#include "nir/nir_deref.h"
#include "spirv_info.h"
+#include "util/u_math.h"
+
#include <stdio.h>
void
nir_load_const_instr *load =
nir_load_const_instr_create(b->shader, num_components, bit_size);
- load->value = constant->values[0];
+ memcpy(load->value, constant->values[0],
+ sizeof(nir_const_value) * load->def.num_components);
nir_instr_insert_before_cf_list(&b->nb.impl->body, &load->instr);
val->def = &load->def;
nir_load_const_instr *load =
nir_load_const_instr_create(b->shader, rows, bit_size);
- load->value = constant->values[i];
+ memcpy(load->value, constant->values[i],
+ sizeof(nir_const_value) * load->def.num_components);
nir_instr_insert_before_cf_list(&b->nb.impl->body, &load->instr);
col_val->def = &load->def;
} else if ((strcmp(ext, "SPV_AMD_shader_trinary_minmax") == 0)
&& (b->options && b->options->caps.trinary_minmax)) {
val->ext_handler = vtn_handle_amd_shader_trinary_minmax_instruction;
+ } else if (strcmp(ext, "OpenCL.std") == 0) {
+ val->ext_handler = vtn_handle_opencl_instruction;
} else {
vtn_fail("Unsupported extension: %s", ext);
}
}
default:
- vtn_fail("Unhandled opcode");
+ vtn_fail_with_opcode("Unhandled opcode", opcode);
}
}
break;
case SpvDecorationCPacked:
+ if (b->shader->info.stage != MESA_SHADER_KERNEL)
+ vtn_warn("Decoration only allowed for CL-style kernels: %s",
+ spirv_decoration_to_string(dec->decoration));
+ else
+ ctx->type->packed = true;
+ break;
+
case SpvDecorationSaturatedConversion:
case SpvDecorationFuncParamAttr:
case SpvDecorationFPRoundingMode:
break;
default:
- vtn_fail("Unhandled decoration");
+ vtn_fail_with_decoration("Unhandled decoration", dec->decoration);
}
}
ctx->fields[member].type = ctx->type->members[member]->type;
}
+static void
+struct_block_decoration_cb(struct vtn_builder *b,
+ struct vtn_value *val, int member,
+ const struct vtn_decoration *dec, void *ctx)
+{
+ if (member != -1)
+ return;
+
+ struct vtn_type *type = val->type;
+ if (dec->decoration == SpvDecorationBlock)
+ type->block = true;
+ else if (dec->decoration == SpvDecorationBufferBlock)
+ type->buffer_block = true;
+}
+
static void
type_decoration_cb(struct vtn_builder *b,
struct vtn_value *val, int member,
break;
case SpvDecorationBlock:
vtn_assert(type->base_type == vtn_base_type_struct);
- type->block = true;
+ vtn_assert(type->block);
break;
case SpvDecorationBufferBlock:
vtn_assert(type->base_type == vtn_base_type_struct);
- type->buffer_block = true;
+ vtn_assert(type->buffer_block);
break;
case SpvDecorationGLSLShared:
case SpvDecorationGLSLPacked:
break;
case SpvDecorationCPacked:
+ if (b->shader->info.stage != MESA_SHADER_KERNEL)
+ vtn_warn("Decoration only allowed for CL-style kernels: %s",
+ spirv_decoration_to_string(dec->decoration));
+ else
+ type->packed = true;
+ break;
+
case SpvDecorationSaturatedConversion:
case SpvDecorationFuncParamAttr:
case SpvDecorationFPRoundingMode:
break;
default:
- vtn_fail("Unhandled decoration");
+ vtn_fail_with_decoration("Unhandled decoration", dec->decoration);
}
}
case SpvImageFormatR16ui: return 0x8234; /* GL_R16UI */
case SpvImageFormatR8ui: return 0x8232; /* GL_R8UI */
default:
- vtn_fail("Invalid image format");
+ vtn_fail("Invalid image format: %s (%u)",
+ spirv_imageformat_to_string(format), format);
}
}
val->type->type = (signedness ? glsl_int8_t_type() : glsl_uint8_t_type());
break;
default:
- vtn_fail("Invalid int bit size");
+ vtn_fail("Invalid int bit size: %u", bit_size);
}
val->type->length = 1;
break;
val->type->type = glsl_double_type();
break;
default:
- vtn_fail("Invalid float bit size");
+ vtn_fail("Invalid float bit size: %u", bit_size);
}
val->type->length = 1;
break;
val->type->length = 0;
} else {
val->type->length =
- vtn_value(b, w[3], vtn_value_type_constant)->constant->values[0].u32[0];
+ vtn_value(b, w[3], vtn_value_type_constant)->constant->values[0][0].u32;
}
val->type->base_type = vtn_base_type_array;
val->type->array_element = array_element;
- val->type->stride = 0;
+ if (b->shader->info.stage == MESA_SHADER_KERNEL)
+ val->type->stride = glsl_get_cl_size(array_element->type);
vtn_foreach_decoration(b, val, array_stride_decoration_cb, NULL);
val->type->type = glsl_array_type(array_element->type, val->type->length,
val->type->length = num_fields;
val->type->members = ralloc_array(b, struct vtn_type *, num_fields);
val->type->offsets = ralloc_array(b, unsigned, num_fields);
+ val->type->packed = false;
NIR_VLA(struct glsl_struct_field, fields, count);
for (unsigned i = 0; i < num_fields; i++) {
};
}
+ if (b->shader->info.stage == MESA_SHADER_KERNEL) {
+ unsigned offset = 0;
+ for (unsigned i = 0; i < num_fields; i++) {
+ offset = align(offset, glsl_get_cl_alignment(fields[i].type));
+ fields[i].offset = offset;
+ offset += glsl_get_cl_size(fields[i].type);
+ }
+ }
+
struct member_decoration_ctx ctx = {
.num_fields = num_fields,
.fields = fields,
vtn_foreach_decoration(b, val, struct_member_decoration_cb, &ctx);
vtn_foreach_decoration(b, val, struct_member_matrix_stride_cb, &ctx);
- const char *name = val->name ? val->name : "struct";
+ vtn_foreach_decoration(b, val, struct_block_decoration_cb, NULL);
+
+ const char *name = val->name;
- val->type->type = glsl_struct_type(fields, num_fields, name);
+ if (val->type->block || val->type->buffer_block) {
+ /* Packing will be ignored since types coming from SPIR-V are
+ * explicitly laid out.
+ */
+ val->type->type = glsl_interface_type(fields, num_fields,
+ /* packing */ 0, false,
+ name ? name : "block");
+ } else {
+ val->type->type = glsl_struct_type(fields, num_fields,
+ name ? name : "struct", false);
+ }
break;
}
case SpvStorageClassWorkgroup:
val->type->type = b->options->shared_ptr_type;
break;
+ case SpvStorageClassCrossWorkgroup:
+ val->type->type = b->options->global_ptr_type;
+ break;
+ case SpvStorageClassFunction:
+ if (b->physical_ptrs)
+ val->type->type = b->options->temp_ptr_type;
+ break;
default:
/* In this case, no variable pointers are allowed so all deref
* chains are complete back to the variable and it doesn't matter
vtn_foreach_decoration(b, val, array_stride_decoration_cb, NULL);
+ if (b->physical_ptrs) {
+ switch (storage_class) {
+ case SpvStorageClassFunction:
+ case SpvStorageClassWorkgroup:
+ case SpvStorageClassCrossWorkgroup:
+ val->type->stride = align(glsl_get_cl_size(val->type->deref->type),
+ glsl_get_cl_alignment(val->type->deref->type));
+ break;
+ default:
+ break;
+ }
+ }
+
if (storage_class == SpvStorageClassWorkgroup &&
b->options->lower_workgroup_access_to_offsets) {
uint32_t size, align;
case SpvDimBuffer: dim = GLSL_SAMPLER_DIM_BUF; break;
case SpvDimSubpassData: dim = GLSL_SAMPLER_DIM_SUBPASS; break;
default:
- vtn_fail("Invalid SPIR-V image dimensionality");
+ vtn_fail("Invalid SPIR-V image dimensionality: %s (%u)",
+ spirv_dim_to_string((SpvDim)w[3]), w[3]);
}
- bool is_shadow = w[4];
+ /* w[4]: as per Vulkan spec "Validation Rules within a Module",
+ * The “Depth” operand of OpTypeImage is ignored.
+ */
bool is_array = w[5];
bool multisampled = w[6];
unsigned sampled = w[7];
glsl_get_base_type(sampled_type->type);
if (sampled == 1) {
val->type->sampled = true;
- val->type->type = glsl_sampler_type(dim, is_shadow, is_array,
+ val->type->type = glsl_sampler_type(dim, false, is_array,
sampled_base_type);
} else if (sampled == 2) {
- vtn_assert(!is_shadow);
val->type->sampled = false;
val->type->type = glsl_image_type(dim, is_array, sampled_base_type);
} else {
case SpvOpTypeQueue:
case SpvOpTypePipe:
default:
- vtn_fail("Unhandled opcode");
+ vtn_fail_with_opcode("Unhandled opcode", opcode);
}
vtn_foreach_decoration(b, val, type_decoration_cb, NULL);
opcode == SpvOpSpecConstantFalse)
int_val = get_specialization(b, val, int_val);
- val->constant->values[0].b[0] = int_val != 0;
+ val->constant->values[0][0].b = int_val != 0;
break;
}
int bit_size = glsl_get_bit_size(val->type->type);
switch (bit_size) {
case 64:
- val->constant->values->u64[0] = vtn_u64_literal(&w[3]);
+ val->constant->values[0][0].u64 = vtn_u64_literal(&w[3]);
break;
case 32:
- val->constant->values->u32[0] = w[3];
+ val->constant->values[0][0].u32 = w[3];
break;
case 16:
- val->constant->values->u16[0] = w[3];
+ val->constant->values[0][0].u16 = w[3];
break;
case 8:
- val->constant->values->u8[0] = w[3];
+ val->constant->values[0][0].u8 = w[3];
break;
default:
- vtn_fail("Unsupported SpvOpConstant bit size");
+ vtn_fail("Unsupported SpvOpConstant bit size: %u", bit_size);
}
break;
}
int bit_size = glsl_get_bit_size(val->type->type);
switch (bit_size) {
case 64:
- val->constant->values[0].u64[0] =
+ val->constant->values[0][0].u64 =
get_specialization64(b, val, vtn_u64_literal(&w[3]));
break;
case 32:
- val->constant->values[0].u32[0] = get_specialization(b, val, w[3]);
+ val->constant->values[0][0].u32 = get_specialization(b, val, w[3]);
break;
case 16:
- val->constant->values[0].u16[0] = get_specialization(b, val, w[3]);
+ val->constant->values[0][0].u16 = get_specialization(b, val, w[3]);
break;
case 8:
- val->constant->values[0].u8[0] = get_specialization(b, val, w[3]);
+ val->constant->values[0][0].u8 = get_specialization(b, val, w[3]);
break;
default:
vtn_fail("Unsupported SpvOpSpecConstant bit size");
switch (val->type->base_type) {
case vtn_base_type_vector: {
assert(glsl_type_is_vector(val->type->type));
- int bit_size = glsl_get_bit_size(val->type->type);
- for (unsigned i = 0; i < elem_count; i++) {
- switch (bit_size) {
- case 64:
- val->constant->values[0].u64[i] = elems[i]->values[0].u64[0];
- break;
- case 32:
- val->constant->values[0].u32[i] = elems[i]->values[0].u32[0];
- break;
- case 16:
- val->constant->values[0].u16[i] = elems[i]->values[0].u16[0];
- break;
- case 8:
- val->constant->values[0].u8[i] = elems[i]->values[0].u8[0];
- break;
- case 1:
- val->constant->values[0].b[i] = elems[i]->values[0].b[0];
- break;
- default:
- vtn_fail("Invalid SpvOpConstantComposite bit size");
- }
- }
+ for (unsigned i = 0; i < elem_count; i++)
+ val->constant->values[0][i] = elems[i]->values[0][0];
break;
}
case vtn_base_type_matrix:
assert(glsl_type_is_matrix(val->type->type));
- for (unsigned i = 0; i < elem_count; i++)
- val->constant->values[i] = elems[i]->values[0];
+ for (unsigned i = 0; i < elem_count; i++) {
+ unsigned components =
+ glsl_get_components(glsl_get_column_type(val->type->type));
+ memcpy(val->constant->values[i], elems[i]->values,
+ sizeof(nir_const_value) * components);
+ }
break;
case vtn_base_type_struct:
uint64_t u64[8];
if (v0->value_type == vtn_value_type_constant) {
for (unsigned i = 0; i < len0; i++)
- u64[i] = v0->constant->values[0].u64[i];
+ u64[i] = v0->constant->values[0][i].u64;
}
if (v1->value_type == vtn_value_type_constant) {
for (unsigned i = 0; i < len1; i++)
- u64[len0 + i] = v1->constant->values[0].u64[i];
+ u64[len0 + i] = v1->constant->values[0][i].u64;
}
for (unsigned i = 0, j = 0; i < count - 6; i++, j++) {
* to detect if it is wrongly used.
*/
if (comp == (uint32_t)-1)
- val->constant->values[0].u64[j] = 0xdeadbeefdeadbeef;
+ val->constant->values[0][j].u64 = 0xdeadbeefdeadbeef;
else
- val->constant->values[0].u64[j] = u64[comp];
+ val->constant->values[0][j].u64 = u64[comp];
}
} else {
/* This is for both 32-bit and 16-bit values */
uint32_t u32[8];
if (v0->value_type == vtn_value_type_constant) {
for (unsigned i = 0; i < len0; i++)
- u32[i] = v0->constant->values[0].u32[i];
+ u32[i] = v0->constant->values[0][i].u32;
}
if (v1->value_type == vtn_value_type_constant) {
for (unsigned i = 0; i < len1; i++)
- u32[len0 + i] = v1->constant->values[0].u32[i];
+ u32[len0 + i] = v1->constant->values[0][i].u32;
}
for (unsigned i = 0, j = 0; i < count - 6; i++, j++) {
* to detect if it is wrongly used.
*/
if (comp == (uint32_t)-1)
- val->constant->values[0].u32[j] = 0xdeadbeef;
+ val->constant->values[0][j].u32 = 0xdeadbeef;
else
- val->constant->values[0].u32[j] = u32[comp];
+ val->constant->values[0][j].u32 = u32[comp];
}
}
break;
val->constant = *c;
} else {
unsigned num_components = type->length;
- unsigned bit_size = glsl_get_bit_size(type->type);
for (unsigned i = 0; i < num_components; i++)
- switch(bit_size) {
- case 64:
- val->constant->values[0].u64[i] = (*c)->values[col].u64[elem + i];
- break;
- case 32:
- val->constant->values[0].u32[i] = (*c)->values[col].u32[elem + i];
- break;
- case 16:
- val->constant->values[0].u16[i] = (*c)->values[col].u16[elem + i];
- break;
- case 8:
- val->constant->values[0].u8[i] = (*c)->values[col].u8[elem + i];
- break;
- case 1:
- val->constant->values[0].b[i] = (*c)->values[col].b[elem + i];
- break;
- default:
- vtn_fail("Invalid SpvOpCompositeExtract bit size");
- }
+ val->constant->values[0][i] = (*c)->values[col][elem + i];
}
} else {
struct vtn_value *insert =
*c = insert->constant;
} else {
unsigned num_components = type->length;
- unsigned bit_size = glsl_get_bit_size(type->type);
for (unsigned i = 0; i < num_components; i++)
- switch (bit_size) {
- case 64:
- (*c)->values[col].u64[elem + i] = insert->constant->values[0].u64[i];
- break;
- case 32:
- (*c)->values[col].u32[elem + i] = insert->constant->values[0].u32[i];
- break;
- case 16:
- (*c)->values[col].u16[elem + i] = insert->constant->values[0].u16[i];
- break;
- case 8:
- (*c)->values[col].u8[elem + i] = insert->constant->values[0].u8[i];
- break;
- case 1:
- (*c)->values[col].b[elem + i] = insert->constant->values[0].b[i];
- break;
- default:
- vtn_fail("Invalid SpvOpCompositeInsert bit size");
- }
+ (*c)->values[col][elem + i] = insert->constant->values[0][i];
}
}
break;
nir_op op = vtn_nir_alu_op_for_spirv_opcode(b, opcode, &swap,
nir_alu_type_get_type_size(src_alu_type),
nir_alu_type_get_type_size(dst_alu_type));
- nir_const_value src[4];
+ nir_const_value src[3][NIR_MAX_VEC_COMPONENTS];
for (unsigned i = 0; i < count - 4; i++) {
struct vtn_value *src_val =
bit_size = glsl_get_bit_size(src_val->type->type);
unsigned j = swap ? 1 - i : i;
- src[j] = src_val->constant->values[0];
+ memcpy(src[j], src_val->constant->values[0], sizeof(src[j]));
}
/* fix up fixed size sources */
break;
for (unsigned i = 0; i < num_components; ++i) {
switch (bit_size) {
- case 64: src[1].u32[i] = src[1].u64[i]; break;
- case 16: src[1].u32[i] = src[1].u16[i]; break;
- case 8: src[1].u32[i] = src[1].u8[i]; break;
+ case 64: src[1][i].u32 = src[1][i].u64; break;
+ case 16: src[1][i].u32 = src[1][i].u16; break;
+ case 8: src[1][i].u32 = src[1][i].u8; break;
}
}
break;
break;
}
- val->constant->values[0] =
- nir_eval_const_opcode(op, num_components, bit_size, src);
+ nir_const_value *srcs[3] = {
+ src[0], src[1], src[2],
+ };
+ nir_eval_const_opcode(op, val->constant->values[0], num_components, bit_size, srcs);
break;
} /* default */
}
break;
default:
- vtn_fail("Unhandled opcode");
+ vtn_fail_with_opcode("Unhandled opcode", opcode);
}
/* Now that we have the value, update the workgroup size if needed */
child_type = glsl_get_array_element(type);
break;
case GLSL_TYPE_STRUCT:
+ case GLSL_TYPE_INTERFACE:
child_type = glsl_get_struct_field(type, i);
break;
default:
break;
default:
- vtn_fail("Unhandled opcode");
+ vtn_fail_with_opcode("Unhandled opcode", opcode);
}
nir_tex_src srcs[10]; /* 10 should be enough */
case nir_texop_txl:
case nir_texop_txd:
case nir_texop_tg4:
+ case nir_texop_lod:
/* These operations require a sampler */
p->src = nir_src_for_ssa(&sampler->dest.ssa);
p->src_type = nir_tex_src_sampler_deref;
case nir_texop_txf:
case nir_texop_txf_ms:
case nir_texop_txs:
- case nir_texop_lod:
case nir_texop_query_levels:
case nir_texop_texture_samples:
case nir_texop_samples_identical:
case SpvOpImageGather:
/* This has a component as its next source */
gather_component =
- vtn_value(b, w[idx++], vtn_value_type_constant)->constant->values[0].u32[0];
+ vtn_value(b, w[idx++], vtn_value_type_constant)->constant->values[0][0].u32;
break;
default:
(*p++) = vtn_tex_src(b, w[idx++], nir_tex_src_lod);
/* Now we need to handle some number of optional arguments */
- const struct vtn_ssa_value *gather_offsets = NULL;
+ struct vtn_value *gather_offsets = NULL;
if (idx < count) {
uint32_t operands = w[idx++];
(*p++) = vtn_tex_src(b, w[idx++], nir_tex_src_offset);
if (operands & SpvImageOperandsConstOffsetsMask) {
- nir_tex_src none = {0};
- gather_offsets = vtn_ssa_value(b, w[idx++]);
- (*p++) = none;
+ vtn_assert(texop == nir_texop_tg4);
+ gather_offsets = vtn_value(b, w[idx++], vtn_value_type_constant);
}
if (operands & SpvImageOperandsSampleMask) {
is_shadow && glsl_get_components(ret_type->type) == 1;
instr->component = gather_component;
+ if (sampled.image && (sampled.image->access & ACCESS_NON_UNIFORM))
+ instr->texture_non_uniform = true;
+
+ if (sampled.sampler && (sampled.sampler->access & ACCESS_NON_UNIFORM))
+ instr->sampler_non_uniform = true;
+
switch (glsl_get_sampler_result_type(image_type)) {
case GLSL_TYPE_FLOAT: instr->dest_type = nir_type_float; break;
case GLSL_TYPE_INT: instr->dest_type = nir_type_int; break;
vtn_assert(glsl_get_vector_elements(ret_type->type) ==
nir_tex_instr_dest_size(instr));
- nir_ssa_def *def;
- nir_instr *instruction;
if (gather_offsets) {
- vtn_assert(glsl_get_base_type(gather_offsets->type) == GLSL_TYPE_ARRAY);
- vtn_assert(glsl_get_length(gather_offsets->type) == 4);
- nir_tex_instr *instrs[4] = {instr, NULL, NULL, NULL};
-
- /* Copy the current instruction 4x */
- for (uint32_t i = 1; i < 4; i++) {
- instrs[i] = nir_tex_instr_create(b->shader, instr->num_srcs);
- instrs[i]->op = instr->op;
- instrs[i]->coord_components = instr->coord_components;
- instrs[i]->sampler_dim = instr->sampler_dim;
- instrs[i]->is_array = instr->is_array;
- instrs[i]->is_shadow = instr->is_shadow;
- instrs[i]->is_new_style_shadow = instr->is_new_style_shadow;
- instrs[i]->component = instr->component;
- instrs[i]->dest_type = instr->dest_type;
-
- memcpy(instrs[i]->src, srcs, instr->num_srcs * sizeof(*instr->src));
-
- nir_ssa_dest_init(&instrs[i]->instr, &instrs[i]->dest,
- nir_tex_instr_dest_size(instr), 32, NULL);
- }
-
- /* Fill in the last argument with the offset from the passed in offsets
- * and insert the instruction into the stream.
- */
- for (uint32_t i = 0; i < 4; i++) {
- nir_tex_src src;
- src.src = nir_src_for_ssa(gather_offsets->elems[i]->def);
- src.src_type = nir_tex_src_offset;
- instrs[i]->src[instrs[i]->num_srcs - 1] = src;
- nir_builder_instr_insert(&b->nb, &instrs[i]->instr);
- }
-
- /* Combine the results of the 4 instructions by taking their .w
- * components
- */
- nir_alu_instr *vec4 = nir_alu_instr_create(b->shader, nir_op_vec4);
- nir_ssa_dest_init(&vec4->instr, &vec4->dest.dest, 4, 32, NULL);
- vec4->dest.write_mask = 0xf;
+ vtn_fail_if(gather_offsets->type->base_type != vtn_base_type_array ||
+ gather_offsets->type->length != 4,
+ "ConstOffsets must be an array of size four of vectors "
+ "of two integer components");
+
+ struct vtn_type *vec_type = gather_offsets->type->array_element;
+ vtn_fail_if(vec_type->base_type != vtn_base_type_vector ||
+ vec_type->length != 2 ||
+ !glsl_type_is_integer(vec_type->type),
+ "ConstOffsets must be an array of size four of vectors "
+ "of two integer components");
+
+ unsigned bit_size = glsl_get_bit_size(vec_type->type);
for (uint32_t i = 0; i < 4; i++) {
- vec4->src[i].src = nir_src_for_ssa(&instrs[i]->dest.ssa);
- vec4->src[i].swizzle[0] = 3;
+ const nir_const_value *cvec =
+ gather_offsets->constant->elements[i]->values[0];
+ for (uint32_t j = 0; j < 2; j++) {
+ switch (bit_size) {
+ case 8: instr->tg4_offsets[i][j] = cvec[j].i8; break;
+ case 16: instr->tg4_offsets[i][j] = cvec[j].i16; break;
+ case 32: instr->tg4_offsets[i][j] = cvec[j].i32; break;
+ case 64: instr->tg4_offsets[i][j] = cvec[j].i64; break;
+ default:
+ vtn_fail("Unsupported bit size: %u", bit_size);
+ }
+ }
}
- def = &vec4->dest.dest.ssa;
- instruction = &vec4->instr;
- } else {
- def = &instr->dest.ssa;
- instruction = &instr->instr;
}
val->ssa = vtn_create_ssa_value(b, ret_type->type);
- val->ssa->def = def;
+ val->ssa->def = &instr->dest.ssa;
- nir_builder_instr_insert(&b->nb, instruction);
+ nir_builder_instr_insert(&b->nb, &instr->instr);
}
static void
break;
default:
- vtn_fail("Invalid SPIR-V atomic");
+ vtn_fail_with_opcode("Invalid SPIR-V atomic", opcode);
}
}
break;
default:
- vtn_fail("Invalid image opcode");
+ vtn_fail_with_opcode("Invalid image opcode", opcode);
}
nir_intrinsic_op op;
OP(AtomicXor, atomic_xor)
#undef OP
default:
- vtn_fail("Invalid image opcode");
+ vtn_fail_with_opcode("Invalid image opcode", opcode);
}
nir_intrinsic_instr *intrin = nir_intrinsic_instr_create(b->shader, op);
break;
default:
- vtn_fail("Invalid image opcode");
+ vtn_fail_with_opcode("Invalid image opcode", opcode);
}
if (opcode != SpvOpImageWrite && opcode != SpvOpAtomicStore) {
OP(AtomicXor, atomic_xor)
#undef OP
default:
- vtn_fail("Invalid SSBO atomic");
+ vtn_fail_with_opcode("Invalid SSBO atomic", opcode);
}
}
OP(AtomicXor, atomic_xor)
#undef OP
default:
- vtn_fail("Invalid shared atomic");
+ vtn_fail_with_opcode("Invalid shared atomic", opcode);
}
}
OP(AtomicXor, atomic_xor)
#undef OP
default:
- vtn_fail("Invalid shared atomic");
+ vtn_fail_with_opcode("Invalid shared atomic", opcode);
}
}
break;
default:
- vtn_fail("Invalid SPIR-V atomic");
+ vtn_fail_with_opcode("Invalid SPIR-V atomic", opcode);
}
/*
break;
default:
- vtn_fail("Invalid SPIR-V atomic");
+ vtn_fail_with_opcode("Invalid SPIR-V atomic", opcode);
}
} else {
nir_deref_instr *deref = vtn_pointer_to_deref(b, ptr);
break;
default:
- vtn_fail("Invalid SPIR-V atomic");
+ vtn_fail_with_opcode("Invalid SPIR-V atomic", opcode);
}
}
case 2: op = nir_op_vec2; break;
case 3: op = nir_op_vec3; break;
case 4: op = nir_op_vec4; break;
- default: vtn_fail("bad vector size");
+ default: vtn_fail("bad vector size: %u", num_components);
}
nir_alu_instr *vec = nir_alu_instr_create(b->shader, op);
vtn_vector_extract_dynamic(struct vtn_builder *b, nir_ssa_def *src,
nir_ssa_def *index)
{
- nir_ssa_def *dest = vtn_vector_extract(b, src, 0);
- for (unsigned i = 1; i < src->num_components; i++)
- dest = nir_bcsel(&b->nb, nir_ieq_imm(&b->nb, index, i),
- vtn_vector_extract(b, src, i), dest);
-
- return dest;
+ return nir_vector_extract(&b->nb, src, nir_i2i(&b->nb, index, 32));
}
nir_ssa_def *
break;
default:
- vtn_fail("unknown composite operation");
+ vtn_fail_with_opcode("unknown composite operation", opcode);
}
}
case SpvExecutionModeOutputTriangleStrip:
return 5; /* GL_TRIANGLE_STRIP */
default:
- vtn_fail("Invalid primitive type");
+ vtn_fail("Invalid primitive type: %s (%u)",
+ spirv_executionmode_to_string(mode), mode);
}
}
case SpvExecutionModeInputTrianglesAdjacency:
return 6;
default:
- vtn_fail("Invalid GS input mode");
+ vtn_fail("Invalid GS input mode: %s (%u)",
+ spirv_executionmode_to_string(mode), mode);
}
}
case SpvExecutionModelKernel:
return MESA_SHADER_KERNEL;
default:
- vtn_fail("Unsupported execution model");
+ vtn_fail("Unsupported execution model: %s (%u)",
+ spirv_executionmodel_to_string(model), model);
}
}
-#define spv_check_supported(name, cap) do { \
- if (!(b->options && b->options->caps.name)) \
- vtn_warn("Unsupported SPIR-V capability: %s", \
- spirv_capability_to_string(cap)); \
+#define spv_check_supported(name, cap) do { \
+ if (!(b->options && b->options->caps.name)) \
+ vtn_warn("Unsupported SPIR-V capability: %s (%u)", \
+ spirv_capability_to_string(cap), cap); \
} while(0)
case SpvCapabilityLinkage:
case SpvCapabilityVector16:
case SpvCapabilityFloat16Buffer:
- case SpvCapabilityFloat16:
case SpvCapabilitySparseResidency:
vtn_warn("Unsupported SPIR-V capability: %s",
spirv_capability_to_string(cap));
spv_check_supported(physical_storage_buffer_address, cap);
break;
+ case SpvCapabilityComputeDerivativeGroupQuadsNV:
+ case SpvCapabilityComputeDerivativeGroupLinearNV:
+ spv_check_supported(derivative_group, cap);
+ break;
+
+ case SpvCapabilityFloat16:
+ spv_check_supported(float16, cap);
+ break;
+
default:
- vtn_fail("Unhandled capability");
+ vtn_fail("Unhandled capability: %s (%u)",
+ spirv_capability_to_string(cap), cap);
}
break;
}
break;
case SpvOpMemoryModel:
- vtn_assert(w[1] == SpvAddressingModelLogical ||
- (b->options &&
- b->options->caps.physical_storage_buffer_address &&
- w[1] == SpvAddressingModelPhysicalStorageBuffer64EXT));
+ switch (w[1]) {
+ case SpvAddressingModelPhysical32:
+ vtn_fail_if(b->shader->info.stage != MESA_SHADER_KERNEL,
+ "AddressingModelPhysical32 only supported for kernels");
+ b->shader->info.cs.ptr_size = 32;
+ b->physical_ptrs = true;
+ b->options->shared_ptr_type = glsl_uint_type();
+ b->options->global_ptr_type = glsl_uint_type();
+ b->options->temp_ptr_type = glsl_uint_type();
+ break;
+ case SpvAddressingModelPhysical64:
+ vtn_fail_if(b->shader->info.stage != MESA_SHADER_KERNEL,
+ "AddressingModelPhysical64 only supported for kernels");
+ b->shader->info.cs.ptr_size = 64;
+ b->physical_ptrs = true;
+ b->options->shared_ptr_type = glsl_uint64_t_type();
+ b->options->global_ptr_type = glsl_uint64_t_type();
+ b->options->temp_ptr_type = glsl_uint64_t_type();
+ break;
+ case SpvAddressingModelLogical:
+ vtn_fail_if(b->shader->info.stage >= MESA_SHADER_STAGES,
+ "AddressingModelLogical only supported for shaders");
+ b->shader->info.cs.ptr_size = 0;
+ b->physical_ptrs = false;
+ break;
+ case SpvAddressingModelPhysicalStorageBuffer64EXT:
+ vtn_fail_if(!b->options ||
+ !b->options->caps.physical_storage_buffer_address,
+ "AddressingModelPhysicalStorageBuffer64EXT not supported");
+ break;
+ default:
+ vtn_fail("Unknown addressing model: %s (%u)",
+ spirv_addressingmodel_to_string(w[1]), w[1]);
+ break;
+ }
+
vtn_assert(w[2] == SpvMemoryModelSimple ||
- w[2] == SpvMemoryModelGLSL450);
+ w[2] == SpvMemoryModelGLSL450 ||
+ w[2] == SpvMemoryModelOpenCL);
break;
case SpvOpEntryPoint:
vtn_assert(b->shader->info.stage == MESA_SHADER_FRAGMENT);
break;
+ case SpvExecutionModeDerivativeGroupQuadsNV:
+ vtn_assert(b->shader->info.stage == MESA_SHADER_COMPUTE);
+ b->shader->info.cs.derivative_group = DERIVATIVE_GROUP_QUADS;
+ break;
+
+ case SpvExecutionModeDerivativeGroupLinearNV:
+ vtn_assert(b->shader->info.stage == MESA_SHADER_COMPUTE);
+ b->shader->info.cs.derivative_group = DERIVATIVE_GROUP_LINEAR;
+ break;
+
default:
- vtn_fail("Unhandled execution mode");
+ vtn_fail("Unhandled execution mode: %s (%u)",
+ spirv_executionmode_to_string(mode->exec_mode),
+ mode->exec_mode);
}
}
case SpvOpAccessChain:
case SpvOpPtrAccessChain:
case SpvOpInBoundsAccessChain:
+ case SpvOpInBoundsPtrAccessChain:
case SpvOpArrayLength:
case SpvOpConvertPtrToU:
case SpvOpConvertUToPtr:
break;
default:
- vtn_fail("Unhandled opcode");
+ vtn_fail_with_opcode("Unhandled opcode", opcode);
}
return true;
{
/* Initialize the vtn_builder object */
struct vtn_builder *b = rzalloc(NULL, struct vtn_builder);
+ struct spirv_to_nir_options *dup_options =
+ ralloc(b, struct spirv_to_nir_options);
+ *dup_options = *options;
+
b->spirv = words;
b->spirv_word_count = word_count;
b->file = NULL;
exec_list_make_empty(&b->functions);
b->entry_point_stage = stage;
b->entry_point_name = entry_point_name;
- b->options = options;
+ b->options = dup_options;
/*
* Handle the SPIR-V header (first 5 dwords).
return NULL;
}
+static nir_function *
+vtn_emit_kernel_entry_point_wrapper(struct vtn_builder *b,
+ nir_function *entry_point)
+{
+ vtn_assert(entry_point == b->entry_point->func->impl->function);
+ vtn_fail_if(!entry_point->name, "entry points are required to have a name");
+ const char *func_name =
+ ralloc_asprintf(b->shader, "__wrapped_%s", entry_point->name);
+
+ /* we shouldn't have any inputs yet */
+ vtn_assert(!entry_point->shader->num_inputs);
+ vtn_assert(b->shader->info.stage == MESA_SHADER_KERNEL);
+
+ nir_function *main_entry_point = nir_function_create(b->shader, func_name);
+ main_entry_point->impl = nir_function_impl_create(main_entry_point);
+ nir_builder_init(&b->nb, main_entry_point->impl);
+ b->nb.cursor = nir_after_cf_list(&main_entry_point->impl->body);
+ b->func_param_idx = 0;
+
+ nir_call_instr *call = nir_call_instr_create(b->nb.shader, entry_point);
+
+ for (unsigned i = 0; i < entry_point->num_params; ++i) {
+ struct vtn_type *param_type = b->entry_point->func->type->params[i];
+
+ /* consider all pointers to function memory to be parameters passed
+ * by value
+ */
+ bool is_by_val = param_type->base_type == vtn_base_type_pointer &&
+ param_type->storage_class == SpvStorageClassFunction;
+
+ /* input variable */
+ nir_variable *in_var = rzalloc(b->nb.shader, nir_variable);
+ in_var->data.mode = nir_var_shader_in;
+ in_var->data.read_only = true;
+ in_var->data.location = i;
+
+ if (is_by_val)
+ in_var->type = param_type->deref->type;
+ else
+ in_var->type = param_type->type;
+
+ nir_shader_add_variable(b->nb.shader, in_var);
+ b->nb.shader->num_inputs++;
+
+ /* we have to copy the entire variable into function memory */
+ if (is_by_val) {
+ nir_variable *copy_var =
+ nir_local_variable_create(main_entry_point->impl, in_var->type,
+ "copy_in");
+ nir_copy_var(&b->nb, copy_var, in_var);
+ call->params[i] =
+ nir_src_for_ssa(&nir_build_deref_var(&b->nb, copy_var)->dest.ssa);
+ } else {
+ call->params[i] = nir_src_for_ssa(nir_load_var(&b->nb, in_var));
+ }
+ }
+
+ nir_builder_instr_insert(&b->nb, &call->instr);
+
+ return main_entry_point;
+}
+
nir_function *
spirv_to_nir(const uint32_t *words, size_t word_count,
struct nir_spirv_specialization *spec, unsigned num_spec,
/* Skip the SPIR-V header, handled at vtn_create_builder */
words+= 5;
+ b->shader = nir_shader_create(b, stage, nir_options, NULL);
+
/* Handle all the preamble instructions */
words = vtn_foreach_instruction(b, words, word_end,
vtn_handle_preamble_instruction);
return NULL;
}
- b->shader = nir_shader_create(b, stage, nir_options, NULL);
-
/* Set shader info defaults */
b->shader->info.gs.invocations = 1;
glsl_vector_type(GLSL_TYPE_UINT, 3));
nir_const_value *const_size =
- &b->workgroup_size_builtin->constant->values[0];
+ b->workgroup_size_builtin->constant->values[0];
- b->shader->info.cs.local_size[0] = const_size->u32[0];
- b->shader->info.cs.local_size[1] = const_size->u32[1];
- b->shader->info.cs.local_size[2] = const_size->u32[2];
+ b->shader->info.cs.local_size[0] = const_size[0].u32;
+ b->shader->info.cs.local_size[1] = const_size[1].u32;
+ b->shader->info.cs.local_size[2] = const_size[2].u32;
}
/* Set types on all vtn_values */
nir_function *entry_point = b->entry_point->func->impl->function;
vtn_assert(entry_point);
+ /* post process entry_points with input params */
+ if (entry_point->num_params && b->shader->info.stage == MESA_SHADER_KERNEL)
+ entry_point = vtn_emit_kernel_entry_point_wrapper(b, entry_point);
+
entry_point->is_entrypoint = true;
/* When multiple shader stages exist in the same SPIR-V module, we
projects / mesa.git / blobdiff