*/
#include <pthread.h>
-#include "main/imports.h"
#include "main/glspirv.h"
#include "program/prog_parameter.h"
#include "program/prog_print.h"
}
}
-static struct gl_program *brwNewProgram(struct gl_context *ctx, GLenum target,
+static struct gl_program *brwNewProgram(struct gl_context *ctx,
+ gl_shader_stage stage,
GLuint id, bool is_arb_asm);
nir_shader *
nir = _mesa_spirv_to_nir(ctx, shader_prog, stage, options);
} else {
nir = glsl_to_nir(ctx, shader_prog, stage, options);
+
+ /* Remap the locations to slots so those requiring two slots will
+ * occupy two locations. For instance, if we have in the IR code a
+ * dvec3 attr0 in location 0 and vec4 attr1 in location 1, in NIR attr0
+ * will use locations/slots 0 and 1, and attr1 will use location/slot 2
+ */
+ if (nir->info.stage == MESA_SHADER_VERTEX)
+ nir_remap_dual_slot_attributes(nir, &prog->DualSlotInputs);
}
assert (nir);
- nir_remove_dead_variables(nir, nir_var_shader_in | nir_var_shader_out);
+ nir_remove_dead_variables(nir, nir_var_shader_in | nir_var_shader_out,
+ NULL);
nir_validate_shader(nir, "after glsl_to_nir or spirv_to_nir");
NIR_PASS_V(nir, nir_lower_io_to_temporaries,
nir_shader_get_entrypoint(nir), true, false);
}
}
- NIR_PASS_V(nir, brw_nir_lower_uniforms, is_scalar);
-
return nir;
}
+static void
+shared_type_info(const struct glsl_type *type, unsigned *size, unsigned *align)
+{
+ assert(glsl_type_is_vector_or_scalar(type));
+
+ uint32_t comp_size = glsl_type_is_boolean(type)
+ ? 4 : glsl_get_bit_size(type) / 8;
+ unsigned length = glsl_get_vector_elements(type);
+ *size = comp_size * length,
+ *align = comp_size * (length == 3 ? 4 : length);
+}
+
void
brw_nir_lower_resources(nir_shader *nir, struct gl_shader_program *shader_prog,
struct gl_program *prog,
const struct gen_device_info *devinfo)
{
+ NIR_PASS_V(nir, brw_nir_lower_uniforms, nir->options->lower_to_scalar);
NIR_PASS_V(prog->nir, gl_nir_lower_samplers, shader_prog);
prog->info.textures_used = prog->nir->info.textures_used;
prog->info.textures_used_by_txf = prog->nir->info.textures_used_by_txf;
- NIR_PASS_V(prog->nir, brw_nir_lower_image_load_store, devinfo);
+ NIR_PASS_V(prog->nir, brw_nir_lower_image_load_store, devinfo, NULL);
+
+ if (prog->nir->info.stage == MESA_SHADER_COMPUTE &&
+ shader_prog->data->spirv) {
+ NIR_PASS_V(prog->nir, nir_lower_vars_to_explicit_types,
+ nir_var_mem_shared, shared_type_info);
+ NIR_PASS_V(prog->nir, nir_lower_explicit_io,
+ nir_var_mem_shared, nir_address_format_32bit_offset);
+ }
NIR_PASS_V(prog->nir, gl_nir_lower_buffers, shader_prog);
/* Do a round of constant folding to clean up address calculations */
return p_atomic_inc_return(&screen->program_id);
}
-static struct gl_program *brwNewProgram(struct gl_context *ctx, GLenum target,
+static struct gl_program *brwNewProgram(struct gl_context *ctx,
+ gl_shader_stage stage,
GLuint id, bool is_arb_asm)
{
struct brw_context *brw = brw_context(ctx);
if (prog) {
prog->id = get_new_program_id(brw->screen);
- return _mesa_init_gl_program(&prog->program, target, id, is_arb_asm);
+ return _mesa_init_gl_program(&prog->program, stage, id, is_arb_asm);
}
return NULL;
* brw->screen->subslice_total is the TOTAL number of subslices
* and we wish to view that there are 4 subslices per slice
* instead of the actual number of subslices per slice.
+ *
+ * For, ICL, scratch space allocation is based on the number of threads
+ * in the base configuration.
*/
- if (devinfo->gen >= 9 && devinfo->gen < 11)
+ if (devinfo->gen == 11)
+ subslices = 8;
+ else if (devinfo->gen >= 9 && devinfo->gen < 11)
subslices = 4 * brw->screen->devinfo.num_slices;
unsigned scratch_ids_per_subslice;
- if (devinfo->is_haswell) {
+ if (devinfo->gen >= 11) {
+ /* The MEDIA_VFE_STATE docs say:
+ *
+ * "Starting with this configuration, the Maximum Number of
+ * Threads must be set to (#EU * 8) for GPGPU dispatches.
+ *
+ * Although there are only 7 threads per EU in the configuration,
+ * the FFTID is calculated as if there are 8 threads per EU,
+ * which in turn requires a larger amount of Scratch Space to be
+ * allocated by the driver."
+ */
+ scratch_ids_per_subslice = 8 * 8;
+ } else if (devinfo->is_haswell) {
/* WaCSScratchSize:hsw
*
* Haswell's scratch space address calculation appears to be sparse