+ processed_outputs |= ((uint64_t)1 << i);
+ num_outputs++;
+
info->output_semantic_name[i] = semantic_name;
info->output_semantic_index[i] = semantic_index;
info->output_usagemask[i] = TGSI_WRITEMASK_XYZW;
+ unsigned num_components = 4;
+ unsigned vector_elements = glsl_get_vector_elements(glsl_without_array(variable->type));
+ if (vector_elements)
+ num_components = vector_elements;
+
+ unsigned gs_out_streams;
+ if (variable->data.stream & (1u << 31)) {
+ gs_out_streams = variable->data.stream & ~(1u << 31);
+ } else {
+ assert(variable->data.stream < 4);
+ gs_out_streams = 0;
+ for (unsigned j = 0; j < num_components; ++j)
+ gs_out_streams |= variable->data.stream << (2 * (variable->data.location_frac + j));
+ }
+
+ unsigned streamx = gs_out_streams & 3;
+ unsigned streamy = (gs_out_streams >> 2) & 3;
+ unsigned streamz = (gs_out_streams >> 4) & 3;
+ unsigned streamw = (gs_out_streams >> 6) & 3;
+
+ if (info->output_usagemask[i] & TGSI_WRITEMASK_X) {
+ info->output_streams[i] |= streamx;
+ info->num_stream_output_components[streamx]++;
+ }
+ if (info->output_usagemask[i] & TGSI_WRITEMASK_Y) {
+ info->output_streams[i] |= streamy << 2;
+ info->num_stream_output_components[streamy]++;
+ }
+ if (info->output_usagemask[i] & TGSI_WRITEMASK_Z) {
+ info->output_streams[i] |= streamz << 4;
+ info->num_stream_output_components[streamz]++;
+ }
+ if (info->output_usagemask[i] & TGSI_WRITEMASK_W) {
+ info->output_streams[i] |= streamw << 6;
+ info->num_stream_output_components[streamw]++;
+ }
+
switch (semantic_name) {
case TGSI_SEMANTIC_PRIMID:
info->writes_primid = true;
info->writes_position = true;
break;
}
-
- i++;
}
+ info->num_outputs = num_outputs;
+
nir_foreach_variable(variable, &nir->uniforms) {
const struct glsl_type *type = variable->type;
enum glsl_base_type base_type =
info->num_written_clipdistance = nir->info.clip_distance_array_size;
info->num_written_culldistance = nir->info.cull_distance_array_size;
info->clipdist_writemask = u_bit_consecutive(0, info->num_written_clipdistance);
- info->culldist_writemask = u_bit_consecutive(info->num_written_clipdistance,
- info->num_written_culldistance);
+ info->culldist_writemask = u_bit_consecutive(0, info->num_written_culldistance);
if (info->processor == PIPE_SHADER_FRAGMENT)
info->uses_kill = nir->info.fs.uses_discard;
}
}
+/**
+ * Perform "lowering" operations on the NIR that are run once when the shader
+ * selector is created.
+ */
+void
+si_lower_nir(struct si_shader_selector* sel)
+{
+ /* Adjust the driver location of inputs and outputs. The state tracker
+ * interprets them as slots, while the ac/nir backend interprets them
+ * as individual components.
+ */
+ nir_foreach_variable(variable, &sel->nir->inputs)
+ variable->data.driver_location *= 4;
+
+ nir_foreach_variable(variable, &sel->nir->outputs) {
+ variable->data.driver_location *= 4;
+
+ if (sel->nir->info.stage == MESA_SHADER_FRAGMENT) {
+ if (variable->data.location == FRAG_RESULT_DEPTH)
+ variable->data.driver_location += 2;
+ else if (variable->data.location == FRAG_RESULT_STENCIL)
+ variable->data.driver_location += 1;
+ }
+ }
+
+ /* Perform lowerings (and optimizations) of code.
+ *
+ * Performance considerations aside, we must:
+ * - lower certain ALU operations
+ * - ensure constant offsets for texture instructions are folded
+ * and copy-propagated
+ */
+ NIR_PASS_V(sel->nir, nir_lower_io, nir_var_uniform, type_size,
+ (nir_lower_io_options)0);
+ NIR_PASS_V(sel->nir, nir_lower_uniforms_to_ubo);
+
+ NIR_PASS_V(sel->nir, nir_lower_returns);
+ NIR_PASS_V(sel->nir, nir_lower_vars_to_ssa);
+ NIR_PASS_V(sel->nir, nir_lower_alu_to_scalar);
+ NIR_PASS_V(sel->nir, nir_lower_phis_to_scalar);
+
+ static const struct nir_lower_tex_options lower_tex_options = {
+ .lower_txp = ~0u,
+ };
+ NIR_PASS_V(sel->nir, nir_lower_tex, &lower_tex_options);
+
+ bool progress;
+ do {
+ progress = false;
+
+ /* (Constant) copy propagation is needed for txf with offsets. */
+ NIR_PASS(progress, sel->nir, nir_copy_prop);
+ NIR_PASS(progress, sel->nir, nir_opt_remove_phis);
+ NIR_PASS(progress, sel->nir, nir_opt_dce);
+ if (nir_opt_trivial_continues(sel->nir)) {
+ progress = true;
+ NIR_PASS(progress, sel->nir, nir_copy_prop);
+ NIR_PASS(progress, sel->nir, nir_opt_dce);
+ }
+ NIR_PASS(progress, sel->nir, nir_opt_if);
+ NIR_PASS(progress, sel->nir, nir_opt_dead_cf);
+ NIR_PASS(progress, sel->nir, nir_opt_cse);
+ NIR_PASS(progress, sel->nir, nir_opt_peephole_select, 8);
+
+ /* Needed for algebraic lowering */
+ NIR_PASS(progress, sel->nir, nir_opt_algebraic);
+ NIR_PASS(progress, sel->nir, nir_opt_constant_folding);
+
+ NIR_PASS(progress, sel->nir, nir_opt_undef);
+ NIR_PASS(progress, sel->nir, nir_opt_conditional_discard);
+ if (sel->nir->options->max_unroll_iterations) {
+ NIR_PASS(progress, sel->nir, nir_opt_loop_unroll, 0);
+ }
+ } while (progress);
+}
+
static void declare_nir_input_vs(struct si_shader_context *ctx,
- struct nir_variable *variable, unsigned rel,
+ struct nir_variable *variable,
LLVMValueRef out[4])
{
- si_llvm_load_input_vs(ctx, variable->data.driver_location / 4 + rel, out);
+ si_llvm_load_input_vs(ctx, variable->data.driver_location / 4, out);
}
static void declare_nir_input_fs(struct si_shader_context *ctx,
- struct nir_variable *variable, unsigned rel,
- unsigned *fs_attr_idx,
+ struct nir_variable *variable,
+ unsigned input_index,
LLVMValueRef out[4])
{
- unsigned slot = variable->data.location + rel;
-
- assert(variable->data.location >= VARYING_SLOT_VAR0 || rel == 0);
-
+ unsigned slot = variable->data.location;
if (slot == VARYING_SLOT_POS) {
out[0] = LLVMGetParam(ctx->main_fn, SI_PARAM_POS_X_FLOAT);
out[1] = LLVMGetParam(ctx->main_fn, SI_PARAM_POS_Y_FLOAT);
return;
}
- si_llvm_load_input_fs(ctx, *fs_attr_idx, out);
- (*fs_attr_idx)++;
+ si_llvm_load_input_fs(ctx, input_index, out);
+}
+
+static LLVMValueRef
+si_nir_load_sampler_desc(struct ac_shader_abi *abi,
+ unsigned descriptor_set, unsigned base_index,
+ unsigned constant_index, LLVMValueRef dynamic_index,
+ enum ac_descriptor_type desc_type, bool image,
+ bool write)
+{
+ struct si_shader_context *ctx = si_shader_context_from_abi(abi);
+ LLVMBuilderRef builder = ctx->ac.builder;
+ LLVMValueRef list = LLVMGetParam(ctx->main_fn, ctx->param_samplers_and_images);
+ LLVMValueRef index = dynamic_index;
+
+ assert(!descriptor_set);
+
+ if (!index)
+ index = ctx->ac.i32_0;
+
+ index = LLVMBuildAdd(builder, index,
+ LLVMConstInt(ctx->ac.i32, base_index + constant_index, false),
+ "");
+
+ if (image) {
+ assert(desc_type == AC_DESC_IMAGE || desc_type == AC_DESC_BUFFER);
+ assert(base_index + constant_index < ctx->num_images);
+
+ if (dynamic_index)
+ index = si_llvm_bound_index(ctx, index, ctx->num_images);
+
+ index = LLVMBuildSub(ctx->gallivm.builder,
+ LLVMConstInt(ctx->i32, SI_NUM_IMAGES - 1, 0),
+ index, "");
+
+ /* TODO: be smarter about when we use dcc_off */
+ return si_load_image_desc(ctx, list, index, desc_type, write);
+ }
+
+ assert(base_index + constant_index < ctx->num_samplers);
+
+ if (dynamic_index)
+ index = si_llvm_bound_index(ctx, index, ctx->num_samplers);
+
+ index = LLVMBuildAdd(ctx->gallivm.builder, index,
+ LLVMConstInt(ctx->i32, SI_NUM_IMAGES / 2, 0), "");
+
+ return si_load_sampler_desc(ctx, list, index, desc_type);
}
bool si_nir_build_llvm(struct si_shader_context *ctx, struct nir_shader *nir)
{
- unsigned fs_attr_idx = 0;
+ struct tgsi_shader_info *info = &ctx->shader->selector->info;
+
+ uint64_t processed_inputs = 0;
nir_foreach_variable(variable, &nir->inputs) {
unsigned attrib_count = glsl_count_attribute_slots(variable->type,
- nir->stage == MESA_SHADER_VERTEX);
+ nir->info.stage == MESA_SHADER_VERTEX);
unsigned input_idx = variable->data.driver_location;
- for (unsigned i = 0; i < attrib_count; ++i) {
- LLVMValueRef data[4];
+ assert(attrib_count == 1);
- if (nir->stage == MESA_SHADER_VERTEX)
- declare_nir_input_vs(ctx, variable, i, data);
- else if (nir->stage == MESA_SHADER_FRAGMENT)
- declare_nir_input_fs(ctx, variable, i, &fs_attr_idx, data);
+ LLVMValueRef data[4];
+ unsigned loc = variable->data.location;
- for (unsigned chan = 0; chan < 4; chan++) {
- ctx->inputs[input_idx + chan] =
- LLVMBuildBitCast(ctx->ac.builder, data[chan], ctx->ac.i32, "");
- }
+ /* Packed components share the same location so skip
+ * them if we have already processed the location.
+ */
+ if (processed_inputs & ((uint64_t)1 << loc))
+ continue;
+
+ if (nir->info.stage == MESA_SHADER_VERTEX)
+ declare_nir_input_vs(ctx, variable, data);
+ else if (nir->info.stage == MESA_SHADER_FRAGMENT)
+ declare_nir_input_fs(ctx, variable, input_idx / 4, data);
+
+ for (unsigned chan = 0; chan < 4; chan++) {
+ ctx->inputs[input_idx + chan] =
+ LLVMBuildBitCast(ctx->ac.builder, data[chan], ctx->ac.i32, "");
}
+ processed_inputs |= ((uint64_t)1 << loc);
}
ctx->abi.inputs = &ctx->inputs[0];
+ ctx->abi.load_sampler_desc = si_nir_load_sampler_desc;
+ ctx->abi.clamp_shadow_reference = true;
+
+ ctx->num_samplers = util_last_bit(info->samplers_declared);
+ ctx->num_images = util_last_bit(info->images_declared);
ac_nir_translate(&ctx->ac, &ctx->abi, nir, NULL);