Merge branch 'master' into asm-shader-rework-1

Conflicts:
	src/mesa/shader/arbprogparse.c

4 files changed, 112 insertions, 21 deletions

diff --git a/src/gallium/winsys/drm/intel/gem/intel_be_batchbuffer.c b/src/gallium/winsys/drm/intel/gem/intel_be_batchbuffer.c
index d5e63c3bae..c4a79586e6 100644
--- a/src/gallium/winsys/drm/intel/gem/intel_be_batchbuffer.c
+++ b/src/gallium/winsys/drm/intel/gem/intel_be_batchbuffer.c
@@ -21,12 +21,10 @@ intel_be_batchbuffer_alloc(struct intel_be_context *intel)
 	batch->base.size = 0;
 	batch->base.actual_size = intel->device->max_batch_size;
 	batch->base.relocs = 0;
-	batch->base.max_relocs = 500;/*INTEL_DEFAULT_RELOCS;*/
+	batch->base.max_relocs = 100;/*INTEL_DEFAULT_RELOCS;*/
 
-	batch->base.map = malloc(batch->base.actual_size);
-	memset(batch->base.map, 0, batch->base.actual_size);
-
-	batch->base.ptr = batch->base.map;
+	batch->intel = intel;
+	batch->device = intel->device;
 
 	intel_be_batchbuffer_reset(batch);
 
@@ -41,16 +39,17 @@ intel_be_batchbuffer_reset(struct intel_be_batchbuffer *batch)
 
 	if (batch->bo)
 		drm_intel_bo_unreference(batch->bo);
+	batch->bo = drm_intel_bo_alloc(dev->pools.gem,
+	                               "gallium3d_batch_buffer",
+	                               batch->base.actual_size,
+	                               4096);
+	drm_intel_bo_map(batch->bo, TRUE);
+	batch->base.map = batch->bo->virtual;
 
 	memset(batch->base.map, 0, batch->base.actual_size);
 	batch->base.ptr = batch->base.map;
 	batch->base.size = batch->base.actual_size - BATCH_RESERVED;
-
 	batch->base.relocs = 0;
-
-	batch->bo = drm_intel_bo_alloc(dev->pools.gem,
-	                               "gallium3d_batch_buffer",
-	                               batch->base.actual_size, 0);
 }
 
 int
@@ -88,6 +87,7 @@ intel_be_batchbuffer_flush(struct intel_be_batchbuffer *batch,
 	struct i915_batchbuffer *i915 = &batch->base;
 	unsigned used = 0;
 	int ret = 0;
+	int i;
 
 	assert(i915_batchbuffer_space(i915) >= 0);
 
@@ -105,11 +105,29 @@ intel_be_batchbuffer_flush(struct intel_be_batchbuffer *batch,
 
 	used = batch->base.ptr - batch->base.map;
 
-	drm_intel_bo_subdata(batch->bo, 0, used, batch->base.map);
-	ret = drm_intel_bo_exec(batch->bo, used, NULL, 0, 0);
+	drm_intel_bo_unmap(batch->bo);
 
+	/* Do the sending to HW */
+	ret = drm_intel_bo_exec(batch->bo, used, NULL, 0, 0);
 	assert(ret == 0);
 
+	if (batch->device->dump_cmd) {
+		unsigned *ptr;
+		drm_intel_bo_map(batch->bo, FALSE);
+		ptr = (unsigned*)batch->bo->virtual;
+
+		debug_printf("%s:\n", __func__);
+		for (i = 0; i < used / 4; i++, ptr++) {
+			debug_printf("\t%08x:    %08x\n", i*4, *ptr);
+		}
+
+		drm_intel_bo_unmap(batch->bo);
+	} else {
+		/* TODO figgure out why the gpu hangs if we don't run sync */
+		drm_intel_bo_map(batch->bo, FALSE);
+		drm_intel_bo_unmap(batch->bo);
+	}
+
 	intel_be_batchbuffer_reset(batch);
 
 	if (fence) {
@@ -134,6 +152,5 @@ intel_be_batchbuffer_free(struct intel_be_batchbuffer *batch)
 	if (batch->bo)
 		drm_intel_bo_unreference(batch->bo);
 
-	free(batch->base.map);
 	free(batch);
 }
diff --git a/src/gallium/winsys/drm/intel/gem/intel_be_context.c b/src/gallium/winsys/drm/intel/gem/intel_be_context.c
index db84f9af51..629987c6f9 100644
--- a/src/gallium/winsys/drm/intel/gem/intel_be_context.c
+++ b/src/gallium/winsys/drm/intel/gem/intel_be_context.c
@@ -36,7 +36,7 @@ intel_be_batch_reloc(struct i915_winsys *sws,
 	}
 
 	if (access_flags & I915_BUFFER_ACCESS_READ) {
-		read |= I915_GEM_DOMAIN_VERTEX;
+		read |= I915_GEM_DOMAIN_SAMPLER;
 	}
 
 	ret = intel_be_offset_relocation(intel->batch,
diff --git a/src/gallium/winsys/drm/intel/gem/intel_be_device.c b/src/gallium/winsys/drm/intel/gem/intel_be_device.c
index e3630f5d12..5312865a03 100644
--- a/src/gallium/winsys/drm/intel/gem/intel_be_device.c
+++ b/src/gallium/winsys/drm/intel/gem/intel_be_device.c
@@ -7,6 +7,7 @@
 #include "pipe/p_inlines.h"
 #include "util/u_memory.h"
 #include "util/u_debug.h"
+#include "util/u_math.h"
 
 #include "intel_be_fence.h"
 
@@ -16,6 +17,8 @@
 #include "intel_be_api.h"
 #include <stdio.h>
 
+#define I915_TILING_X 1
+
 /*
  * Buffer
  */
@@ -25,9 +28,10 @@ intel_be_buffer_map(struct pipe_winsys *winsys,
 		    struct pipe_buffer *buf,
 		    unsigned flags)
 {
+	struct intel_be_buffer *buffer = intel_be_buffer(buf);
 	drm_intel_bo *bo = intel_bo(buf);
 	int write = 0;
-	int ret;
+	int ret = 0;
 
         if (flags & PIPE_BUFFER_USAGE_DONTBLOCK) {
            /* Remove this when drm_intel_bo_map supports DONTBLOCK 
@@ -38,19 +42,37 @@ intel_be_buffer_map(struct pipe_winsys *winsys,
 	if (flags & PIPE_BUFFER_USAGE_CPU_WRITE)
 		write = 1;
 
-	ret = drm_intel_bo_map(bo, write);
+	if (buffer->map_count)
+		goto out;
+
+	if (buffer->map_gtt)
+		ret = drm_intel_gem_bo_map_gtt(bo);
+	else
+		ret = drm_intel_bo_map(bo, write);
+
+	buffer->ptr = bo->virtual;
 
+out:
 	if (ret)
 		return NULL;
 
-	return bo->virtual;
+	buffer->map_count++;
+	return buffer->ptr;
 }
 
 static void
 intel_be_buffer_unmap(struct pipe_winsys *winsys,
 		      struct pipe_buffer *buf)
 {
-	drm_intel_bo_unmap(intel_bo(buf));
+	struct intel_be_buffer *buffer = intel_be_buffer(buf);
+
+	if (--buffer->map_count)
+		return;
+
+	if (buffer->map_gtt)
+		drm_intel_gem_bo_unmap_gtt(intel_bo(buf));
+	else
+		drm_intel_bo_unmap(intel_bo(buf));
 }
 
 static void
@@ -80,8 +102,13 @@ intel_be_buffer_create(struct pipe_winsys *winsys,
 	buffer->base.size = size;
 	buffer->flinked = FALSE;
 	buffer->flink = 0;
+	buffer->map_gtt = FALSE;
 
-	if (usage & (PIPE_BUFFER_USAGE_VERTEX | PIPE_BUFFER_USAGE_CONSTANT)) {
+	if (usage & I915_BUFFER_USAGE_SCANOUT) {
+		/* Scanout buffer */
+		name = "gallium3d_scanout";
+		pool = dev->pools.gem;
+	} else if (usage & (PIPE_BUFFER_USAGE_VERTEX | PIPE_BUFFER_USAGE_CONSTANT)) {
 		/* Local buffer */
 		name = "gallium3d_local";
 		pool = dev->pools.gem;
@@ -96,6 +123,12 @@ intel_be_buffer_create(struct pipe_winsys *winsys,
 	}
 
 	buffer->bo = drm_intel_bo_alloc(pool, name, size, alignment);
+	if (usage & I915_BUFFER_USAGE_SCANOUT) {
+		unsigned tiling = I915_TILING_X;
+		unsigned stride = 2048 * 4; /* TODO do something smarter here */
+		drm_intel_bo_set_tiling(buffer->bo, &tiling, stride);
+		buffer->map_gtt = TRUE;
+	}
 
 	if (!buffer->bo)
 		goto err;
@@ -142,6 +175,40 @@ err:
 	return NULL;
 }
 
+static struct pipe_buffer *
+intel_be_surface_buffer_create(struct pipe_winsys *winsys,
+                               unsigned width, unsigned height,
+                               enum pipe_format format,
+                               unsigned usage,
+                               unsigned tex_usage,
+                               unsigned *stride)
+{
+	struct pipe_format_block block;
+	unsigned buf_usage = 0;
+	unsigned buf_stride = 0;
+	unsigned buf_size = 0;
+
+	pf_get_block(format, &block);
+	buf_stride = pf_get_stride(&block, width);
+	buf_stride = align(buf_stride, 64);
+
+	if (tex_usage & PIPE_TEXTURE_USAGE_PRIMARY) {
+		/* TODO more checks */
+		assert(buf_stride <= 2048*4);
+		assert(height % 8 == 0);
+		buf_stride = 2048 * 4;
+		buf_usage |= I915_BUFFER_USAGE_SCANOUT;
+	}
+
+	buf_size = buf_stride * height;
+	*stride = buf_stride;
+
+	return intel_be_buffer_create(winsys,
+	                              0,
+	                              buf_usage,
+	                              buf_size);
+}
+
 boolean
 intel_be_get_texture_buffer(struct drm_api *api,
                             struct pipe_texture *texture,
@@ -225,6 +292,7 @@ intel_be_global_handle_from_buffer(struct drm_api *api,
 	*handle = buf->flink;
 	return TRUE;
 }
+
 /*
  * Fence
  */
@@ -296,8 +364,8 @@ intel_be_init_device(struct intel_be_device *dev, int fd, unsigned id)
 	dev->base.buffer_unmap = intel_be_buffer_unmap;
 	dev->base.buffer_destroy = intel_be_buffer_destroy;
 
-	/* Not used anymore */
-	dev->base.surface_buffer_create = NULL;
+	/* Used by softpipe */
+	dev->base.surface_buffer_create = intel_be_surface_buffer_create;
 
 	dev->base.fence_reference = intel_be_fence_refunref;
 	dev->base.fence_signalled = intel_be_fence_signalled;
@@ -308,6 +376,7 @@ intel_be_init_device(struct intel_be_device *dev, int fd, unsigned id)
 	dev->pools.gem = drm_intel_bufmgr_gem_init(dev->fd, dev->max_batch_size);
 
 	dev->softpipe = debug_get_bool_option("INTEL_SOFTPIPE", FALSE);
+	dev->dump_cmd = debug_get_bool_option("INTEL_DUMP_CMD", FALSE);
 
 	return true;
 }
diff --git a/src/gallium/winsys/drm/intel/gem/intel_be_device.h b/src/gallium/winsys/drm/intel/gem/intel_be_device.h
index 56d95bd7fe..c397048f8c 100644
--- a/src/gallium/winsys/drm/intel/gem/intel_be_device.h
+++ b/src/gallium/winsys/drm/intel/gem/intel_be_device.h
@@ -19,6 +19,7 @@ struct intel_be_device
 	struct pipe_winsys base;
 
 	boolean softpipe;
+	boolean dump_cmd;
 
 	int fd; /**< Drm file discriptor */
 
@@ -47,6 +48,10 @@ intel_be_init_device(struct intel_be_device *device, int fd, unsigned id);
 
 struct intel_be_buffer {
 	struct pipe_buffer base;
+	void *ptr;
+	unsigned map_count;
+	boolean map_gtt;
+
 	drm_intel_bo *bo;
 	boolean flinked;
 	unsigned flink;