Merge remote branch 'origin/master' into glsl2

This pulls in multiple i965 driver fixes which will help ensure better
testing coverage during development, and also gets past the conflicts
of the src/mesa/shader -> src/mesa/program move.

Conflicts:
	src/mesa/Makefile
	src/mesa/main/shaderapi.c
	src/mesa/main/shaderobj.h

1 files changed, 22 insertions, 17 deletions

diff --git a/src/gallium/drivers/llvmpipe/lp_perf.c b/src/gallium/drivers/llvmpipe/lp_perf.c
index a316597675c..083e7e30a5b 100644
--- a/src/gallium/drivers/llvmpipe/lp_perf.c
+++ b/src/gallium/drivers/llvmpipe/lp_perf.c
@@ -46,10 +46,10 @@ lp_print_counters(void)
 {
    if (LP_DEBUG & DEBUG_COUNTERS) {
       unsigned total_64, total_16, total_4;
-      float p1, p2, p3;
+      float p1, p2, p3, p4;
 
-      debug_printf("llvmpipe: nr_triangles:               %9u\n", lp_count.nr_tris);
-      debug_printf("llvmpipe: nr_culled_triangles:        %9u\n", lp_count.nr_culled_tris);
+      debug_printf("llvmpipe: nr_triangles:                 %9u\n", lp_count.nr_tris);
+      debug_printf("llvmpipe: nr_culled_triangles:          %9u\n", lp_count.nr_culled_tris);
 
       total_64 = (lp_count.nr_empty_64 + 
                   lp_count.nr_fully_covered_64 +
@@ -58,10 +58,13 @@ lp_print_counters(void)
       p1 = 100.0 * (float) lp_count.nr_empty_64 / (float) total_64;
       p2 = 100.0 * (float) lp_count.nr_fully_covered_64 / (float) total_64;
       p3 = 100.0 * (float) lp_count.nr_partially_covered_64 / (float) total_64;
+      p4 = 100.0 * (float) lp_count.nr_shade_opaque_64 / (float) total_64;
 
-      debug_printf("llvmpipe: nr_empty_64x64:             %9u (%2.0f%% of %u)\n", lp_count.nr_empty_64, p1, total_64);
-      debug_printf("llvmpipe: nr_fully_covered_64x64:     %9u (%2.0f%% of %u)\n", lp_count.nr_fully_covered_64, p2, total_64);
-      debug_printf("llvmpipe: nr_partially_covered_64x64: %9u (%2.0f%% of %u)\n", lp_count.nr_partially_covered_64, p3, total_64);
+      debug_printf("llvmpipe: nr_64x64:                     %9u\n", total_64);
+      debug_printf("llvmpipe:   nr_fully_covered_64x64:     %9u (%3.0f%% of %u)\n", lp_count.nr_fully_covered_64, p2, total_64);
+      debug_printf("llvmpipe:     nr_shade_opaque_64x64:    %9u (%3.0f%% of %u)\n", lp_count.nr_shade_opaque_64, p4, total_64);
+      debug_printf("llvmpipe:   nr_partially_covered_64x64: %9u (%3.0f%% of %u)\n", lp_count.nr_partially_covered_64, p3, total_64);
+      debug_printf("llvmpipe:   nr_empty_64x64:             %9u (%3.0f%% of %u)\n", lp_count.nr_empty_64, p1, total_64);
 
       total_16 = (lp_count.nr_empty_16 + 
                   lp_count.nr_fully_covered_16 +
@@ -71,25 +74,27 @@ lp_print_counters(void)
       p2 = 100.0 * (float) lp_count.nr_fully_covered_16 / (float) total_16;
       p3 = 100.0 * (float) lp_count.nr_partially_covered_16 / (float) total_16;
 
-      debug_printf("llvmpipe: nr_empty_16x16:             %9u (%2.0f%% of %u)\n", lp_count.nr_empty_16, p1, total_16);
-      debug_printf("llvmpipe: nr_fully_covered_16x16:     %9u (%2.0f%% of %u)\n", lp_count.nr_fully_covered_16, p2, total_16);
-      debug_printf("llvmpipe: nr_partially_covered_16x16: %9u (%2.0f%% of %u)\n", lp_count.nr_partially_covered_16, p3, total_16);
+      debug_printf("llvmpipe: nr_16x16:                     %9u\n", total_16);
+      debug_printf("llvmpipe:   nr_fully_covered_16x16:     %9u (%3.0f%% of %u)\n", lp_count.nr_fully_covered_16, p2, total_16);
+      debug_printf("llvmpipe:   nr_partially_covered_16x16: %9u (%3.0f%% of %u)\n", lp_count.nr_partially_covered_16, p3, total_16);
+      debug_printf("llvmpipe:   nr_empty_16x16:             %9u (%3.0f%% of %u)\n", lp_count.nr_empty_16, p1, total_16);
 
       total_4 = (lp_count.nr_empty_4 + lp_count.nr_non_empty_4);
 
       p1 = 100.0 * (float) lp_count.nr_empty_4 / (float) total_4;
       p2 = 100.0 * (float) lp_count.nr_non_empty_4 / (float) total_4;
 
-      debug_printf("llvmpipe: nr_empty_4x4:               %9u (%2.0f%% of %u)\n", lp_count.nr_empty_4, p1, total_4);
-      debug_printf("llvmpipe: nr_non_empty_4x4:           %9u (%2.0f%% of %u)\n", lp_count.nr_non_empty_4, p2, total_4);
+      debug_printf("llvmpipe: nr_4x4:                       %9u\n", total_4);
+      debug_printf("llvmpipe:   nr_empty_4x4:               %9u (%3.0f%% of %u)\n", lp_count.nr_empty_4, p1, total_4);
+      debug_printf("llvmpipe:   nr_non_empty_4x4:           %9u (%3.0f%% of %u)\n", lp_count.nr_non_empty_4, p2, total_4);
 
-      debug_printf("llvmpipe: nr_color_tile_clear:        %9u\n", lp_count.nr_color_tile_clear);
-      debug_printf("llvmpipe: nr_color_tile_load:         %9u\n", lp_count.nr_color_tile_load);
-      debug_printf("llvmpipe: nr_color_tile_store:        %9u\n", lp_count.nr_color_tile_store);
+      debug_printf("llvmpipe: nr_color_tile_clear:          %9u\n", lp_count.nr_color_tile_clear);
+      debug_printf("llvmpipe: nr_color_tile_load:           %9u\n", lp_count.nr_color_tile_load);
+      debug_printf("llvmpipe: nr_color_tile_store:          %9u\n", lp_count.nr_color_tile_store);
 
-      debug_printf("llvmpipe: nr_llvm_compiles:           %u\n", lp_count.nr_llvm_compiles);
-      debug_printf("llvmpipe: total LLVM compile time:    %.2f sec\n", lp_count.llvm_compile_time / 1000000.0);
-      debug_printf("llvmpipe: average LLVM compile time:  %.2f sec\n", lp_count.llvm_compile_time / 1000000.0 / lp_count.nr_llvm_compiles);
+      debug_printf("llvmpipe: nr_llvm_compiles:             %u\n", lp_count.nr_llvm_compiles);
+      debug_printf("llvmpipe: total LLVM compile time:      %.2f sec\n", lp_count.llvm_compile_time / 1000000.0);
+      debug_printf("llvmpipe: average LLVM compile time:    %.2f sec\n", lp_count.llvm_compile_time / 1000000.0 / lp_count.nr_llvm_compiles);
 
    }
 }