glsl: Use link_calculate_matrix_stride in lower_buffer_access and friends

Signed-off-by: Ian Romanick <[email protected]>
Reviewed-by: Thomas Helland <[email protected]>

4 files changed, 42 insertions, 70 deletions

diff --git a/src/compiler/glsl/lower_buffer_access.cpp b/src/compiler/glsl/lower_buffer_access.cpp
index caffaf9bfb2..219e03e550d 100644
--- a/src/compiler/glsl/lower_buffer_access.cpp
+++ b/src/compiler/glsl/lower_buffer_access.cpp
@@ -36,6 +36,7 @@
 #include "main/macros.h"
 #include "util/list.h"
 #include "glsl_parser_extras.h"
+#include "linker.h"
 
 using namespace ir_builder;
 
@@ -58,8 +59,8 @@ lower_buffer_access::emit_access(void *mem_ctx,
                                  ir_variable *base_offset,
                                  unsigned int deref_offset,
                                  bool row_major,
-                                 int matrix_columns,
-                                 unsigned int packing,
+                                 const glsl_type *matrix_type,
+                                 enum glsl_interface_packing packing,
                                  unsigned int write_mask)
 {
    if (deref->type->is_record()) {
@@ -78,7 +79,7 @@ lower_buffer_access::emit_access(void *mem_ctx,
 
          emit_access(mem_ctx, is_write, field_deref, base_offset,
                      deref_offset + field_offset,
-                     row_major, 1, packing,
+                     row_major, NULL, packing,
                      writemask_for_size(field_deref->type->vector_elements));
 
          field_offset += field->type->std140_size(row_major);
@@ -98,7 +99,7 @@ lower_buffer_access::emit_access(void *mem_ctx,
                                               element);
          emit_access(mem_ctx, is_write, element_deref, base_offset,
                      deref_offset + i * array_stride,
-                     row_major, 1, packing,
+                     row_major, NULL, packing,
                      writemask_for_size(element_deref->type->vector_elements));
       }
       return;
@@ -117,7 +118,7 @@ lower_buffer_access::emit_access(void *mem_ctx,
             int size_mul = deref->type->is_64bit() ? 8 : 4;
             emit_access(mem_ctx, is_write, col_deref, base_offset,
                         deref_offset + i * size_mul,
-                        row_major, deref->type->matrix_columns, packing,
+                        row_major, deref->type, packing,
                         writemask_for_size(col_deref->type->vector_elements));
          } else {
             int size_mul;
@@ -143,7 +144,7 @@ lower_buffer_access::emit_access(void *mem_ctx,
 
             emit_access(mem_ctx, is_write, col_deref, base_offset,
                         deref_offset + i * size_mul,
-                        row_major, deref->type->matrix_columns, packing,
+                        row_major, deref->type, packing,
                         writemask_for_size(col_deref->type->vector_elements));
          }
       }
@@ -159,45 +160,14 @@ lower_buffer_access::emit_access(void *mem_ctx,
          is_write ? write_mask : (1 << deref->type->vector_elements) - 1;
       insert_buffer_access(mem_ctx, deref, deref->type, offset, mask, -1);
    } else {
-      unsigned N = deref->type->is_64bit() ? 8 : 4;
-
       /* We're dereffing a column out of a row-major matrix, so we
        * gather the vector from each stored row.
-      */
+       */
       assert(deref->type->is_float() || deref->type->is_double());
+      assert(matrix_type != NULL);
 
-      /* Matrices, row_major or not, are stored as if they were
-       * arrays of vectors of the appropriate size in std140.
-       * Arrays have their strides rounded up to a vec4, so the
-       * matrix stride is always 16. However a double matrix may either be 16
-       * or 32 depending on the number of columns.
-       */
-      assert(matrix_columns <= 4);
-      unsigned matrix_stride = 0;
-      /* Matrix stride for std430 mat2xY matrices are not rounded up to
-       * vec4 size. From OpenGL 4.3 spec, section 7.6.2.2 "Standard Uniform
-       * Block Layout":
-       *
-       * "2. If the member is a two- or four-component vector with components
-       * consuming N basic machine units, the base alignment is 2N or 4N,
-       * respectively." [...]
-       * "4. If the member is an array of scalars or vectors, the base alignment
-       * and array stride are set to match the base alignment of a single array
-       * element, according to rules (1), (2), and (3), and rounded up to the
-       * base alignment of a vec4." [...]
-       * "7. If the member is a row-major matrix with C columns and R rows, the
-       * matrix is stored identically to an array of R row vectors with C
-       * components each, according to rule (4)." [...]
-       * "When using the std430 storage layout, shader storage blocks will be
-       * laid out in buffer storage identically to uniform and shader storage
-       * blocks using the std140 layout, except that the base alignment and
-       * stride of arrays of scalars and vectors in rule 4 and of structures in
-       * rule 9 are not rounded up a multiple of the base alignment of a vec4."
-       */
-      if (packing == GLSL_INTERFACE_PACKING_STD430 && matrix_columns == 2)
-         matrix_stride = 2 * N;
-      else
-         matrix_stride = glsl_align(matrix_columns * N, 16);
+      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;
@@ -325,13 +295,13 @@ lower_buffer_access::setup_buffer_access(void *mem_ctx,
                                          ir_rvalue **offset,
                                          unsigned *const_offset,
                                          bool *row_major,
-                                         int *matrix_columns,
+                                         const glsl_type **matrix_type,
                                          const glsl_struct_field **struct_field,
                                          enum glsl_interface_packing packing)
 {
    *offset = new(mem_ctx) ir_constant(0u);
    *row_major = is_dereferenced_thing_row_major(deref);
-   *matrix_columns = 1;
+   *matrix_type = NULL;
 
    /* Calculate the offset to the start of the region of the UBO
     * dereferenced by *rvalue.  This may be a variable offset if an
@@ -368,7 +338,7 @@ lower_buffer_access::setup_buffer_access(void *mem_ctx,
             array_stride = 4;
             if (deref_array->array->type->is_64bit())
                array_stride *= 2;
-            *matrix_columns = deref_array->array->type->matrix_columns;
+            *matrix_type = deref_array->array->type;
          } else if (deref_array->type->without_array()->is_interface()) {
             /* We're processing an array dereference of an interface instance
              * array. The thing being dereferenced *must* be a variable
diff --git a/src/compiler/glsl/lower_buffer_access.h b/src/compiler/glsl/lower_buffer_access.h
index 9e89d324bed..bd177ca142f 100644
--- a/src/compiler/glsl/lower_buffer_access.h
+++ b/src/compiler/glsl/lower_buffer_access.h
@@ -48,14 +48,16 @@ public:
 
    void emit_access(void *mem_ctx, bool is_write, ir_dereference *deref,
                     ir_variable *base_offset, unsigned int deref_offset,
-                    bool row_major, int matrix_columns,
-                    unsigned int packing, unsigned int write_mask);
+                    bool row_major, const glsl_type *matrix_type,
+                    enum glsl_interface_packing packing,
+                    unsigned int write_mask);
 
    bool is_dereferenced_thing_row_major(const ir_rvalue *deref);
 
    void setup_buffer_access(void *mem_ctx, ir_rvalue *deref,
                             ir_rvalue **offset, unsigned *const_offset,
-                            bool *row_major, int *matrix_columns,
+                            bool *row_major,
+                            const glsl_type **matrix_type,
                             const glsl_struct_field **struct_field,
                             enum glsl_interface_packing packing);
 
diff --git a/src/compiler/glsl/lower_shared_reference.cpp b/src/compiler/glsl/lower_shared_reference.cpp
index 86afc5de924..5de269fde82 100644
--- a/src/compiler/glsl/lower_shared_reference.cpp
+++ b/src/compiler/glsl/lower_shared_reference.cpp
@@ -137,13 +137,13 @@ lower_shared_reference_visitor::handle_rvalue(ir_rvalue **rvalue)
    ir_rvalue *offset = NULL;
    unsigned const_offset = get_shared_offset(var);
    bool row_major;
-   int matrix_columns;
+   const glsl_type *matrix_type;
    assert(var->get_interface_type() == NULL);
    const enum glsl_interface_packing packing = GLSL_INTERFACE_PACKING_STD430;
 
    setup_buffer_access(mem_ctx, deref,
                        &offset, &const_offset,
-                       &row_major, &matrix_columns, NULL, packing);
+                       &row_major, &matrix_type, NULL, packing);
 
    /* Now that we've calculated the offset to the start of the
     * dereference, walk over the type and emit loads into a temporary.
@@ -163,7 +163,7 @@ lower_shared_reference_visitor::handle_rvalue(ir_rvalue **rvalue)
    deref = new(mem_ctx) ir_dereference_variable(load_var);
 
    emit_access(mem_ctx, false, deref, load_offset, const_offset, row_major,
-               matrix_columns, packing, 0);
+               matrix_type, packing, 0);
 
    *rvalue = deref;
 
@@ -205,13 +205,13 @@ lower_shared_reference_visitor::handle_assignment(ir_assignment *ir)
    ir_rvalue *offset = NULL;
    unsigned const_offset = get_shared_offset(var);
    bool row_major;
-   int matrix_columns;
+   const glsl_type *matrix_type;
    assert(var->get_interface_type() == NULL);
    const enum glsl_interface_packing packing = GLSL_INTERFACE_PACKING_STD430;
 
    setup_buffer_access(mem_ctx, deref,
                        &offset, &const_offset,
-                       &row_major, &matrix_columns, NULL, packing);
+                       &row_major, &matrix_type, NULL, packing);
 
    deref = new(mem_ctx) ir_dereference_variable(store_var);
 
@@ -223,7 +223,7 @@ lower_shared_reference_visitor::handle_assignment(ir_assignment *ir)
 
    /* Now we have to write the value assigned to the temporary back to memory */
    emit_access(mem_ctx, true, deref, store_offset, const_offset, row_major,
-               matrix_columns, packing, ir->write_mask);
+               matrix_type, packing, ir->write_mask);
 
    progress = true;
 }
@@ -364,18 +364,18 @@ lower_shared_reference_visitor::lower_shared_atomic_intrinsic(ir_call *ir)
    ir_rvalue *offset = NULL;
    unsigned const_offset = get_shared_offset(var);
    bool row_major;
-   int matrix_columns;
+   const glsl_type *matrix_type;
    assert(var->get_interface_type() == NULL);
    const enum glsl_interface_packing packing = GLSL_INTERFACE_PACKING_STD430;
    buffer_access_type = shared_atomic_access;
 
    setup_buffer_access(mem_ctx, deref,
                        &offset, &const_offset,
-                       &row_major, &matrix_columns, NULL, packing);
+                       &row_major, &matrix_type, NULL, packing);
 
    assert(offset);
    assert(!row_major);
-   assert(matrix_columns == 1);
+   assert(matrix_type == NULL);
 
    ir_rvalue *deref_offset =
       add(offset, new(mem_ctx) ir_constant(const_offset));
diff --git a/src/compiler/glsl/lower_ubo_reference.cpp b/src/compiler/glsl/lower_ubo_reference.cpp
index e0745670df0..1e35b44d5fe 100644
--- a/src/compiler/glsl/lower_ubo_reference.cpp
+++ b/src/compiler/glsl/lower_ubo_reference.cpp
@@ -62,7 +62,7 @@ public:
                                 ir_rvalue **offset,
                                 unsigned *const_offset,
                                 bool *row_major,
-                                int *matrix_columns,
+                                const glsl_type **matrix_type,
                                 enum glsl_interface_packing packing);
    uint32_t ssbo_access_params();
    ir_expression *ubo_load(void *mem_ctx, const struct glsl_type *type,
@@ -274,7 +274,7 @@ lower_ubo_reference_visitor::setup_for_load_or_store(void *mem_ctx,
                                                      ir_rvalue **offset,
                                                      unsigned *const_offset,
                                                      bool *row_major,
-                                                     int *matrix_columns,
+                                                     const glsl_type **matrix_type,
                                                      enum glsl_interface_packing packing)
 {
    /* Determine the name of the interface block */
@@ -324,7 +324,7 @@ lower_ubo_reference_visitor::setup_for_load_or_store(void *mem_ctx,
 
    this->struct_field = NULL;
    setup_buffer_access(mem_ctx, deref, offset, const_offset, row_major,
-                       matrix_columns, &this->struct_field, packing);
+                       matrix_type, &this->struct_field, packing);
 }
 
 void
@@ -346,7 +346,7 @@ lower_ubo_reference_visitor::handle_rvalue(ir_rvalue **rvalue)
    ir_rvalue *offset = NULL;
    unsigned const_offset;
    bool row_major;
-   int matrix_columns;
+   const glsl_type *matrix_type;
 
    enum glsl_interface_packing packing =
       var->get_interface_type()->
@@ -362,7 +362,7 @@ lower_ubo_reference_visitor::handle_rvalue(ir_rvalue **rvalue)
     */
    setup_for_load_or_store(mem_ctx, var, deref,
                            &offset, &const_offset,
-                           &row_major, &matrix_columns,
+                           &row_major, &matrix_type,
                            packing);
    assert(offset);
 
@@ -383,7 +383,7 @@ lower_ubo_reference_visitor::handle_rvalue(ir_rvalue **rvalue)
 
    deref = new(mem_ctx) ir_dereference_variable(load_var);
    emit_access(mem_ctx, false, deref, load_offset, const_offset,
-               row_major, matrix_columns, packing, 0);
+               row_major, matrix_type, packing, 0);
    *rvalue = deref;
 
    progress = true;
@@ -562,7 +562,7 @@ lower_ubo_reference_visitor::write_to_memory(void *mem_ctx,
    ir_rvalue *offset = NULL;
    unsigned const_offset;
    bool row_major;
-   int matrix_columns;
+   const glsl_type *matrix_type;
 
    enum glsl_interface_packing packing =
       var->get_interface_type()->
@@ -576,7 +576,7 @@ lower_ubo_reference_visitor::write_to_memory(void *mem_ctx,
     */
    setup_for_load_or_store(mem_ctx, var, deref,
                            &offset, &const_offset,
-                           &row_major, &matrix_columns,
+                           &row_major, &matrix_type,
                            packing);
    assert(offset);
 
@@ -591,7 +591,7 @@ lower_ubo_reference_visitor::write_to_memory(void *mem_ctx,
 
    deref = new(mem_ctx) ir_dereference_variable(write_var);
    emit_access(mem_ctx, true, deref, write_offset, const_offset,
-               row_major, matrix_columns, packing, write_mask);
+               row_major, matrix_type, packing, write_mask);
 }
 
 ir_visitor_status
@@ -744,7 +744,7 @@ lower_ubo_reference_visitor::process_ssbo_unsized_array_length(ir_rvalue **rvalu
    ir_rvalue *base_offset = NULL;
    unsigned const_offset;
    bool row_major;
-   int matrix_columns;
+   const glsl_type *matrix_type;
 
    enum glsl_interface_packing packing =
       var->get_interface_type()->
@@ -760,7 +760,7 @@ lower_ubo_reference_visitor::process_ssbo_unsized_array_length(ir_rvalue **rvalu
     */
    setup_for_load_or_store(mem_ctx, var, deref,
                            &base_offset, &const_offset,
-                           &row_major, &matrix_columns,
+                           &row_major, &matrix_type,
                            packing);
    /* array.length() =
     *  max((buffer_object_size - offset_of_array) / stride_of_array, 0)
@@ -982,7 +982,7 @@ lower_ubo_reference_visitor::lower_ssbo_atomic_intrinsic(ir_call *ir)
    ir_rvalue *offset = NULL;
    unsigned const_offset;
    bool row_major;
-   int matrix_columns;
+   const glsl_type *matrix_type;
 
    enum glsl_interface_packing packing =
       var->get_interface_type()->
@@ -993,11 +993,11 @@ lower_ubo_reference_visitor::lower_ssbo_atomic_intrinsic(ir_call *ir)
 
    setup_for_load_or_store(mem_ctx, var, deref,
                            &offset, &const_offset,
-                           &row_major, &matrix_columns,
+                           &row_major, &matrix_type,
                            packing);
    assert(offset);
    assert(!row_major);
-   assert(matrix_columns == 1);
+   assert(matrix_type == NULL);
 
    ir_rvalue *deref_offset =
       add(offset, new(mem_ctx) ir_constant(const_offset));