<div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">On Wed, Feb 28, 2018 at 11:51 AM, Rob Clark <<a href="mailto:robdclark@gmail.com" target="_blank">robdclark@gmail.com</a>> wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">From: Karol Herbst <<a href="mailto:kherbst@redhat.com">kherbst@redhat.com</a>> OpenCL kernels also have int8/uint8. Signed-off-by: Rob Clark <<a href="mailto:robdclark@gmail.com">robdclark@gmail.com</a>> --- src/compiler/builtin_type_macros.h | 10 ++++++++ src/compiler/glsl/ast_to_hir.cpp | 2 ++ src/compiler/glsl/ir_clone.cpp | 2 ++ src/compiler/glsl/link_uniform_initializers.cpp | 2 ++ src/compiler/glsl_types.cpp | 33 +++++++++++++++++++++++++ src/compiler/glsl_types.h | 4 +++ src/compiler/nir/nir.h | 4 +++ src/compiler/nir/nir_search.c | 2 ++ src/compiler/nir_types.cpp | 12 +++++++++ src/compiler/nir_types.h | 6 +++++ src/compiler/spirv/spirv_to_nir.c | 24 ++++++++++++++++++ src/compiler/spirv/vtn_variables.c | 14 +++++++++++ src/intel/compiler/brw_fs.cpp | 3 +++ src/intel/compiler/brw_shader.cpp | 4 +++ src/intel/compiler/brw_vec4_visitor.cpp | 2 ++ src/mesa/program/ir_to_mesa.cpp | 4 +++ src/mesa/state_tracker/st_glsl_types.cpp | 2 ++ 17 files changed, 130 insertions(+) diff --git a/src/compiler/builtin_type_macros.h b/src/compiler/builtin_type_macros.h index e3a1cd29c87..807691824d3 100644 --- a/src/compiler/builtin_type_macros.h +++ b/src/compiler/builtin_type_macros.h @@ -114,6 +114,16 @@ DECL_TYPE(u16vec2, GL_UNSIGNED_INT16_VEC2_NV, GLSL_TYPE_UINT16, 2, 1) DECL_TYPE(u16vec3, GL_UNSIGNED_INT16_VEC3_NV, GLSL_TYPE_UINT16, 3, 1) DECL_TYPE(u16vec4, GL_UNSIGNED_INT16_VEC4_NV, GLSL_TYPE_UINT16, 4, 1) +DECL_TYPE(int8_t, GL_INT8_NV, GLSL_TYPE_INT8, 1, 1) +DECL_TYPE(i8vec2, GL_INT8_VEC2_NV, GLSL_TYPE_INT8, 2, 1) +DECL_TYPE(i8vec3, GL_INT8_VEC3_NV, GLSL_TYPE_INT8, 3, 1) +DECL_TYPE(i8vec4, GL_INT8_VEC4_NV, GLSL_TYPE_INT8, 4, 1) + +DECL_TYPE(uint8_t, GL_UNSIGNED_INT8_NV, GLSL_TYPE_UINT8, 1, 1) +DECL_TYPE(u8vec2, GL_UNSIGNED_INT8_VEC2_NV, GLSL_TYPE_UINT8, 2, 1) +DECL_TYPE(u8vec3, GL_UNSIGNED_INT8_VEC3_NV, GLSL_TYPE_UINT8, 3, 1) +DECL_TYPE(u8vec4, GL_UNSIGNED_INT8_VEC4_NV, GLSL_TYPE_UINT8, 4, 1) + DECL_TYPE(sampler, GL_SAMPLER_1D, GLSL_TYPE_SAMPLER, GLSL_SAMPLER_DIM_1D, 0, 0, GLSL_TYPE_VOID) DECL_TYPE(sampler1D, GL_SAMPLER_1D, GLSL_TYPE_SAMPLER, GLSL_SAMPLER_DIM_1D, 0, 0, GLSL_TYPE_FLOAT) DECL_TYPE(sampler2D, GL_SAMPLER_2D, GLSL_TYPE_SAMPLER, GLSL_SAMPLER_DIM_2D, 0, 0, GLSL_TYPE_FLOAT) diff --git a/src/compiler/glsl/ast_to_hir.cpp b/src/compiler/glsl/ast_to_hir.cpp index 41e74815f31..815e571097c 100644 --- a/src/compiler/glsl/ast_to_hir.cpp +++ b/src/compiler/glsl/ast_to_hir.cpp @@ -1117,6 +1117,8 @@ do_comparison(void *mem_ctx, int operation, ir_rvalue *op0, ir_rvalue *op1) case GLSL_TYPE_INT64: case GLSL_TYPE_UINT16: case GLSL_TYPE_INT16: + case GLSL_TYPE_UINT8: + case GLSL_TYPE_INT8: return new(mem_ctx) ir_expression(operation, op0, op1); case GLSL_TYPE_ARRAY: { diff --git a/src/compiler/glsl/ir_clone.cpp b/src/compiler/glsl/ir_clone.cpp index f088b6aebd5..69441fae7de 100644 --- a/src/compiler/glsl/ir_clone.cpp +++ b/src/compiler/glsl/ir_clone.cpp @@ -344,6 +344,8 @@ ir_constant::clone(void *mem_ctx, struct hash_table *ht) const case GLSL_TYPE_INT64: case GLSL_TYPE_UINT16: case GLSL_TYPE_INT16: + case GLSL_TYPE_UINT8: + case GLSL_TYPE_INT8: case GLSL_TYPE_SAMPLER: case GLSL_TYPE_IMAGE: return new(mem_ctx) ir_constant(this->type, &this->value); diff --git a/src/compiler/glsl/link_uniform_initializers.cpp b/src/compiler/glsl/link_uniform_initializers.cpp index 35522f76467..d6d63bd61fc 100644 --- a/src/compiler/glsl/link_uniform_initializers.cpp +++ b/src/compiler/glsl/link_uniform_initializers.cpp @@ -83,6 +83,8 @@ copy_constant_to_storage(union gl_constant_value *storage, case GLSL_TYPE_ERROR: case GLSL_TYPE_UINT16: case GLSL_TYPE_INT16: + case GLSL_TYPE_UINT8: + case GLSL_TYPE_INT8: case GLSL_TYPE_FLOAT16: /* All other types should have already been filtered by other * paths in the caller. diff --git a/src/compiler/glsl_types.cpp b/src/compiler/glsl_types.cpp index 3cc5eb0495c..b1438300746 100644 --- a/src/compiler/glsl_types.cpp +++ b/src/compiler/glsl_types.cpp @@ -623,6 +623,31 @@ glsl_type::u16vec(unsigned components) return ts[components - 1]; } +const glsl_type * +glsl_type::i8vec(unsigned components) +{ + if (components == 0 || components > 4) + return error_type; + + static const glsl_type *const ts[] = { + int8_t_type, i8vec2_type, i8vec3_type, i8vec4_type + }; + return ts[components - 1]; +} + + +const glsl_type * +glsl_type::u8vec(unsigned components) +{ + if (components == 0 || components > 4) + return error_type; + + static const glsl_type *const ts[] = { + uint8_t_type, u8vec2_type, u8vec3_type, u8vec4_type + }; + return ts[components - 1]; +} + const glsl_type * glsl_type::get_instance(unsigned base_type, unsigned rows, unsigned columns) { @@ -656,6 +681,10 @@ glsl_type::get_instance(unsigned base_type, unsigned rows, unsigned columns) return u16vec(rows); case GLSL_TYPE_INT16: return i16vec(rows); + case GLSL_TYPE_UINT8: + return u8vec(rows); + case GLSL_TYPE_INT8: + return i8vec(rows); default: return error_type; } @@ -1357,6 +1386,8 @@ glsl_type::component_slots() const switch (this->base_type) { case GLSL_TYPE_UINT: case GLSL_TYPE_INT: + case GLSL_TYPE_UINT8: + case GLSL_TYPE_INT8: case GLSL_TYPE_UINT16: case GLSL_TYPE_INT16: case GLSL_TYPE_FLOAT: @@ -2058,6 +2089,8 @@ glsl_type::count_attribute_slots(bool is_vertex_input) const switch (this->base_type) { case GLSL_TYPE_UINT: case GLSL_TYPE_INT: + case GLSL_TYPE_UINT8: + case GLSL_TYPE_INT8: case GLSL_TYPE_UINT16: case GLSL_TYPE_INT16: case GLSL_TYPE_FLOAT: diff --git a/src/compiler/glsl_types.h b/src/compiler/glsl_types.h index ab0b2637649..224f754c0dd 100644 --- a/src/compiler/glsl_types.h +++ b/src/compiler/glsl_types.h @@ -63,6 +63,8 @@ enum glsl_base_type { GLSL_TYPE_FLOAT, GLSL_TYPE_FLOAT16, GLSL_TYPE_DOUBLE, + GLSL_TYPE_UINT8, + GLSL_TYPE_INT8, GLSL_TYPE_UINT16, GLSL_TYPE_INT16, GLSL_TYPE_UINT64, @@ -264,6 +266,8 @@ public: static const glsl_type *u64vec(unsigned components); static const glsl_type *i16vec(unsigned components); static const glsl_type *u16vec(unsigned components); + static const glsl_type *i8vec(unsigned components); + static const glsl_type *u8vec(unsigned components); /**@}*/ /** diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h index 29181694a35..ce6114424b3 100644 --- a/src/compiler/nir/nir.h +++ b/src/compiler/nir/nir.h @@ -735,6 +735,10 @@ nir_get_nir_type_for_glsl_base_type(enum glsl_base_type base_type) case GLSL_TYPE_INT16: return nir_type_int16; break; + case GLSL_TYPE_UINT8: + return nir_type_uint8; + case GLSL_TYPE_INT8: + return nir_type_int8; case GLSL_TYPE_UINT64: return nir_type_uint64; break; diff --git a/src/compiler/nir/nir_search.c b/src/compiler/nir/nir_search.c index dec56fee747..dda7878b372 100644 --- a/src/compiler/nir/nir_search.c +++ b/src/compiler/nir/nir_search.c @@ -213,6 +213,8 @@ match_value(const nir_search_value *value, nir_alu_instr *instr, unsigned src, case nir_type_uint: case nir_type_bool32: switch (load->def.bit_size) { + case 8: + case 16: </blockquote><div> </div><div>This is wrong. We need to do different things for 8 and 16. I found another broken case above for fp16. I'll send a patch. </div><div>Other than that, I don't see anything amiss. There may be something missing but I double-checked nir_constant_expressions and it should be doing 8-bit correctly already. Assuming you rebase on my nir_search patch, this patch is </div><div>Reviewed-by: Jason Ekstrand <<a href="mailto:jason@jlekstrand.net">jason@jlekstrand.net</a>> </div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> case 32: for (unsigned i = 0; i < num_components; ++i) { if (load->value.u32[new_swizzle[i]] != diff --git a/src/compiler/nir_types.cpp b/src/compiler/nir_types.cpp index 0085a19248a..718cec5e032 100644 --- a/src/compiler/nir_types.cpp +++ b/src/compiler/nir_types.cpp @@ -351,6 +351,18 @@ glsl_uint16_t_type(void) return glsl_type::uint16_t_type; } +const glsl_type * +glsl_int8_t_type(void) +{ + return glsl_type::int8_t_type; +} + +const glsl_type * +glsl_uint8_t_type(void) +{ + return glsl_type::uint8_t_type; +} + const glsl_type * glsl_bool_type(void) { diff --git a/src/compiler/nir_types.h b/src/compiler/nir_types.h index 5b5e09d137f..b2a51c6df5d 100644 --- a/src/compiler/nir_types.h +++ b/src/compiler/nir_types.h @@ -103,6 +103,10 @@ glsl_get_bit_size(const struct glsl_type *type) case GLSL_TYPE_INT16: return 16; + case GLSL_TYPE_UINT8: + case GLSL_TYPE_INT8: + return 8; + case GLSL_TYPE_DOUBLE: case GLSL_TYPE_INT64: case GLSL_TYPE_UINT64: @@ -149,6 +153,8 @@ const struct glsl_type *glsl_int64_t_type(void); const struct glsl_type *glsl_uint64_t_type(void); const struct glsl_type *glsl_int16_t_type(void); const struct glsl_type *glsl_uint16_t_type(void); +const struct glsl_type *glsl_int8_t_type(void); +const struct glsl_type *glsl_uint8_t_type(void); const struct glsl_type *glsl_bool_type(void); const struct glsl_type *glsl_scalar_type(enum glsl_base_type base_type); diff --git a/src/compiler/spirv/spirv_to_nir.c b/src/compiler/spirv/spirv_to_nir.c index b9e5198e7bc..2f261fbadd7 100644 --- a/src/compiler/spirv/spirv_to_nir.c +++ b/src/compiler/spirv/spirv_to_nir.c @@ -207,6 +207,8 @@ vtn_const_ssa_value(struct vtn_builder *b, nir_constant *constant, case GLSL_TYPE_UINT: case GLSL_TYPE_INT16: case GLSL_TYPE_UINT16: + case GLSL_TYPE_UINT8: + case GLSL_TYPE_INT8: case GLSL_TYPE_INT64: case GLSL_TYPE_UINT64: case GLSL_TYPE_BOOL: @@ -1007,6 +1009,9 @@ vtn_handle_type(struct vtn_builder *b, SpvOp opcode, case 16: val->type->type = (signedness ? glsl_int16_t_type() : glsl_uint16_t_type()); break; + case 8: + val->type->type = (signedness ? glsl_int8_t_type() : glsl_uint8_t_type()); + break; default: vtn_fail("Invalid int bit size"); } @@ -1281,6 +1286,8 @@ vtn_null_constant(struct vtn_builder *b, const struct glsl_type *type) case GLSL_TYPE_UINT: case GLSL_TYPE_INT16: case GLSL_TYPE_UINT16: + case GLSL_TYPE_UINT8: + case GLSL_TYPE_INT8: case GLSL_TYPE_INT64: case GLSL_TYPE_UINT64: case GLSL_TYPE_BOOL: @@ -1420,6 +1427,9 @@ vtn_handle_constant(struct vtn_builder *b, SpvOp opcode, case 16: val->constant->values->u16[0] = w[3]; break; + case 8: + val->constant->values->u8[0] = w[3]; + break; default: vtn_fail("Unsupported SpvOpConstant bit size"); } @@ -1442,6 +1452,9 @@ vtn_handle_constant(struct vtn_builder *b, SpvOp opcode, case 16: val->constant->values[0].u16[0] = get_specialization(b, val, w[3]); break; + case 8: + val->constant->values[0].u8[0] = get_specialization(b, val, w[3]); + break; default: vtn_fail("Unsupported SpvOpSpecConstant bit size"); } @@ -1474,6 +1487,9 @@ vtn_handle_constant(struct vtn_builder *b, SpvOp opcode, 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; default: vtn_fail("Invalid SpvOpConstantComposite bit size"); } @@ -1644,6 +1660,9 @@ vtn_handle_constant(struct vtn_builder *b, SpvOp opcode, 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; default: vtn_fail("Invalid SpvOpCompositeExtract bit size"); } @@ -1668,6 +1687,9 @@ vtn_handle_constant(struct vtn_builder *b, SpvOp opcode, 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; default: vtn_fail("Invalid SpvOpCompositeInsert bit size"); } @@ -1802,6 +1824,8 @@ vtn_create_ssa_value(struct vtn_builder *b, const struct glsl_type *type) case GLSL_TYPE_UINT: case GLSL_TYPE_INT16: case GLSL_TYPE_UINT16: + case GLSL_TYPE_UINT8: + case GLSL_TYPE_INT8: case GLSL_TYPE_INT64: case GLSL_TYPE_UINT64: case GLSL_TYPE_BOOL: diff --git a/src/compiler/spirv/vtn_variables.c b/src/compiler/spirv/vtn_variables.c index 9680d074131..b156451b3ec 100644 --- a/src/compiler/spirv/vtn_variables.c +++ b/src/compiler/spirv/vtn_variables.c @@ -297,6 +297,8 @@ vtn_ssa_offset_pointer_dereference(struct vtn_builder *b, case GLSL_TYPE_INT: case GLSL_TYPE_UINT16: case GLSL_TYPE_INT16: + case GLSL_TYPE_UINT8: + case GLSL_TYPE_INT8: case GLSL_TYPE_UINT64: case GLSL_TYPE_INT64: case GLSL_TYPE_FLOAT: @@ -413,6 +415,8 @@ vtn_pointer_to_deref(struct vtn_builder *b, struct vtn_pointer *ptr) case GLSL_TYPE_INT: case GLSL_TYPE_UINT16: case GLSL_TYPE_INT16: + case GLSL_TYPE_UINT8: + case GLSL_TYPE_INT8: case GLSL_TYPE_UINT64: case GLSL_TYPE_INT64: case GLSL_TYPE_FLOAT: @@ -624,6 +628,8 @@ vtn_pointer_to_offset(struct vtn_builder *b, struct vtn_pointer *ptr, case GLSL_TYPE_INT: case GLSL_TYPE_UINT16: case GLSL_TYPE_INT16: + case GLSL_TYPE_UINT8: + case GLSL_TYPE_INT8: case GLSL_TYPE_UINT64: case GLSL_TYPE_INT64: case GLSL_TYPE_FLOAT: @@ -672,6 +678,8 @@ vtn_type_block_size(struct vtn_builder *b, struct vtn_type *type) case GLSL_TYPE_INT: case GLSL_TYPE_UINT16: case GLSL_TYPE_INT16: + case GLSL_TYPE_UINT8: + case GLSL_TYPE_INT8: case GLSL_TYPE_UINT64: case GLSL_TYPE_INT64: case GLSL_TYPE_FLOAT: @@ -807,6 +815,8 @@ _vtn_block_load_store(struct vtn_builder *b, nir_intrinsic_op op, bool load, case GLSL_TYPE_INT: case GLSL_TYPE_UINT16: case GLSL_TYPE_INT16: + case GLSL_TYPE_UINT8: + case GLSL_TYPE_INT8: case GLSL_TYPE_UINT64: case GLSL_TYPE_INT64: case GLSL_TYPE_FLOAT: @@ -992,6 +1002,8 @@ _vtn_variable_load_store(struct vtn_builder *b, bool load, case GLSL_TYPE_INT: case GLSL_TYPE_UINT16: case GLSL_TYPE_INT16: + case GLSL_TYPE_UINT8: + case GLSL_TYPE_INT8: case GLSL_TYPE_UINT64: case GLSL_TYPE_INT64: case GLSL_TYPE_FLOAT: @@ -1076,6 +1088,8 @@ _vtn_variable_copy(struct vtn_builder *b, struct vtn_pointer *dest, case GLSL_TYPE_INT: case GLSL_TYPE_UINT16: case GLSL_TYPE_INT16: + case GLSL_TYPE_UINT8: + case GLSL_TYPE_INT8: case GLSL_TYPE_UINT64: case GLSL_TYPE_INT64: case GLSL_TYPE_FLOAT: diff --git a/src/intel/compiler/brw_fs.cpp b/src/intel/compiler/brw_fs.cpp index bed632d21b9..cde63b89e86 100644 --- a/src/intel/compiler/brw_fs.cpp +++ b/src/intel/compiler/brw_fs.cpp @@ -467,6 +467,9 @@ type_size_scalar(const struct glsl_type *type) case GLSL_TYPE_INT16: case GLSL_TYPE_FLOAT16: return DIV_ROUND_UP(type->components(), 2); + case GLSL_TYPE_UINT8: + case GLSL_TYPE_INT8: + return DIV_ROUND_UP(type->components(), 4); case GLSL_TYPE_DOUBLE: case GLSL_TYPE_UINT64: case GLSL_TYPE_INT64: diff --git a/src/intel/compiler/brw_shader.cpp b/src/intel/compiler/brw_shader.cpp index 1df4f35cd8e..9aeea7276d7 100644 --- a/src/intel/compiler/brw_shader.cpp +++ b/src/intel/compiler/brw_shader.cpp @@ -44,10 +44,14 @@ brw_type_for_base_type(const struct glsl_type *type) return BRW_REGISTER_TYPE_D; case GLSL_TYPE_INT16: return BRW_REGISTER_TYPE_W; + case GLSL_TYPE_INT8: + return BRW_REGISTER_TYPE_B; case GLSL_TYPE_UINT: return BRW_REGISTER_TYPE_UD; case GLSL_TYPE_UINT16: return BRW_REGISTER_TYPE_UW; + case GLSL_TYPE_UINT8: + return BRW_REGISTER_TYPE_UB; case GLSL_TYPE_ARRAY: return brw_type_for_base_type(type->fields.array); case GLSL_TYPE_STRUCT: diff --git a/src/intel/compiler/brw_vec4_visitor.cpp b/src/intel/compiler/brw_vec4_visitor.cpp index 53f6a5ed546..cea6a85ff10 100644 --- a/src/intel/compiler/brw_vec4_visitor.cpp +++ b/src/intel/compiler/brw_vec4_visitor.cpp @@ -589,6 +589,8 @@ type_size_xvec4(const struct glsl_type *type, bool as_vec4) case GLSL_TYPE_DOUBLE: case GLSL_TYPE_UINT16: case GLSL_TYPE_INT16: + case GLSL_TYPE_UINT8: + case GLSL_TYPE_INT8: case GLSL_TYPE_UINT64: case GLSL_TYPE_INT64: if (type->is_matrix()) { diff --git a/src/mesa/program/ir_to_mesa.cpp b/src/mesa/program/ir_to_mesa.cpp index 74089605029..29025d1c177 100644 --- a/src/mesa/program/ir_to_mesa.cpp +++ b/src/mesa/program/ir_to_mesa.cpp @@ -506,6 +506,8 @@ storage_type_size(const struct glsl_type *type, bool bindless) switch (type->base_type) { case GLSL_TYPE_UINT: case GLSL_TYPE_INT: + case GLSL_TYPE_UINT8: + case GLSL_TYPE_INT8: case GLSL_TYPE_UINT16: case GLSL_TYPE_INT16: case GLSL_TYPE_FLOAT: @@ -2534,6 +2536,7 @@ _mesa_associate_uniform_storage(struct gl_context *ctx, /* fallthrough */ case GLSL_TYPE_UINT: case GLSL_TYPE_UINT16: + case GLSL_TYPE_UINT8: assert(ctx->Const.NativeIntegers); format = uniform_native; columns = 1; @@ -2544,6 +2547,7 @@ _mesa_associate_uniform_storage(struct gl_context *ctx, /* fallthrough */ case GLSL_TYPE_INT: case GLSL_TYPE_INT16: + case GLSL_TYPE_INT8: format = (ctx->Const.NativeIntegers) ? uniform_native : uniform_int_float; columns = 1; diff --git a/src/mesa/state_tracker/st_glsl_types.cpp b/src/mesa/state_tracker/st_glsl_types.cpp index e57fbc8f314..d4d2139d9da 100644 --- a/src/mesa/state_tracker/st_glsl_types.cpp +++ b/src/mesa/state_tracker/st_glsl_types.cpp @@ -101,6 +101,8 @@ st_glsl_storage_type_size(const struct glsl_type *type, bool is_bindless) case GLSL_TYPE_FLOAT16: case GLSL_TYPE_UINT16: case GLSL_TYPE_INT16: + case GLSL_TYPE_UINT8: + case GLSL_TYPE_INT8: assert(!"Invalid type in type_size"); break; } -- 2.14.3 </blockquote></div> </div></div>