[Mesa-dev] [PATCH 02/45] glsl: Add 16-bit types
Jason Ekstrand
jason at jlekstrand.net
Tue Aug 15 20:56:16 UTC 2017
On Thu, Jul 13, 2017 at 7:35 AM, Alejandro Piñeiro <apinheiro at igalia.com>
wrote:
> From: Eduardo Lima Mitev <elima at igalia.com>
>
> Adds new INT16, UINT16 and HALF FLOAT base types.
>
> The corresponding GL types for half floats were reused from the
> AMD_gpu_shader_half_float extension. The int16 and uint16 types come from
> NV_gpu_shader_5 extension.
>
> This adds the builtins and the lexer support.
>
> To avoid a bunch of warnings due to cases not handled in switch, the
> new types have been added to quite a few places using same behavior as
> their 32-bit counterparts, except for a few trivial cases where they are
> already handled properly. Subsequent patches in this set will provide
> correct 16-bit implementations when needed.
>
> Signed-off-by: Jose Maria Casanova Crespo <jmcasanova at igalia.com>
> Signed-off-by: Eduardo Lima <elima at igalia.com>
> Signed-off-by: Alejandro Piñeiro <apinheiro at igalia.com>
> ---
> src/compiler/builtin_type_macros.h | 26 +++++++
> src/compiler/glsl/ast_to_hir.cpp | 3 +
> src/compiler/glsl/builtin_types.cpp | 1 +
> src/compiler/glsl/ir_clone.cpp | 3 +
> src/compiler/glsl/link_uniform_initializers.cpp | 3 +
> src/compiler/glsl/lower_buffer_access.cpp | 16 ++++-
> src/compiler/glsl_types.cpp | 93
> ++++++++++++++++++++++++-
> src/compiler/glsl_types.h | 10 ++-
> src/mesa/program/ir_to_mesa.cpp | 6 ++
> 9 files changed, 155 insertions(+), 6 deletions(-)
>
> diff --git a/src/compiler/builtin_type_macros.h
> b/src/compiler/builtin_type_macros.h
> index a275617..52bae13 100644
> --- a/src/compiler/builtin_type_macros.h
> +++ b/src/compiler/builtin_type_macros.h
> @@ -62,6 +62,22 @@ DECL_TYPE(mat3x4, GL_FLOAT_MAT3x4, GLSL_TYPE_FLOAT, 4,
> 3)
> DECL_TYPE(mat4x2, GL_FLOAT_MAT4x2, GLSL_TYPE_FLOAT, 2, 4)
> DECL_TYPE(mat4x3, GL_FLOAT_MAT4x3, GLSL_TYPE_FLOAT, 3, 4)
>
> +DECL_TYPE(float16_t, GL_HALF_FLOAT, GLSL_TYPE_HALF_FLOAT, 1, 1)
> +DECL_TYPE(f16vec2, GL_FLOAT16_VEC2_NV, GLSL_TYPE_HALF_FLOAT, 2, 1)
> +DECL_TYPE(f16vec3, GL_FLOAT16_VEC3_NV, GLSL_TYPE_HALF_FLOAT, 3, 1)
> +DECL_TYPE(f16vec4, GL_FLOAT16_VEC4_NV, GLSL_TYPE_HALF_FLOAT, 4, 1)
> +
> +DECL_TYPE(f16mat2, GL_FLOAT16_MAT2_AMD, GLSL_TYPE_HALF_FLOAT, 2, 2)
> +DECL_TYPE(f16mat3, GL_FLOAT16_MAT3_AMD, GLSL_TYPE_HALF_FLOAT, 3, 3)
> +DECL_TYPE(f16mat4, GL_FLOAT16_MAT4_AMD, GLSL_TYPE_HALF_FLOAT, 4, 4)
> +
> +DECL_TYPE(f16mat2x3, GL_FLOAT16_MAT2x3_AMD, GLSL_TYPE_HALF_FLOAT, 3, 2)
> +DECL_TYPE(f16mat2x4, GL_FLOAT16_MAT2x4_AMD, GLSL_TYPE_HALF_FLOAT, 4, 2)
> +DECL_TYPE(f16mat3x2, GL_FLOAT16_MAT3x2_AMD, GLSL_TYPE_HALF_FLOAT, 2, 3)
> +DECL_TYPE(f16mat3x4, GL_FLOAT16_MAT3x4_AMD, GLSL_TYPE_HALF_FLOAT, 4, 3)
> +DECL_TYPE(f16mat4x2, GL_FLOAT16_MAT4x2_AMD, GLSL_TYPE_HALF_FLOAT, 2, 4)
> +DECL_TYPE(f16mat4x3, GL_FLOAT16_MAT4x3_AMD, GLSL_TYPE_HALF_FLOAT, 3, 4)
> +
> DECL_TYPE(double, GL_DOUBLE, GLSL_TYPE_DOUBLE, 1, 1)
> DECL_TYPE(dvec2, GL_DOUBLE_VEC2, GLSL_TYPE_DOUBLE, 2, 1)
> DECL_TYPE(dvec3, GL_DOUBLE_VEC3, GLSL_TYPE_DOUBLE, 3, 1)
> @@ -88,6 +104,16 @@ DECL_TYPE(u64vec2, GL_UNSIGNED_INT64_VEC2_ARB,
> GLSL_TYPE_UINT64, 2, 1)
> DECL_TYPE(u64vec3, GL_UNSIGNED_INT64_VEC3_ARB, GLSL_TYPE_UINT64, 3, 1)
> DECL_TYPE(u64vec4, GL_UNSIGNED_INT64_VEC4_ARB, GLSL_TYPE_UINT64, 4, 1)
>
> +DECL_TYPE(int16_t, GL_INT16_NV, GLSL_TYPE_INT16, 1, 1)
> +DECL_TYPE(i16vec2, GL_INT16_VEC2_NV, GLSL_TYPE_INT16, 2, 1)
> +DECL_TYPE(i16vec3, GL_INT16_VEC3_NV, GLSL_TYPE_INT16, 3, 1)
> +DECL_TYPE(i16vec4, GL_INT16_VEC4_NV, GLSL_TYPE_INT16, 4, 1)
> +
> +DECL_TYPE(uint16_t, GL_UNSIGNED_INT16_NV, GLSL_TYPE_UINT16, 1, 1)
> +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(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 c338ad7..4f6a324 100644
> --- a/src/compiler/glsl/ast_to_hir.cpp
> +++ b/src/compiler/glsl/ast_to_hir.cpp
> @@ -1108,12 +1108,15 @@ do_comparison(void *mem_ctx, int operation,
> ir_rvalue *op0, ir_rvalue *op1)
>
> switch (op0->type->base_type) {
> case GLSL_TYPE_FLOAT:
> + case GLSL_TYPE_HALF_FLOAT:
>
How would you feel about calling this FLOAT16? That's more consistent with
the integer enums though it is a bit different than double. In general,
I'm a fan of base+bits naming.
> case GLSL_TYPE_UINT:
> case GLSL_TYPE_INT:
> case GLSL_TYPE_BOOL:
> case GLSL_TYPE_DOUBLE:
> case GLSL_TYPE_UINT64:
> case GLSL_TYPE_INT64:
> + case GLSL_TYPE_UINT16:
> + case GLSL_TYPE_INT16:
> return new(mem_ctx) ir_expression(operation, op0, op1);
>
> case GLSL_TYPE_ARRAY: {
> diff --git a/src/compiler/glsl/builtin_types.cpp
> b/src/compiler/glsl/builtin_types.cpp
> index cae972b..4071e31 100644
> --- a/src/compiler/glsl/builtin_types.cpp
> +++ b/src/compiler/glsl/builtin_types.cpp
> @@ -144,6 +144,7 @@ static const struct builtin_type_versions {
> T(uvec3, 130, 300)
> T(uvec4, 130, 300)
> T(float, 110, 100)
> + T(float16_t, 330, 300)
>
I don't think float16_t is a builtin type in any version.
> T(vec2, 110, 100)
> T(vec3, 110, 100)
> T(vec4, 110, 100)
> diff --git a/src/compiler/glsl/ir_clone.cpp b/src/compiler/glsl/ir_clone.
> cpp
> index a64c7af..54421a3 100644
> --- a/src/compiler/glsl/ir_clone.cpp
> +++ b/src/compiler/glsl/ir_clone.cpp
> @@ -335,10 +335,13 @@ ir_constant::clone(void *mem_ctx, struct hash_table
> *ht) const
> case GLSL_TYPE_UINT:
> case GLSL_TYPE_INT:
> case GLSL_TYPE_FLOAT:
> + case GLSL_TYPE_HALF_FLOAT:
> case GLSL_TYPE_DOUBLE:
> case GLSL_TYPE_BOOL:
> case GLSL_TYPE_UINT64:
> case GLSL_TYPE_INT64:
> + case GLSL_TYPE_UINT16:
> + case GLSL_TYPE_INT16:
> 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 e7f9c9d..448e31c 100644
> --- a/src/compiler/glsl/link_uniform_initializers.cpp
> +++ b/src/compiler/glsl/link_uniform_initializers.cpp
> @@ -54,13 +54,16 @@ copy_constant_to_storage(union gl_constant_value
> *storage,
> for (unsigned int i = 0; i < elements; i++) {
> switch (base_type) {
> case GLSL_TYPE_UINT:
> + case GLSL_TYPE_UINT16:
> storage[i].u = val->value.u[i];
> break;
> case GLSL_TYPE_INT:
> + case GLSL_TYPE_INT16:
> case GLSL_TYPE_SAMPLER:
> storage[i].i = val->value.i[i];
> break;
> case GLSL_TYPE_FLOAT:
> + case GLSL_TYPE_HALF_FLOAT:
>
This doesn't look right. I think it's better to just let these hit the
unreachable().
> storage[i].f = val->value.f[i];
> break;
> case GLSL_TYPE_DOUBLE:
> diff --git a/src/compiler/glsl/lower_buffer_access.cpp
> b/src/compiler/glsl/lower_buffer_access.cpp
> index 24a96e2..51a540c 100644
> --- a/src/compiler/glsl/lower_buffer_access.cpp
> +++ b/src/compiler/glsl/lower_buffer_access.cpp
> @@ -199,8 +199,20 @@ lower_buffer_access::emit_access(void *mem_ctx,
> else
> matrix_stride = glsl_align(matrix_columns * N, 16);
>
> - const glsl_type *deref_type = deref->type->is_float() ?
> - glsl_type::float_type : glsl_type::double_type;
> + const glsl_type *deref_type;
> + switch (deref->type->base_type) {
> + case GLSL_TYPE_FLOAT:
> + deref_type = glsl_type::float_type;
> + break;
> + case GLSL_TYPE_HALF_FLOAT:
> + deref_type = glsl_type::float16_t_type;
> + break;
> + case GLSL_TYPE_DOUBLE:
> + deref_type = glsl_type::double_type;
> + break;
> + default:
> + assert(!"Invalid type");
>
Why not just make this deref->type->get_scalar_type() or get_base_type()?
> + }
>
> for (unsigned i = 0; i < deref->type->vector_elements; i++) {
> ir_rvalue *chan_offset =
> diff --git a/src/compiler/glsl_types.cpp b/src/compiler/glsl_types.cpp
> index 188b72f..8af0e25 100644
> --- a/src/compiler/glsl_types.cpp
> +++ b/src/compiler/glsl_types.cpp
> @@ -355,10 +355,16 @@ const glsl_type *glsl_type::get_base_type() const
> switch (base_type) {
> case GLSL_TYPE_UINT:
> return uint_type;
> + case GLSL_TYPE_UINT16:
> + return uint16_t_type;
> case GLSL_TYPE_INT:
> return int_type;
> + case GLSL_TYPE_INT16:
> + return int16_t_type;
> case GLSL_TYPE_FLOAT:
> return float_type;
> + case GLSL_TYPE_HALF_FLOAT:
> + return float16_t_type;
> case GLSL_TYPE_DOUBLE:
> return double_type;
> case GLSL_TYPE_BOOL:
> @@ -385,10 +391,16 @@ const glsl_type *glsl_type::get_scalar_type() const
> switch (type->base_type) {
> case GLSL_TYPE_UINT:
> return uint_type;
> + case GLSL_TYPE_UINT16:
> + return uint16_t_type;
> case GLSL_TYPE_INT:
> return int_type;
> + case GLSL_TYPE_INT16:
> + return int16_t_type;
> case GLSL_TYPE_FLOAT:
> return float_type;
> + case GLSL_TYPE_HALF_FLOAT:
> + return float16_t_type;
> case GLSL_TYPE_DOUBLE:
> return double_type;
> case GLSL_TYPE_BOOL:
> @@ -499,6 +511,18 @@ glsl_type::vec(unsigned components)
> }
>
> const glsl_type *
> +glsl_type::f16vec(unsigned components)
> +{
> + if (components == 0 || components > 4)
> + return error_type;
> +
> + static const glsl_type *const ts[] = {
> + float16_t_type, f16vec2_type, f16vec3_type, f16vec4_type
> + };
> + return ts[components - 1];
> +}
> +
> +const glsl_type *
> glsl_type::dvec(unsigned components)
> {
> if (components == 0 || components > 4)
> @@ -575,6 +599,31 @@ glsl_type::u64vec(unsigned components)
> }
>
> const glsl_type *
> +glsl_type::i16vec(unsigned components)
> +{
> + if (components == 0 || components > 4)
> + return error_type;
> +
> + static const glsl_type *const ts[] = {
> + int16_t_type, i16vec2_type, i16vec3_type, i16vec4_type
> + };
> + return ts[components - 1];
> +}
> +
> +
> +const glsl_type *
> +glsl_type::u16vec(unsigned components)
> +{
> + if (components == 0 || components > 4)
> + return error_type;
> +
> + static const glsl_type *const ts[] = {
> + uint16_t_type, u16vec2_type, u16vec3_type, u16vec4_type
> + };
> + return ts[components - 1];
> +}
> +
> +const glsl_type *
> glsl_type::get_instance(unsigned base_type, unsigned rows, unsigned
> columns)
> {
> if (base_type == GLSL_TYPE_VOID)
> @@ -593,6 +642,8 @@ glsl_type::get_instance(unsigned base_type, unsigned
> rows, unsigned columns)
> return ivec(rows);
> case GLSL_TYPE_FLOAT:
> return vec(rows);
> + case GLSL_TYPE_HALF_FLOAT:
> + return f16vec(rows);
> case GLSL_TYPE_DOUBLE:
> return dvec(rows);
> case GLSL_TYPE_BOOL:
> @@ -601,11 +652,17 @@ glsl_type::get_instance(unsigned base_type,
> unsigned rows, unsigned columns)
> return u64vec(rows);
> case GLSL_TYPE_INT64:
> return i64vec(rows);
> + case GLSL_TYPE_UINT16:
> + return u16vec(rows);
> + case GLSL_TYPE_INT16:
> + return i16vec(rows);
> default:
> return error_type;
> }
> } else {
> - if ((base_type != GLSL_TYPE_FLOAT && base_type != GLSL_TYPE_DOUBLE)
> || (rows == 1))
> + if ((base_type != GLSL_TYPE_FLOAT &&
> + base_type != GLSL_TYPE_DOUBLE &&
> + base_type != GLSL_TYPE_HALF_FLOAT) || (rows == 1))
> return error_type;
>
> /* GLSL matrix types are named mat{COLUMNS}x{ROWS}. Only the
> following
> @@ -619,7 +676,8 @@ glsl_type::get_instance(unsigned base_type, unsigned
> rows, unsigned columns)
> */
> #define IDX(c,r) (((c-1)*3) + (r-1))
>
> - if (base_type == GLSL_TYPE_DOUBLE) {
> + switch (base_type) {
> + case GLSL_TYPE_DOUBLE: {
> switch (IDX(columns, rows)) {
> case IDX(2,2): return dmat2_type;
> case IDX(2,3): return dmat2x3_type;
> @@ -632,7 +690,8 @@ glsl_type::get_instance(unsigned base_type, unsigned
> rows, unsigned columns)
> case IDX(4,4): return dmat4_type;
> default: return error_type;
> }
> - } else {
> + }
> + case GLSL_TYPE_FLOAT: {
> switch (IDX(columns, rows)) {
> case IDX(2,2): return mat2_type;
> case IDX(2,3): return mat2x3_type;
> @@ -646,6 +705,22 @@ glsl_type::get_instance(unsigned base_type, unsigned
> rows, unsigned columns)
> default: return error_type;
> }
> }
> + case GLSL_TYPE_HALF_FLOAT: {
> + switch (IDX(columns, rows)) {
> + case IDX(2,2): return f16mat2_type;
> + case IDX(2,3): return f16mat2x3_type;
> + case IDX(2,4): return f16mat2x4_type;
> + case IDX(3,2): return f16mat3x2_type;
> + case IDX(3,3): return f16mat3_type;
> + case IDX(3,4): return f16mat3x4_type;
> + case IDX(4,2): return f16mat4x2_type;
> + case IDX(4,3): return f16mat4x3_type;
> + case IDX(4,4): return f16mat4_type;
> + default: return error_type;
> + }
> + }
> + default: return error_type;
> + }
> }
>
> assert(!"Should not get here.");
> @@ -1282,7 +1357,10 @@ glsl_type::component_slots() const
> switch (this->base_type) {
> case GLSL_TYPE_UINT:
> case GLSL_TYPE_INT:
> + case GLSL_TYPE_UINT16:
> + case GLSL_TYPE_INT16:
> case GLSL_TYPE_FLOAT:
> + case GLSL_TYPE_HALF_FLOAT:
> case GLSL_TYPE_BOOL:
> return this->components();
>
> @@ -1371,7 +1449,10 @@ glsl_type::uniform_locations() const
> case GLSL_TYPE_UINT:
> case GLSL_TYPE_INT:
> case GLSL_TYPE_FLOAT:
> + case GLSL_TYPE_HALF_FLOAT:
> case GLSL_TYPE_DOUBLE:
> + case GLSL_TYPE_UINT16:
> + case GLSL_TYPE_INT16:
> case GLSL_TYPE_UINT64:
> case GLSL_TYPE_INT64:
> case GLSL_TYPE_BOOL:
> @@ -1401,8 +1482,11 @@ glsl_type::varying_count() const
> case GLSL_TYPE_UINT:
> case GLSL_TYPE_INT:
> case GLSL_TYPE_FLOAT:
> + case GLSL_TYPE_HALF_FLOAT:
> case GLSL_TYPE_DOUBLE:
> case GLSL_TYPE_BOOL:
> + case GLSL_TYPE_UINT16:
> + case GLSL_TYPE_INT16:
> case GLSL_TYPE_UINT64:
> case GLSL_TYPE_INT64:
> return 1;
> @@ -1974,7 +2058,10 @@ 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_UINT16:
> + case GLSL_TYPE_INT16:
> case GLSL_TYPE_FLOAT:
> + case GLSL_TYPE_HALF_FLOAT:
> case GLSL_TYPE_BOOL:
> case GLSL_TYPE_SAMPLER:
> case GLSL_TYPE_IMAGE:
> diff --git a/src/compiler/glsl_types.h b/src/compiler/glsl_types.h
> index 2857dc9..8ebf2f4 100644
> --- a/src/compiler/glsl_types.h
> +++ b/src/compiler/glsl_types.h
> @@ -52,7 +52,10 @@ enum glsl_base_type {
> GLSL_TYPE_UINT = 0,
> GLSL_TYPE_INT,
> GLSL_TYPE_FLOAT,
> + GLSL_TYPE_HALF_FLOAT,
> GLSL_TYPE_DOUBLE,
> + GLSL_TYPE_UINT16,
> + GLSL_TYPE_INT16,
> GLSL_TYPE_UINT64,
> GLSL_TYPE_INT64,
> GLSL_TYPE_BOOL,
> @@ -237,12 +240,15 @@ struct glsl_type {
> * @{
> */
> static const glsl_type *vec(unsigned components);
> + static const glsl_type *f16vec(unsigned components);
> static const glsl_type *dvec(unsigned components);
> static const glsl_type *ivec(unsigned components);
> static const glsl_type *uvec(unsigned components);
> static const glsl_type *bvec(unsigned components);
> static const glsl_type *i64vec(unsigned components);
> static const glsl_type *u64vec(unsigned components);
> + static const glsl_type *i16vec(unsigned components);
> + static const glsl_type *u16vec(unsigned components);
> /**@}*/
>
> /**
> @@ -472,7 +478,9 @@ struct glsl_type {
> bool is_matrix() const
> {
> /* GLSL only has float matrices. */
> - return (matrix_columns > 1) && (base_type == GLSL_TYPE_FLOAT ||
> base_type == GLSL_TYPE_DOUBLE);
> + return (matrix_columns > 1) && (base_type == GLSL_TYPE_FLOAT
> + || base_type == GLSL_TYPE_DOUBLE
> + || base_type ==
> GLSL_TYPE_HALF_FLOAT);
> }
>
> /**
> diff --git a/src/mesa/program/ir_to_mesa.cpp
> b/src/mesa/program/ir_to_mesa.cpp
> index ac12b59..c92bdbf 100644
> --- a/src/mesa/program/ir_to_mesa.cpp
> +++ b/src/mesa/program/ir_to_mesa.cpp
> @@ -507,7 +507,10 @@ type_size(const struct glsl_type *type)
> switch (type->base_type) {
> case GLSL_TYPE_UINT:
> case GLSL_TYPE_INT:
> + case GLSL_TYPE_UINT16:
> + case GLSL_TYPE_INT16:
> case GLSL_TYPE_FLOAT:
> + case GLSL_TYPE_HALF_FLOAT:
> case GLSL_TYPE_BOOL:
> if (type->is_matrix()) {
> return type->matrix_columns;
> @@ -2536,6 +2539,7 @@ _mesa_associate_uniform_storage(struct gl_context
> *ctx,
> dmul *= 2;
> /* fallthrough */
> case GLSL_TYPE_UINT:
> + case GLSL_TYPE_UINT16:
> assert(ctx->Const.NativeIntegers);
> format = uniform_native;
> columns = 1;
> @@ -2545,6 +2549,7 @@ _mesa_associate_uniform_storage(struct gl_context
> *ctx,
> dmul *= 2;
> /* fallthrough */
> case GLSL_TYPE_INT:
> + case GLSL_TYPE_INT16:
> format =
> (ctx->Const.NativeIntegers) ? uniform_native :
> uniform_int_float;
> columns = 1;
> @@ -2554,6 +2559,7 @@ _mesa_associate_uniform_storage(struct gl_context
> *ctx,
> dmul *= 2;
> /* fallthrough */
> case GLSL_TYPE_FLOAT:
> + case GLSL_TYPE_HALF_FLOAT:
> format = uniform_native;
> columns = storage->type->matrix_columns;
> break;
> --
> 2.9.3
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20170815/b27fb2c1/attachment-0001.html>
More information about the mesa-dev
mailing list