Merge commit 'origin/gallium-0.1' into gallium-0.2

9 files changed, 175 insertions, 9 deletions

diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c
index b98c0a0ecf..3c175f31d8 100644
--- a/src/gallium/auxiliary/draw/draw_pt.c
+++ b/src/gallium/auxiliary/draw/draw_pt.c
@@ -178,6 +178,92 @@ void draw_pt_destroy( struct draw_context *draw )
 }
 
 
+/**
+ * Debug- print the first 'count' vertices.
+ */
+static void
+draw_print_arrays(struct draw_context *draw, uint prim, int start, uint count)
+{
+   uint i;
+
+   debug_printf("Draw arrays(prim = %u, start = %u, count = %u)\n",
+                prim, start, count);
+
+   for (i = 0; i < count; i++) {
+      uint ii, j;
+
+      if (draw->pt.user.elts) {
+         /* indexed arrays */
+         switch (draw->pt.user.eltSize) {
+         case 1:
+            {
+               const ubyte *elem = (const ubyte *) draw->pt.user.elts;
+               ii = elem[start + i];
+            }
+            break;
+         case 2:
+            {
+               const ushort *elem = (const ushort *) draw->pt.user.elts;
+               ii = elem[start + i];
+            }
+            break;
+         case 4:
+            {
+               const uint *elem = (const uint *) draw->pt.user.elts;
+               ii = elem[start + i];
+            }
+            break;
+         default:
+            assert(0);
+         }
+         debug_printf("Element[%u + %u] -> Vertex %u:\n", start, i, ii);
+      }
+      else {
+         /* non-indexed arrays */
+         ii = start + i;
+         debug_printf("Vertex %u:\n", ii);
+      }
+
+      for (j = 0; j < draw->pt.nr_vertex_elements; j++) {
+         uint buf = draw->pt.vertex_element[j].vertex_buffer_index;
+         ubyte *ptr = (ubyte *) draw->pt.user.vbuffer[buf];
+         ptr += draw->pt.vertex_buffer[buf].pitch * ii;
+         ptr += draw->pt.vertex_element[j].src_offset;
+
+         debug_printf("  Attr %u: ", j);
+         switch (draw->pt.vertex_element[j].src_format) {
+         case PIPE_FORMAT_R32_FLOAT:
+            {
+               float *v = (float *) ptr;
+               debug_printf("%f  @ %p\n", v[0], (void *) v);
+            }
+            break;
+         case PIPE_FORMAT_R32G32_FLOAT:
+            {
+               float *v = (float *) ptr;
+               debug_printf("%f %f  @ %p\n", v[0], v[1], (void *) v);
+            }
+            break;
+         case PIPE_FORMAT_R32G32B32_FLOAT:
+            {
+               float *v = (float *) ptr;
+               debug_printf("%f %f %f  @ %p\n", v[0], v[1], v[2], (void *) v);
+            }
+            break;
+         case PIPE_FORMAT_R32G32B32A32_FLOAT:
+            {
+               float *v = (float *) ptr;
+               debug_printf("%f %f %f %f  @ %p\n", v[0], v[1], v[2], v[3],
+                            (void *) v);
+            }
+            break;
+         default:
+            debug_printf("other format (fix me)\n");
+            ;
+         }
+      }
+   }
+}
 
 
 /**
@@ -197,6 +283,9 @@ draw_arrays(struct draw_context *draw, unsigned prim,
       draw->reduced_prim = reduced_prim;
    }
 
+   if (0)
+      draw_print_arrays(draw, prim, start, MIN2(count, 20));
+
 #if 0
    {
       int i;
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index e32779123f..9d2c87a2f9 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -1045,12 +1045,16 @@ fetch_source(
    if (reg->SrcRegister.Indirect) {
       union tgsi_exec_channel index2;
       union tgsi_exec_channel indir_index;
+      const uint execmask = mach->ExecMask;
+      uint i;
 
+      /* which address register (always zero now) */
       index2.i[0] =
       index2.i[1] =
       index2.i[2] =
       index2.i[3] = reg->SrcRegisterInd.Index;
 
+      /* get current value of address register[swizzle] */
       swizzle = tgsi_util_get_src_register_swizzle( &reg->SrcRegisterInd, CHAN_X );
       fetch_src_file_channel(
          mach,
@@ -1059,10 +1063,19 @@ fetch_source(
          &index2,
          &indir_index );
 
+      /* add value of address register to the offset */
       index.i[0] += indir_index.i[0];
       index.i[1] += indir_index.i[1];
       index.i[2] += indir_index.i[2];
       index.i[3] += indir_index.i[3];
+
+      /* for disabled execution channels, zero-out the index to
+       * avoid using a potential garbage value.
+       */
+      for (i = 0; i < QUAD_SIZE; i++) {
+         if ((execmask & (1 << i)) == 0)
+            index.i[i] = 0;
+      }
    }
 
    if( reg->SrcRegister.Dimension ) {
@@ -1091,6 +1104,8 @@ fetch_source(
       if (reg->SrcRegisterDim.Indirect) {
          union tgsi_exec_channel index2;
          union tgsi_exec_channel indir_index;
+         const uint execmask = mach->ExecMask;
+         uint i;
 
          index2.i[0] =
          index2.i[1] =
@@ -1109,6 +1124,14 @@ fetch_source(
          index.i[1] += indir_index.i[1];
          index.i[2] += indir_index.i[2];
          index.i[3] += indir_index.i[3];
+
+         /* for disabled execution channels, zero-out the index to
+          * avoid using a potential garbage value.
+          */
+         for (i = 0; i < QUAD_SIZE; i++) {
+            if ((execmask & (1 << i)) == 0)
+               index.i[i] = 0;
+         }
       }
    }
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sanity.c b/src/gallium/auxiliary/tgsi/tgsi_sanity.c
index 11659247c0..bc7b941b78 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_sanity.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_sanity.c
@@ -153,17 +153,21 @@ check_register_usage(
    if (!check_file_name( ctx, file ))
       return FALSE;
 
-   if (index < 0 || index > MAX_REGISTERS) {
-      report_error( ctx, "%s[%i]: Invalid index %s", file_names[file], index, name );
-      return FALSE;
-   }
-
    if (indirect_access) {
+      /* Note that 'index' is an offset relative to the value of the
+       * address register.  No range checking done here.
+       */
       if (!is_any_register_declared( ctx, file ))
          report_error( ctx, "%s: Undeclared %s register", file_names[file], name );
       ctx->regs_ind_used[file] = TRUE;
    }
    else {
+      if (index < 0 || index > MAX_REGISTERS) {
+         report_error( ctx, "%s[%i]: Invalid index %s",
+                       file_names[file], index, name );
+         return FALSE;
+      }
+
       if (!is_register_declared( ctx, file, index ))
          report_error( ctx, "%s[%d]: Undeclared %s register", file_names[file], index, name );
       ctx->regs_used[file][index / BITS_IN_REG_FLAG] |= (1 << (index % BITS_IN_REG_FLAG));
diff --git a/src/gallium/drivers/softpipe/sp_quad_output.c b/src/gallium/drivers/softpipe/sp_quad_output.c
index d05e12d1d9..b7aac7f84a 100644
--- a/src/gallium/drivers/softpipe/sp_quad_output.c
+++ b/src/gallium/drivers/softpipe/sp_quad_output.c
@@ -64,6 +64,14 @@ output_quad(struct quad_stage *qs, struct quad_header *quad)
             for (i = 0; i < 4; i++) { /* loop over color chans */
                tile->data.color[y][x][i] = quadColor[i][j];
             }
+            if (0) {
+               debug_printf("sp write pixel %d,%d: %g, %g, %g\n",
+                            quad->input.x0 + x,
+                            quad->input.y0 + y,
+                            quadColor[0][j],
+                            quadColor[1][j],
+                            quadColor[2][j]);
+            }
          }
       }
    }
diff --git a/src/mesa/shader/shader_api.c b/src/mesa/shader/shader_api.c
index 7f020fe5d3..7a8501dcdd 100644
--- a/src/mesa/shader/shader_api.c
+++ b/src/mesa/shader/shader_api.c
@@ -1422,10 +1422,22 @@ _mesa_use_program(GLcontext *ctx, GLuint program)
 
 
 /**
- * Update the vertex and fragment program's TexturesUsed arrays.
+ * Update the vertex/fragment program's TexturesUsed array.
+ *
+ * This needs to be called after glUniform(set sampler var) is called.
+ * A call to glUniform(samplerVar, value) causes a sampler to point to a
+ * particular texture unit.  We know the sampler's texture target
+ * (1D/2D/3D/etc) from compile time but the sampler's texture unit is
+ * set by glUniform() calls.
+ *
+ * So, scan the program->SamplerUnits[] and program->SamplerTargets[]
+ * information to update the prog->TexturesUsed[] values.
+ * Each value of TexturesUsed[unit] is one of zero, TEXTURE_1D_INDEX,
+ * TEXTURE_2D_INDEX, TEXTURE_3D_INDEX, etc.
+ * We'll use that info for state validation before rendering.
  */
-static void
-update_textures_used(struct gl_program *prog)
+void
+_mesa_update_shader_textures_used(struct gl_program *prog)
 {
    GLuint s;
 
@@ -1551,7 +1563,7 @@ set_program_uniform(GLcontext *ctx, struct gl_program *program,
 
       /* This maps a sampler to a texture unit: */
       program->SamplerUnits[sampler] = texUnit;
-      update_textures_used(program);
+      _mesa_update_shader_textures_used(program);
 
       FLUSH_VERTICES(ctx, _NEW_TEXTURE);
    }
diff --git a/src/mesa/shader/shader_api.h b/src/mesa/shader/shader_api.h
index e7f1266915..ec1996ee98 100644
--- a/src/mesa/shader/shader_api.h
+++ b/src/mesa/shader/shader_api.h
@@ -80,6 +80,10 @@ _mesa_lookup_shader(GLcontext *ctx, GLuint name);
 
 
 extern void
+_mesa_update_shader_textures_used(struct gl_program *prog);
+
+
+extern void
 _mesa_use_program(GLcontext *ctx, GLuint program);
 
 
diff --git a/src/mesa/shader/slang/slang_link.c b/src/mesa/shader/slang/slang_link.c
index 11fdf3b9dc..2074e84209 100644
--- a/src/mesa/shader/slang/slang_link.c
+++ b/src/mesa/shader/slang/slang_link.c
@@ -561,6 +561,9 @@ _slang_link(GLcontext *ctx,
 
 
    if (fragProg && shProg->FragmentProgram) {
+      /* Compute initial program's TexturesUsed info */
+      _mesa_update_shader_textures_used(&shProg->FragmentProgram->Base);
+
       /* notify driver that a new fragment program has been compiled/linked */
       ctx->Driver.ProgramStringNotify(ctx, GL_FRAGMENT_PROGRAM_ARB,
                                       &shProg->FragmentProgram->Base);
@@ -576,6 +579,9 @@ _slang_link(GLcontext *ctx,
    }
 
    if (vertProg && shProg->VertexProgram) {
+      /* Compute initial program's TexturesUsed info */
+      _mesa_update_shader_textures_used(&shProg->VertexProgram->Base);
+
       /* notify driver that a new vertex program has been compiled/linked */
       ctx->Driver.ProgramStringNotify(ctx, GL_VERTEX_PROGRAM_ARB,
                                       &shProg->VertexProgram->Base);
diff --git a/src/mesa/state_tracker/st_program.c b/src/mesa/state_tracker/st_program.c
index 55b52c3745..af0df22dc5 100644
--- a/src/mesa/state_tracker/st_program.c
+++ b/src/mesa/state_tracker/st_program.c
@@ -512,3 +512,19 @@ st_translate_fragment_program(struct st_context *st,
       tgsi_dump( fs.tokens, 0/*TGSI_DUMP_VERBOSE*/ );
 }
 
+
+/**
+ * Debug- print current shader text
+ */
+void
+st_print_shaders(GLcontext *ctx)
+{
+   struct gl_shader_program *shProg = ctx->Shader.CurrentProgram;
+   if (shProg) {
+      GLuint i;
+      for (i = 0; i < shProg->NumShaders; i++) {
+         printf("GLSL shader %u of %u:\n", i, shProg->NumShaders);
+         printf("%s\n", shProg->Shaders[i]->Source);
+      }
+   }
+}
diff --git a/src/mesa/state_tracker/st_program.h b/src/mesa/state_tracker/st_program.h
index 078e2c42a6..e2e5eddef2 100644
--- a/src/mesa/state_tracker/st_program.h
+++ b/src/mesa/state_tracker/st_program.h
@@ -151,4 +151,8 @@ st_translate_vertex_program(struct st_context *st,
                             const ubyte *fs_input_semantic_index);
 
 
+extern void
+st_print_shaders(GLcontext *ctx);
+
+
 #endif