ilo: add 3D pipeline for GEN6
authorChia-I Wu <olv@lunarg.com>
Wed, 12 Dec 2012 21:28:42 +0000 (05:28 +0800)
committerChia-I Wu <olvaffe@gmail.com>
Fri, 26 Apr 2013 08:16:43 +0000 (16:16 +0800)
The 3D pipeline is a high-level interface to emit 3D commands and states.  It
uses GEN6 GPE to do the real work.

src/gallium/drivers/ilo/Makefile.sources
src/gallium/drivers/ilo/ilo_3d_pipeline.c [new file with mode: 0644]
src/gallium/drivers/ilo/ilo_3d_pipeline.h [new file with mode: 0644]
src/gallium/drivers/ilo/ilo_3d_pipeline_dump.c [new file with mode: 0644]
src/gallium/drivers/ilo/ilo_3d_pipeline_gen6.c [new file with mode: 0644]
src/gallium/drivers/ilo/ilo_3d_pipeline_gen6.h [new file with mode: 0644]

index f897166f763028a0c709f09e7f36bf86288d77a2..6c7ebe31512abc5938136886da541edfaf4ea67e 100644 (file)
@@ -1,5 +1,8 @@
 C_SOURCES := \
        ilo_3d.c \
+       ilo_3d_pipeline.c \
+       ilo_3d_pipeline_dump.c \
+       ilo_3d_pipeline_gen6.c \
        ilo_blit.c \
        ilo_context.c \
        ilo_cp.c \
diff --git a/src/gallium/drivers/ilo/ilo_3d_pipeline.c b/src/gallium/drivers/ilo/ilo_3d_pipeline.c
new file mode 100644 (file)
index 0000000..089db06
--- /dev/null
@@ -0,0 +1,348 @@
+/*
+ * Mesa 3-D graphics library
+ *
+ * Copyright (C) 2013 LunarG, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Chia-I Wu <olv@lunarg.com>
+ */
+
+#include "util/u_prim.h"
+#include "intel_winsys.h"
+
+#include "ilo_context.h"
+#include "ilo_cp.h"
+#include "ilo_state.h"
+#include "ilo_3d_pipeline_gen6.h"
+#include "ilo_3d_pipeline.h"
+
+/* in U0.4 */
+struct sample_position {
+   uint8_t x, y;
+};
+
+/* \see gen6_get_sample_position() */
+static const struct sample_position sample_position_1x[1] = {
+   {  8,  8 },
+};
+
+static const struct sample_position sample_position_4x[4] = {
+   {  6,  2 }, /* distance from the center is sqrt(40) */
+   { 14,  6 }, /* distance from the center is sqrt(40) */
+   {  2, 10 }, /* distance from the center is sqrt(40) */
+   { 10, 14 }, /* distance from the center is sqrt(40) */
+};
+
+static const struct sample_position sample_position_8x[8] = {
+   {  7,  9 }, /* distance from the center is sqrt(2) */
+   {  9, 13 }, /* distance from the center is sqrt(26) */
+   { 11,  3 }, /* distance from the center is sqrt(34) */
+   { 13, 11 }, /* distance from the center is sqrt(34) */
+   {  1,  7 }, /* distance from the center is sqrt(50) */
+   {  5,  1 }, /* distance from the center is sqrt(58) */
+   { 15,  5 }, /* distance from the center is sqrt(58) */
+   {  3, 15 }, /* distance from the center is sqrt(74) */
+};
+
+struct ilo_3d_pipeline *
+ilo_3d_pipeline_create(struct ilo_cp *cp, int gen, int gt)
+{
+   struct ilo_3d_pipeline *p;
+   int i;
+
+   p = CALLOC_STRUCT(ilo_3d_pipeline);
+   if (!p)
+      return NULL;
+
+   p->cp = cp;
+   p->gen = gen;
+
+   switch (p->gen) {
+   case ILO_GEN(6):
+      ilo_3d_pipeline_init_gen6(p);
+      break;
+   default:
+      assert(!"unsupported GEN");
+      FREE(p);
+      return NULL;
+      break;
+   }
+
+   p->gpe.gen = p->gen;
+   p->gpe.gt = gt;
+
+   p->invalidate_flags = ILO_3D_PIPELINE_INVALIDATE_ALL;
+
+   p->workaround_bo = p->cp->winsys->alloc_buffer(p->cp->winsys,
+         "PIPE_CONTROL workaround", 4096, 0);
+   if (!p->workaround_bo) {
+      ilo_warn("failed to allocate PIPE_CONTROL workaround bo\n");
+      FREE(p);
+      return NULL;
+   }
+
+   p->packed_sample_position_1x =
+      sample_position_1x[0].x << 4 |
+      sample_position_1x[0].y;
+
+   /* pack into dwords */
+   for (i = 0; i < 4; i++) {
+      p->packed_sample_position_4x |=
+         sample_position_4x[i].x << (8 * i + 4) |
+         sample_position_4x[i].y << (8 * i);
+
+      p->packed_sample_position_8x[0] |=
+         sample_position_8x[i].x << (8 * i + 4) |
+         sample_position_8x[i].y << (8 * i);
+
+      p->packed_sample_position_8x[1] |=
+         sample_position_8x[4 + i].x << (8 * i + 4) |
+         sample_position_8x[4 + i].y << (8 * i);
+   }
+
+   return p;
+}
+
+void
+ilo_3d_pipeline_destroy(struct ilo_3d_pipeline *p)
+{
+   if (p->workaround_bo)
+      p->workaround_bo->unreference(p->workaround_bo);
+
+   FREE(p);
+}
+
+static void
+handle_invalid_batch_bo(struct ilo_3d_pipeline *p, bool unset)
+{
+   if (p->invalidate_flags & ILO_3D_PIPELINE_INVALIDATE_BATCH_BO) {
+      if (p->gen == ILO_GEN(6))
+         p->state.has_gen6_wa_pipe_control = false;
+
+      if (unset)
+         p->invalidate_flags &= ~ILO_3D_PIPELINE_INVALIDATE_BATCH_BO;
+   }
+}
+
+/* XXX move to u_prim.h */
+static unsigned
+prim_count(unsigned prim, unsigned num_verts)
+{
+   unsigned num_prims;
+
+   u_trim_pipe_prim(prim, &num_verts);
+
+   switch (prim) {
+   case PIPE_PRIM_POINTS:
+      num_prims = num_verts;
+      break;
+   case PIPE_PRIM_LINES:
+      num_prims = num_verts / 2;
+      break;
+   case PIPE_PRIM_LINE_LOOP:
+      num_prims = num_verts;
+      break;
+   case PIPE_PRIM_LINE_STRIP:
+      num_prims = num_verts - 1;
+      break;
+   case PIPE_PRIM_TRIANGLES:
+      num_prims = num_verts / 3;
+      break;
+   case PIPE_PRIM_TRIANGLE_STRIP:
+   case PIPE_PRIM_TRIANGLE_FAN:
+      num_prims = num_verts - 2;
+      break;
+   case PIPE_PRIM_QUADS:
+      num_prims = (num_verts / 4) * 2;
+      break;
+   case PIPE_PRIM_QUAD_STRIP:
+      num_prims = (num_verts / 2 - 1) * 2;
+      break;
+   case PIPE_PRIM_POLYGON:
+      num_prims = num_verts - 2;
+      break;
+   case PIPE_PRIM_LINES_ADJACENCY:
+      num_prims = num_verts / 4;
+      break;
+   case PIPE_PRIM_LINE_STRIP_ADJACENCY:
+      num_prims = num_verts - 3;
+      break;
+   case PIPE_PRIM_TRIANGLES_ADJACENCY:
+      /* u_trim_pipe_prim is wrong? */
+      num_verts += 1;
+
+      num_prims = num_verts / 6;
+      break;
+   case PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY:
+      /* u_trim_pipe_prim is wrong? */
+      if (num_verts >= 6)
+         num_verts -= (num_verts % 2);
+      else
+         num_verts = 0;
+
+      num_prims = (num_verts / 2 - 2);
+      break;
+   default:
+      assert(!"unknown pipe prim");
+      num_prims = 0;
+      break;
+   }
+
+   return num_prims;
+}
+
+/**
+ * Emit context states and 3DPRIMITIVE.
+ */
+bool
+ilo_3d_pipeline_emit_draw(struct ilo_3d_pipeline *p,
+                          const struct ilo_context *ilo,
+                          const struct pipe_draw_info *info,
+                          int *prim_generated, int *prim_emitted)
+{
+   bool success;
+
+   /*
+    * We keep track of the SVBI in the driver, so that we can restore it when
+    * the HW context is invalidated (by another process).  The value needs to
+    * be reset when the stream output targets are changed.
+    */
+   if (ilo->dirty & ILO_DIRTY_STREAM_OUTPUT_TARGETS)
+      p->state.so_num_vertices = 0;
+
+   while (true) {
+      struct ilo_cp_jmp_buf jmp;
+      int err;
+
+      /* we will rewind if aperture check below fails */
+      ilo_cp_setjmp(p->cp, &jmp);
+
+      handle_invalid_batch_bo(p, false);
+
+      /* draw! */
+      ilo_cp_assert_no_implicit_flush(p->cp, true);
+      p->emit_draw(p, ilo, info);
+      ilo_cp_assert_no_implicit_flush(p->cp, false);
+
+      err = ilo->winsys->check_aperture_space(ilo->winsys, &p->cp->bo, 1);
+      if (!err) {
+         success = true;
+         break;
+      }
+
+      /* rewind */
+      ilo_cp_longjmp(p->cp, &jmp);
+
+      if (ilo_cp_empty(p->cp)) {
+         success = false;
+         break;
+      }
+      else {
+         /* flush and try again */
+         ilo_cp_flush(p->cp);
+      }
+   }
+
+   if (success) {
+      const int num_verts = u_vertices_per_prim(u_reduced_prim(info->mode));
+      const int max_emit =
+         (p->state.so_max_vertices - p->state.so_num_vertices) / num_verts;
+      const int generated = prim_count(info->mode, info->count);
+      const int emitted = MIN2(generated, max_emit);
+
+      p->state.so_num_vertices += emitted * num_verts;
+
+      if (prim_generated)
+         *prim_generated = generated;
+
+      if (prim_emitted)
+         *prim_emitted = emitted;
+   }
+
+   p->invalidate_flags = 0x0;
+
+   return success;
+}
+
+/**
+ * Emit PIPE_CONTROL to flush all caches.
+ */
+void
+ilo_3d_pipeline_emit_flush(struct ilo_3d_pipeline *p)
+{
+   handle_invalid_batch_bo(p, true);
+   p->emit_flush(p);
+}
+
+/**
+ * Emit PIPE_CONTROL with PIPE_CONTROL_WRITE_TIMESTAMP post-sync op.
+ */
+void
+ilo_3d_pipeline_emit_write_timestamp(struct ilo_3d_pipeline *p,
+                                     struct intel_bo *bo, int index)
+{
+   handle_invalid_batch_bo(p, true);
+   p->emit_write_timestamp(p, bo, index);
+}
+
+/**
+ * Emit PIPE_CONTROL with PIPE_CONTROL_WRITE_DEPTH_COUNT post-sync op.
+ */
+void
+ilo_3d_pipeline_emit_write_depth_count(struct ilo_3d_pipeline *p,
+                                       struct intel_bo *bo, int index)
+{
+   handle_invalid_batch_bo(p, true);
+   p->emit_write_depth_count(p, bo, index);
+}
+
+void
+ilo_3d_pipeline_get_sample_position(struct ilo_3d_pipeline *p,
+                                    unsigned sample_count,
+                                    unsigned sample_index,
+                                    float *x, float *y)
+{
+   const struct sample_position *pos;
+
+   switch (sample_count) {
+   case 1:
+      assert(sample_index < Elements(sample_position_1x));
+      pos = sample_position_1x;
+      break;
+   case 4:
+      assert(sample_index < Elements(sample_position_4x));
+      pos = sample_position_4x;
+      break;
+   case 8:
+      assert(sample_index < Elements(sample_position_8x));
+      pos = sample_position_8x;
+      break;
+   default:
+      assert(!"unknown sample count");
+      *x = 0.5f;
+      *y = 0.5f;
+      return;
+      break;
+   }
+
+   *x = (float) pos[sample_index].x / 16.0f;
+   *y = (float) pos[sample_index].y / 16.0f;
+}
diff --git a/src/gallium/drivers/ilo/ilo_3d_pipeline.h b/src/gallium/drivers/ilo/ilo_3d_pipeline.h
new file mode 100644 (file)
index 0000000..3ed3304
--- /dev/null
@@ -0,0 +1,242 @@
+/*
+ * Mesa 3-D graphics library
+ *
+ * Copyright (C) 2013 LunarG, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Chia-I Wu <olv@lunarg.com>
+ */
+
+#ifndef ILO_3D_PIPELINE_H
+#define ILO_3D_PIPELINE_H
+
+#include "ilo_common.h"
+#include "ilo_context.h"
+#include "ilo_gpe_gen6.h"
+
+struct pipe_draw_info;
+struct intel_bo;
+struct ilo_cp;
+struct ilo_context;
+
+enum ilo_3d_pipeline_invalidate_flags {
+   ILO_3D_PIPELINE_INVALIDATE_HW         = 1 << 0,
+   ILO_3D_PIPELINE_INVALIDATE_BATCH_BO   = 1 << 1,
+   ILO_3D_PIPELINE_INVALIDATE_STATE_BO   = 1 << 2,
+   ILO_3D_PIPELINE_INVALIDATE_KERNEL_BO  = 1 << 3,
+
+   ILO_3D_PIPELINE_INVALIDATE_ALL        = 0xffffffff,
+};
+
+enum ilo_3d_pipeline_action {
+   ILO_3D_PIPELINE_DRAW,
+   ILO_3D_PIPELINE_FLUSH,
+   ILO_3D_PIPELINE_WRITE_TIMESTAMP,
+   ILO_3D_PIPELINE_WRITE_DEPTH_COUNT,
+};
+
+/**
+ * 3D pipeline.
+ */
+struct ilo_3d_pipeline {
+   struct ilo_cp *cp;
+   int gen;
+
+   struct ilo_gpe gpe;
+
+   uint32_t invalidate_flags;
+
+   struct intel_bo *workaround_bo;
+
+   uint32_t packed_sample_position_1x;
+   uint32_t packed_sample_position_4x;
+   uint32_t packed_sample_position_8x[2];
+
+   int (*estimate_size)(struct ilo_3d_pipeline *pipeline,
+                        enum ilo_3d_pipeline_action action,
+                        const void *arg);
+
+   void (*emit_draw)(struct ilo_3d_pipeline *pipeline,
+                     const struct ilo_context *ilo,
+                     const struct pipe_draw_info *info);
+
+   void (*emit_flush)(struct ilo_3d_pipeline *pipeline);
+
+   void (*emit_write_timestamp)(struct ilo_3d_pipeline *pipeline,
+                                struct intel_bo *bo, int index);
+
+   void (*emit_write_depth_count)(struct ilo_3d_pipeline *pipeline,
+                                  struct intel_bo *bo, int index);
+
+   /**
+    * all GPE functions of all GENs
+    */
+#define GEN6_EMIT(name) ilo_gpe_gen6_ ## name gen6_ ## name
+   GEN6_EMIT(STATE_BASE_ADDRESS);
+   GEN6_EMIT(STATE_SIP);
+   GEN6_EMIT(PIPELINE_SELECT);
+   GEN6_EMIT(3DSTATE_BINDING_TABLE_POINTERS);
+   GEN6_EMIT(3DSTATE_SAMPLER_STATE_POINTERS);
+   GEN6_EMIT(3DSTATE_URB);
+   GEN6_EMIT(3DSTATE_VERTEX_BUFFERS);
+   GEN6_EMIT(3DSTATE_VERTEX_ELEMENTS);
+   GEN6_EMIT(3DSTATE_INDEX_BUFFER);
+   GEN6_EMIT(3DSTATE_VF_STATISTICS);
+   GEN6_EMIT(3DSTATE_VIEWPORT_STATE_POINTERS);
+   GEN6_EMIT(3DSTATE_CC_STATE_POINTERS);
+   GEN6_EMIT(3DSTATE_SCISSOR_STATE_POINTERS);
+   GEN6_EMIT(3DSTATE_VS);
+   GEN6_EMIT(3DSTATE_GS);
+   GEN6_EMIT(3DSTATE_CLIP);
+   GEN6_EMIT(3DSTATE_SF);
+   GEN6_EMIT(3DSTATE_WM);
+   GEN6_EMIT(3DSTATE_CONSTANT_VS);
+   GEN6_EMIT(3DSTATE_CONSTANT_GS);
+   GEN6_EMIT(3DSTATE_CONSTANT_PS);
+   GEN6_EMIT(3DSTATE_SAMPLE_MASK);
+   GEN6_EMIT(3DSTATE_DRAWING_RECTANGLE);
+   GEN6_EMIT(3DSTATE_DEPTH_BUFFER);
+   GEN6_EMIT(3DSTATE_POLY_STIPPLE_OFFSET);
+   GEN6_EMIT(3DSTATE_POLY_STIPPLE_PATTERN);
+   GEN6_EMIT(3DSTATE_LINE_STIPPLE);
+   GEN6_EMIT(3DSTATE_AA_LINE_PARAMETERS);
+   GEN6_EMIT(3DSTATE_GS_SVB_INDEX);
+   GEN6_EMIT(3DSTATE_MULTISAMPLE);
+   GEN6_EMIT(3DSTATE_STENCIL_BUFFER);
+   GEN6_EMIT(3DSTATE_HIER_DEPTH_BUFFER);
+   GEN6_EMIT(3DSTATE_CLEAR_PARAMS);
+   GEN6_EMIT(PIPE_CONTROL);
+   GEN6_EMIT(3DPRIMITIVE);
+   GEN6_EMIT(INTERFACE_DESCRIPTOR_DATA);
+   GEN6_EMIT(SF_VIEWPORT);
+   GEN6_EMIT(CLIP_VIEWPORT);
+   GEN6_EMIT(CC_VIEWPORT);
+   GEN6_EMIT(COLOR_CALC_STATE);
+   GEN6_EMIT(BLEND_STATE);
+   GEN6_EMIT(DEPTH_STENCIL_STATE);
+   GEN6_EMIT(SCISSOR_RECT);
+   GEN6_EMIT(BINDING_TABLE_STATE);
+   GEN6_EMIT(surf_SURFACE_STATE);
+   GEN6_EMIT(view_SURFACE_STATE);
+   GEN6_EMIT(cbuf_SURFACE_STATE);
+   GEN6_EMIT(so_SURFACE_STATE);
+   GEN6_EMIT(SAMPLER_STATE);
+   GEN6_EMIT(SAMPLER_BORDER_COLOR_STATE);
+   GEN6_EMIT(push_constant_buffer);
+#undef GEN6_EMIT
+
+   /**
+    * HW states.
+    */
+   struct ilo_3d_pipeline_state {
+      bool has_gen6_wa_pipe_control;
+
+      int reduced_prim;
+      int so_num_vertices, so_max_vertices;
+
+      uint32_t SF_VIEWPORT;
+      uint32_t CLIP_VIEWPORT;
+      uint32_t CC_VIEWPORT;
+
+      uint32_t COLOR_CALC_STATE;
+      uint32_t BLEND_STATE;
+      uint32_t DEPTH_STENCIL_STATE;
+
+      uint32_t SCISSOR_RECT;
+
+      struct {
+         uint32_t BINDING_TABLE_STATE;
+         int BINDING_TABLE_STATE_size;
+         uint32_t SURFACE_STATE[ILO_MAX_VS_SURFACES];
+         uint32_t SAMPLER_STATE;
+         uint32_t SAMPLER_BORDER_COLOR_STATE[ILO_MAX_SAMPLERS];
+         uint32_t PUSH_CONSTANT_BUFFER;
+         int PUSH_CONSTANT_BUFFER_size;
+      } vs;
+
+      struct {
+         uint32_t BINDING_TABLE_STATE;
+         int BINDING_TABLE_STATE_size;
+         uint32_t SURFACE_STATE[ILO_MAX_GS_SURFACES];
+         bool active;
+      } gs;
+
+      struct {
+         uint32_t BINDING_TABLE_STATE;
+         int BINDING_TABLE_STATE_size;
+         uint32_t SURFACE_STATE[ILO_MAX_WM_SURFACES];
+         uint32_t SAMPLER_STATE;
+         uint32_t SAMPLER_BORDER_COLOR_STATE[ILO_MAX_SAMPLERS];
+      } wm;
+   } state;
+};
+
+struct ilo_3d_pipeline *
+ilo_3d_pipeline_create(struct ilo_cp *cp, int gen, int gt);
+
+void
+ilo_3d_pipeline_destroy(struct ilo_3d_pipeline *pipeline);
+
+
+static inline void
+ilo_3d_pipeline_invalidate(struct ilo_3d_pipeline *p, uint32_t flags)
+{
+   p->invalidate_flags |= flags;
+}
+
+/**
+ * Estimate the size of an action.
+ */
+static inline int
+ilo_3d_pipeline_estimate_size(struct ilo_3d_pipeline *pipeline,
+                              enum ilo_3d_pipeline_action action,
+                              const void *arg)
+{
+   return pipeline->estimate_size(pipeline, action, arg);
+}
+
+bool
+ilo_3d_pipeline_emit_draw(struct ilo_3d_pipeline *p,
+                          const struct ilo_context *ilo,
+                          const struct pipe_draw_info *info,
+                          int *prim_generated, int *prim_emitted);
+
+void
+ilo_3d_pipeline_emit_flush(struct ilo_3d_pipeline *p);
+
+void
+ilo_3d_pipeline_emit_write_timestamp(struct ilo_3d_pipeline *p,
+                                     struct intel_bo *bo, int index);
+
+void
+ilo_3d_pipeline_emit_write_depth_count(struct ilo_3d_pipeline *p,
+                                       struct intel_bo *bo, int index);
+
+void
+ilo_3d_pipeline_get_sample_position(struct ilo_3d_pipeline *p,
+                                    unsigned sample_count,
+                                    unsigned sample_index,
+                                    float *x, float *y);
+
+void
+ilo_3d_pipeline_dump(struct ilo_3d_pipeline *p);
+
+#endif /* ILO_3D_PIPELINE_H */
diff --git a/src/gallium/drivers/ilo/ilo_3d_pipeline_dump.c b/src/gallium/drivers/ilo/ilo_3d_pipeline_dump.c
new file mode 100644 (file)
index 0000000..7f78852
--- /dev/null
@@ -0,0 +1,643 @@
+/*
+ * Copyright Â© 2007 Intel Corporation
+ * Copyright (C) 2012-2013 LunarG, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Eric Anholt <eric@anholt.net>
+ *    Chia-I Wu <olv@lunarg.com>
+ */
+
+#include "intel_winsys.h"
+
+#include "ilo_cp.h"
+#include "ilo_3d_pipeline.h"
+
+#define PRINTFLIKE(f, a) _util_printf_format(f, a)
+typedef short GLshort;
+typedef int GLint;
+typedef unsigned char GLubyte;
+typedef unsigned int GLuint;
+typedef float GLfloat;
+#include <stdint.h>
+#include <stdarg.h>
+#include <stdio.h>
+#include "brw_structs.h"
+#include "brw_defines.h"
+
+struct intel_context {
+   int gen;
+
+   struct {
+      struct {
+         void *virtual;
+      } *bo, bo_dst;
+   } batch;
+};
+
+struct brw_context {
+   struct intel_context intel;
+};
+
+static void
+batch_out(struct brw_context *brw, const char *name, uint32_t offset,
+         int index, char *fmt, ...) PRINTFLIKE(5, 6);
+
+static void
+batch_out(struct brw_context *brw, const char *name, uint32_t offset,
+         int index, char *fmt, ...)
+{
+   struct intel_context *intel = &brw->intel;
+   uint32_t *data = intel->batch.bo->virtual + offset;
+   va_list va;
+
+   fprintf(stderr, "0x%08x:      0x%08x: %8s: ",
+          offset + index * 4, data[index], name);
+   va_start(va, fmt);
+   vfprintf(stderr, fmt, va);
+   va_end(va);
+}
+
+static const char *
+get_965_surfacetype(unsigned int surfacetype)
+{
+    switch (surfacetype) {
+    case 0: return "1D";
+    case 1: return "2D";
+    case 2: return "3D";
+    case 3: return "CUBE";
+    case 4: return "BUFFER";
+    case 7: return "NULL";
+    default: return "unknown";
+    }
+}
+
+static const char *
+get_965_surface_format(unsigned int surface_format)
+{
+    switch (surface_format) {
+    case 0x000: return "r32g32b32a32_float";
+    case 0x0c1: return "b8g8r8a8_unorm";
+    case 0x100: return "b5g6r5_unorm";
+    case 0x102: return "b5g5r5a1_unorm";
+    case 0x104: return "b4g4r4a4_unorm";
+    default: return "unknown";
+    }
+}
+
+static void dump_vs_state(struct brw_context *brw, uint32_t offset)
+{
+   struct intel_context *intel = &brw->intel;
+   const char *name = "VS_STATE";
+   struct brw_vs_unit_state *vs = intel->batch.bo->virtual + offset;
+
+   batch_out(brw, name, offset, 0, "thread0\n");
+   batch_out(brw, name, offset, 1, "thread1\n");
+   batch_out(brw, name, offset, 2, "thread2\n");
+   batch_out(brw, name, offset, 3, "thread3\n");
+   batch_out(brw, name, offset, 4, "thread4: %d threads\n",
+            vs->thread4.max_threads + 1);
+   batch_out(brw, name, offset, 5, "vs5\n");
+   batch_out(brw, name, offset, 6, "vs6\n");
+}
+
+static void dump_gs_state(struct brw_context *brw, uint32_t offset)
+{
+   struct intel_context *intel = &brw->intel;
+   const char *name = "GS_STATE";
+   struct brw_gs_unit_state *gs = intel->batch.bo->virtual + offset;
+
+   batch_out(brw, name, offset, 0, "thread0\n");
+   batch_out(brw, name, offset, 1, "thread1\n");
+   batch_out(brw, name, offset, 2, "thread2\n");
+   batch_out(brw, name, offset, 3, "thread3\n");
+   batch_out(brw, name, offset, 4, "thread4: %d threads\n",
+            gs->thread4.max_threads + 1);
+   batch_out(brw, name, offset, 5, "vs5\n");
+   batch_out(brw, name, offset, 6, "vs6\n");
+}
+
+static void dump_clip_state(struct brw_context *brw, uint32_t offset)
+{
+   struct intel_context *intel = &brw->intel;
+   const char *name = "CLIP_STATE";
+   struct brw_clip_unit_state *clip = intel->batch.bo->virtual + offset;
+
+   batch_out(brw, name, offset, 0, "thread0\n");
+   batch_out(brw, name, offset, 1, "thread1\n");
+   batch_out(brw, name, offset, 2, "thread2\n");
+   batch_out(brw, name, offset, 3, "thread3\n");
+   batch_out(brw, name, offset, 4, "thread4: %d threads\n",
+            clip->thread4.max_threads + 1);
+   batch_out(brw, name, offset, 5, "clip5\n");
+   batch_out(brw, name, offset, 6, "clip6\n");
+   batch_out(brw, name, offset, 7, "vp xmin %f\n", clip->viewport_xmin);
+   batch_out(brw, name, offset, 8, "vp xmax %f\n", clip->viewport_xmax);
+   batch_out(brw, name, offset, 9, "vp ymin %f\n", clip->viewport_ymin);
+   batch_out(brw, name, offset, 10, "vp ymax %f\n", clip->viewport_ymax);
+}
+
+static void dump_sf_state(struct brw_context *brw, uint32_t offset)
+{
+   struct intel_context *intel = &brw->intel;
+   const char *name = "SF_STATE";
+   struct brw_sf_unit_state *sf = intel->batch.bo->virtual + offset;
+
+   batch_out(brw, name, offset, 0, "thread0\n");
+   batch_out(brw, name, offset, 1, "thread1\n");
+   batch_out(brw, name, offset, 2, "thread2\n");
+   batch_out(brw, name, offset, 3, "thread3\n");
+   batch_out(brw, name, offset, 4, "thread4: %d threads\n",
+            sf->thread4.max_threads + 1);
+   batch_out(brw, name, offset, 5, "sf5: viewport offset\n");
+   batch_out(brw, name, offset, 6, "sf6\n");
+   batch_out(brw, name, offset, 7, "sf7\n");
+}
+
+static void dump_wm_state(struct brw_context *brw, uint32_t offset)
+{
+   struct intel_context *intel = &brw->intel;
+   const char *name = "WM_STATE";
+   struct brw_wm_unit_state *wm = intel->batch.bo->virtual + offset;
+
+   batch_out(brw, name, offset, 0, "thread0\n");
+   batch_out(brw, name, offset, 1, "thread1\n");
+   batch_out(brw, name, offset, 2, "thread2\n");
+   batch_out(brw, name, offset, 3, "thread3\n");
+   batch_out(brw, name, offset, 4, "wm4\n");
+   batch_out(brw, name, offset, 5, "wm5: %s%s%s%s%s%s, %d threads\n",
+            wm->wm5.enable_8_pix ? "8pix" : "",
+            wm->wm5.enable_16_pix ? "16pix" : "",
+            wm->wm5.program_uses_depth ? ", uses depth" : "",
+            wm->wm5.program_computes_depth ? ", computes depth" : "",
+            wm->wm5.program_uses_killpixel ? ", kills" : "",
+            wm->wm5.thread_dispatch_enable ? "" : ", no dispatch",
+            wm->wm5.max_threads + 1);
+   batch_out(brw, name, offset, 6, "depth offset constant %f\n",
+            wm->global_depth_offset_constant);
+   batch_out(brw, name, offset, 7, "depth offset scale %f\n",
+            wm->global_depth_offset_scale);
+   batch_out(brw, name, offset, 8, "wm8: kernel 1 (gen5+)\n");
+   batch_out(brw, name, offset, 9, "wm9: kernel 2 (gen5+)\n");
+   batch_out(brw, name, offset, 10, "wm10: kernel 3 (gen5+)\n");
+}
+
+static void dump_surface_state(struct brw_context *brw, uint32_t offset)
+{
+   const char *name = "SURF";
+   uint32_t *surf = brw->intel.batch.bo->virtual + offset;
+
+   batch_out(brw, name, offset, 0, "%s %s\n",
+            get_965_surfacetype(GET_FIELD(surf[0], BRW_SURFACE_TYPE)),
+            get_965_surface_format(GET_FIELD(surf[0], BRW_SURFACE_FORMAT)));
+   batch_out(brw, name, offset, 1, "offset\n");
+   batch_out(brw, name, offset, 2, "%dx%d size, %d mips\n",
+            GET_FIELD(surf[2], BRW_SURFACE_WIDTH) + 1,
+            GET_FIELD(surf[2], BRW_SURFACE_HEIGHT) + 1,
+            GET_FIELD(surf[2], BRW_SURFACE_LOD));
+   batch_out(brw, name, offset, 3, "pitch %d, %s tiled\n",
+            GET_FIELD(surf[3], BRW_SURFACE_PITCH) + 1,
+            (surf[3] & BRW_SURFACE_TILED) ?
+            ((surf[3] & BRW_SURFACE_TILED_Y) ? "Y" : "X") : "not");
+   batch_out(brw, name, offset, 4, "mip base %d\n",
+            GET_FIELD(surf[4], BRW_SURFACE_MIN_LOD));
+   batch_out(brw, name, offset, 5, "x,y offset: %d,%d\n",
+            GET_FIELD(surf[5], BRW_SURFACE_X_OFFSET),
+            GET_FIELD(surf[5], BRW_SURFACE_Y_OFFSET));
+}
+
+static void dump_gen7_surface_state(struct brw_context *brw, uint32_t offset)
+{
+   const char *name = "SURF";
+   uint32_t *surf = brw->intel.batch.bo->virtual + offset;
+
+   batch_out(brw, name, offset, 0, "%s %s\n",
+             get_965_surfacetype(GET_FIELD(surf[0], BRW_SURFACE_TYPE)),
+             get_965_surface_format(GET_FIELD(surf[0], BRW_SURFACE_FORMAT)));
+   batch_out(brw, name, offset, 1, "offset\n");
+   batch_out(brw, name, offset, 2, "%dx%d size, %d mips\n",
+             GET_FIELD(surf[2], GEN7_SURFACE_WIDTH) + 1,
+             GET_FIELD(surf[2], GEN7_SURFACE_HEIGHT) + 1,
+             surf[5] & INTEL_MASK(3, 0));
+   batch_out(brw, name, offset, 3, "pitch %d, %stiled\n",
+            (surf[3] & INTEL_MASK(17, 0)) + 1,
+             (surf[0] & (1 << 14)) ? "" : "not ");
+   batch_out(brw, name, offset, 4, "mip base %d\n",
+             GET_FIELD(surf[5], GEN7_SURFACE_MIN_LOD));
+   batch_out(brw, name, offset, 5, "x,y offset: %d,%d\n",
+             GET_FIELD(surf[5], BRW_SURFACE_X_OFFSET),
+             GET_FIELD(surf[5], BRW_SURFACE_Y_OFFSET));
+}
+
+static void
+dump_sdc(struct brw_context *brw, uint32_t offset)
+{
+   const char *name = "SDC";
+   struct intel_context *intel = &brw->intel;
+
+   if (intel->gen >= 5 && intel->gen <= 6) {
+      struct gen5_sampler_default_color *sdc = (intel->batch.bo->virtual +
+                                               offset);
+      batch_out(brw, name, offset, 0, "unorm rgba\n");
+      batch_out(brw, name, offset, 1, "r %f\n", sdc->f[0]);
+      batch_out(brw, name, offset, 2, "b %f\n", sdc->f[1]);
+      batch_out(brw, name, offset, 3, "g %f\n", sdc->f[2]);
+      batch_out(brw, name, offset, 4, "a %f\n", sdc->f[3]);
+      batch_out(brw, name, offset, 5, "half float rg\n");
+      batch_out(brw, name, offset, 6, "half float ba\n");
+      batch_out(brw, name, offset, 7, "u16 rg\n");
+      batch_out(brw, name, offset, 8, "u16 ba\n");
+      batch_out(brw, name, offset, 9, "s16 rg\n");
+      batch_out(brw, name, offset, 10, "s16 ba\n");
+      batch_out(brw, name, offset, 11, "s8 rgba\n");
+   } else {
+      struct brw_sampler_default_color *sdc = (intel->batch.bo->virtual +
+                                              offset);
+      batch_out(brw, name, offset, 0, "r %f\n", sdc->color[0]);
+      batch_out(brw, name, offset, 1, "g %f\n", sdc->color[1]);
+      batch_out(brw, name, offset, 2, "b %f\n", sdc->color[2]);
+      batch_out(brw, name, offset, 3, "a %f\n", sdc->color[3]);
+   }
+}
+
+static void dump_sampler_state(struct brw_context *brw,
+                              uint32_t offset, uint32_t size)
+{
+   struct intel_context *intel = &brw->intel;
+   int i;
+   struct brw_sampler_state *samp = intel->batch.bo->virtual + offset;
+
+   assert(intel->gen < 7);
+
+   for (i = 0; i < size / sizeof(*samp); i++) {
+      char name[20];
+
+      sprintf(name, "WM SAMP%d", i);
+      batch_out(brw, name, offset, 0, "filtering\n");
+      batch_out(brw, name, offset, 1, "wrapping, lod\n");
+      batch_out(brw, name, offset, 2, "default color pointer\n");
+      batch_out(brw, name, offset, 3, "chroma key, aniso\n");
+
+      samp++;
+      offset += sizeof(*samp);
+   }
+}
+
+static void dump_gen7_sampler_state(struct brw_context *brw,
+                                   uint32_t offset, uint32_t size)
+{
+   struct intel_context *intel = &brw->intel;
+   struct gen7_sampler_state *samp = intel->batch.bo->virtual + offset;
+   int i;
+
+   assert(intel->gen >= 7);
+
+   for (i = 0; i < size / sizeof(*samp); i++) {
+      char name[20];
+
+      sprintf(name, "WM SAMP%d", i);
+      batch_out(brw, name, offset, 0, "filtering\n");
+      batch_out(brw, name, offset, 1, "wrapping, lod\n");
+      batch_out(brw, name, offset, 2, "default color pointer\n");
+      batch_out(brw, name, offset, 3, "chroma key, aniso\n");
+
+      samp++;
+      offset += sizeof(*samp);
+   }
+}
+
+
+static void dump_sf_viewport_state(struct brw_context *brw,
+                                  uint32_t offset)
+{
+   struct intel_context *intel = &brw->intel;
+   const char *name = "SF VP";
+   struct brw_sf_viewport *vp = intel->batch.bo->virtual + offset;
+
+   assert(intel->gen < 7);
+
+   batch_out(brw, name, offset, 0, "m00 = %f\n", vp->viewport.m00);
+   batch_out(brw, name, offset, 1, "m11 = %f\n", vp->viewport.m11);
+   batch_out(brw, name, offset, 2, "m22 = %f\n", vp->viewport.m22);
+   batch_out(brw, name, offset, 3, "m30 = %f\n", vp->viewport.m30);
+   batch_out(brw, name, offset, 4, "m31 = %f\n", vp->viewport.m31);
+   batch_out(brw, name, offset, 5, "m32 = %f\n", vp->viewport.m32);
+
+   batch_out(brw, name, offset, 6, "top left = %d,%d\n",
+            vp->scissor.xmin, vp->scissor.ymin);
+   batch_out(brw, name, offset, 7, "bottom right = %d,%d\n",
+            vp->scissor.xmax, vp->scissor.ymax);
+}
+
+static void dump_clip_viewport_state(struct brw_context *brw,
+                                    uint32_t offset)
+{
+   struct intel_context *intel = &brw->intel;
+   const char *name = "CLIP VP";
+   struct brw_clipper_viewport *vp = intel->batch.bo->virtual + offset;
+
+   assert(intel->gen < 7);
+
+   batch_out(brw, name, offset, 0, "xmin = %f\n", vp->xmin);
+   batch_out(brw, name, offset, 1, "xmax = %f\n", vp->xmax);
+   batch_out(brw, name, offset, 2, "ymin = %f\n", vp->ymin);
+   batch_out(brw, name, offset, 3, "ymax = %f\n", vp->ymax);
+}
+
+static void dump_sf_clip_viewport_state(struct brw_context *brw,
+                                       uint32_t offset)
+{
+   struct intel_context *intel = &brw->intel;
+   const char *name = "SF_CLIP VP";
+   struct gen7_sf_clip_viewport *vp = intel->batch.bo->virtual + offset;
+
+   assert(intel->gen >= 7);
+
+   batch_out(brw, name, offset, 0, "m00 = %f\n", vp->viewport.m00);
+   batch_out(brw, name, offset, 1, "m11 = %f\n", vp->viewport.m11);
+   batch_out(brw, name, offset, 2, "m22 = %f\n", vp->viewport.m22);
+   batch_out(brw, name, offset, 3, "m30 = %f\n", vp->viewport.m30);
+   batch_out(brw, name, offset, 4, "m31 = %f\n", vp->viewport.m31);
+   batch_out(brw, name, offset, 5, "m32 = %f\n", vp->viewport.m32);
+   batch_out(brw, name, offset, 6, "guardband xmin = %f\n", vp->guardband.xmin);
+   batch_out(brw, name, offset, 7, "guardband xmax = %f\n", vp->guardband.xmax);
+   batch_out(brw, name, offset, 8, "guardband ymin = %f\n", vp->guardband.ymin);
+   batch_out(brw, name, offset, 9, "guardband ymax = %f\n", vp->guardband.ymax);
+}
+
+
+static void dump_cc_viewport_state(struct brw_context *brw, uint32_t offset)
+{
+   const char *name = "CC VP";
+   struct brw_cc_viewport *vp = brw->intel.batch.bo->virtual + offset;
+
+   batch_out(brw, name, offset, 0, "min_depth = %f\n", vp->min_depth);
+   batch_out(brw, name, offset, 1, "max_depth = %f\n", vp->max_depth);
+}
+
+static void dump_depth_stencil_state(struct brw_context *brw, uint32_t offset)
+{
+   const char *name = "D_S";
+   struct gen6_depth_stencil_state *ds = brw->intel.batch.bo->virtual + offset;
+
+   batch_out(brw, name, offset, 0,
+            "stencil %sable, func %d, write %sable\n",
+            ds->ds0.stencil_enable ? "en" : "dis",
+            ds->ds0.stencil_func,
+            ds->ds0.stencil_write_enable ? "en" : "dis");
+   batch_out(brw, name, offset, 1,
+            "stencil test mask 0x%x, write mask 0x%x\n",
+            ds->ds1.stencil_test_mask, ds->ds1.stencil_write_mask);
+   batch_out(brw, name, offset, 2,
+            "depth test %sable, func %d, write %sable\n",
+            ds->ds2.depth_test_enable ? "en" : "dis",
+            ds->ds2.depth_test_func,
+            ds->ds2.depth_write_enable ? "en" : "dis");
+}
+
+static void dump_cc_state_gen4(struct brw_context *brw, uint32_t offset)
+{
+   const char *name = "CC";
+
+   batch_out(brw, name, offset, 0, "cc0\n");
+   batch_out(brw, name, offset, 1, "cc1\n");
+   batch_out(brw, name, offset, 2, "cc2\n");
+   batch_out(brw, name, offset, 3, "cc3\n");
+   batch_out(brw, name, offset, 4, "cc4: viewport offset\n");
+   batch_out(brw, name, offset, 5, "cc5\n");
+   batch_out(brw, name, offset, 6, "cc6\n");
+   batch_out(brw, name, offset, 7, "cc7\n");
+}
+
+static void dump_cc_state_gen6(struct brw_context *brw, uint32_t offset)
+{
+   const char *name = "CC";
+   struct gen6_color_calc_state *cc = brw->intel.batch.bo->virtual + offset;
+
+   batch_out(brw, name, offset, 0,
+            "alpha test format %s, round disable %d, stencil ref %d, "
+            "bf stencil ref %d\n",
+            cc->cc0.alpha_test_format ? "FLOAT32" : "UNORM8",
+            cc->cc0.round_disable,
+            cc->cc0.stencil_ref,
+            cc->cc0.bf_stencil_ref);
+   batch_out(brw, name, offset, 1, "\n");
+   batch_out(brw, name, offset, 2, "constant red %f\n", cc->constant_r);
+   batch_out(brw, name, offset, 3, "constant green %f\n", cc->constant_g);
+   batch_out(brw, name, offset, 4, "constant blue %f\n", cc->constant_b);
+   batch_out(brw, name, offset, 5, "constant alpha %f\n", cc->constant_a);
+}
+
+static void dump_blend_state(struct brw_context *brw, uint32_t offset)
+{
+   const char *name = "BLEND";
+
+   batch_out(brw, name, offset, 0, "\n");
+   batch_out(brw, name, offset, 1, "\n");
+}
+
+static void
+dump_scissor(struct brw_context *brw, uint32_t offset)
+{
+   const char *name = "SCISSOR";
+   struct intel_context *intel = &brw->intel;
+   struct gen6_scissor_rect *scissor = intel->batch.bo->virtual + offset;
+
+   batch_out(brw, name, offset, 0, "xmin %d, ymin %d\n",
+            scissor->xmin, scissor->ymin);
+   batch_out(brw, name, offset, 1, "xmax %d, ymax %d\n",
+            scissor->xmax, scissor->ymax);
+}
+
+static void
+dump_vs_constants(struct brw_context *brw, uint32_t offset, uint32_t size)
+{
+   const char *name = "VS_CONST";
+   struct intel_context *intel = &brw->intel;
+   uint32_t *as_uint = intel->batch.bo->virtual + offset;
+   float *as_float = intel->batch.bo->virtual + offset;
+   int i;
+
+   for (i = 0; i < size / 4; i += 4) {
+      batch_out(brw, name, offset, i, "%3d: (% f % f % f % f) (0x%08x 0x%08x 0x%08x 0x%08x)\n",
+               i / 4,
+               as_float[i], as_float[i + 1], as_float[i + 2], as_float[i + 3],
+               as_uint[i], as_uint[i + 1], as_uint[i + 2], as_uint[i + 3]);
+   }
+}
+
+static void
+dump_wm_constants(struct brw_context *brw, uint32_t offset, uint32_t size)
+{
+   const char *name = "WM_CONST";
+   struct intel_context *intel = &brw->intel;
+   uint32_t *as_uint = intel->batch.bo->virtual + offset;
+   float *as_float = intel->batch.bo->virtual + offset;
+   int i;
+
+   for (i = 0; i < size / 4; i += 4) {
+      batch_out(brw, name, offset, i, "%3d: (% f % f % f % f) (0x%08x 0x%08x 0x%08x 0x%08x)\n",
+               i / 4,
+               as_float[i], as_float[i + 1], as_float[i + 2], as_float[i + 3],
+               as_uint[i], as_uint[i + 1], as_uint[i + 2], as_uint[i + 3]);
+   }
+}
+
+static void dump_binding_table(struct brw_context *brw, uint32_t offset,
+                              uint32_t size)
+{
+   char name[20];
+   int i;
+   uint32_t *data = brw->intel.batch.bo->virtual + offset;
+
+   for (i = 0; i < size / 4; i++) {
+      if (data[i] == 0)
+        continue;
+
+      sprintf(name, "BIND%d", i);
+      batch_out(brw, name, offset, i, "surface state address\n");
+   }
+}
+
+static void
+init_brw(struct brw_context *brw, struct ilo_3d_pipeline *p)
+{
+   brw->intel.gen = ILO_GEN_GET_MAJOR(p->gen);
+   brw->intel.batch.bo_dst.virtual = p->cp->bo->get_virtual(p->cp->bo);
+   brw->intel.batch.bo = &brw->intel.batch.bo_dst;
+}
+
+static void
+dump_3d_state(struct ilo_3d_pipeline *p)
+{
+   struct brw_context brw;
+   int num_states, i;
+
+   init_brw(&brw, p);
+
+   if (brw.intel.gen >= 7) {
+      dump_cc_viewport_state(&brw, p->state.CC_VIEWPORT);
+      dump_sf_clip_viewport_state(&brw, p->state.SF_VIEWPORT);
+   }
+   else {
+      dump_clip_viewport_state(&brw, p->state.CLIP_VIEWPORT);
+      dump_sf_viewport_state(&brw, p->state.SF_VIEWPORT);
+      dump_cc_viewport_state(&brw, p->state.CC_VIEWPORT);
+   }
+
+   dump_blend_state(&brw, p->state.BLEND_STATE);
+   dump_cc_state_gen6(&brw, p->state.COLOR_CALC_STATE);
+   dump_depth_stencil_state(&brw, p->state.DEPTH_STENCIL_STATE);
+
+   /* VS */
+   num_states = p->state.vs.BINDING_TABLE_STATE_size;
+   for (i = 0; i < num_states; i++) {
+      if (brw.intel.gen < 7)
+         dump_surface_state(&brw, p->state.vs.SURFACE_STATE[i]);
+      else
+         dump_gen7_surface_state(&brw, p->state.vs.SURFACE_STATE[i]);
+   }
+   dump_binding_table(&brw, p->state.vs.BINDING_TABLE_STATE, num_states * 4);
+
+   num_states = 0;
+   for (i = 0; i < Elements(p->state.vs.SAMPLER_BORDER_COLOR_STATE); i++) {
+      if (!p->state.vs.SAMPLER_BORDER_COLOR_STATE[i])
+         continue;
+
+      dump_sdc(&brw, p->state.vs.SAMPLER_BORDER_COLOR_STATE[i]);
+      num_states++;
+   }
+   if (brw.intel.gen < 7)
+      dump_sampler_state(&brw, p->state.vs.SAMPLER_STATE, num_states * 16);
+   else
+      dump_gen7_sampler_state(&brw, p->state.vs.SAMPLER_STATE, num_states * 16);
+
+   if (p->state.vs.PUSH_CONSTANT_BUFFER_size) {
+      dump_vs_constants(&brw, p->state.vs.PUSH_CONSTANT_BUFFER,
+            p->state.vs.PUSH_CONSTANT_BUFFER_size);
+   }
+
+   /* GS */
+   num_states = p->state.gs.BINDING_TABLE_STATE_size;
+   for (i = 0; i < num_states; i++) {
+      if (!p->state.gs.SURFACE_STATE[i])
+         continue;
+
+      if (brw.intel.gen < 7)
+         dump_surface_state(&brw, p->state.gs.SURFACE_STATE[i]);
+      else
+         dump_gen7_surface_state(&brw, p->state.gs.SURFACE_STATE[i]);
+   }
+   dump_binding_table(&brw, p->state.gs.BINDING_TABLE_STATE, num_states * 4);
+
+   /* WM */
+   num_states = p->state.wm.BINDING_TABLE_STATE_size;
+   for (i = 0; i < num_states; i++) {
+      if (!p->state.wm.SURFACE_STATE[i])
+         continue;
+
+      if (brw.intel.gen < 7)
+         dump_surface_state(&brw, p->state.wm.SURFACE_STATE[i]);
+      else
+         dump_gen7_surface_state(&brw, p->state.wm.SURFACE_STATE[i]);
+   }
+   dump_binding_table(&brw, p->state.wm.BINDING_TABLE_STATE, num_states * 4);
+
+   num_states = 0;
+   for (i = 0; i < Elements(p->state.wm.SAMPLER_BORDER_COLOR_STATE); i++) {
+      if (!p->state.wm.SAMPLER_BORDER_COLOR_STATE[i])
+         continue;
+
+      dump_sdc(&brw, p->state.wm.SAMPLER_BORDER_COLOR_STATE[i]);
+      num_states++;
+   }
+   if (brw.intel.gen < 7)
+      dump_sampler_state(&brw, p->state.wm.SAMPLER_STATE, num_states * 16);
+   else
+      dump_gen7_sampler_state(&brw, p->state.wm.SAMPLER_STATE, num_states * 16);
+
+   dump_scissor(&brw, p->state.SCISSOR_RECT);
+
+   (void) dump_vs_state;
+   (void) dump_gs_state;
+   (void) dump_clip_state;
+   (void) dump_sf_state;
+   (void) dump_wm_state;
+   (void) dump_cc_state_gen4;
+   (void) dump_wm_constants;
+}
+
+/**
+ * Dump the pipeline.
+ */
+void
+ilo_3d_pipeline_dump(struct ilo_3d_pipeline *p)
+{
+   int err;
+
+   ilo_cp_dump(p->cp);
+
+   err = p->cp->bo->map(p->cp->bo, false);
+   if (!err) {
+      dump_3d_state(p);
+      p->cp->bo->unmap(p->cp->bo);
+   }
+}
diff --git a/src/gallium/drivers/ilo/ilo_3d_pipeline_gen6.c b/src/gallium/drivers/ilo/ilo_3d_pipeline_gen6.c
new file mode 100644 (file)
index 0000000..16bd422
--- /dev/null
@@ -0,0 +1,1637 @@
+/*
+ * Mesa 3-D graphics library
+ *
+ * Copyright (C) 2013 LunarG, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Chia-I Wu <olv@lunarg.com>
+ */
+
+#include "util/u_dual_blend.h"
+#include "util/u_prim.h"
+#include "intel_reg.h"
+
+#include "ilo_context.h"
+#include "ilo_cp.h"
+#include "ilo_gpe_gen6.h"
+#include "ilo_shader.h"
+#include "ilo_state.h"
+#include "ilo_3d_pipeline.h"
+#include "ilo_3d_pipeline_gen6.h"
+
+/**
+ * This should be called before any depth stall flush (including those
+ * produced by non-pipelined state commands) or cache flush on GEN6.
+ *
+ * \see intel_emit_post_sync_nonzero_flush()
+ */
+static void
+gen6_wa_pipe_control_post_sync(struct ilo_3d_pipeline *p,
+                               bool caller_post_sync)
+{
+   assert(p->gen == ILO_GEN(6));
+
+   /* emit once */
+   if (p->state.has_gen6_wa_pipe_control)
+      return;
+
+   p->state.has_gen6_wa_pipe_control = true;
+
+   /*
+    * From the Sandy Bridge PRM, volume 2 part 1, page 60:
+    *
+    *     "Pipe-control with CS-stall bit set must be sent BEFORE the
+    *      pipe-control with a post-sync op and no write-cache flushes."
+    *
+    * The workaround below necessitates this workaround.
+    */
+   p->gen6_PIPE_CONTROL(&p->gpe,
+         PIPE_CONTROL_CS_STALL |
+         PIPE_CONTROL_STALL_AT_SCOREBOARD,
+         NULL, 0, false, p->cp);
+
+   /* the caller will emit the post-sync op */
+   if (caller_post_sync)
+      return;
+
+   /*
+    * From the Sandy Bridge PRM, volume 2 part 1, page 60:
+    *
+    *     "Before any depth stall flush (including those produced by
+    *      non-pipelined state commands), software needs to first send a
+    *      PIPE_CONTROL with no bits set except Post-Sync Operation != 0."
+    *
+    *     "Before a PIPE_CONTROL with Write Cache Flush Enable =1, a
+    *      PIPE_CONTROL with any non-zero post-sync-op is required."
+    */
+   p->gen6_PIPE_CONTROL(&p->gpe,
+         PIPE_CONTROL_WRITE_IMMEDIATE,
+         p->workaround_bo, 0, false, p->cp);
+}
+
+static void
+gen6_wa_pipe_control_wm_multisample_flush(struct ilo_3d_pipeline *p)
+{
+   assert(p->gen == ILO_GEN(6));
+
+   gen6_wa_pipe_control_post_sync(p, false);
+
+   /*
+    * From the Sandy Bridge PRM, volume 2 part 1, page 305:
+    *
+    *     "Driver must guarentee that all the caches in the depth pipe are
+    *      flushed before this command (3DSTATE_MULTISAMPLE) is parsed. This
+    *      requires driver to send a PIPE_CONTROL with a CS stall along with a
+    *      Depth Flush prior to this command."
+    */
+   p->gen6_PIPE_CONTROL(&p->gpe,
+         PIPE_CONTROL_DEPTH_CACHE_FLUSH |
+         PIPE_CONTROL_CS_STALL,
+         0, 0, false, p->cp);
+}
+
+static void
+gen6_wa_pipe_control_wm_depth_flush(struct ilo_3d_pipeline *p)
+{
+   assert(p->gen == ILO_GEN(6));
+
+   gen6_wa_pipe_control_post_sync(p, false);
+
+   /*
+    * According to intel_emit_depth_stall_flushes() of classic i965, we need
+    * to emit a sequence of PIPE_CONTROLs prior to emitting depth related
+    * commands.
+    */
+   p->gen6_PIPE_CONTROL(&p->gpe,
+         PIPE_CONTROL_DEPTH_STALL,
+         NULL, 0, false, p->cp);
+
+   p->gen6_PIPE_CONTROL(&p->gpe,
+         PIPE_CONTROL_DEPTH_CACHE_FLUSH,
+         NULL, 0, false, p->cp);
+
+   p->gen6_PIPE_CONTROL(&p->gpe,
+         PIPE_CONTROL_DEPTH_STALL,
+         NULL, 0, false, p->cp);
+}
+
+static void
+gen6_wa_pipe_control_wm_max_threads_stall(struct ilo_3d_pipeline *p)
+{
+   assert(p->gen == ILO_GEN(6));
+
+   /* the post-sync workaround should cover this already */
+   if (p->state.has_gen6_wa_pipe_control)
+      return;
+
+   /*
+    * From the Sandy Bridge PRM, volume 2 part 1, page 274:
+    *
+    *     "A PIPE_CONTROL command, with only the Stall At Pixel Scoreboard
+    *      field set (DW1 Bit 1), must be issued prior to any change to the
+    *      value in this field (Maximum Number of Threads in 3DSTATE_WM)"
+    */
+   p->gen6_PIPE_CONTROL(&p->gpe,
+         PIPE_CONTROL_STALL_AT_SCOREBOARD,
+         NULL, 0, false, p->cp);
+
+}
+
+static void
+gen6_wa_pipe_control_vs_const_flush(struct ilo_3d_pipeline *p)
+{
+   assert(p->gen == ILO_GEN(6));
+
+   gen6_wa_pipe_control_post_sync(p, false);
+
+   /*
+    * According to upload_vs_state() of classic i965, we need to emit
+    * PIPE_CONTROL after 3DSTATE_CONSTANT_VS so that the command is kept being
+    * buffered by VS FF, to the point that the FF dies.
+    */
+   p->gen6_PIPE_CONTROL(&p->gpe,
+         PIPE_CONTROL_DEPTH_STALL |
+         PIPE_CONTROL_INSTRUCTION_FLUSH |
+         PIPE_CONTROL_STATE_CACHE_INVALIDATE,
+         NULL, 0, false, p->cp);
+}
+
+#define DIRTY(state) (session->pipe_dirty & ILO_DIRTY_ ## state)
+
+void
+gen6_pipeline_common_select(struct ilo_3d_pipeline *p,
+                            const struct ilo_context *ilo,
+                            struct gen6_pipeline_session *session)
+{
+   /* PIPELINE_SELECT */
+   if (session->hw_ctx_changed) {
+      if (p->gen == ILO_GEN(6))
+         gen6_wa_pipe_control_post_sync(p, false);
+
+      p->gen6_PIPELINE_SELECT(&p->gpe, 0x0, p->cp);
+   }
+}
+
+void
+gen6_pipeline_common_sip(struct ilo_3d_pipeline *p,
+                         const struct ilo_context *ilo,
+                         struct gen6_pipeline_session *session)
+{
+   /* STATE_SIP */
+   if (session->hw_ctx_changed) {
+      if (p->gen == ILO_GEN(6))
+         gen6_wa_pipe_control_post_sync(p, false);
+
+      p->gen6_STATE_SIP(&p->gpe, 0, p->cp);
+   }
+}
+
+void
+gen6_pipeline_common_base_address(struct ilo_3d_pipeline *p,
+                                  const struct ilo_context *ilo,
+                                  struct gen6_pipeline_session *session)
+{
+   /* STATE_BASE_ADDRESS */
+   if (session->state_bo_changed || session->instruction_bo_changed) {
+      if (p->gen == ILO_GEN(6))
+         gen6_wa_pipe_control_post_sync(p, false);
+
+      p->gen6_STATE_BASE_ADDRESS(&p->gpe,
+            NULL, p->cp->bo, p->cp->bo, NULL, ilo->shader_cache->bo,
+            0, 0, 0, 0, p->cp);
+
+      /*
+       * From the Sandy Bridge PRM, volume 1 part 1, page 28:
+       *
+       *     "The following commands must be reissued following any change to
+       *      the base addresses:
+       *
+       *       * 3DSTATE_BINDING_TABLE_POINTERS
+       *       * 3DSTATE_SAMPLER_STATE_POINTERS
+       *       * 3DSTATE_VIEWPORT_STATE_POINTERS
+       *       * 3DSTATE_CC_POINTERS
+       *       * MEDIA_STATE_POINTERS"
+       *
+       * 3DSTATE_SCISSOR_STATE_POINTERS is not on the list, but it is
+       * reasonable to also reissue the command.  Same to PCB.
+       */
+      session->viewport_state_changed = true;
+
+      session->cc_state_blend_changed = true;
+      session->cc_state_dsa_changed = true;
+      session->cc_state_cc_changed = true;
+
+      session->scissor_state_changed = true;
+
+      session->binding_table_vs_changed = true;
+      session->binding_table_gs_changed = true;
+      session->binding_table_fs_changed = true;
+
+      session->sampler_state_vs_changed = true;
+      session->sampler_state_gs_changed = true;
+      session->sampler_state_fs_changed = true;
+
+      session->pcb_state_vs_changed = true;
+      session->pcb_state_gs_changed = true;
+      session->pcb_state_fs_changed = true;
+   }
+}
+
+static void
+gen6_pipeline_common_urb(struct ilo_3d_pipeline *p,
+                         const struct ilo_context *ilo,
+                         struct gen6_pipeline_session *session)
+{
+   /* 3DSTATE_URB */
+   if (DIRTY(VERTEX_ELEMENTS) || DIRTY(VS) || DIRTY(GS)) {
+      const struct ilo_shader *vs = (ilo->vs) ? ilo->vs->shader : NULL;
+      const struct ilo_shader *gs = (ilo->gs) ? ilo->gs->shader : NULL;
+      const bool gs_active = (gs || (vs && vs->stream_output));
+      int vs_entry_size, gs_entry_size;
+      int vs_total_size, gs_total_size;
+
+      vs_entry_size = (vs) ? vs->out.count : 0;
+
+      /*
+       * As indicated by 2e712e41db0c0676e9f30fc73172c0e8de8d84d4, VF and VS
+       * share VUE handles.  The VUE allocation size must be large enough to
+       * store either VF outputs (number of VERTEX_ELEMENTs) and VS outputs.
+       *
+       * I am not sure if the PRM explicitly states that VF and VS share VUE
+       * handles.  But here is a citation that implies so:
+       *
+       * From the Sandy Bridge PRM, volume 2 part 1, page 44:
+       *
+       *     "Once a FF stage that spawn threads has sufficient input to
+       *      initiate a thread, it must guarantee that it is safe to request
+       *      the thread initiation. For all these FF stages, this check is
+       *      based on :
+       *
+       *      - The availability of output URB entries:
+       *        - VS: As the input URB entries are overwritten with the
+       *          VS-generated output data, output URB availability isn't a
+       *          factor."
+       */
+      if (vs_entry_size < ilo->vertex_elements->num_elements)
+         vs_entry_size = ilo->vertex_elements->num_elements;
+
+      gs_entry_size = (gs) ? gs->out.count :
+         (vs && vs->stream_output) ? vs_entry_size : 0;
+
+      /* in bytes */
+      vs_entry_size *= sizeof(float) * 4;
+      gs_entry_size *= sizeof(float) * 4;
+      vs_total_size = ilo->urb.size * 1024;
+
+      if (gs_active) {
+         vs_total_size /= 2;
+         gs_total_size = vs_total_size;
+      }
+      else {
+         gs_total_size = 0;
+      }
+
+      p->gen6_3DSTATE_URB(&p->gpe, vs_total_size, gs_total_size,
+            vs_entry_size, gs_entry_size, p->cp);
+
+      /*
+       * From the Sandy Bridge PRM, volume 2 part 1, page 27:
+       *
+       *     "Because of a urb corruption caused by allocating a previous
+       *      gsunit's urb entry to vsunit software is required to send a
+       *      "GS NULL Fence" (Send URB fence with VS URB size == 1 and GS URB
+       *      size == 0) plus a dummy DRAW call before any case where VS will
+       *      be taking over GS URB space."
+       */
+      if (p->state.gs.active && !gs_active)
+         ilo_3d_pipeline_emit_flush_gen6(p);
+
+      p->state.gs.active = gs_active;
+   }
+}
+
+static void
+gen6_pipeline_common_pointers_1(struct ilo_3d_pipeline *p,
+                                const struct ilo_context *ilo,
+                                struct gen6_pipeline_session *session)
+{
+   /* 3DSTATE_VIEWPORT_STATE_POINTERS */
+   if (session->viewport_state_changed) {
+      p->gen6_3DSTATE_VIEWPORT_STATE_POINTERS(&p->gpe,
+            p->state.CLIP_VIEWPORT,
+            p->state.SF_VIEWPORT,
+            p->state.CC_VIEWPORT, p->cp);
+   }
+}
+
+static void
+gen6_pipeline_common_pointers_2(struct ilo_3d_pipeline *p,
+                                const struct ilo_context *ilo,
+                                struct gen6_pipeline_session *session)
+{
+   /* 3DSTATE_CC_STATE_POINTERS */
+   if (session->cc_state_blend_changed ||
+       session->cc_state_dsa_changed ||
+       session->cc_state_cc_changed) {
+      p->gen6_3DSTATE_CC_STATE_POINTERS(&p->gpe,
+            p->state.BLEND_STATE,
+            p->state.DEPTH_STENCIL_STATE,
+            p->state.COLOR_CALC_STATE, p->cp);
+   }
+
+   /* 3DSTATE_SAMPLER_STATE_POINTERS */
+   if (session->sampler_state_vs_changed ||
+       session->sampler_state_gs_changed ||
+       session->sampler_state_fs_changed) {
+      p->gen6_3DSTATE_SAMPLER_STATE_POINTERS(&p->gpe,
+            p->state.vs.SAMPLER_STATE,
+            0,
+            p->state.wm.SAMPLER_STATE, p->cp);
+   }
+}
+
+static void
+gen6_pipeline_common_pointers_3(struct ilo_3d_pipeline *p,
+                                const struct ilo_context *ilo,
+                                struct gen6_pipeline_session *session)
+{
+   /* 3DSTATE_SCISSOR_STATE_POINTERS */
+   if (session->scissor_state_changed) {
+      p->gen6_3DSTATE_SCISSOR_STATE_POINTERS(&p->gpe,
+            p->state.SCISSOR_RECT, p->cp);
+   }
+
+   /* 3DSTATE_BINDING_TABLE_POINTERS */
+   if (session->binding_table_vs_changed ||
+       session->binding_table_gs_changed ||
+       session->binding_table_fs_changed) {
+      p->gen6_3DSTATE_BINDING_TABLE_POINTERS(&p->gpe,
+            p->state.vs.BINDING_TABLE_STATE,
+            p->state.gs.BINDING_TABLE_STATE,
+            p->state.wm.BINDING_TABLE_STATE, p->cp);
+   }
+}
+
+void
+gen6_pipeline_vf(struct ilo_3d_pipeline *p,
+                 const struct ilo_context *ilo,
+                 struct gen6_pipeline_session *session)
+{
+   /* 3DSTATE_INDEX_BUFFER */
+   if (DIRTY(INDEX_BUFFER)) {
+      p->gen6_3DSTATE_INDEX_BUFFER(&p->gpe,
+            &ilo->index_buffer, false, p->cp);
+   }
+
+   /* 3DSTATE_VERTEX_BUFFERS */
+   if (DIRTY(VERTEX_BUFFERS)) {
+      p->gen6_3DSTATE_VERTEX_BUFFERS(&p->gpe,
+            ilo->vertex_buffers.buffers, NULL,
+            (1 << ilo->vertex_buffers.num_buffers) - 1, p->cp);
+   }
+
+   /* 3DSTATE_VERTEX_ELEMENTS */
+   if (DIRTY(VERTEX_ELEMENTS) || DIRTY(VS)) {
+      const struct ilo_vertex_element *ive = ilo->vertex_elements;
+      bool last_velement_edgeflag = false;
+      bool prepend_generate_ids = false;
+
+      if (ilo->vs) {
+         const struct ilo_shader_info *info = &ilo->vs->info;
+
+         if (info->edgeflag_in >= 0) {
+            /* we rely on the state tracker here */
+            assert(info->edgeflag_in ==  ive->num_elements - 1);
+            last_velement_edgeflag = true;
+         }
+
+         prepend_generate_ids = (info->has_instanceid || info->has_vertexid);
+      }
+
+      p->gen6_3DSTATE_VERTEX_ELEMENTS(&p->gpe,
+            ive->elements, ive->num_elements,
+            last_velement_edgeflag, prepend_generate_ids, p->cp);
+   }
+}
+
+void
+gen6_pipeline_vf_statistics(struct ilo_3d_pipeline *p,
+                            const struct ilo_context *ilo,
+                            struct gen6_pipeline_session *session)
+{
+   /* 3DSTATE_VF_STATISTICS */
+   if (session->hw_ctx_changed)
+      p->gen6_3DSTATE_VF_STATISTICS(&p->gpe, false, p->cp);
+}
+
+void
+gen6_pipeline_vf_draw(struct ilo_3d_pipeline *p,
+                      const struct ilo_context *ilo,
+                      struct gen6_pipeline_session *session)
+{
+   /* 3DPRIMITIVE */
+   p->gen6_3DPRIMITIVE(&p->gpe, session->info, false, p->cp);
+   p->state.has_gen6_wa_pipe_control = false;
+}
+
+void
+gen6_pipeline_vs(struct ilo_3d_pipeline *p,
+                 const struct ilo_context *ilo,
+                 struct gen6_pipeline_session *session)
+{
+   const bool emit_3dstate_vs = (DIRTY(VS) || DIRTY(VERTEX_SAMPLERS));
+   const bool emit_3dstate_constant_vs = session->pcb_state_vs_changed;
+
+   /*
+    * the classic i965 does this in upload_vs_state(), citing a spec that I
+    * cannot find
+    */
+   if (emit_3dstate_vs && p->gen == ILO_GEN(6))
+      gen6_wa_pipe_control_post_sync(p, false);
+
+   /* 3DSTATE_CONSTANT_VS */
+   if (emit_3dstate_constant_vs) {
+      p->gen6_3DSTATE_CONSTANT_VS(&p->gpe,
+            &p->state.vs.PUSH_CONSTANT_BUFFER,
+            &p->state.vs.PUSH_CONSTANT_BUFFER_size,
+            1, p->cp);
+   }
+
+   /* 3DSTATE_VS */
+   if (emit_3dstate_vs) {
+      const struct ilo_shader *vs = (ilo->vs)? ilo->vs->shader : NULL;
+      const int num_samplers = ilo->samplers[PIPE_SHADER_VERTEX].num_samplers;
+
+      p->gen6_3DSTATE_VS(&p->gpe,
+            vs, ilo->max_vs_threads, num_samplers, p->cp);
+   }
+
+   if (emit_3dstate_constant_vs && p->gen == ILO_GEN(6))
+      gen6_wa_pipe_control_vs_const_flush(p);
+}
+
+static void
+gen6_pipeline_gs(struct ilo_3d_pipeline *p,
+                 const struct ilo_context *ilo,
+                 struct gen6_pipeline_session *session)
+{
+   /* 3DSTATE_CONSTANT_GS */
+   if (session->pcb_state_gs_changed)
+      p->gen6_3DSTATE_CONSTANT_GS(&p->gpe, NULL, NULL, 0, p->cp);
+
+   /* 3DSTATE_GS */
+   if (DIRTY(GS) || DIRTY(VS) || session->prim_changed) {
+      const struct ilo_shader *gs = (ilo->gs)? ilo->gs->shader : NULL;
+      const struct ilo_shader *vs = (ilo->vs)? ilo->vs->shader : NULL;
+      const int num_vertices = u_vertices_per_prim(session->reduced_prim);
+
+      if (gs)
+         assert(!gs->pcb.clip_state_size);
+
+      p->gen6_3DSTATE_GS(&p->gpe,
+            gs, ilo->max_gs_threads, vs,
+            (vs) ? vs->cache_offset + vs->gs_offsets[num_vertices - 1] : 0,
+            p->cp);
+   }
+}
+
+static void
+gen6_pipeline_gs_svbi(struct ilo_3d_pipeline *p,
+                      const struct ilo_context *ilo,
+                      struct gen6_pipeline_session *session)
+{
+   /* 3DSTATE_GS_SVB_INDEX */
+   if (DIRTY(VS) || DIRTY(GS) || DIRTY(STREAM_OUTPUT_TARGETS)) {
+      const struct pipe_stream_output_info *so_info =
+         (ilo->gs) ? &ilo->gs->info.stream_output :
+         (ilo->vs) ? &ilo->vs->info.stream_output : NULL;
+      unsigned max_svbi = 0xffffffff;
+      int i;
+
+      /* get max_svbi */
+      for (i = 0; so_info && i < so_info->num_outputs; i++) {
+         const int output_buffer = so_info->output[i].output_buffer;
+         const struct pipe_stream_output_target *so =
+            ilo->stream_output_targets.targets[output_buffer];
+         const int struct_size = so_info->stride[output_buffer] * 4;
+         const int elem_size = so_info->output[i].num_components * 4;
+         int buf_size, count;
+
+         if (!so)
+            continue;
+
+         buf_size = so->buffer_size - so_info->output[i].dst_offset * 4;
+
+         count = buf_size / struct_size;
+         if (buf_size % struct_size >= elem_size)
+            count++;
+
+         if (count < max_svbi)
+            max_svbi = count;
+      }
+
+      if (p->gen == ILO_GEN(6))
+         gen6_wa_pipe_control_post_sync(p, false);
+
+      p->gen6_3DSTATE_GS_SVB_INDEX(&p->gpe,
+            0, p->state.so_num_vertices, max_svbi,
+            false, p->cp);
+
+      if (session->hw_ctx_changed) {
+         /*
+          * From the Sandy Bridge PRM, volume 2 part 1, page 148:
+          *
+          *     "If a buffer is not enabled then the SVBI must be set to 0x0
+          *      in order to not cause overflow in that SVBI."
+          *
+          *     "If a buffer is not enabled then the MaxSVBI must be set to
+          *      0xFFFFFFFF in order to not cause overflow in that SVBI."
+          */
+         for (i = 1; i < 4; i++) {
+            p->gen6_3DSTATE_GS_SVB_INDEX(&p->gpe,
+                  i, 0, 0xffffffff, false, p->cp);
+         }
+      }
+
+      /* remember the state for calculating primtive emitted in software */
+      p->state.so_max_vertices = max_svbi;
+   }
+}
+
+void
+gen6_pipeline_clip(struct ilo_3d_pipeline *p,
+                   const struct ilo_context *ilo,
+                   struct gen6_pipeline_session *session)
+{
+   /* 3DSTATE_CLIP */
+   if (DIRTY(RASTERIZER) || DIRTY(FS) ||
+       DIRTY(VIEWPORT) || DIRTY(FRAMEBUFFER)) {
+      bool enable_guardband;
+      float x1, x2, y1, y2;
+
+      /*
+       * We do not do 2D clipping yet.  Guard band test should only be enabled
+       * when the viewport is larger than the framebuffer.
+       */
+      x1 = fabs(ilo->viewport.scale[0]) * -1.0f + ilo->viewport.translate[0];
+      x2 = fabs(ilo->viewport.scale[0]) *  1.0f + ilo->viewport.translate[0];
+      y1 = fabs(ilo->viewport.scale[1]) * -1.0f + ilo->viewport.translate[1];
+      y2 = fabs(ilo->viewport.scale[1]) *  1.0f + ilo->viewport.translate[1];
+      enable_guardband =
+         (x1 <= 0.0f && x2 >= (float) ilo->framebuffer.width &&
+          y1 <= 0.0f && y2 >= (float) ilo->framebuffer.height);
+
+      p->gen6_3DSTATE_CLIP(&p->gpe,
+            ilo->rasterizer,
+            (ilo->fs && ilo->fs->shader->in.has_linear_interp),
+            enable_guardband, 1, p->cp);
+   }
+}
+
+static void
+gen6_pipeline_sf(struct ilo_3d_pipeline *p,
+                 const struct ilo_context *ilo,
+                 struct gen6_pipeline_session *session)
+{
+   /* 3DSTATE_SF */
+   if (DIRTY(RASTERIZER) || DIRTY(VS) || DIRTY(GS) || DIRTY(FS)) {
+      const struct ilo_shader *fs = (ilo->fs)? ilo->fs->shader : NULL;
+      const struct ilo_shader *last_sh =
+         (ilo->gs)? ilo->gs->shader :
+         (ilo->vs)? ilo->vs->shader : NULL;
+
+      p->gen6_3DSTATE_SF(&p->gpe,
+            ilo->rasterizer, fs, last_sh, p->cp);
+   }
+}
+
+void
+gen6_pipeline_sf_rect(struct ilo_3d_pipeline *p,
+                      const struct ilo_context *ilo,
+                      struct gen6_pipeline_session *session)
+{
+   /* 3DSTATE_DRAWING_RECTANGLE */
+   if (DIRTY(FRAMEBUFFER)) {
+      if (p->gen == ILO_GEN(6))
+         gen6_wa_pipe_control_post_sync(p, false);
+
+      p->gen6_3DSTATE_DRAWING_RECTANGLE(&p->gpe, 0, 0,
+            ilo->framebuffer.width, ilo->framebuffer.height, p->cp);
+   }
+}
+
+static void
+gen6_pipeline_wm(struct ilo_3d_pipeline *p,
+                 const struct ilo_context *ilo,
+                 struct gen6_pipeline_session *session)
+{
+   /* 3DSTATE_CONSTANT_PS */
+   if (session->pcb_state_fs_changed)
+      p->gen6_3DSTATE_CONSTANT_PS(&p->gpe, NULL, NULL, 0, p->cp);
+
+   /* 3DSTATE_WM */
+   if (DIRTY(FS) || DIRTY(FRAGMENT_SAMPLERS) ||
+       DIRTY(BLEND) || DIRTY(DEPTH_STENCIL_ALPHA) ||
+       DIRTY(RASTERIZER)) {
+      const struct ilo_shader *fs = (ilo->fs)? ilo->fs->shader : NULL;
+      const int num_samplers =
+         ilo->samplers[PIPE_SHADER_FRAGMENT].num_samplers;
+      const bool dual_blend = (!ilo->blend->logicop_enable &&
+                                  ilo->blend->rt[0].blend_enable &&
+                                  util_blend_state_is_dual(ilo->blend, 0));
+      const bool cc_may_kill = (ilo->depth_stencil_alpha->alpha.enabled ||
+                                   ilo->blend->alpha_to_coverage);
+
+      if (fs)
+         assert(!fs->pcb.clip_state_size);
+
+      if (p->gen == ILO_GEN(6) && session->hw_ctx_changed)
+         gen6_wa_pipe_control_wm_max_threads_stall(p);
+
+      p->gen6_3DSTATE_WM(&p->gpe,
+            fs, ilo->max_wm_threads, num_samplers,
+            ilo->rasterizer, dual_blend, cc_may_kill, p->cp);
+   }
+}
+
+static void
+gen6_pipeline_wm_multisample(struct ilo_3d_pipeline *p,
+                             const struct ilo_context *ilo,
+                             struct gen6_pipeline_session *session)
+{
+   /* 3DSTATE_MULTISAMPLE and 3DSTATE_SAMPLE_MASK */
+   if (DIRTY(SAMPLE_MASK) || DIRTY(FRAMEBUFFER)) {
+      const uint32_t *packed_sample_pos;
+      int num_samples = 1;
+
+      if (ilo->framebuffer.nr_cbufs)
+         num_samples = ilo->framebuffer.cbufs[0]->texture->nr_samples;
+
+      packed_sample_pos = (num_samples > 1) ?
+         &p->packed_sample_position_4x : &p->packed_sample_position_1x;
+
+      if (p->gen == ILO_GEN(6)) {
+         gen6_wa_pipe_control_post_sync(p, false);
+         gen6_wa_pipe_control_wm_multisample_flush(p);
+      }
+
+      p->gen6_3DSTATE_MULTISAMPLE(&p->gpe, num_samples, packed_sample_pos,
+            ilo->rasterizer->half_pixel_center, p->cp);
+
+      p->gen6_3DSTATE_SAMPLE_MASK(&p->gpe,
+            (num_samples > 1) ? ilo->sample_mask : 0x1, p->cp);
+   }
+}
+
+static void
+gen6_pipeline_wm_depth(struct ilo_3d_pipeline *p,
+                       const struct ilo_context *ilo,
+                       struct gen6_pipeline_session *session)
+{
+   /* 3DSTATE_DEPTH_BUFFER and 3DSTATE_CLEAR_PARAMS */
+   if (DIRTY(FRAMEBUFFER)) {
+      if (p->gen == ILO_GEN(6)) {
+         gen6_wa_pipe_control_post_sync(p, false);
+         gen6_wa_pipe_control_wm_depth_flush(p);
+      }
+
+      p->gen6_3DSTATE_DEPTH_BUFFER(&p->gpe,
+            ilo->framebuffer.zsbuf, false, p->cp);
+
+      /* TODO */
+      p->gen6_3DSTATE_CLEAR_PARAMS(&p->gpe, 0, p->cp);
+   }
+}
+
+void
+gen6_pipeline_wm_raster(struct ilo_3d_pipeline *p,
+                        const struct ilo_context *ilo,
+                        struct gen6_pipeline_session *session)
+{
+   /* 3DSTATE_POLY_STIPPLE_PATTERN and 3DSTATE_POLY_STIPPLE_OFFSET */
+   if ((DIRTY(RASTERIZER) || DIRTY(POLY_STIPPLE)) &&
+       ilo->rasterizer->poly_stipple_enable) {
+      if (p->gen == ILO_GEN(6))
+         gen6_wa_pipe_control_post_sync(p, false);
+
+      p->gen6_3DSTATE_POLY_STIPPLE_PATTERN(&p->gpe,
+            &ilo->poly_stipple, p->cp);
+
+      p->gen6_3DSTATE_POLY_STIPPLE_OFFSET(&p->gpe, 0, 0, p->cp);
+   }
+
+   /* 3DSTATE_LINE_STIPPLE */
+   if (DIRTY(RASTERIZER) && ilo->rasterizer->line_stipple_enable) {
+      if (p->gen == ILO_GEN(6))
+         gen6_wa_pipe_control_post_sync(p, false);
+
+      p->gen6_3DSTATE_LINE_STIPPLE(&p->gpe,
+            ilo->rasterizer->line_stipple_pattern,
+            ilo->rasterizer->line_stipple_factor + 1, p->cp);
+   }
+
+   /* 3DSTATE_AA_LINE_PARAMETERS */
+   if (DIRTY(RASTERIZER) && ilo->rasterizer->line_smooth) {
+      if (p->gen == ILO_GEN(6))
+         gen6_wa_pipe_control_post_sync(p, false);
+
+      p->gen6_3DSTATE_AA_LINE_PARAMETERS(&p->gpe, p->cp);
+   }
+}
+
+static void
+gen6_pipeline_state_viewports(struct ilo_3d_pipeline *p,
+                              const struct ilo_context *ilo,
+                              struct gen6_pipeline_session *session)
+{
+   /* SF_VIEWPORT, CLIP_VIEWPORT, and CC_VIEWPORT */
+   if (DIRTY(VIEWPORT)) {
+      p->state.CLIP_VIEWPORT = p->gen6_CLIP_VIEWPORT(&p->gpe,
+            &ilo->viewport, 1, p->cp);
+
+      p->state.SF_VIEWPORT = p->gen6_SF_VIEWPORT(&p->gpe,
+            &ilo->viewport, 1, p->cp);
+
+      p->state.CC_VIEWPORT = p->gen6_CC_VIEWPORT(&p->gpe,
+            &ilo->viewport, 1, p->cp);
+
+      session->viewport_state_changed = true;
+   }
+}
+
+static void
+gen6_pipeline_state_cc(struct ilo_3d_pipeline *p,
+                       const struct ilo_context *ilo,
+                       struct gen6_pipeline_session *session)
+{
+   /* BLEND_STATE */
+   if (DIRTY(BLEND) || DIRTY(FRAMEBUFFER) || DIRTY(DEPTH_STENCIL_ALPHA)) {
+      p->state.BLEND_STATE = p->gen6_BLEND_STATE(&p->gpe,
+            ilo->blend, &ilo->framebuffer,
+            &ilo->depth_stencil_alpha->alpha, p->cp);
+
+      session->cc_state_blend_changed = true;
+   }
+
+   /* COLOR_CALC_STATE */
+   if (DIRTY(DEPTH_STENCIL_ALPHA) || DIRTY(STENCIL_REF) || DIRTY(BLEND_COLOR)) {
+      p->state.COLOR_CALC_STATE = p->gen6_COLOR_CALC_STATE(&p->gpe,
+            &ilo->stencil_ref,
+            ilo->depth_stencil_alpha->alpha.ref_value,
+            &ilo->blend_color, p->cp);
+
+      session->cc_state_cc_changed = true;
+   }
+
+   /* DEPTH_STENCIL_STATE */
+   if (DIRTY(DEPTH_STENCIL_ALPHA)) {
+      p->state.DEPTH_STENCIL_STATE =
+         p->gen6_DEPTH_STENCIL_STATE(&p->gpe,
+               ilo->depth_stencil_alpha, p->cp);
+
+      session->cc_state_dsa_changed = true;
+   }
+}
+
+static void
+gen6_pipeline_state_scissors(struct ilo_3d_pipeline *p,
+                             const struct ilo_context *ilo,
+                             struct gen6_pipeline_session *session)
+{
+   /* SCISSOR_RECT */
+   if (DIRTY(SCISSOR)) {
+      p->state.SCISSOR_RECT = p->gen6_SCISSOR_RECT(&p->gpe,
+            &ilo->scissor, 1, p->cp);
+
+      session->scissor_state_changed = true;
+   }
+}
+
+static void
+gen6_pipeline_state_surfaces_rt(struct ilo_3d_pipeline *p,
+                                const struct ilo_context *ilo,
+                                struct gen6_pipeline_session *session)
+{
+   /* SURFACE_STATEs for render targets */
+   if (DIRTY(FRAMEBUFFER)) {
+      const int offset = ILO_WM_DRAW_SURFACE(0);
+      uint32_t *surface_state = &p->state.wm.SURFACE_STATE[offset];
+      int i;
+
+      for (i = 0; i < ilo->framebuffer.nr_cbufs; i++) {
+         const struct pipe_surface *surface = ilo->framebuffer.cbufs[i];
+
+         assert(surface);
+         surface_state[i] =
+            p->gen6_surf_SURFACE_STATE(&p->gpe, surface, p->cp);
+      }
+
+      /*
+       * Upload at least one render target, as
+       * brw_update_renderbuffer_surfaces() does.  I don't know why.
+       */
+      if (i == 0) {
+         struct pipe_surface null_surface;
+
+         memset(&null_surface, 0, sizeof(null_surface));
+         null_surface.width = ilo->framebuffer.width;
+         null_surface.height = ilo->framebuffer.height;
+
+         surface_state[i] =
+            p->gen6_surf_SURFACE_STATE(&p->gpe, &null_surface, p->cp);
+
+         i++;
+      }
+
+      memset(&surface_state[i], 0, (ILO_MAX_DRAW_BUFFERS - i) * 4);
+
+      if (i && session->num_surfaces[PIPE_SHADER_FRAGMENT] < offset + i)
+         session->num_surfaces[PIPE_SHADER_FRAGMENT] = offset + i;
+
+      session->binding_table_fs_changed = true;
+   }
+}
+
+static void
+gen6_pipeline_state_surfaces_so(struct ilo_3d_pipeline *p,
+                                const struct ilo_context *ilo,
+                                struct gen6_pipeline_session *session)
+{
+   const struct ilo_shader_state *vs = ilo->vs;
+   const struct ilo_shader_state *gs = ilo->gs;
+   const struct pipe_stream_output_target **so_targets =
+      (const struct pipe_stream_output_target **)
+      ilo->stream_output_targets.targets;
+   const int num_so_targets = ilo->stream_output_targets.num_targets;
+
+   if (p->gen != ILO_GEN(6))
+      return;
+
+   /* SURFACE_STATEs for stream output targets */
+   if (DIRTY(VS) || DIRTY(GS) || DIRTY(STREAM_OUTPUT_TARGETS)) {
+      const struct pipe_stream_output_info *so_info =
+         (gs) ? &gs->info.stream_output :
+         (vs) ? &vs->info.stream_output : NULL;
+      const int offset = ILO_GS_SO_SURFACE(0);
+      uint32_t *surface_state = &p->state.gs.SURFACE_STATE[offset];
+      int i;
+
+      for (i = 0; so_info && i < so_info->num_outputs; i++) {
+         const int target = so_info->output[i].output_buffer;
+         const struct pipe_stream_output_target *so_target =
+            (target < num_so_targets) ? so_targets[target] : NULL;
+
+         if (so_target) {
+            surface_state[i] = p->gen6_so_SURFACE_STATE(&p->gpe,
+                  so_target, so_info, i, p->cp);
+         }
+         else {
+            surface_state[i] = 0;
+         }
+      }
+
+      memset(&surface_state[i], 0, (ILO_MAX_SO_BINDINGS - i) * 4);
+
+      if (i && session->num_surfaces[PIPE_SHADER_GEOMETRY] < offset + i)
+         session->num_surfaces[PIPE_SHADER_GEOMETRY] = offset + i;
+
+      session->binding_table_gs_changed = true;
+   }
+}
+
+static void
+gen6_pipeline_state_surfaces_view(struct ilo_3d_pipeline *p,
+                                  const struct ilo_context *ilo,
+                                  int shader_type,
+                                  struct gen6_pipeline_session *session)
+{
+   const struct pipe_sampler_view **views =
+      (const struct pipe_sampler_view **)
+      ilo->sampler_views[shader_type].views;
+   const int num_views = ilo->sampler_views[shader_type].num_views;
+   uint32_t *surface_state;
+   int offset, i;
+   bool skip = false;
+
+   /* SURFACE_STATEs for sampler views */
+   switch (shader_type) {
+   case PIPE_SHADER_VERTEX:
+      if (DIRTY(VERTEX_SAMPLER_VIEWS)) {
+         offset = ILO_VS_TEXTURE_SURFACE(0);
+         surface_state = &p->state.vs.SURFACE_STATE[offset];
+
+         session->binding_table_vs_changed = true;
+      }
+      else {
+         skip = true;
+      }
+      break;
+   case PIPE_SHADER_FRAGMENT:
+      if (DIRTY(FRAGMENT_SAMPLER_VIEWS)) {
+         offset = ILO_WM_TEXTURE_SURFACE(0);
+         surface_state = &p->state.wm.SURFACE_STATE[offset];
+
+         session->binding_table_fs_changed = true;
+      }
+      else {
+         skip = true;
+      }
+      break;
+   default:
+      skip = true;
+      break;
+   }
+
+   if (skip)
+      return;
+
+   for (i = 0; i < num_views; i++) {
+      if (views[i]) {
+         surface_state[i] =
+            p->gen6_view_SURFACE_STATE(&p->gpe, views[i], p->cp);
+      }
+      else {
+         surface_state[i] = 0;
+      }
+   }
+
+   memset(&surface_state[i], 0, (ILO_MAX_SAMPLER_VIEWS - i) * 4);
+
+   if (i && session->num_surfaces[shader_type] < offset + i)
+      session->num_surfaces[shader_type] = offset + i;
+}
+
+static void
+gen6_pipeline_state_surfaces_const(struct ilo_3d_pipeline *p,
+                                   const struct ilo_context *ilo,
+                                   int shader_type,
+                                   struct gen6_pipeline_session *session)
+{
+   const struct pipe_constant_buffer *buffers =
+      ilo->constant_buffers[shader_type].buffers;
+   const int num_buffers = ilo->constant_buffers[shader_type].num_buffers;
+   uint32_t *surface_state;
+   int offset, i;
+   bool skip = false;
+
+   /* SURFACE_STATEs for constant buffers */
+   switch (shader_type) {
+   case PIPE_SHADER_VERTEX:
+      if (DIRTY(CONSTANT_BUFFER)) {
+         offset = ILO_VS_CONST_SURFACE(0);
+         surface_state = &p->state.vs.SURFACE_STATE[offset];
+
+         session->binding_table_vs_changed = true;
+      }
+      else {
+         skip = true;
+      }
+      break;
+   case PIPE_SHADER_FRAGMENT:
+      if (DIRTY(CONSTANT_BUFFER)) {
+         offset = ILO_WM_CONST_SURFACE(0);
+         surface_state = &p->state.wm.SURFACE_STATE[offset];
+
+         session->binding_table_fs_changed = true;
+      }
+      else {
+         skip = true;
+      }
+      break;
+   default:
+      skip = true;
+      break;
+   }
+
+   if (skip)
+      return;
+
+   for (i = 0; i < num_buffers; i++) {
+      if (buffers[i].buffer) {
+         surface_state[i] =
+            p->gen6_cbuf_SURFACE_STATE(&p->gpe, &buffers[i], p->cp);
+      }
+      else {
+         surface_state[i] = 0;
+      }
+   }
+
+   memset(&surface_state[i], 0, (ILO_MAX_CONST_BUFFERS - i) * 4);
+
+   if (i && session->num_surfaces[shader_type] < offset + i)
+      session->num_surfaces[shader_type] = offset + i;
+}
+
+static void
+gen6_pipeline_state_binding_tables(struct ilo_3d_pipeline *p,
+                                   const struct ilo_context *ilo,
+                                   int shader_type,
+                                   struct gen6_pipeline_session *session)
+{
+   uint32_t *binding_table_state, *surface_state;
+   int *binding_table_state_size, size;
+   bool skip = false;
+
+   /* BINDING_TABLE_STATE */
+   switch (shader_type) {
+   case PIPE_SHADER_VERTEX:
+      surface_state = p->state.vs.SURFACE_STATE;
+      binding_table_state = &p->state.vs.BINDING_TABLE_STATE;
+      binding_table_state_size = &p->state.vs.BINDING_TABLE_STATE_size;
+
+      skip = !session->binding_table_vs_changed;
+      break;
+   case PIPE_SHADER_GEOMETRY:
+      surface_state = p->state.gs.SURFACE_STATE;
+      binding_table_state = &p->state.gs.BINDING_TABLE_STATE;
+      binding_table_state_size = &p->state.gs.BINDING_TABLE_STATE_size;
+
+      skip = !session->binding_table_gs_changed;
+      break;
+   case PIPE_SHADER_FRAGMENT:
+      surface_state = p->state.wm.SURFACE_STATE;
+      binding_table_state = &p->state.wm.BINDING_TABLE_STATE;
+      binding_table_state_size = &p->state.wm.BINDING_TABLE_STATE_size;
+
+      skip = !session->binding_table_fs_changed;
+      break;
+   default:
+      skip = true;
+      break;
+   }
+
+   if (skip)
+      return;
+
+   /*
+    * If we have seemingly less SURFACE_STATEs than before, it could be that
+    * we did not touch those reside at the tail in this upload.  Loop over
+    * them to figure out the real number of SURFACE_STATEs.
+    */
+   for (size = *binding_table_state_size;
+         size > session->num_surfaces[shader_type]; size--) {
+      if (surface_state[size - 1])
+         break;
+   }
+   if (size < session->num_surfaces[shader_type])
+      size = session->num_surfaces[shader_type];
+
+   *binding_table_state = p->gen6_BINDING_TABLE_STATE(&p->gpe,
+         surface_state, size, p->cp);
+   *binding_table_state_size = size;
+}
+
+static void
+gen6_pipeline_state_samplers(struct ilo_3d_pipeline *p,
+                             const struct ilo_context *ilo,
+                             int shader_type,
+                             struct gen6_pipeline_session *session)
+{
+   const struct pipe_sampler_state **samplers =
+      (const struct pipe_sampler_state **)
+      ilo->samplers[shader_type].samplers;
+   const struct pipe_sampler_view **views =
+      (const struct pipe_sampler_view **)
+      ilo->sampler_views[shader_type].views;
+   const int num_samplers = ilo->samplers[shader_type].num_samplers;
+   const int num_views = ilo->sampler_views[shader_type].num_views;
+   uint32_t *sampler_state, *border_color_state;
+   bool emit_border_color = false;
+   bool skip = false;
+
+   /* SAMPLER_BORDER_COLOR_STATE and SAMPLER_STATE */
+   switch (shader_type) {
+   case PIPE_SHADER_VERTEX:
+      if (DIRTY(VERTEX_SAMPLERS) || DIRTY(VERTEX_SAMPLER_VIEWS)) {
+         sampler_state = &p->state.vs.SAMPLER_STATE;
+         border_color_state = p->state.vs.SAMPLER_BORDER_COLOR_STATE;
+
+         if (DIRTY(VERTEX_SAMPLERS))
+            emit_border_color = true;
+
+         session->sampler_state_vs_changed = true;
+      }
+      else {
+         skip = true;
+      }
+      break;
+   case PIPE_SHADER_FRAGMENT:
+      if (DIRTY(FRAGMENT_SAMPLERS) || DIRTY(FRAGMENT_SAMPLER_VIEWS)) {
+         sampler_state = &p->state.wm.SAMPLER_STATE;
+         border_color_state = p->state.wm.SAMPLER_BORDER_COLOR_STATE;
+
+         if (DIRTY(FRAGMENT_SAMPLERS))
+            emit_border_color = true;
+
+         session->sampler_state_fs_changed = true;
+      }
+      else {
+         skip = true;
+      }
+      break;
+   default:
+      skip = true;
+      break;
+   }
+
+   if (skip)
+      return;
+
+   if (emit_border_color) {
+      int i;
+
+      for (i = 0; i < num_samplers; i++) {
+         border_color_state[i] = (samplers[i]) ?
+            p->gen6_SAMPLER_BORDER_COLOR_STATE(&p->gpe,
+                  &samplers[i]->border_color, p->cp) : 0;
+      }
+   }
+
+   /* should we take the minimum of num_samplers and num_views? */
+   *sampler_state = p->gen6_SAMPLER_STATE(&p->gpe,
+         samplers, views,
+         border_color_state,
+         MIN2(num_samplers, num_views), p->cp);
+}
+
+static void
+gen6_pipeline_state_pcb(struct ilo_3d_pipeline *p,
+                        const struct ilo_context *ilo,
+                        struct gen6_pipeline_session *session)
+{
+   /* push constant buffer for VS */
+   if (DIRTY(VS) || DIRTY(CLIP)) {
+      const struct ilo_shader *vs = (ilo->vs)? ilo->vs->shader : NULL;
+
+      if (vs && vs->pcb.clip_state_size) {
+         void *pcb;
+
+         p->state.vs.PUSH_CONSTANT_BUFFER_size = vs->pcb.clip_state_size;
+         p->state.vs.PUSH_CONSTANT_BUFFER =
+            p->gen6_push_constant_buffer(&p->gpe,
+                  p->state.vs.PUSH_CONSTANT_BUFFER_size, &pcb, p->cp);
+
+         memcpy(pcb, &ilo->clip, vs->pcb.clip_state_size);
+      }
+      else {
+         p->state.vs.PUSH_CONSTANT_BUFFER_size = 0;
+         p->state.vs.PUSH_CONSTANT_BUFFER = 0;
+      }
+
+      session->pcb_state_vs_changed = true;
+   }
+}
+
+#undef DIRTY
+
+static void
+gen6_pipeline_commands(struct ilo_3d_pipeline *p,
+                       const struct ilo_context *ilo,
+                       struct gen6_pipeline_session *session)
+{
+   /*
+    * We try to keep the order of the commands match, as closely as possible,
+    * that of the classic i965 driver.  It allows us to compare the command
+    * streams easily.
+    */
+   gen6_pipeline_common_select(p, ilo, session);
+   gen6_pipeline_gs_svbi(p, ilo, session);
+   gen6_pipeline_common_sip(p, ilo, session);
+   gen6_pipeline_vf_statistics(p, ilo, session);
+   gen6_pipeline_common_base_address(p, ilo, session);
+   gen6_pipeline_common_pointers_1(p, ilo, session);
+   gen6_pipeline_common_urb(p, ilo, session);
+   gen6_pipeline_common_pointers_2(p, ilo, session);
+   gen6_pipeline_wm_multisample(p, ilo, session);
+   gen6_pipeline_vs(p, ilo, session);
+   gen6_pipeline_gs(p, ilo, session);
+   gen6_pipeline_clip(p, ilo, session);
+   gen6_pipeline_sf(p, ilo, session);
+   gen6_pipeline_wm(p, ilo, session);
+   gen6_pipeline_common_pointers_3(p, ilo, session);
+   gen6_pipeline_wm_depth(p, ilo, session);
+   gen6_pipeline_wm_raster(p, ilo, session);
+   gen6_pipeline_sf_rect(p, ilo, session);
+   gen6_pipeline_vf(p, ilo, session);
+   gen6_pipeline_vf_draw(p, ilo, session);
+}
+
+void
+gen6_pipeline_states(struct ilo_3d_pipeline *p,
+                     const struct ilo_context *ilo,
+                     struct gen6_pipeline_session *session)
+{
+   int shader_type;
+
+   gen6_pipeline_state_viewports(p, ilo, session);
+   gen6_pipeline_state_cc(p, ilo, session);
+   gen6_pipeline_state_scissors(p, ilo, session);
+   gen6_pipeline_state_pcb(p, ilo, session);
+
+   /*
+    * upload all SURAFCE_STATEs together so that we know there are minimal
+    * paddings
+    */
+   gen6_pipeline_state_surfaces_rt(p, ilo, session);
+   gen6_pipeline_state_surfaces_so(p, ilo, session);
+   for (shader_type = 0; shader_type < PIPE_SHADER_TYPES; shader_type++) {
+      gen6_pipeline_state_surfaces_view(p, ilo, shader_type, session);
+      gen6_pipeline_state_surfaces_const(p, ilo, shader_type, session);
+   }
+
+   for (shader_type = 0; shader_type < PIPE_SHADER_TYPES; shader_type++) {
+      gen6_pipeline_state_samplers(p, ilo, shader_type, session);
+      /* this must be called after all SURFACE_STATEs are uploaded */
+      gen6_pipeline_state_binding_tables(p, ilo, shader_type, session);
+   }
+}
+
+void
+gen6_pipeline_prepare(const struct ilo_3d_pipeline *p,
+                      const struct ilo_context *ilo,
+                      const struct pipe_draw_info *info,
+                      struct gen6_pipeline_session *session)
+{
+   memset(session, 0, sizeof(*session));
+   session->info = info;
+   session->pipe_dirty = ilo->dirty;
+   session->reduced_prim = u_reduced_prim(info->mode);
+
+   /* available space before the session */
+   session->init_cp_space = ilo_cp_space(p->cp);
+
+   session->hw_ctx_changed =
+      (p->invalidate_flags & ILO_3D_PIPELINE_INVALIDATE_HW);
+
+   if (session->hw_ctx_changed) {
+      /* these should be enough to make everything uploaded */
+      session->state_bo_changed = true;
+      session->instruction_bo_changed = true;
+      session->prim_changed = true;
+   }
+   else {
+      session->state_bo_changed =
+         (p->invalidate_flags & ILO_3D_PIPELINE_INVALIDATE_STATE_BO);
+      session->instruction_bo_changed =
+         (p->invalidate_flags & ILO_3D_PIPELINE_INVALIDATE_KERNEL_BO);
+      session->prim_changed = (p->state.reduced_prim != session->reduced_prim);
+   }
+}
+
+void
+gen6_pipeline_draw(struct ilo_3d_pipeline *p,
+                   const struct ilo_context *ilo,
+                   struct gen6_pipeline_session *session)
+{
+   /* force all states to be uploaded if the state bo changed */
+   if (session->state_bo_changed)
+      session->pipe_dirty = ILO_DIRTY_ALL;
+   else
+      session->pipe_dirty = ilo->dirty;
+
+   session->emit_draw_states(p, ilo, session);
+
+   /* force all commands to be uploaded if the HW context changed */
+   if (session->hw_ctx_changed)
+      session->pipe_dirty = ILO_DIRTY_ALL;
+   else
+      session->pipe_dirty = ilo->dirty;
+
+   session->emit_draw_commands(p, ilo, session);
+}
+
+void
+gen6_pipeline_end(struct ilo_3d_pipeline *p,
+                  const struct ilo_context *ilo,
+                  struct gen6_pipeline_session *session)
+{
+   int used, estimate;
+
+   /* sanity check size estimation */
+   used = session->init_cp_space - ilo_cp_space(p->cp);
+   estimate = ilo_3d_pipeline_estimate_size(p, ILO_3D_PIPELINE_DRAW, ilo);
+   assert(used <= estimate);
+
+   p->state.reduced_prim = session->reduced_prim;
+}
+
+static void
+ilo_3d_pipeline_emit_draw_gen6(struct ilo_3d_pipeline *p,
+                               const struct ilo_context *ilo,
+                               const struct pipe_draw_info *info)
+{
+   struct gen6_pipeline_session session;
+
+   gen6_pipeline_prepare(p, ilo, info, &session);
+
+   session.emit_draw_states = gen6_pipeline_states;
+   session.emit_draw_commands = gen6_pipeline_commands;
+
+   gen6_pipeline_draw(p, ilo, &session);
+   gen6_pipeline_end(p, ilo, &session);
+}
+
+void
+ilo_3d_pipeline_emit_flush_gen6(struct ilo_3d_pipeline *p)
+{
+   if (p->gen == ILO_GEN(6))
+      gen6_wa_pipe_control_post_sync(p, false);
+
+   p->gen6_PIPE_CONTROL(&p->gpe,
+         PIPE_CONTROL_INSTRUCTION_FLUSH |
+         PIPE_CONTROL_WRITE_FLUSH |
+         PIPE_CONTROL_DEPTH_CACHE_FLUSH |
+         PIPE_CONTROL_VF_CACHE_INVALIDATE |
+         PIPE_CONTROL_TC_FLUSH |
+         PIPE_CONTROL_NO_WRITE |
+         PIPE_CONTROL_CS_STALL,
+         0, 0, false, p->cp);
+}
+
+void
+ilo_3d_pipeline_emit_write_timestamp_gen6(struct ilo_3d_pipeline *p,
+                                          struct intel_bo *bo, int index)
+{
+   if (p->gen == ILO_GEN(6))
+      gen6_wa_pipe_control_post_sync(p, true);
+
+   p->gen6_PIPE_CONTROL(&p->gpe,
+         PIPE_CONTROL_WRITE_TIMESTAMP,
+         bo, index * sizeof(uint64_t) | PIPE_CONTROL_GLOBAL_GTT_WRITE,
+         true, p->cp);
+}
+
+void
+ilo_3d_pipeline_emit_write_depth_count_gen6(struct ilo_3d_pipeline *p,
+                                            struct intel_bo *bo, int index)
+{
+   if (p->gen == ILO_GEN(6))
+      gen6_wa_pipe_control_post_sync(p, false);
+
+   p->gen6_PIPE_CONTROL(&p->gpe,
+         PIPE_CONTROL_DEPTH_STALL |
+         PIPE_CONTROL_WRITE_DEPTH_COUNT,
+         bo, index * sizeof(uint64_t) | PIPE_CONTROL_GLOBAL_GTT_WRITE,
+         true, p->cp);
+}
+
+static int
+gen6_pipeline_estimate_commands(const struct ilo_3d_pipeline *p,
+                                const struct ilo_gpe_gen6 *gen6,
+                                const struct ilo_context *ilo)
+{
+   static int size;
+   enum ilo_gpe_gen6_command cmd;
+
+   if (size)
+      return size;
+
+   for (cmd = 0; cmd < ILO_GPE_GEN6_COMMAND_COUNT; cmd++) {
+      int count;
+
+      switch (cmd) {
+      case ILO_GPE_GEN6_PIPE_CONTROL:
+         /* for the workaround */
+         count = 2;
+         /* another one after 3DSTATE_URB */
+         count += 1;
+         /* and another one after 3DSTATE_CONSTANT_VS */
+         count += 1;
+         break;
+      case ILO_GPE_GEN6_3DSTATE_GS_SVB_INDEX:
+         /* there are 4 SVBIs */
+         count = 4;
+         break;
+      case ILO_GPE_GEN6_3DSTATE_VERTEX_BUFFERS:
+         count = 33;
+         break;
+      case ILO_GPE_GEN6_3DSTATE_VERTEX_ELEMENTS:
+         count = 34;
+         break;
+      case ILO_GPE_GEN6_MEDIA_VFE_STATE:
+      case ILO_GPE_GEN6_MEDIA_CURBE_LOAD:
+      case ILO_GPE_GEN6_MEDIA_INTERFACE_DESCRIPTOR_LOAD:
+      case ILO_GPE_GEN6_MEDIA_GATEWAY_STATE:
+      case ILO_GPE_GEN6_MEDIA_STATE_FLUSH:
+      case ILO_GPE_GEN6_MEDIA_OBJECT_WALKER:
+         /* media commands */
+         count = 0;
+         break;
+      default:
+         count = 1;
+         break;
+      }
+
+      if (count)
+         size += gen6->estimate_command_size(&p->gpe, cmd, count);
+   }
+
+   return size;
+}
+
+static int
+gen6_pipeline_estimate_states(const struct ilo_3d_pipeline *p,
+                              const struct ilo_gpe_gen6 *gen6,
+                              const struct ilo_context *ilo)
+{
+   static int static_size;
+   int shader_type, count, size;
+
+   if (!static_size) {
+      struct {
+         enum ilo_gpe_gen6_state state;
+         int count;
+      } static_states[] = {
+         /* viewports */
+         { ILO_GPE_GEN6_SF_VIEWPORT,                 1 },
+         { ILO_GPE_GEN6_CLIP_VIEWPORT,               1 },
+         { ILO_GPE_GEN6_CC_VIEWPORT,                 1 },
+         /* cc */
+         { ILO_GPE_GEN6_COLOR_CALC_STATE,            1 },
+         { ILO_GPE_GEN6_BLEND_STATE,                 ILO_MAX_DRAW_BUFFERS },
+         { ILO_GPE_GEN6_DEPTH_STENCIL_STATE,         1 },
+         /* scissors */
+         { ILO_GPE_GEN6_SCISSOR_RECT,                1 },
+         /* binding table (vs, gs, fs) */
+         { ILO_GPE_GEN6_BINDING_TABLE_STATE,         ILO_MAX_VS_SURFACES },
+         { ILO_GPE_GEN6_BINDING_TABLE_STATE,         ILO_MAX_GS_SURFACES },
+         { ILO_GPE_GEN6_BINDING_TABLE_STATE,         ILO_MAX_WM_SURFACES },
+      };
+      int i;
+
+      for (i = 0; i < Elements(static_states); i++) {
+         static_size += gen6->estimate_state_size(&p->gpe,
+               static_states[i].state,
+               static_states[i].count);
+      }
+   }
+
+   size = static_size;
+
+   /*
+    * render targets (fs)
+    * stream outputs (gs)
+    * sampler views (vs, fs)
+    * constant buffers (vs, fs)
+    */
+   count = ilo->framebuffer.nr_cbufs;
+
+   if (ilo->gs)
+      count += ilo->gs->info.stream_output.num_outputs;
+   else if (ilo->vs)
+      count += ilo->vs->info.stream_output.num_outputs;
+
+   for (shader_type = 0; shader_type < PIPE_SHADER_TYPES; shader_type++) {
+      count += ilo->sampler_views[shader_type].num_views;
+      count += ilo->constant_buffers[shader_type].num_buffers;
+   }
+
+   if (count) {
+      size += gen6->estimate_state_size(&p->gpe,
+            ILO_GPE_GEN6_SURFACE_STATE, count);
+   }
+
+   /* samplers (vs, fs) */
+   for (shader_type = 0; shader_type < PIPE_SHADER_TYPES; shader_type++) {
+      count = ilo->samplers[shader_type].num_samplers;
+      if (count) {
+         size += gen6->estimate_state_size(&p->gpe,
+               ILO_GPE_GEN6_SAMPLER_BORDER_COLOR_STATE, count);
+         size += gen6->estimate_state_size(&p->gpe,
+               ILO_GPE_GEN6_SAMPLER_STATE, count);
+      }
+   }
+
+   /* pcb (vs) */
+   if (ilo->vs && ilo->vs->shader->pcb.clip_state_size) {
+      const int pcb_size = ilo->vs->shader->pcb.clip_state_size;
+
+      size += gen6->estimate_state_size(&p->gpe,
+            ILO_GPE_GEN6_PUSH_CONSTANT_BUFFER, pcb_size);
+   }
+
+   return size;
+}
+
+static int
+ilo_3d_pipeline_estimate_size_gen6(struct ilo_3d_pipeline *p,
+                                   enum ilo_3d_pipeline_action action,
+                                   const void *arg)
+{
+   const struct ilo_gpe_gen6 *gen6 = ilo_gpe_gen6_get();
+   int size;
+
+   switch (action) {
+   case ILO_3D_PIPELINE_DRAW:
+      {
+         const struct ilo_context *ilo = arg;
+
+         size = gen6_pipeline_estimate_commands(p, gen6, ilo) +
+            gen6_pipeline_estimate_states(p, gen6, ilo);
+      }
+      break;
+   case ILO_3D_PIPELINE_FLUSH:
+      size = gen6->estimate_command_size(&p->gpe,
+            ILO_GPE_GEN6_PIPE_CONTROL, 1) * 3;
+      break;
+   case ILO_3D_PIPELINE_WRITE_TIMESTAMP:
+      size = gen6->estimate_command_size(&p->gpe,
+            ILO_GPE_GEN6_PIPE_CONTROL, 1) * 2;
+      break;
+   case ILO_3D_PIPELINE_WRITE_DEPTH_COUNT:
+      size = gen6->estimate_command_size(&p->gpe,
+            ILO_GPE_GEN6_PIPE_CONTROL, 1) * 3;
+      break;
+   default:
+      assert(!"unknown 3D pipeline action");
+      size = 0;
+      break;
+   }
+
+   return size;
+}
+
+void
+ilo_3d_pipeline_init_gen6(struct ilo_3d_pipeline *p)
+{
+   const struct ilo_gpe_gen6 *gen6 = ilo_gpe_gen6_get();
+
+   p->estimate_size = ilo_3d_pipeline_estimate_size_gen6;
+   p->emit_draw = ilo_3d_pipeline_emit_draw_gen6;
+   p->emit_flush = ilo_3d_pipeline_emit_flush_gen6;
+   p->emit_write_timestamp = ilo_3d_pipeline_emit_write_timestamp_gen6;
+   p->emit_write_depth_count = ilo_3d_pipeline_emit_write_depth_count_gen6;
+
+#define GEN6_USE(p, name, from) \
+   p->gen6_ ## name = from->emit_ ## name
+   GEN6_USE(p, STATE_BASE_ADDRESS, gen6);
+   GEN6_USE(p, STATE_SIP, gen6);
+   GEN6_USE(p, PIPELINE_SELECT, gen6);
+   GEN6_USE(p, 3DSTATE_BINDING_TABLE_POINTERS, gen6);
+   GEN6_USE(p, 3DSTATE_SAMPLER_STATE_POINTERS, gen6);
+   GEN6_USE(p, 3DSTATE_URB, gen6);
+   GEN6_USE(p, 3DSTATE_VERTEX_BUFFERS, gen6);
+   GEN6_USE(p, 3DSTATE_VERTEX_ELEMENTS, gen6);
+   GEN6_USE(p, 3DSTATE_INDEX_BUFFER, gen6);
+   GEN6_USE(p, 3DSTATE_VF_STATISTICS, gen6);
+   GEN6_USE(p, 3DSTATE_VIEWPORT_STATE_POINTERS, gen6);
+   GEN6_USE(p, 3DSTATE_CC_STATE_POINTERS, gen6);
+   GEN6_USE(p, 3DSTATE_SCISSOR_STATE_POINTERS, gen6);
+   GEN6_USE(p, 3DSTATE_VS, gen6);
+   GEN6_USE(p, 3DSTATE_GS, gen6);
+   GEN6_USE(p, 3DSTATE_CLIP, gen6);
+   GEN6_USE(p, 3DSTATE_SF, gen6);
+   GEN6_USE(p, 3DSTATE_WM, gen6);
+   GEN6_USE(p, 3DSTATE_CONSTANT_VS, gen6);
+   GEN6_USE(p, 3DSTATE_CONSTANT_GS, gen6);
+   GEN6_USE(p, 3DSTATE_CONSTANT_PS, gen6);
+   GEN6_USE(p, 3DSTATE_SAMPLE_MASK, gen6);
+   GEN6_USE(p, 3DSTATE_DRAWING_RECTANGLE, gen6);
+   GEN6_USE(p, 3DSTATE_DEPTH_BUFFER, gen6);
+   GEN6_USE(p, 3DSTATE_POLY_STIPPLE_OFFSET, gen6);
+   GEN6_USE(p, 3DSTATE_POLY_STIPPLE_PATTERN, gen6);
+   GEN6_USE(p, 3DSTATE_LINE_STIPPLE, gen6);
+   GEN6_USE(p, 3DSTATE_AA_LINE_PARAMETERS, gen6);
+   GEN6_USE(p, 3DSTATE_GS_SVB_INDEX, gen6);
+   GEN6_USE(p, 3DSTATE_MULTISAMPLE, gen6);
+   GEN6_USE(p, 3DSTATE_STENCIL_BUFFER, gen6);
+   GEN6_USE(p, 3DSTATE_HIER_DEPTH_BUFFER, gen6);
+   GEN6_USE(p, 3DSTATE_CLEAR_PARAMS, gen6);
+   GEN6_USE(p, PIPE_CONTROL, gen6);
+   GEN6_USE(p, 3DPRIMITIVE, gen6);
+   GEN6_USE(p, INTERFACE_DESCRIPTOR_DATA, gen6);
+   GEN6_USE(p, SF_VIEWPORT, gen6);
+   GEN6_USE(p, CLIP_VIEWPORT, gen6);
+   GEN6_USE(p, CC_VIEWPORT, gen6);
+   GEN6_USE(p, COLOR_CALC_STATE, gen6);
+   GEN6_USE(p, BLEND_STATE, gen6);
+   GEN6_USE(p, DEPTH_STENCIL_STATE, gen6);
+   GEN6_USE(p, SCISSOR_RECT, gen6);
+   GEN6_USE(p, BINDING_TABLE_STATE, gen6);
+   GEN6_USE(p, surf_SURFACE_STATE, gen6);
+   GEN6_USE(p, view_SURFACE_STATE, gen6);
+   GEN6_USE(p, cbuf_SURFACE_STATE, gen6);
+   GEN6_USE(p, so_SURFACE_STATE, gen6);
+   GEN6_USE(p, SAMPLER_STATE, gen6);
+   GEN6_USE(p, SAMPLER_BORDER_COLOR_STATE, gen6);
+   GEN6_USE(p, push_constant_buffer, gen6);
+#undef GEN6_USE
+}
diff --git a/src/gallium/drivers/ilo/ilo_3d_pipeline_gen6.h b/src/gallium/drivers/ilo/ilo_3d_pipeline_gen6.h
new file mode 100644 (file)
index 0000000..8cc6ab2
--- /dev/null
@@ -0,0 +1,161 @@
+/*
+ * Mesa 3-D graphics library
+ *
+ * Copyright (C) 2012-2013 LunarG, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Chia-I Wu <olv@lunarg.com>
+ */
+
+#ifndef ILO_3D_PIPELINE_GEN6_H
+#define ILO_3D_PIPELINE_GEN6_H
+
+#include "ilo_common.h"
+
+struct ilo_3d_pipeline;
+struct ilo_context;
+
+struct gen6_pipeline_session {
+   const struct pipe_draw_info *info;
+
+   uint32_t pipe_dirty;
+
+   int reduced_prim;
+   int init_cp_space;
+
+   bool hw_ctx_changed;
+   bool state_bo_changed;
+   bool instruction_bo_changed;
+   bool prim_changed;
+
+   void (*emit_draw_states)(struct ilo_3d_pipeline *p,
+                            const struct ilo_context *ilo,
+                            struct gen6_pipeline_session *session);
+
+   void (*emit_draw_commands)(struct ilo_3d_pipeline *p,
+                              const struct ilo_context *ilo,
+                              struct gen6_pipeline_session *session);
+
+   /* indirect states */
+   bool viewport_state_changed;
+   bool cc_state_blend_changed;
+   bool cc_state_dsa_changed;
+   bool cc_state_cc_changed;
+   bool scissor_state_changed;
+   bool binding_table_vs_changed;
+   bool binding_table_gs_changed;
+   bool binding_table_fs_changed;
+   bool sampler_state_vs_changed;
+   bool sampler_state_gs_changed;
+   bool sampler_state_fs_changed;
+   bool pcb_state_vs_changed;
+   bool pcb_state_gs_changed;
+   bool pcb_state_fs_changed;
+
+   int num_surfaces[PIPE_SHADER_TYPES];
+};
+
+void
+gen6_pipeline_prepare(const struct ilo_3d_pipeline *p,
+                      const struct ilo_context *ilo,
+                      const struct pipe_draw_info *info,
+                      struct gen6_pipeline_session *session);
+
+void
+gen6_pipeline_draw(struct ilo_3d_pipeline *p,
+                   const struct ilo_context *ilo,
+                   struct gen6_pipeline_session *session);
+
+void
+gen6_pipeline_end(struct ilo_3d_pipeline *p,
+                  const struct ilo_context *ilo,
+                  struct gen6_pipeline_session *session);
+
+void
+gen6_pipeline_common_select(struct ilo_3d_pipeline *p,
+                            const struct ilo_context *ilo,
+                            struct gen6_pipeline_session *session);
+
+void
+gen6_pipeline_common_sip(struct ilo_3d_pipeline *p,
+                         const struct ilo_context *ilo,
+                         struct gen6_pipeline_session *session);
+
+void
+gen6_pipeline_common_base_address(struct ilo_3d_pipeline *p,
+                                  const struct ilo_context *ilo,
+                                  struct gen6_pipeline_session *session);
+
+void
+gen6_pipeline_vf(struct ilo_3d_pipeline *p,
+                 const struct ilo_context *ilo,
+                 struct gen6_pipeline_session *session);
+
+void
+gen6_pipeline_vf_statistics(struct ilo_3d_pipeline *p,
+                            const struct ilo_context *ilo,
+                            struct gen6_pipeline_session *session);
+
+void
+gen6_pipeline_vf_draw(struct ilo_3d_pipeline *p,
+                      const struct ilo_context *ilo,
+                      struct gen6_pipeline_session *session);
+
+void
+gen6_pipeline_vs(struct ilo_3d_pipeline *p,
+                 const struct ilo_context *ilo,
+                 struct gen6_pipeline_session *session);
+
+void
+gen6_pipeline_clip(struct ilo_3d_pipeline *p,
+                   const struct ilo_context *ilo,
+                   struct gen6_pipeline_session *session);
+
+void
+gen6_pipeline_sf_rect(struct ilo_3d_pipeline *p,
+                      const struct ilo_context *ilo,
+                      struct gen6_pipeline_session *session);
+
+void
+gen6_pipeline_wm_raster(struct ilo_3d_pipeline *p,
+                        const struct ilo_context *ilo,
+                        struct gen6_pipeline_session *session);
+
+void
+gen6_pipeline_states(struct ilo_3d_pipeline *p,
+                     const struct ilo_context *ilo,
+                     struct gen6_pipeline_session *session);
+
+void
+ilo_3d_pipeline_emit_flush_gen6(struct ilo_3d_pipeline *p);
+
+void
+ilo_3d_pipeline_emit_write_timestamp_gen6(struct ilo_3d_pipeline *p,
+                                          struct intel_bo *bo, int index);
+
+void
+ilo_3d_pipeline_emit_write_depth_count_gen6(struct ilo_3d_pipeline *p,
+                                            struct intel_bo *bo, int index);
+
+void
+ilo_3d_pipeline_init_gen6(struct ilo_3d_pipeline *p);
+
+#endif /* ILO_3D_PIPELINE_GEN6_H */