intel: Do the drawable x/y offset in intel_renderbuffer_map() for spans.
[mesa.git] / src / mesa / drivers / dri / intel / intel_span.c
index 725ba5c97d6bf97824255ca898cdd18e576f2b8d..9343f40eef0e6f1ffeef41617e1e290e8bfff6d0 100644 (file)
@@ -1,6 +1,7 @@
 /**************************************************************************
  * 
  * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * Copyright 2011 Intel Corporation
  * All Rights Reserved.
  * 
  * Permission is hereby granted, free of charge, to any person obtaining a
  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  * 
+ * Authors:
+ *     Chad Versace <chad@chad-versace.us>
+ *
  **************************************************************************/
 
+#include <stdbool.h>
+#include <stdint.h>
 #include "main/glheader.h"
 #include "main/macros.h"
 #include "main/mtypes.h"
 #include "main/colormac.h"
+#include "main/renderbuffer.h"
 
 #include "intel_buffers.h"
 #include "intel_fbo.h"
@@ -43,368 +50,128 @@ static void
 intel_set_span_functions(struct intel_context *intel,
                         struct gl_renderbuffer *rb);
 
-#define SPAN_CACHE_SIZE                4096
-
-static void
-get_span_cache(struct intel_renderbuffer *irb, uint32_t offset)
-{
-   if (irb->span_cache == NULL) {
-      irb->span_cache = _mesa_malloc(SPAN_CACHE_SIZE);
-      irb->span_cache_offset = -1;
-   }
-
-   if ((offset & ~(SPAN_CACHE_SIZE - 1)) != irb->span_cache_offset) {
-      irb->span_cache_offset = offset & ~(SPAN_CACHE_SIZE - 1);
-      dri_bo_get_subdata(irb->region->buffer, irb->span_cache_offset,
-                        SPAN_CACHE_SIZE, irb->span_cache);
-   }
-}
-
-static void
-clear_span_cache(struct intel_renderbuffer *irb)
-{
-   irb->span_cache_offset = -1;
-}
-
-static uint32_t
-pread_32(struct intel_renderbuffer *irb, uint32_t offset)
-{
-   get_span_cache(irb, offset);
-
-   return *(uint32_t *)(irb->span_cache + (offset & (SPAN_CACHE_SIZE - 1)));
-}
-
-static uint32_t
-pread_xrgb8888(struct intel_renderbuffer *irb, uint32_t offset)
-{
-   get_span_cache(irb, offset);
-
-   return *(uint32_t *)(irb->span_cache + (offset & (SPAN_CACHE_SIZE - 1))) |
-      0xff000000;
-}
-
-static uint16_t
-pread_16(struct intel_renderbuffer *irb, uint32_t offset)
-{
-   get_span_cache(irb, offset);
-
-   return *(uint16_t *)(irb->span_cache + (offset & (SPAN_CACHE_SIZE - 1)));
-}
-
-static uint8_t
-pread_8(struct intel_renderbuffer *irb, uint32_t offset)
-{
-   get_span_cache(irb, offset);
-
-   return *(uint8_t *)(irb->span_cache + (offset & (SPAN_CACHE_SIZE - 1)));
-}
-
-static void
-pwrite_32(struct intel_renderbuffer *irb, uint32_t offset, uint32_t val)
-{
-   clear_span_cache(irb);
-
-   dri_bo_subdata(irb->region->buffer, offset, 4, &val);
-}
-
-static void
-pwrite_xrgb8888(struct intel_renderbuffer *irb, uint32_t offset, uint32_t val)
-{
-   clear_span_cache(irb);
-
-   dri_bo_subdata(irb->region->buffer, offset, 3, &val);
-}
-
-static void
-pwrite_16(struct intel_renderbuffer *irb, uint32_t offset, uint16_t val)
-{
-   clear_span_cache(irb);
-
-   dri_bo_subdata(irb->region->buffer, offset, 2, &val);
-}
-
-static void
-pwrite_8(struct intel_renderbuffer *irb, uint32_t offset, uint8_t val)
-{
-   clear_span_cache(irb);
-
-   dri_bo_subdata(irb->region->buffer, offset, 1, &val);
-}
-
-static uint32_t no_tile_swizzle(struct intel_renderbuffer *irb,
-                               int x, int y)
-{
-       return (y * irb->region->pitch + x) * irb->region->cpp;
-}
-
-/*
- * Deal with tiled surfaces
- */
-
-static uint32_t x_tile_swizzle(struct intel_renderbuffer *irb,
-                              int x, int y)
-{
-       int     tile_stride;
-       int     xbyte;
-       int     x_tile_off, y_tile_off;
-       int     x_tile_number, y_tile_number;
-       int     tile_off, tile_base;
-       
-        x += irb->region->draw_x;
-        y += irb->region->draw_y;
-
-       tile_stride = (irb->region->pitch * irb->region->cpp) << 3;
-
-       xbyte = x * irb->region->cpp;
-
-       x_tile_off = xbyte & 0x1ff;
-       y_tile_off = y & 7;
-
-       x_tile_number = xbyte >> 9;
-       y_tile_number = y >> 3;
-
-       tile_off = (y_tile_off << 9) + x_tile_off;
-
-       switch (irb->region->bit_6_swizzle) {
-       case I915_BIT_6_SWIZZLE_NONE:
-          break;
-       case I915_BIT_6_SWIZZLE_9:
-          tile_off ^= ((tile_off >> 3) & 64);
-          break;
-       case I915_BIT_6_SWIZZLE_9_10:
-          tile_off ^= ((tile_off >> 3) & 64) ^ ((tile_off >> 4) & 64);
-          break;
-       case I915_BIT_6_SWIZZLE_9_11:
-          tile_off ^= ((tile_off >> 3) & 64) ^ ((tile_off >> 5) & 64);
-          break;
-       case I915_BIT_6_SWIZZLE_9_10_11:
-          tile_off ^= ((tile_off >> 3) & 64) ^ ((tile_off >> 4) & 64) ^
-             ((tile_off >> 5) & 64);
-          break;
-       default:
-          fprintf(stderr, "Unknown tile swizzling mode %d\n",
-                  irb->region->bit_6_swizzle);
-          exit(1);
-       }
-
-       tile_base = (x_tile_number << 12) + y_tile_number * tile_stride;
-
-#if 0
-       printf("(%d,%d) -> %d + %d = %d (pitch = %d, tstride = %d)\n",
-              x, y, tile_off, tile_base,
-              tile_off + tile_base,
-              irb->region->pitch, tile_stride);
-#endif
-
-       return tile_base + tile_off;
-}
-
-static uint32_t y_tile_swizzle(struct intel_renderbuffer *irb,
-                              int x, int y)
-{
-       int     tile_stride;
-       int     xbyte;
-       int     x_tile_off, y_tile_off;
-       int     x_tile_number, y_tile_number;
-       int     tile_off, tile_base;
-       
-        x += irb->region->draw_x;
-        y += irb->region->draw_y;
-
-       tile_stride = (irb->region->pitch * irb->region->cpp) << 5;
-
-       xbyte = x * irb->region->cpp;
-
-       x_tile_off = xbyte & 0x7f;
-       y_tile_off = y & 0x1f;
-
-       x_tile_number = xbyte >> 7;
-       y_tile_number = y >> 5;
-
-       tile_off = ((x_tile_off & ~0xf) << 5) + (y_tile_off << 4) +
-          (x_tile_off & 0xf);
-
-       switch (irb->region->bit_6_swizzle) {
-       case I915_BIT_6_SWIZZLE_NONE:
-          break;
-       case I915_BIT_6_SWIZZLE_9:
-          tile_off ^= ((tile_off >> 3) & 64);
-          break;
-       case I915_BIT_6_SWIZZLE_9_10:
-          tile_off ^= ((tile_off >> 3) & 64) ^ ((tile_off >> 4) & 64);
-          break;
-       case I915_BIT_6_SWIZZLE_9_11:
-          tile_off ^= ((tile_off >> 3) & 64) ^ ((tile_off >> 5) & 64);
-          break;
-       case I915_BIT_6_SWIZZLE_9_10_11:
-          tile_off ^= ((tile_off >> 3) & 64) ^ ((tile_off >> 4) & 64) ^
-             ((tile_off >> 5) & 64);
-          break;
-       default:
-          fprintf(stderr, "Unknown tile swizzling mode %d\n",
-                  irb->region->bit_6_swizzle);
-          exit(1);
-       }
-
-       tile_base = (x_tile_number << 12) + y_tile_number * tile_stride;
-
-       return tile_base + tile_off;
-}
-
-/*
-  break intelWriteRGBASpan_ARGB8888
-*/
-
 #undef DBG
 #define DBG 0
 
 #define LOCAL_VARS                                                     \
-   struct intel_context *intel = intel_context(ctx);                   \
    struct intel_renderbuffer *irb = intel_renderbuffer(rb);            \
-   const GLint yScale = ctx->DrawBuffer->Name ? 1 : -1;                        \
-   const GLint yBias = ctx->DrawBuffer->Name ? 0 : irb->Base.Height - 1;\
-   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;                           \
+   int minx = 0, miny = 0;                                             \
+   int maxx = rb->Width;                                               \
+   int maxy = rb->Height;                                              \
+   int pitch = rb->RowStride * irb->region->cpp;                       \
+   void *buf = rb->Data;                                               \
    GLuint p;                                                           \
-   (void) p;                                                           \
-   (void)buf; (void)pitch; /* unused for non-gttmap. */                        \
-   intel_get_cliprects(intel, &cliprects, &num_cliprects, &x_off, &y_off);
+   (void) p;
 
-/* XXX FBO: this is identical to the macro in spantmp2.h except we get
- * the cliprect info from the context, not the driDrawable.
- * Move this into spantmp2.h someday.
- */
-#define HW_CLIPLOOP()                                                  \
-   do {                                                                        \
-      int _nc = num_cliprects;                                         \
-      while ( _nc-- ) {                                                        \
-        int minx = cliprects[_nc].x1 - x_off;                          \
-        int miny = cliprects[_nc].y1 - y_off;                          \
-        int maxx = cliprects[_nc].x2 - x_off;                          \
-        int maxy = cliprects[_nc].y2 - y_off;
-       
-#if 0
-      }}
-#endif
-
-#define Y_FLIP(_y) ((_y) * yScale + yBias)
+#define HW_CLIPLOOP()
+#define HW_ENDCLIPLOOP()
+
+#define Y_FLIP(_y) (_y)
 
 #define HW_LOCK()
 
 #define HW_UNLOCK()
 
-/* Convenience macros to avoid typing the swizzle argument over and over */
-#define NO_TILE(_X, _Y) no_tile_swizzle(irb, (_X) + x_off, (_Y) + y_off)
-#define X_TILE(_X, _Y) x_tile_swizzle(irb, (_X) + x_off, (_Y) + y_off)
-#define Y_TILE(_X, _Y) y_tile_swizzle(irb, (_X) + x_off, (_Y) + y_off)
-
 /* r5g6b5 color span and pixel functions */
-#define INTEL_PIXEL_FMT GL_RGB
-#define INTEL_PIXEL_TYPE GL_UNSIGNED_SHORT_5_6_5
-#define INTEL_READ_VALUE(offset) pread_16(irb, offset)
-#define INTEL_WRITE_VALUE(offset, v) pwrite_16(irb, offset, v)
-#define INTEL_TAG(x) x##_RGB565
-#include "intel_spantmp.h"
+#define SPANTMP_PIXEL_FMT GL_RGB
+#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_SHORT_5_6_5
+#define TAG(x) intel_##x##_RGB565
+#define TAG2(x,y) intel_##x##y_RGB565
+#include "spantmp2.h"
 
 /* a4r4g4b4 color span and pixel functions */
-#define INTEL_PIXEL_FMT GL_BGRA
-#define INTEL_PIXEL_TYPE GL_UNSIGNED_SHORT_4_4_4_4_REV
-#define INTEL_READ_VALUE(offset) pread_16(irb, offset)
-#define INTEL_WRITE_VALUE(offset, v) pwrite_16(irb, offset, v)
-#define INTEL_TAG(x) x##_ARGB4444
-#include "intel_spantmp.h"
+#define SPANTMP_PIXEL_FMT GL_BGRA
+#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_SHORT_4_4_4_4_REV
+#define TAG(x) intel_##x##_ARGB4444
+#define TAG2(x,y) intel_##x##y_ARGB4444
+#include "spantmp2.h"
 
 /* a1r5g5b5 color span and pixel functions */
-#define INTEL_PIXEL_FMT GL_BGRA
-#define INTEL_PIXEL_TYPE GL_UNSIGNED_SHORT_1_5_5_5_REV
-#define INTEL_READ_VALUE(offset) pread_16(irb, offset)
-#define INTEL_WRITE_VALUE(offset, v) pwrite_16(irb, offset, v)
-#define INTEL_TAG(x) x##_ARGB1555
-#include "intel_spantmp.h"
+#define SPANTMP_PIXEL_FMT GL_BGRA
+#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_SHORT_1_5_5_5_REV
+#define TAG(x) intel_##x##_ARGB1555
+#define TAG2(x,y) intel_##x##y##_ARGB1555
+#include "spantmp2.h"
 
 /* a8r8g8b8 color span and pixel functions */
-#define INTEL_PIXEL_FMT GL_BGRA
-#define INTEL_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV
-#define INTEL_READ_VALUE(offset) pread_32(irb, offset)
-#define INTEL_WRITE_VALUE(offset, v) pwrite_32(irb, offset, v)
-#define INTEL_TAG(x) x##_ARGB8888
-#include "intel_spantmp.h"
+#define SPANTMP_PIXEL_FMT GL_BGRA
+#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV
+#define TAG(x) intel_##x##_ARGB8888
+#define TAG2(x,y) intel_##x##y##_ARGB8888
+#include "spantmp2.h"
 
 /* x8r8g8b8 color span and pixel functions */
-#define INTEL_PIXEL_FMT GL_BGR
-#define INTEL_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV
-#define INTEL_READ_VALUE(offset) pread_xrgb8888(irb, offset)
-#define INTEL_WRITE_VALUE(offset, v) pwrite_xrgb8888(irb, offset, v)
-#define INTEL_TAG(x) x##_xRGB8888
-#include "intel_spantmp.h"
-
-#define LOCAL_DEPTH_VARS                                               \
-   struct intel_context *intel = intel_context(ctx);                   \
-   struct intel_renderbuffer *irb = intel_renderbuffer(rb);            \
-   const GLint yScale = ctx->DrawBuffer->Name ? 1 : -1;                        \
-   const GLint yBias = ctx->DrawBuffer->Name ? 0 : irb->Base.Height - 1;\
-   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);
-
-
-#define LOCAL_STENCIL_VARS LOCAL_DEPTH_VARS
-
-/* z16 depthbuffer functions. */
-#define INTEL_VALUE_TYPE GLushort
-#define INTEL_WRITE_DEPTH(offset, d) pwrite_16(irb, offset, d)
-#define INTEL_READ_DEPTH(offset) pread_16(irb, offset)
-#define INTEL_TAG(name) name##_z16
-#include "intel_depthtmp.h"
-
-/* z24x8 depthbuffer functions. */
-#define INTEL_VALUE_TYPE GLuint
-#define INTEL_WRITE_DEPTH(offset, d) pwrite_32(irb, offset, d)
-#define INTEL_READ_DEPTH(offset) pread_32(irb, offset)
-#define INTEL_TAG(name) name##_z24_x8
-#include "intel_depthtmp.h"
+#define SPANTMP_PIXEL_FMT GL_BGR
+#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV
+#define TAG(x) intel_##x##_xRGB8888
+#define TAG2(x,y) intel_##x##y##_xRGB8888
+#include "spantmp2.h"
+
+/* a8 color span and pixel functions */
+#define SPANTMP_PIXEL_FMT GL_ALPHA
+#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_BYTE
+#define TAG(x) intel_##x##_A8
+#define TAG2(x,y) intel_##x##y##_A8
+#include "spantmp2.h"
+
+/* ------------------------------------------------------------------------- */
+/* s8 stencil span and pixel functions                                       */
+/* ------------------------------------------------------------------------- */
 
+/*
+ * HAVE_HW_STENCIL_SPANS determines if stencil buffer read/writes are done with
+ * memcpy or for loops. Since the stencil buffer is interleaved, memcpy won't
+ * work.
+ */
+#define HAVE_HW_STENCIL_SPANS 0
 
-/**
- ** 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"
+#define LOCAL_STENCIL_VARS                                             \
+   (void) ctx;                                                         \
+   int minx = 0;                                                       \
+   int miny = 0;                                                       \
+   int maxx = rb->Width;                                               \
+   int maxy = rb->Height;                                              \
+   int stride = rb->RowStride;                                         \
+   uint8_t *buf = rb->Data;                                            \
 
-/**
- ** 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
-#include "stenciltmp.h"
+/* Don't flip y. */
+#undef Y_FLIP
+#define Y_FLIP(y) y
 
 /**
- ** 8-bit x-tile stencil function (XXX FBO: This is obsolete)
- **/
-#define WRITE_STENCIL(_x, _y, d) pwrite_8(irb, X_TILE(_x, _y) + 3, d)
-#define READ_STENCIL(d, _x, _y) d = pread_8(irb, X_TILE(_x, _y) + 3);
-#define TAG(x) intel_XTile_##x##_z24_s8
-#include "stenciltmp.h"
+ * \brief Get pointer offset into stencil buffer.
+ *
+ * The stencil buffer interleaves two rows into one. Yay for crazy hardware.
+ * The table below demonstrates how the pointer arithmetic behaves for a buffer
+ * with positive stride (s=stride).
+ *
+ *     x    | y     | byte offset
+ *     --------------------------
+ *     0    | 0     | 0
+ *     0    | 0     | 1
+ *     1    | 0     | 2
+ *     1    | 1     | 3
+ *     ...  | ...   | ...
+ *     0    | 2     | s
+ *     0    | 3     | s + 1
+ *     1    | 2     | s + 2
+ *     1    | 3     | s + 3
+ *
+ *
+ */
+static inline intptr_t
+intel_offset_S8(int stride, GLint x, GLint y)
+{
+   return 2 * ((y / 2) * stride + x) + y % 2;
+}
 
-/**
- ** 8-bit y-tile stencil function (XXX FBO: This is obsolete)
- **/
-#define WRITE_STENCIL(_x, _y, d) pwrite_8(irb, Y_TILE(_x, _y) + 3, d)
-#define READ_STENCIL(d, _x, _y) d = pread_8(irb, Y_TILE(_x, _y) + 3)
-#define TAG(x) intel_YTile_##x##_z24_s8
+#define WRITE_STENCIL(x, y, src)  buf[intel_offset_S8(stride, x, y)] = src;
+#define READ_STENCIL(dest, x, y) dest = buf[intel_offset_S8(stride, x, y)]
+#define TAG(x) intel_##x##_S8
 #include "stenciltmp.h"
 
+/* ------------------------------------------------------------------------- */
+
 void
 intel_renderbuffer_map(struct intel_context *intel, struct gl_renderbuffer *rb)
 {
@@ -413,8 +180,22 @@ 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);
+   drm_intel_gem_bo_map_gtt(irb->region->buffer);
+
+   rb->Data = irb->region->buffer->virtual;
+   rb->RowStride = irb->region->pitch;
+
+   if (!rb->Name) {
+      /* Flip orientation of the window system buffer */
+      rb->Data += rb->RowStride * (irb->region->height - 1) * irb->region->cpp;
+      rb->RowStride = -rb->RowStride;
+   } else {
+      /* Adjust the base pointer of a texture image drawbuffer to the image
+       * within the miptree region (all else has draw_x/y = 0).
+       */
+      rb->Data += irb->draw_x * irb->region->cpp;
+      rb->Data += irb->draw_y * rb->RowStride * irb->region->cpp;
+   }
 
    intel_set_span_functions(intel, rb);
 }
@@ -428,80 +209,34 @@ intel_renderbuffer_unmap(struct intel_context *intel,
    if (irb == NULL || irb->region == NULL)
       return;
 
-   if (intel->intelScreen->kernel_exec_fencing)
-      drm_intel_gem_bo_unmap_gtt(irb->region->buffer);
-   else
-      clear_span_cache(irb);
+   drm_intel_gem_bo_unmap_gtt(irb->region->buffer);
 
    rb->GetRow = NULL;
    rb->PutRow = NULL;
+   rb->Data = NULL;
+   rb->RowStride = 0;
 }
 
-/**
- * Map or unmap all the renderbuffers which we may need during
- * software rendering.
- * XXX in the future, we could probably convey extra information to
- * reduce the number of mappings needed.  I.e. if doing a glReadPixels
- * from the depth buffer, we really only need one mapping.
- *
- * XXX Rewrite this function someday.
- * We can probably just loop over all the renderbuffer attachments,
- * map/unmap all of them, and not worry about the _ColorDrawBuffers
- * _ColorReadBuffer, _DepthBuffer or _StencilBuffer fields.
- */
 static void
-intel_map_unmap_framebuffer(struct intel_context *intel,
-                           struct gl_framebuffer *fb,
-                           GLboolean map)
+intel_framebuffer_map(struct intel_context *intel, struct gl_framebuffer *fb)
 {
-   GLuint i;
-
-   /* color draw buffers */
-   for (i = 0; i < fb->_NumColorDrawBuffers; i++) {
-      if (map)
-         intel_renderbuffer_map(intel, fb->_ColorDrawBuffers[i]);
-      else
-         intel_renderbuffer_unmap(intel, fb->_ColorDrawBuffers[i]);
-   }
-
-   /* color read buffer */
-   if (map)
-      intel_renderbuffer_map(intel, fb->_ColorReadBuffer);
-   else
-      intel_renderbuffer_unmap(intel, fb->_ColorReadBuffer);
+   int i;
 
-   /* check for render to textures */
    for (i = 0; i < BUFFER_COUNT; i++) {
-      struct gl_renderbuffer_attachment *att =
-         fb->Attachment + i;
-      struct gl_texture_object *tex = att->Texture;
-      if (tex) {
-         /* render to texture */
-         ASSERT(att->Renderbuffer);
-         if (map)
-            intel_tex_map_images(intel, intel_texture_object(tex));
-         else
-            intel_tex_unmap_images(intel, intel_texture_object(tex));
-      }
+      intel_renderbuffer_map(intel, fb->Attachment[i].Renderbuffer);
    }
 
-   /* depth buffer (Note wrapper!) */
-   if (fb->_DepthBuffer) {
-      if (map)
-         intel_renderbuffer_map(intel, fb->_DepthBuffer->Wrapped);
-      else
-         intel_renderbuffer_unmap(intel, fb->_DepthBuffer->Wrapped);
-   }
+   intel_check_front_buffer_rendering(intel);
+}
 
-   /* stencil buffer (Note wrapper!) */
-   if (fb->_StencilBuffer) {
-      if (map)
-         intel_renderbuffer_map(intel, fb->_StencilBuffer->Wrapped);
-      else
-         intel_renderbuffer_unmap(intel, fb->_StencilBuffer->Wrapped);
-   }
+static void
+intel_framebuffer_unmap(struct intel_context *intel, struct gl_framebuffer *fb)
+{
+   int i;
 
-   intel_check_front_buffer_rendering(intel);
+   for (i = 0; i < BUFFER_COUNT; i++) {
+      intel_renderbuffer_unmap(intel, fb->Attachment[i].Renderbuffer);
+   }
 }
 
 /**
@@ -511,24 +246,27 @@ intel_map_unmap_framebuffer(struct intel_context *intel,
  * Old note: Moved locking out to get reasonable span performance.
  */
 void
-intelSpanRenderStart(GLcontext * ctx)
+intelSpanRenderStart(struct gl_context * ctx)
 {
    struct intel_context *intel = intel_context(ctx);
    GLuint i;
 
-   intelFlush(&intel->ctx);
-   LOCK_HARDWARE(intel);
+   intel_flush(&intel->ctx);
+   intel_prepare_render(intel);
 
    for (i = 0; i < ctx->Const.MaxTextureImageUnits; i++) {
       if (ctx->Texture.Unit[i]._ReallyEnabled) {
          struct gl_texture_object *texObj = ctx->Texture.Unit[i]._Current;
+
+         intel_finalize_mipmap_tree(intel, i);
          intel_tex_map_images(intel, intel_texture_object(texObj));
       }
    }
 
-   intel_map_unmap_framebuffer(intel, ctx->DrawBuffer, GL_TRUE);
-   if (ctx->ReadBuffer != ctx->DrawBuffer)
-      intel_map_unmap_framebuffer(intel, ctx->ReadBuffer, GL_TRUE);
+   intel_framebuffer_map(intel, ctx->DrawBuffer);
+   if (ctx->ReadBuffer != ctx->DrawBuffer) {
+      intel_framebuffer_map(intel, ctx->ReadBuffer);
+   }
 }
 
 /**
@@ -536,7 +274,7 @@ intelSpanRenderStart(GLcontext * ctx)
  * the above function.
  */
 void
-intelSpanRenderFinish(GLcontext * ctx)
+intelSpanRenderFinish(struct gl_context * ctx)
 {
    struct intel_context *intel = intel_context(ctx);
    GLuint i;
@@ -550,16 +288,15 @@ intelSpanRenderFinish(GLcontext * ctx)
       }
    }
 
-   intel_map_unmap_framebuffer(intel, ctx->DrawBuffer, GL_FALSE);
-   if (ctx->ReadBuffer != ctx->DrawBuffer)
-      intel_map_unmap_framebuffer(intel, ctx->ReadBuffer, GL_FALSE);
-
-   UNLOCK_HARDWARE(intel);
+   intel_framebuffer_unmap(intel, ctx->DrawBuffer);
+   if (ctx->ReadBuffer != ctx->DrawBuffer) {
+      intel_framebuffer_unmap(intel, ctx->ReadBuffer);
+   }
 }
 
 
 void
-intelInitSpanFuncs(GLcontext * ctx)
+intelInitSpanFuncs(struct gl_context * ctx)
 {
    struct swrast_device_driver *swdd = _swrast_GetDeviceDriverReference(ctx);
    swdd->SpanRenderStart = intelSpanRenderStart;
@@ -567,7 +304,7 @@ intelInitSpanFuncs(GLcontext * ctx)
 }
 
 void
-intel_map_vertex_shader_textures(GLcontext *ctx)
+intel_map_vertex_shader_textures(struct gl_context *ctx)
 {
    struct intel_context *intel = intel_context(ctx);
    int i;
@@ -586,7 +323,7 @@ intel_map_vertex_shader_textures(GLcontext *ctx)
 }
 
 void
-intel_unmap_vertex_shader_textures(GLcontext *ctx)
+intel_unmap_vertex_shader_textures(struct gl_context *ctx)
 {
    struct intel_context *intel = intel_context(ctx);
    int i;
@@ -604,6 +341,38 @@ intel_unmap_vertex_shader_textures(GLcontext *ctx)
    }
 }
 
+typedef void (*span_init_func)(struct gl_renderbuffer *rb);
+
+static span_init_func intel_span_init_funcs[MESA_FORMAT_COUNT] =
+{
+   [MESA_FORMAT_A8] = intel_InitPointers_A8,
+   [MESA_FORMAT_RGB565] = intel_InitPointers_RGB565,
+   [MESA_FORMAT_ARGB4444] = intel_InitPointers_ARGB4444,
+   [MESA_FORMAT_ARGB1555] = intel_InitPointers_ARGB1555,
+   [MESA_FORMAT_XRGB8888] = intel_InitPointers_xRGB8888,
+   [MESA_FORMAT_ARGB8888] = intel_InitPointers_ARGB8888,
+   [MESA_FORMAT_SARGB8] = intel_InitPointers_ARGB8888,
+   [MESA_FORMAT_Z16] = _mesa_set_renderbuffer_accessors,
+   [MESA_FORMAT_X8_Z24] = _mesa_set_renderbuffer_accessors,
+   [MESA_FORMAT_S8_Z24] = _mesa_set_renderbuffer_accessors,
+   [MESA_FORMAT_S8] = intel_InitStencilPointers_S8,
+   [MESA_FORMAT_R8] = _mesa_set_renderbuffer_accessors,
+   [MESA_FORMAT_RG88] = _mesa_set_renderbuffer_accessors,
+   [MESA_FORMAT_R16] = _mesa_set_renderbuffer_accessors,
+   [MESA_FORMAT_RG1616] = _mesa_set_renderbuffer_accessors,
+   [MESA_FORMAT_RGBA_FLOAT32] = _mesa_set_renderbuffer_accessors,
+   [MESA_FORMAT_RG_FLOAT32] = _mesa_set_renderbuffer_accessors,
+   [MESA_FORMAT_R_FLOAT32] = _mesa_set_renderbuffer_accessors,
+   [MESA_FORMAT_INTENSITY_FLOAT32] = _mesa_set_renderbuffer_accessors,
+   [MESA_FORMAT_LUMINANCE_FLOAT32] = _mesa_set_renderbuffer_accessors,
+};
+
+bool
+intel_span_supports_format(gl_format format)
+{
+   return intel_span_init_funcs[format] != NULL;
+}
+
 /**
  * Plug in appropriate span read/write functions for the given renderbuffer.
  * These are used for the software fallbacks.
@@ -613,210 +382,7 @@ intel_set_span_functions(struct intel_context *intel,
                         struct gl_renderbuffer *rb)
 {
    struct intel_renderbuffer *irb = (struct intel_renderbuffer *) rb;
-   uint32_t tiling;
-
-   /* If in GEM mode, we need to do the tile address swizzling ourselves,
-    * instead of the fence registers handling it.
-    */
-   if (intel->ttm)
-      tiling = irb->region->tiling;
-   else
-      tiling = I915_TILING_NONE;
-
-   if (intel->intelScreen->kernel_exec_fencing) {
-      switch (irb->Base.Format) {
-      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->Base.Format);
-        break;
-      }
-      return;
-   }
 
-   switch (irb->Base.Format) {
-   case MESA_FORMAT_RGB565:
-      switch (tiling) {
-      case I915_TILING_NONE:
-      default:
-        intelInitPointers_RGB565(rb);
-        break;
-      case I915_TILING_X:
-        intel_XTile_InitPointers_RGB565(rb);
-        break;
-      case I915_TILING_Y:
-        intel_YTile_InitPointers_RGB565(rb);
-        break;
-      }
-      break;
-   case MESA_FORMAT_ARGB4444:
-      switch (tiling) {
-      case I915_TILING_NONE:
-      default:
-        intelInitPointers_ARGB4444(rb);
-        break;
-      case I915_TILING_X:
-        intel_XTile_InitPointers_ARGB4444(rb);
-        break;
-      case I915_TILING_Y:
-        intel_YTile_InitPointers_ARGB4444(rb);
-        break;
-      }
-      break;
-   case MESA_FORMAT_ARGB1555:
-      switch (tiling) {
-      case I915_TILING_NONE:
-      default:
-        intelInitPointers_ARGB1555(rb);
-        break;
-      case I915_TILING_X:
-        intel_XTile_InitPointers_ARGB1555(rb);
-        break;
-      case I915_TILING_Y:
-        intel_YTile_InitPointers_ARGB1555(rb);
-        break;
-      }
-      break;
-   case MESA_FORMAT_XRGB8888:
-      switch (tiling) {
-      case I915_TILING_NONE:
-      default:
-         intelInitPointers_xRGB8888(rb);
-         break;
-      case I915_TILING_X:
-         intel_XTile_InitPointers_xRGB8888(rb);
-         break;
-      case I915_TILING_Y:
-         intel_YTile_InitPointers_xRGB8888(rb);
-         break;
-      }
-      break;
-   case MESA_FORMAT_ARGB8888:
-      if (rb->_BaseFormat == GL_RGB) {
-         /* XXX remove this code someday when we enable XRGB surfaces */
-        /* 8888 RGBx */
-        switch (tiling) {
-        case I915_TILING_NONE:
-        default:
-           intelInitPointers_xRGB8888(rb);
-           break;
-        case I915_TILING_X:
-           intel_XTile_InitPointers_xRGB8888(rb);
-           break;
-        case I915_TILING_Y:
-           intel_YTile_InitPointers_xRGB8888(rb);
-           break;
-        }
-      } else {
-        /* 8888 RGBA */
-        switch (tiling) {
-        case I915_TILING_NONE:
-        default:
-           intelInitPointers_ARGB8888(rb);
-           break;
-        case I915_TILING_X:
-           intel_XTile_InitPointers_ARGB8888(rb);
-           break;
-        case I915_TILING_Y:
-           intel_YTile_InitPointers_ARGB8888(rb);
-           break;
-        }
-      }
-      break;
-   case MESA_FORMAT_Z16:
-      switch (tiling) {
-      case I915_TILING_NONE:
-      default:
-        intelInitDepthPointers_z16(rb);
-        break;
-      case I915_TILING_X:
-        intel_XTile_InitDepthPointers_z16(rb);
-        break;
-      case I915_TILING_Y:
-        intel_YTile_InitDepthPointers_z16(rb);
-        break;
-      }
-      break;
-   case MESA_FORMAT_X8_Z24:
-   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) {
-        switch (tiling) {
-        case I915_TILING_NONE:
-        default:
-           intelInitDepthPointers_z24_x8(rb);
-           break;
-        case I915_TILING_X:
-           intel_XTile_InitDepthPointers_z24_x8(rb);
-           break;
-        case I915_TILING_Y:
-           intel_YTile_InitDepthPointers_z24_x8(rb);
-           break;
-        }
-      } else if (rb->Format == MESA_FORMAT_S8) {
-        switch (tiling) {
-        case I915_TILING_NONE:
-        default:
-           intelInitStencilPointers_z24_s8(rb);
-           break;
-        case I915_TILING_X:
-           intel_XTile_InitStencilPointers_z24_s8(rb);
-           break;
-        case I915_TILING_Y:
-           intel_YTile_InitStencilPointers_z24_s8(rb);
-           break;
-        }
-      } else {
-        _mesa_problem(NULL,
-                      "Unexpected ActualFormat in intelSetSpanFunctions");
-      }
-      break;
-   default:
-      _mesa_problem(NULL,
-                    "Unexpected MesaFormat in intelSetSpanFunctions");
-      break;
-   }
+   assert(intel_span_init_funcs[irb->Base.Format]);
+   intel_span_init_funcs[irb->Base.Format](rb);
 }