intel: Use GTT mapping when available for swrast.

This improves piglit quick.tests runtime from 19:33 minutes to 6:06 on
my GM45.  It should also hide most of the A17 swizzling issues, though
they'll still exist when swapping occurs (which is the kernel's problem
either way).

3 files changed, 88 insertions, 1 deletions

diff --git a/src/mesa/drivers/dri/intel/intel_depthtmp.h b/src/mesa/drivers/dri/intel/intel_depthtmp.h
index 16d77084530..a9c75d44cf3 100644
--- a/src/mesa/drivers/dri/intel/intel_depthtmp.h
+++ b/src/mesa/drivers/dri/intel/intel_depthtmp.h
@@ -31,6 +31,16 @@
  */
 
 #define VALUE_TYPE INTEL_VALUE_TYPE
+#define WRITE_DEPTH(_x, _y, d) \
+   (*(INTEL_VALUE_TYPE *)(irb->region->buffer->virtual +	\
+			  NO_TILE(_x, _y)) = d)
+#define READ_DEPTH(d, _x, _y) \
+   d = *(INTEL_VALUE_TYPE *)(irb->region->buffer->virtual +	\
+			     NO_TILE(_x, _y))
+#define TAG(x) INTEL_TAG(intel_gttmap_##x)
+#include "depthtmp.h"
+
+#define VALUE_TYPE INTEL_VALUE_TYPE
 #define WRITE_DEPTH(_x, _y, d) INTEL_WRITE_DEPTH(NO_TILE(_x, _y), d)
 #define READ_DEPTH(d, _x, _y) d = INTEL_READ_DEPTH(NO_TILE(_x, _y))
 #define TAG(x) INTEL_TAG(intel##x)
diff --git a/src/mesa/drivers/dri/intel/intel_span.c b/src/mesa/drivers/dri/intel/intel_span.c
index b5e3cad3ff8..638e05f2add 100644
--- a/src/mesa/drivers/dri/intel/intel_span.c
+++ b/src/mesa/drivers/dri/intel/intel_span.c
@@ -266,8 +266,11 @@ static uint32_t y_tile_swizzle(struct intel_renderbuffer *irb,
    unsigned int num_cliprects;						\
    struct drm_clip_rect *cliprects;					\
    int x_off, y_off;							\
+   int pitch = irb->region->pitch * irb->region->cpp;			\
+   void *buf = irb->region->buffer->virtual;				\
    GLuint p;								\
    (void) p;								\
+   (void)buf; (void)pitch; /* unused for non-gttmap. */			\
    intel_get_cliprects(intel, &cliprects, &num_cliprects, &x_off, &y_off);
 
 /* XXX FBO: this is identical to the macro in spantmp2.h except we get
@@ -347,6 +350,9 @@ static uint32_t y_tile_swizzle(struct intel_renderbuffer *irb,
    unsigned int num_cliprects;						\
    struct drm_clip_rect *cliprects;					\
    int x_off, y_off;							\
+   int pitch = irb->region->pitch * irb->region->cpp;			\
+   void *buf = irb->region->buffer->virtual;				\
+   (void)buf; (void)pitch; /* unused for non-gttmap. */			\
    intel_get_cliprects(intel, &cliprects, &num_cliprects, &x_off, &y_off);
 
 
@@ -370,6 +376,15 @@ static uint32_t y_tile_swizzle(struct intel_renderbuffer *irb,
 /**
  ** 8-bit stencil function (XXX FBO: This is obsolete)
  **/
+/* XXX */
+#define WRITE_STENCIL(_x, _y, d) pwrite_8(irb, NO_TILE(_x, _y) + 3, d)
+#define READ_STENCIL(d, _x, _y) d = pread_8(irb, NO_TILE(_x, _y) + 3);
+#define TAG(x) intel_gttmap_##x##_z24_s8
+#include "stenciltmp.h"
+
+/**
+ ** 8-bit stencil function (XXX FBO: This is obsolete)
+ **/
 #define WRITE_STENCIL(_x, _y, d) pwrite_8(irb, NO_TILE(_x, _y) + 3, d)
 #define READ_STENCIL(d, _x, _y) d = pread_8(irb, NO_TILE(_x, _y) + 3);
 #define TAG(x) intel##x##_z24_s8
@@ -399,6 +414,9 @@ intel_renderbuffer_map(struct intel_context *intel, struct gl_renderbuffer *rb)
    if (irb == NULL || irb->region == NULL)
       return;
 
+   if (intel->intelScreen->kernel_exec_fencing)
+      drm_intel_gem_bo_map_gtt(irb->region->buffer);
+
    intel_set_span_functions(intel, rb);
 }
 
@@ -411,7 +429,10 @@ intel_renderbuffer_unmap(struct intel_context *intel,
    if (irb == NULL || irb->region == NULL)
       return;
 
-   clear_span_cache(irb);
+   if (intel->intelScreen->kernel_exec_fencing)
+      drm_intel_gem_bo_unmap_gtt(irb->region->buffer);
+   else
+      clear_span_cache(irb);
 
    rb->GetRow = NULL;
    rb->PutRow = NULL;
@@ -601,6 +622,56 @@ intel_set_span_functions(struct intel_context *intel,
    else
       tiling = I915_TILING_NONE;
 
+   if (intel->intelScreen->kernel_exec_fencing) {
+      switch (irb->texformat) {
+      case MESA_FORMAT_RGB565:
+	 intel_gttmap_InitPointers_RGB565(rb);
+	 break;
+      case MESA_FORMAT_ARGB4444:
+	 intel_gttmap_InitPointers_ARGB4444(rb);
+	 break;
+      case MESA_FORMAT_ARGB1555:
+	 intel_gttmap_InitPointers_ARGB1555(rb);
+	 break;
+      case MESA_FORMAT_XRGB8888:
+         intel_gttmap_InitPointers_xRGB8888(rb);
+	 break;
+      case MESA_FORMAT_ARGB8888:
+	 if (rb->_BaseFormat == GL_RGB) {
+	    /* XXX remove this code someday when we enable XRGB surfaces */
+	    /* 8888 RGBx */
+	    intel_gttmap_InitPointers_xRGB8888(rb);
+	 } else {
+	    intel_gttmap_InitPointers_ARGB8888(rb);
+	 }
+	 break;
+      case MESA_FORMAT_Z16:
+	 intel_gttmap_InitDepthPointers_z16(rb);
+	 break;
+      case MESA_FORMAT_X8_Z24:
+	 intel_gttmap_InitDepthPointers_z24_x8(rb);
+	 break;
+      case MESA_FORMAT_S8_Z24:
+	 /* There are a few different ways SW asks us to access the S8Z24 data:
+	  * Z24 depth-only depth reads
+	  * S8Z24 depth reads
+	  * S8Z24 stencil reads.
+	  */
+	 if (rb->Format == MESA_FORMAT_S8_Z24) {
+	    intel_gttmap_InitDepthPointers_z24_x8(rb);
+	 } else if (rb->Format == MESA_FORMAT_S8) {
+	    intel_gttmap_InitStencilPointers_z24_s8(rb);
+	 }
+	 break;
+      default:
+	 _mesa_problem(NULL,
+		       "Unexpected MesaFormat %d in intelSetSpanFunctions",
+		       irb->texformat);
+	 break;
+      }
+      return;
+   }
+
    switch (irb->texformat) {
    case MESA_FORMAT_RGB565:
       switch (tiling) {
diff --git a/src/mesa/drivers/dri/intel/intel_spantmp.h b/src/mesa/drivers/dri/intel/intel_spantmp.h
index ead0b1c1683..98cc6585f4a 100644
--- a/src/mesa/drivers/dri/intel/intel_spantmp.h
+++ b/src/mesa/drivers/dri/intel/intel_spantmp.h
@@ -32,6 +32,12 @@
 
 #define SPANTMP_PIXEL_FMT INTEL_PIXEL_FMT
 #define SPANTMP_PIXEL_TYPE INTEL_PIXEL_TYPE
+#define TAG(x) INTEL_TAG(intel_gttmap_##x)
+#define TAG2(x, y) INTEL_TAG(intel_gttmap_##x##y)
+#include "spantmp2.h"
+
+#define SPANTMP_PIXEL_FMT INTEL_PIXEL_FMT
+#define SPANTMP_PIXEL_TYPE INTEL_PIXEL_TYPE
 #define PUT_VALUE(_x, _y, v) INTEL_WRITE_VALUE(NO_TILE(_x, _y), v)
 #define GET_VALUE(_x, _y) INTEL_READ_VALUE(NO_TILE(_x, _y))
 #define TAG(x) INTEL_TAG(intel##x)