mesa: move _Used to gl_program
[mesa.git] / src / mesa / main / uniform_query.cpp
index fc2b5f57758bd0156c1d16cc608f48541dac0ae9..3108d348d6fa90f88137c97cae6ef1419a331504 100644 (file)
 
 #include "main/core.h"
 #include "main/context.h"
-#include "ir.h"
-#include "ir_uniform.h"
-#include "program/hash_table.h"
-#include "../glsl/program.h"
-#include "../glsl/ir_uniform.h"
-#include "../glsl/glsl_parser_extras.h"
 #include "main/shaderapi.h"
 #include "main/shaderobj.h"
-#include "uniforms.h"
+#include "main/uniforms.h"
+#include "compiler/glsl/ir.h"
+#include "compiler/glsl/ir_uniform.h"
+#include "compiler/glsl/glsl_parser_extras.h"
+#include "compiler/glsl/program.h"
+#include "util/bitscan.h"
 
 
 extern "C" void GLAPIENTRY
@@ -180,10 +179,10 @@ validate_uniform_parameters(struct gl_context *ctx,
 
    /* Check that the given location is in bounds of uniform remap table.
     * Unlinked programs will have NumUniformRemapTable == 0, so we can take
-    * the shProg->LinkStatus check out of the main path.
+    * the shProg->data->LinkStatus check out of the main path.
     */
    if (unlikely(location >= (GLint) shProg->NumUniformRemapTable)) {
-      if (!shProg->LinkStatus)
+      if (!shProg->data->LinkStatus)
          _mesa_error(ctx, GL_INVALID_OPERATION, "%s(program not linked)",
                      caller);
       else
@@ -194,7 +193,7 @@ validate_uniform_parameters(struct gl_context *ctx,
    }
 
    if (location == -1) {
-      if (!shProg->LinkStatus)
+      if (!shProg->data->LinkStatus)
          _mesa_error(ctx, GL_INVALID_OPERATION, "%s(program not linked)",
                      caller);
 
@@ -319,19 +318,12 @@ _mesa_get_uniform(struct gl_context *ctx, GLuint program, GLint location,
 
       return;
    }
-   if ((uni->type->base_type == GLSL_TYPE_DOUBLE &&
-        returnType != GLSL_TYPE_DOUBLE) ||
-       (uni->type->base_type != GLSL_TYPE_DOUBLE &&
-        returnType == GLSL_TYPE_DOUBLE)) {
-        _mesa_error( ctx, GL_INVALID_OPERATION,
-                    "glGetnUniform*vARB(incompatible uniform types)");
-       return;
-   }
 
    {
       unsigned elements = (uni->type->is_sampler())
         ? 1 : uni->type->components();
-      const int dmul = uni->type->base_type == GLSL_TYPE_DOUBLE ? 2 : 1;
+      const int dmul = uni->type->is_64bit() ? 2 : 1;
+      const int rmul = glsl_base_type_is_64bit(returnType) ? 2 : 1;
 
       /* Calculate the source base address *BEFORE* modifying elements to
        * account for the size of the user's buffer.
@@ -343,7 +335,7 @@ _mesa_get_uniform(struct gl_context *ctx, GLuint program, GLint location,
              returnType == GLSL_TYPE_UINT || returnType == GLSL_TYPE_DOUBLE);
 
       /* doubles have a different size than the other 3 types */
-      unsigned bytes = sizeof(src[0]) * elements * dmul;
+      unsigned bytes = sizeof(src[0]) * elements * rmul;
       if (bufSize < 0 || bytes > (unsigned) bufSize) {
         _mesa_error( ctx, GL_INVALID_OPERATION,
                     "glGetnUniform*vARB(out of bounds: bufSize is %d,"
@@ -367,32 +359,68 @@ _mesa_get_uniform(struct gl_context *ctx, GLuint program, GLint location,
       } else {
         union gl_constant_value *const dst =
            (union gl_constant_value *) paramsOut;
-
         /* This code could be optimized by putting the loop inside the switch
          * statements.  However, this is not expected to be
          * performance-critical code.
          */
         for (unsigned i = 0; i < elements; i++) {
+          int sidx = i * dmul;
+          int didx = i * rmul;
+
            switch (returnType) {
            case GLSL_TYPE_FLOAT:
               switch (uni->type->base_type) {
               case GLSL_TYPE_UINT:
-                 dst[i].f = (float) src[i].u;
+                 dst[didx].f = (float) src[sidx].u;
                  break;
               case GLSL_TYPE_INT:
               case GLSL_TYPE_SAMPLER:
                case GLSL_TYPE_IMAGE:
-                 dst[i].f = (float) src[i].i;
+                 dst[didx].f = (float) src[sidx].i;
                  break;
               case GLSL_TYPE_BOOL:
-                 dst[i].f = src[i].i ? 1.0f : 0.0f;
+                 dst[didx].f = src[sidx].i ? 1.0f : 0.0f;
                  break;
+               case GLSL_TYPE_DOUBLE: {
+                  double tmp;
+                  memcpy(&tmp, &src[sidx].f, sizeof(tmp));
+                  dst[didx].f = tmp;
+                 break;
+               }
+              default:
+                 assert(!"Should not get here.");
+                 break;
+              }
+              break;
+           case GLSL_TYPE_DOUBLE:
+              switch (uni->type->base_type) {
+               case GLSL_TYPE_UINT: {
+                  double tmp = src[sidx].u;
+                  memcpy(&dst[didx].f, &tmp, sizeof(tmp));
+                 break;
+               }
+              case GLSL_TYPE_INT:
+              case GLSL_TYPE_SAMPLER:
+               case GLSL_TYPE_IMAGE: {
+                  double tmp = src[sidx].i;
+                  memcpy(&dst[didx].f, &tmp, sizeof(tmp));
+                 break;
+               }
+               case GLSL_TYPE_BOOL: {
+                  double tmp = src[sidx].i ? 1.0 : 0.0;
+                  memcpy(&dst[didx].f, &tmp, sizeof(tmp));
+                 break;
+               }
+               case GLSL_TYPE_FLOAT: {
+                  double tmp = src[sidx].f;
+                  memcpy(&dst[didx].f, &tmp, sizeof(tmp));
+                 break;
+               }
               default:
                  assert(!"Should not get here.");
                  break;
               }
               break;
-
            case GLSL_TYPE_INT:
            case GLSL_TYPE_UINT:
               switch (uni->type->base_type) {
@@ -414,11 +442,17 @@ _mesa_get_uniform(struct gl_context *ctx, GLuint program, GLint location,
                   *      a floating-point value is rounded to the
                   *      nearest integer..."
                   */
-                 dst[i].i = IROUND(src[i].f);
+                 dst[didx].i = IROUND(src[sidx].f);
                  break;
               case GLSL_TYPE_BOOL:
-                 dst[i].i = src[i].i ? 1 : 0;
+                 dst[didx].i = src[sidx].i ? 1 : 0;
                  break;
+               case GLSL_TYPE_DOUBLE: {
+                  double tmp;
+                  memcpy(&tmp, &src[sidx].f, sizeof(tmp));
+                  dst[didx].i = IROUNDD(tmp);
+                 break;
+               }
               default:
                  assert(!"Should not get here.");
                  break;
@@ -465,9 +499,12 @@ log_uniform(const void *values, enum glsl_base_type basicType,
       case GLSL_TYPE_FLOAT:
         printf("%g ", v[i].f);
         break;
-      case GLSL_TYPE_DOUBLE:
-         printf("%g ", *(double* )&v[i * 2].f);
+      case GLSL_TYPE_DOUBLE: {
+         double tmp;
+         memcpy(&tmp, &v[i * 2].f, sizeof(tmp));
+         printf("%g ", tmp);
          break;
+      }
       default:
         assert(!"Should not get here.");
         break;
@@ -528,7 +565,7 @@ _mesa_propagate_uniforms_to_driver_storage(struct gl_uniform_storage *uni,
     */
    const unsigned components = MAX2(1, uni->type->vector_elements);
    const unsigned vectors = MAX2(1, uni->type->matrix_columns);
-   const int dmul = uni->type->base_type == GLSL_TYPE_DOUBLE ? 2 : 1;
+   const int dmul = uni->type->is_64bit() ? 2 : 1;
 
    /* Store the data in the driver's requested type in the driver's storage
     * areas.
@@ -557,14 +594,31 @@ _mesa_propagate_uniforms_to_driver_storage(struct gl_uniform_storage *uni,
         unsigned j;
         unsigned v;
 
-        for (j = 0; j < count; j++) {
-           for (v = 0; v < vectors; v++) {
-              memcpy(dst, src, src_vector_byte_stride);
-              src += src_vector_byte_stride;
-              dst += store->vector_stride;
+        if (src_vector_byte_stride == store->vector_stride) {
+           if (extra_stride) {
+              for (j = 0; j < count; j++) {
+                 memcpy(dst, src, src_vector_byte_stride * vectors);
+                 src += src_vector_byte_stride * vectors;
+                 dst += store->vector_stride * vectors;
+
+                 dst += extra_stride;
+              }
+           } else {
+              /* Unigine Heaven benchmark gets here */
+              memcpy(dst, src, src_vector_byte_stride * vectors * count);
+              src += src_vector_byte_stride * vectors * count;
+              dst += store->vector_stride * vectors * count;
            }
+        } else {
+           for (j = 0; j < count; j++) {
+              for (v = 0; v < vectors; v++) {
+                 memcpy(dst, src, src_vector_byte_stride);
+                 src += src_vector_byte_stride;
+                 dst += store->vector_stride;
+              }
 
-           dst += extra_stride;
+              dst += extra_stride;
+           }
         }
         break;
       }
@@ -648,7 +702,7 @@ _mesa_uniform(struct gl_context *ctx, struct gl_shader_program *shProg,
               unsigned src_components)
 {
    unsigned offset;
-   int size_mul = basicType == GLSL_TYPE_DOUBLE ? 2 : 1;
+   int size_mul = glsl_base_type_is_64bit(basicType) ? 2 : 1;
 
    struct gl_uniform_storage *const uni =
       validate_uniform_parameters(ctx, shProg, location, count,
@@ -738,6 +792,10 @@ _mesa_uniform(struct gl_context *ctx, struct gl_shader_program *shProg,
             return;
          }
       }
+      /* We need to reset the validate flag on changes to samplers in case
+       * two different sampler types are set to the same texture unit.
+       */
+      ctx->_Shader->Validated = GL_FALSE;
    }
 
    if (uni->type->is_image()) {
@@ -791,8 +849,6 @@ _mesa_uniform(struct gl_context *ctx, struct gl_shader_program *shProg,
       }
    }
 
-   uni->initialized = true;
-
    _mesa_propagate_uniforms_to_driver_storage(uni, offset, count);
 
    /* If the uniform is a sampler, do the extra magic necessary to propagate
@@ -801,15 +857,15 @@ _mesa_uniform(struct gl_context *ctx, struct gl_shader_program *shProg,
    if (uni->type->is_sampler()) {
       bool flushed = false;
       for (int i = 0; i < MESA_SHADER_STAGES; i++) {
-        struct gl_shader *const sh = shProg->_LinkedShaders[i];
+        struct gl_linked_shader *const sh = shProg->_LinkedShaders[i];
 
         /* If the shader stage doesn't use the sampler uniform, skip this.
          */
-        if (sh == NULL || !uni->sampler[i].active)
+        if (sh == NULL || !uni->opaque[i].active)
            continue;
 
          for (int j = 0; j < count; j++) {
-            sh->SamplerUnits[uni->sampler[i].index + offset + j] =
+            sh->SamplerUnits[uni->opaque[i].index + offset + j] =
                ((unsigned *) values)[j];
          }
 
@@ -821,9 +877,10 @@ _mesa_uniform(struct gl_context *ctx, struct gl_shader_program *shProg,
          * been modified.
          */
         bool changed = false;
-        for (unsigned j = 0; j < ARRAY_SIZE(prog->SamplerUnits); j++) {
-           if ((sh->active_samplers & (1U << j)) != 0
-               && (prog->SamplerUnits[j] != sh->SamplerUnits[j])) {
+        GLbitfield mask = sh->active_samplers;
+        while (mask) {
+           const int j = u_bit_scan(&mask);
+           if (prog->SamplerUnits[j] != sh->SamplerUnits[j]) {
               changed = true;
               break;
            }
@@ -835,10 +892,6 @@ _mesa_uniform(struct gl_context *ctx, struct gl_shader_program *shProg,
               flushed = true;
            }
 
-           memcpy(prog->SamplerUnits,
-                  sh->SamplerUnits,
-                  sizeof(sh->SamplerUnits));
-
            _mesa_update_shader_textures_used(shProg, prog);
             if (ctx->Driver.SamplerUniformChange)
               ctx->Driver.SamplerUniformChange(ctx, prog->Target, prog);
@@ -851,11 +904,11 @@ _mesa_uniform(struct gl_context *ctx, struct gl_shader_program *shProg,
     */
    if (uni->type->is_image()) {
       for (int i = 0; i < MESA_SHADER_STAGES; i++) {
-        if (uni->image[i].active) {
-            struct gl_shader *sh = shProg->_LinkedShaders[i];
+        if (uni->opaque[i].active) {
+            struct gl_linked_shader *sh = shProg->_LinkedShaders[i];
 
             for (int j = 0; j < count; j++)
-               sh->ImageUnits[uni->image[i].index + offset + j] =
+               sh->ImageUnits[uni->opaque[i].index + offset + j] =
                   ((GLint *) values)[j];
          }
       }
@@ -919,6 +972,31 @@ _mesa_uniform_matrix(struct gl_context *ctx, struct gl_shader_program *shProg,
       }
    }
 
+   /* Section 2.11.7 (Uniform Variables) of the OpenGL 4.2 Core Profile spec
+    * says:
+    *
+    *     "If any of the following conditions occur, an INVALID_OPERATION
+    *     error is generated by the Uniform* commands, and no uniform values
+    *     are changed:
+    *
+    *     ...
+    *
+    *     - if the uniform declared in the shader is not of type boolean and
+    *       the type indicated in the name of the Uniform* command used does
+    *       not match the type of the uniform"
+    *
+    * There are no Boolean matrix types, so we do not need to allow
+    * GLSL_TYPE_BOOL here (as _mesa_uniform does).
+    */
+   if (uni->type->base_type != basicType) {
+      _mesa_error(ctx, GL_INVALID_OPERATION,
+                  "glUniformMatrix%ux%u(\"%s\"@%d is %s, not %s)",
+                  cols, rows, uni->name, location,
+                  glsl_type_name(uni->type->base_type),
+                  glsl_type_name(basicType));
+      return;
+   }
+
    if (unlikely(ctx->_Shader->Flags & GLSL_UNIFORMS)) {
       log_uniform(values, uni->type->base_type, components, vectors, count,
                  bool(transpose), shProg, location, uni);
@@ -946,7 +1024,7 @@ _mesa_uniform_matrix(struct gl_context *ctx, struct gl_shader_program *shProg,
    elements = components * vectors;
 
    if (!transpose) {
-      memcpy(&uni->storage[elements * offset], values,
+      memcpy(&uni->storage[size_mul * elements * offset], values,
             sizeof(uni->storage[0]) * elements * count * size_mul);
    } else if (basicType == GLSL_TYPE_FLOAT) {
       /* Copy and transpose the matrix.
@@ -981,8 +1059,6 @@ _mesa_uniform_matrix(struct gl_context *ctx, struct gl_shader_program *shProg,
       }
    }
 
-   uni->initialized = true;
-
    _mesa_propagate_uniforms_to_driver_storage(uni, offset, count);
 }
 
@@ -992,7 +1068,7 @@ _mesa_sampler_uniforms_are_valid(const struct gl_shader_program *shProg,
                                 char *errMsg, size_t errMsgLength)
 {
    /* Shader does not have samplers. */
-   if (shProg->NumUniformStorage == 0)
+   if (shProg->data->NumUniformStorage == 0)
       return true;
 
    if (!shProg->SamplersValidated) {
@@ -1021,58 +1097,51 @@ _mesa_sampler_uniforms_pipeline_are_valid(struct gl_pipeline_object *pipeline)
     *         - The number of active samplers in the program exceeds the
     *           maximum number of texture image units allowed."
     */
+
+   GLbitfield mask;
+   GLbitfield TexturesUsed[MAX_COMBINED_TEXTURE_IMAGE_UNITS];
+   struct gl_linked_shader *shader;
    unsigned active_samplers = 0;
    const struct gl_shader_program **shProg =
       (const struct gl_shader_program **) pipeline->CurrentProgram;
 
-   const glsl_type *unit_types[MAX_COMBINED_TEXTURE_IMAGE_UNITS];
-   memset(unit_types, 0, sizeof(unit_types));
+
+   memset(TexturesUsed, 0, sizeof(TexturesUsed));
 
    for (unsigned idx = 0; idx < ARRAY_SIZE(pipeline->CurrentProgram); idx++) {
       if (!shProg[idx])
          continue;
 
-      for (unsigned i = 0; i < shProg[idx]->NumUniformStorage; i++) {
-         const struct gl_uniform_storage *const storage =
-            &shProg[idx]->UniformStorage[i];
+      shader = shProg[idx]->_LinkedShaders[idx];
+      if (!shader || !shader->Program)
+         continue;
 
-         if (!storage->type->is_sampler())
+      mask = shader->Program->SamplersUsed;
+      while (mask) {
+         const int s = u_bit_scan(&mask);
+         GLuint unit = shader->SamplerUnits[s];
+         GLuint tgt = shader->SamplerTargets[s];
+
+         /* FIXME: Samplers are initialized to 0 and Mesa doesn't do a
+          * great job of eliminating unused uniforms currently so for now
+          * don't throw an error if two sampler types both point to 0.
+          */
+         if (unit == 0)
             continue;
 
-         active_samplers++;
-
-         const unsigned count = MAX2(1, storage->array_elements);
-         for (unsigned j = 0; j < count; j++) {
-            const unsigned unit = storage->storage[j].i;
-
-            /* FIXME: Samplers are initialized to 0 and Mesa doesn't do a
-             * great job of eliminating unused uniforms currently so for now
-             * don't throw an error if two sampler types both point to 0.
-             */
-            if (unit == 0)
-               continue;
-
-            /* The types of the samplers associated with a particular texture
-             * unit must be an exact match.  Page 74 (page 89 of the PDF) of
-             * the OpenGL 3.3 core spec says:
-             *
-             *     "It is not allowed to have variables of different sampler
-             *     types pointing to the same texture image unit within a
-             *     program object."
-             */
-            if (unit_types[unit] == NULL) {
-               unit_types[unit] = storage->type;
-            } else if (unit_types[unit] != storage->type) {
-               pipeline->InfoLog =
-                  ralloc_asprintf(pipeline,
-                                  "Texture unit %d is accessed both as %s "
-                                  "and %s",
-                                  unit, unit_types[unit]->name,
-                                  storage->type->name);
-               return false;
-            }
+         if (TexturesUsed[unit] & ~(1 << tgt)) {
+            pipeline->InfoLog =
+               ralloc_asprintf(pipeline,
+                     "Program %d: "
+                     "Texture unit %d is accessed with 2 different types",
+                     shProg[idx]->Name, unit);
+            return false;
          }
+
+         TexturesUsed[unit] |= (1 << tgt);
       }
+
+      active_samplers += shader->num_samplers;
    }
 
    if (active_samplers > MAX_COMBINED_TEXTURE_IMAGE_UNITS) {