glsl: Add 16-bit types

Adds new INT16, UINT16 and FLOAT16 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 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.

v2: * Use FLOAT16 instead of HALF_FLOAT as name of the base type.
    * Removed float16_t from builtin types.
    * Don't copy 16-bit types as if they were 32-bit values in
      copy_constant_to_storage().
    * Use get_scalar_type() instead of adding a new custom switch
      statement.
    (Jason Ekstrand)
v3: Use GL_FLOAT16_NV instead of GL_HALF_FLOAT for consistency
    (Ilia Mirkin)
v4: Add missing 16-bit base types support in glsl_to_nir (Eduardo Lima).
v5: Fix coding style (Topi Poholainen).

Signed-off-by: Jose Maria Casanova Crespo <[email protected]>
Signed-off-by: Eduardo Lima <[email protected]>
Signed-off-by: Alejandro Piñeiro <[email protected]>
Reviewed-by: Jason Ekstrand <[email protected]>
Reviewed-by: Nicolai Hähnle <[email protected]>

9 files changed, 145 insertions, 8 deletions

diff --git a/src/compiler/builtin_type_macros.h b/src/compiler/builtin_type_macros.h
index a275617b343..e3a1cd29c87 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_FLOAT16_NV,        GLSL_TYPE_FLOAT16, 1, 1)
+DECL_TYPE(f16vec2,   GL_FLOAT16_VEC2_NV,   GLSL_TYPE_FLOAT16, 2, 1)
+DECL_TYPE(f16vec3,   GL_FLOAT16_VEC3_NV,   GLSL_TYPE_FLOAT16, 3, 1)
+DECL_TYPE(f16vec4,   GL_FLOAT16_VEC4_NV,   GLSL_TYPE_FLOAT16, 4, 1)
+
+DECL_TYPE(f16mat2,   GL_FLOAT16_MAT2_AMD,   GLSL_TYPE_FLOAT16, 2, 2)
+DECL_TYPE(f16mat3,   GL_FLOAT16_MAT3_AMD,   GLSL_TYPE_FLOAT16, 3, 3)
+DECL_TYPE(f16mat4,   GL_FLOAT16_MAT4_AMD,   GLSL_TYPE_FLOAT16, 4, 4)
+
+DECL_TYPE(f16mat2x3, GL_FLOAT16_MAT2x3_AMD, GLSL_TYPE_FLOAT16, 3, 2)
+DECL_TYPE(f16mat2x4, GL_FLOAT16_MAT2x4_AMD, GLSL_TYPE_FLOAT16, 4, 2)
+DECL_TYPE(f16mat3x2, GL_FLOAT16_MAT3x2_AMD, GLSL_TYPE_FLOAT16, 2, 3)
+DECL_TYPE(f16mat3x4, GL_FLOAT16_MAT3x4_AMD, GLSL_TYPE_FLOAT16, 4, 3)
+DECL_TYPE(f16mat4x2, GL_FLOAT16_MAT4x2_AMD, GLSL_TYPE_FLOAT16, 2, 4)
+DECL_TYPE(f16mat4x3, GL_FLOAT16_MAT4x3_AMD, GLSL_TYPE_FLOAT16, 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 5cdeb94720b..7abb8199e1e 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_FLOAT16:
    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/glsl_to_nir.cpp b/src/compiler/glsl/glsl_to_nir.cpp
index d327f52be60..67f15977ee5 100644
--- a/src/compiler/glsl/glsl_to_nir.cpp
+++ b/src/compiler/glsl/glsl_to_nir.cpp
@@ -1378,13 +1378,15 @@ nir_visitor::evaluate_rvalue(ir_rvalue* ir)
 static bool
 type_is_float(glsl_base_type type)
 {
-   return type == GLSL_TYPE_FLOAT || type == GLSL_TYPE_DOUBLE;
+   return type == GLSL_TYPE_FLOAT || type == GLSL_TYPE_DOUBLE ||
+      type == GLSL_TYPE_FLOAT16;
 }
 
 static bool
 type_is_signed(glsl_base_type type)
 {
-   return type == GLSL_TYPE_INT || type == GLSL_TYPE_INT64;
+   return type == GLSL_TYPE_INT || type == GLSL_TYPE_INT64 ||
+      type == GLSL_TYPE_INT16;
 }
 
 void
diff --git a/src/compiler/glsl/ir_clone.cpp b/src/compiler/glsl/ir_clone.cpp
index 12135070892..f088b6aebd5 100644
--- a/src/compiler/glsl/ir_clone.cpp
+++ b/src/compiler/glsl/ir_clone.cpp
@@ -337,10 +337,13 @@ 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_FLOAT16:
    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 f70d9100e12..be3715cc1a3 100644
--- a/src/compiler/glsl/link_uniform_initializers.cpp
+++ b/src/compiler/glsl/link_uniform_initializers.cpp
@@ -81,6 +81,9 @@ copy_constant_to_storage(union gl_constant_value *storage,
       case GLSL_TYPE_SUBROUTINE:
       case GLSL_TYPE_FUNCTION:
       case GLSL_TYPE_ERROR:
+      case GLSL_TYPE_UINT16:
+      case GLSL_TYPE_INT16:
+      case GLSL_TYPE_FLOAT16:
          /* All other types should have already been filtered by other
           * paths in the caller.
           */
diff --git a/src/compiler/glsl/lower_buffer_access.cpp b/src/compiler/glsl/lower_buffer_access.cpp
index fa6e5f5c066..db6e8e367b7 100644
--- a/src/compiler/glsl/lower_buffer_access.cpp
+++ b/src/compiler/glsl/lower_buffer_access.cpp
@@ -144,8 +144,7 @@ lower_buffer_access::emit_access(void *mem_ctx,
       const unsigned matrix_stride =
          link_calculate_matrix_stride(matrix_type, row_major, packing);
 
-      const glsl_type *deref_type = deref->type->is_float() ?
-         glsl_type::float_type : glsl_type::double_type;
+      const glsl_type *deref_type = deref->type->get_scalar_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 107a81f5e79..3cc5eb0495c 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_FLOAT16:
+      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_FLOAT16:
+      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_FLOAT16:
+         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_FLOAT16) || (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_FLOAT16: {
+         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_FLOAT16:
    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_FLOAT16:
    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_FLOAT16:
    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_FLOAT16:
    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 0cb0f9ee8ae..ab0b2637649 100644
--- a/src/compiler/glsl_types.h
+++ b/src/compiler/glsl_types.h
@@ -61,7 +61,10 @@ enum glsl_base_type {
    GLSL_TYPE_UINT = 0,
    GLSL_TYPE_INT,
    GLSL_TYPE_FLOAT,
+   GLSL_TYPE_FLOAT16,
    GLSL_TYPE_DOUBLE,
+   GLSL_TYPE_UINT16,
+   GLSL_TYPE_INT16,
    GLSL_TYPE_UINT64,
    GLSL_TYPE_INT64,
    GLSL_TYPE_BOOL,
@@ -252,12 +255,15 @@ public:
     * @{
     */
    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);
    /**@}*/
 
    /**
@@ -487,7 +493,9 @@ public:
    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_FLOAT16);
    }
 
    /**
diff --git a/src/mesa/program/ir_to_mesa.cpp b/src/mesa/program/ir_to_mesa.cpp
index aa8b6d7084b..ea74539cd76 100644
--- a/src/mesa/program/ir_to_mesa.cpp
+++ b/src/mesa/program/ir_to_mesa.cpp
@@ -507,7 +507,10 @@ 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_UINT16:
+   case GLSL_TYPE_INT16:
    case GLSL_TYPE_FLOAT:
+   case GLSL_TYPE_FLOAT16:
    case GLSL_TYPE_BOOL:
       if (type->is_matrix()) {
 	 return type->matrix_columns;
@@ -2531,6 +2534,7 @@ _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;
@@ -2540,6 +2544,7 @@ _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;
@@ -2549,6 +2554,7 @@ _mesa_associate_uniform_storage(struct gl_context *ctx,
                dmul *= 2;
             /* fallthrough */
          case GLSL_TYPE_FLOAT:
+         case GLSL_TYPE_FLOAT16:
             format = uniform_native;
             columns = storage->type->matrix_columns;
             break;