*
**************************************************************************/
+#undef NDEBUG
#include "main/glheader.h"
#include "main/bufferobj.h"
#include "main/context.h"
-#include "main/state.h"
-#include "main/api_validate.h"
#include "main/enums.h"
#include "brw_draw.h"
#include "brw_defines.h"
#include "brw_context.h"
#include "brw_state.h"
-#include "brw_fallback.h"
#include "intel_batchbuffer.h"
#include "intel_buffer_objects.h"
-#include "intel_tex.h"
static GLuint double_types[5] = {
0,
BRW_SURFACEFORMAT_R32G32B32A32_FLOAT
};
+static GLuint half_float_types[5] = {
+ 0,
+ BRW_SURFACEFORMAT_R16_FLOAT,
+ BRW_SURFACEFORMAT_R16G16_FLOAT,
+ BRW_SURFACEFORMAT_R16G16B16A16_FLOAT,
+ BRW_SURFACEFORMAT_R16G16B16A16_FLOAT
+};
+
static GLuint uint_types_norm[5] = {
0,
BRW_SURFACEFORMAT_R32_UNORM,
static GLuint get_surface_type( GLenum type, GLuint size,
GLenum format, GLboolean normalized )
{
- if (INTEL_DEBUG & DEBUG_VERTS)
- _mesa_printf("type %s size %d normalized %d\n",
+ if (unlikely(INTEL_DEBUG & DEBUG_VERTS))
+ printf("type %s size %d normalized %d\n",
_mesa_lookup_enum_by_nr(type), size, normalized);
if (normalized) {
switch (type) {
case GL_DOUBLE: return double_types[size];
case GL_FLOAT: return float_types[size];
+ case GL_HALF_FLOAT: return half_float_types[size];
case GL_INT: return int_types_norm[size];
case GL_SHORT: return short_types_norm[size];
case GL_BYTE: return byte_types_norm[size];
switch (type) {
case GL_DOUBLE: return double_types[size];
case GL_FLOAT: return float_types[size];
+ case GL_HALF_FLOAT: return half_float_types[size];
case GL_INT: return int_types_scale[size];
case GL_SHORT: return short_types_scale[size];
case GL_BYTE: return byte_types_scale[size];
switch (type) {
case GL_DOUBLE: return sizeof(GLdouble);
case GL_FLOAT: return sizeof(GLfloat);
+ case GL_HALF_FLOAT: return sizeof(GLhalfARB);
case GL_INT: return sizeof(GLint);
case GL_SHORT: return sizeof(GLshort);
case GL_BYTE: return sizeof(GLbyte);
}
}
-static void wrap_buffers( struct brw_context *brw,
- GLuint size )
-{
- if (size < BRW_UPLOAD_INIT_SIZE)
- size = BRW_UPLOAD_INIT_SIZE;
-
- brw->vb.upload.offset = 0;
-
- if (brw->vb.upload.bo != NULL)
- dri_bo_unreference(brw->vb.upload.bo);
- brw->vb.upload.bo = dri_bo_alloc(brw->intel.bufmgr, "temporary VBO",
- size, 1);
-}
-
-static void get_space( struct brw_context *brw,
- GLuint size,
- dri_bo **bo_return,
- GLuint *offset_return )
-{
- size = ALIGN(size, 64);
-
- if (brw->vb.upload.bo == NULL ||
- brw->vb.upload.offset + size > brw->vb.upload.bo->size) {
- wrap_buffers(brw, size);
- }
-
- assert(*bo_return == NULL);
- dri_bo_reference(brw->vb.upload.bo);
- *bo_return = brw->vb.upload.bo;
- *offset_return = brw->vb.upload.offset;
- brw->vb.upload.offset += size;
-}
-
static void
copy_array_to_vbo_array( struct brw_context *brw,
struct brw_vertex_element *element,
GLuint dst_stride)
{
- struct intel_context *intel = &brw->intel;
GLuint size = element->count * dst_stride;
- get_space(brw, size, &element->bo, &element->offset);
-
if (element->glarray->StrideB == 0) {
assert(element->count == 1);
element->stride = 0;
}
if (dst_stride == element->glarray->StrideB) {
- if (intel->intelScreen->kernel_exec_fencing) {
- drm_intel_gem_bo_map_gtt(element->bo);
- memcpy((char *)element->bo->virtual + element->offset,
- element->glarray->Ptr, size);
- drm_intel_gem_bo_unmap_gtt(element->bo);
- } else {
- dri_bo_subdata(element->bo,
- element->offset,
- size,
- element->glarray->Ptr);
- }
+ intel_upload_data(&brw->intel, element->glarray->Ptr, size,
+ &element->bo, &element->offset);
} else {
- char *dest;
const unsigned char *src = element->glarray->Ptr;
+ char *dst = intel_upload_map(&brw->intel, size,
+ &element->bo, &element->offset);
int i;
- if (intel->intelScreen->kernel_exec_fencing) {
- drm_intel_gem_bo_map_gtt(element->bo);
- dest = element->bo->virtual;
- dest += element->offset;
-
- for (i = 0; i < element->count; i++) {
- memcpy(dest, src, dst_stride);
- src += element->glarray->StrideB;
- dest += dst_stride;
- }
-
- drm_intel_gem_bo_unmap_gtt(element->bo);
- } else {
- void *data;
-
- data = _mesa_malloc(dst_stride * element->count);
- dest = data;
- for (i = 0; i < element->count; i++) {
- memcpy(dest, src, dst_stride);
- src += element->glarray->StrideB;
- dest += dst_stride;
- }
-
- dri_bo_subdata(element->bo,
- element->offset,
- size,
- data);
-
- _mesa_free(data);
+ for (i = 0; i < element->count; i++) {
+ memcpy(dst, src, dst_stride);
+ src += element->glarray->StrideB;
+ dst += dst_stride;
}
}
}
static void brw_prepare_vertices(struct brw_context *brw)
{
- GLcontext *ctx = &brw->intel.ctx;
+ struct gl_context *ctx = &brw->intel.ctx;
struct intel_context *intel = intel_context(ctx);
GLbitfield vs_inputs = brw->vs.prog_data->inputs_read;
GLuint i;
/* First build an array of pointers to ve's in vb.inputs_read
*/
if (0)
- _mesa_printf("%s %d..%d\n", __FUNCTION__, min_index, max_index);
+ printf("%s %d..%d\n", __FUNCTION__, min_index, max_index);
/* Accumulate the list of enabled arrays. */
brw->vb.nr_enabled = 0;
brw->vb.enabled[brw->vb.nr_enabled++] = input;
}
+ if (brw->vb.nr_enabled == 0)
+ return;
+
/* XXX: In the rare cases where this happens we fallback all
* the way to software rasterization, although a tnl fallback
* would be sufficient. I don't know of *any* real world
if (_mesa_is_bufferobj(input->glarray->BufferObj)) {
struct intel_buffer_object *intel_buffer =
intel_buffer_object(input->glarray->BufferObj);
+ GLuint offset;
/* Named buffer object: Just reference its contents directly. */
- dri_bo_unreference(input->bo);
- input->bo = intel_bufferobj_buffer(intel, intel_buffer,
- INTEL_READ);
- dri_bo_reference(input->bo);
- input->offset = (unsigned long)input->glarray->Ptr;
+ drm_intel_bo_unreference(input->bo);
+ input->bo = intel_bufferobj_source(intel, intel_buffer, &offset);
+ drm_intel_bo_reference(input->bo);
+ input->offset = offset + (unsigned long)input->glarray->Ptr;
input->stride = input->glarray->StrideB;
input->count = input->glarray->_MaxElement;
*/
assert(input->offset < input->bo->size);
} else {
- input->count = input->glarray->StrideB ? max_index + 1 - min_index : 1;
+ input->count = input->glarray->StrideB ? max_index + 1 : 1;
if (input->bo != NULL) {
/* Already-uploaded vertex data is present from a previous
* prepare_vertices, but we had to re-validate state due to
}
upload[nr_uploads++] = input;
-
- /* We rebase drawing to start at element zero only when
- * varyings are not in vbos, which means we can end up
- * uploading non-varying arrays (stride != 0) when min_index
- * is zero. This doesn't matter as the amount to upload is
- * the same for these arrays whether the draw call is rebased
- * or not - we just have to upload the one element.
- */
- assert(min_index == 0 || input->glarray->StrideB == 0);
}
}
upload[i]->offset = upload[0]->offset +
((const unsigned char *)upload[i]->glarray->Ptr - ptr);
upload[i]->bo = upload[0]->bo;
- dri_bo_reference(upload[i]->bo);
+ drm_intel_bo_reference(upload[i]->bo);
}
}
else {
static void brw_emit_vertices(struct brw_context *brw)
{
- GLcontext *ctx = &brw->intel.ctx;
+ struct gl_context *ctx = &brw->intel.ctx;
struct intel_context *intel = intel_context(ctx);
GLuint i;
* a VE loads from them.
*/
if (brw->vb.nr_enabled == 0) {
- BEGIN_BATCH(3, IGNORE_CLIPRECTS);
+ BEGIN_BATCH(3);
OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | 1);
- OUT_BATCH((0 << BRW_VE0_INDEX_SHIFT) |
- BRW_VE0_VALID |
- (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) |
- (0 << BRW_VE0_SRC_OFFSET_SHIFT));
+ if (intel->gen >= 6) {
+ OUT_BATCH((0 << GEN6_VE0_INDEX_SHIFT) |
+ GEN6_VE0_VALID |
+ (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) |
+ (0 << BRW_VE0_SRC_OFFSET_SHIFT));
+ } else {
+ OUT_BATCH((0 << BRW_VE0_INDEX_SHIFT) |
+ BRW_VE0_VALID |
+ (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) |
+ (0 << BRW_VE0_SRC_OFFSET_SHIFT));
+ }
OUT_BATCH((BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_0_SHIFT) |
(BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_1_SHIFT) |
(BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_2_SHIFT) |
(BRW_VE1_COMPONENT_STORE_1_FLT << BRW_VE1_COMPONENT_3_SHIFT));
- ADVANCE_BATCH();
+ CACHED_BATCH();
return;
}
* are interleaved or from the same VBO. TBD if this makes a
* performance difference.
*/
- BEGIN_BATCH(1 + brw->vb.nr_enabled * 4, IGNORE_CLIPRECTS);
+ BEGIN_BATCH(1 + brw->vb.nr_enabled * 4);
OUT_BATCH((CMD_VERTEX_BUFFER << 16) |
((1 + brw->vb.nr_enabled * 4) - 2));
for (i = 0; i < brw->vb.nr_enabled; i++) {
struct brw_vertex_element *input = brw->vb.enabled[i];
+ uint32_t dw0;
+
+ if (intel->gen >= 6) {
+ dw0 = GEN6_VB0_ACCESS_VERTEXDATA |
+ (i << GEN6_VB0_INDEX_SHIFT);
+ } else {
+ dw0 = BRW_VB0_ACCESS_VERTEXDATA |
+ (i << BRW_VB0_INDEX_SHIFT);
+ }
- OUT_BATCH((i << BRW_VB0_INDEX_SHIFT) |
- BRW_VB0_ACCESS_VERTEXDATA |
+ OUT_BATCH(dw0 |
(input->stride << BRW_VB0_PITCH_SHIFT));
OUT_RELOC(input->bo,
I915_GEM_DOMAIN_VERTEX, 0,
input->offset);
- if (intel->is_ironlake) {
+ if (intel->gen >= 5) {
OUT_RELOC(input->bo,
I915_GEM_DOMAIN_VERTEX, 0,
input->bo->size - 1);
}
ADVANCE_BATCH();
- BEGIN_BATCH(1 + brw->vb.nr_enabled * 2, IGNORE_CLIPRECTS);
+ BEGIN_BATCH(1 + brw->vb.nr_enabled * 2);
OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | ((1 + brw->vb.nr_enabled * 2) - 2));
for (i = 0; i < brw->vb.nr_enabled; i++) {
struct brw_vertex_element *input = brw->vb.enabled[i];
break;
}
- OUT_BATCH((i << BRW_VE0_INDEX_SHIFT) |
- BRW_VE0_VALID |
- (format << BRW_VE0_FORMAT_SHIFT) |
- (0 << BRW_VE0_SRC_OFFSET_SHIFT));
+ if (intel->gen >= 6) {
+ OUT_BATCH((i << GEN6_VE0_INDEX_SHIFT) |
+ GEN6_VE0_VALID |
+ (format << BRW_VE0_FORMAT_SHIFT) |
+ (0 << BRW_VE0_SRC_OFFSET_SHIFT));
+ } else {
+ OUT_BATCH((i << BRW_VE0_INDEX_SHIFT) |
+ BRW_VE0_VALID |
+ (format << BRW_VE0_FORMAT_SHIFT) |
+ (0 << BRW_VE0_SRC_OFFSET_SHIFT));
+ }
- if (intel->is_ironlake)
+ if (intel->gen >= 5)
OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
(comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
(comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
(comp3 << BRW_VE1_COMPONENT_3_SHIFT) |
((i * 4) << BRW_VE1_DST_OFFSET_SHIFT));
}
- ADVANCE_BATCH();
+ CACHED_BATCH();
}
const struct brw_tracked_state brw_vertices = {
static void brw_prepare_indices(struct brw_context *brw)
{
- GLcontext *ctx = &brw->intel.ctx;
+ struct gl_context *ctx = &brw->intel.ctx;
struct intel_context *intel = &brw->intel;
const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
GLuint ib_size;
- dri_bo *bo = NULL;
+ drm_intel_bo *bo = NULL;
struct gl_buffer_object *bufferobj;
GLuint offset;
GLuint ib_type_size;
ib_type_size = get_size(index_buffer->type);
ib_size = ib_type_size * index_buffer->count;
- bufferobj = index_buffer->obj;;
+ bufferobj = index_buffer->obj;
/* Turn into a proper VBO:
*/
if (!_mesa_is_bufferobj(bufferobj)) {
- brw->ib.start_vertex_offset = 0;
/* Get new bufferobj, offset:
*/
- get_space(brw, ib_size, &bo, &offset);
-
- /* Straight upload
- */
- if (intel->intelScreen->kernel_exec_fencing) {
- drm_intel_gem_bo_map_gtt(bo);
- memcpy((char *)bo->virtual + offset, index_buffer->ptr, ib_size);
- drm_intel_gem_bo_unmap_gtt(bo);
- } else {
- dri_bo_subdata(bo, offset, ib_size, index_buffer->ptr);
- }
+ intel_upload_data(&brw->intel, index_buffer->ptr, ib_size, &bo, &offset);
+ brw->ib.start_vertex_offset = offset / ib_type_size;
+ offset = 0;
} else {
offset = (GLuint) (unsigned long) index_buffer->ptr;
- brw->ib.start_vertex_offset = 0;
/* If the index buffer isn't aligned to its element size, we have to
* rebase it into a temporary.
bufferobj);
map += offset;
- get_space(brw, ib_size, &bo, &offset);
-
- dri_bo_subdata(bo, offset, ib_size, map);
+ intel_upload_data(&brw->intel, map, ib_size, &bo, &offset);
+ brw->ib.start_vertex_offset = offset / ib_type_size;
+ offset = 0;
ctx->Driver.UnmapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER_ARB, bufferobj);
} else {
- bo = intel_bufferobj_buffer(intel, intel_buffer_object(bufferobj),
- INTEL_READ);
- dri_bo_reference(bo);
-
/* Use CMD_3D_PRIM's start_vertex_offset to avoid re-uploading
* the index buffer state when we're just moving the start index
* of our drawing.
*/
brw->ib.start_vertex_offset = offset / ib_type_size;
- offset = 0;
- ib_size = bo->size;
+
+ bo = intel_bufferobj_source(intel, intel_buffer_object(bufferobj),
+ &offset);
+ drm_intel_bo_reference(bo);
}
}
- if (brw->ib.bo != bo ||
- brw->ib.offset != offset ||
- brw->ib.size != ib_size)
- {
+ if (brw->ib.bo != bo || brw->ib.offset != offset) {
drm_intel_bo_unreference(brw->ib.bo);
brw->ib.bo = bo;
brw->ib.offset = offset;
- brw->ib.size = ib_size;
+ brw_add_validated_bo(brw, brw->ib.bo);
brw->state.dirty.brw |= BRW_NEW_INDEX_BUFFER;
} else {
drm_intel_bo_unreference(bo);
}
-
- brw_add_validated_bo(brw, brw->ib.bo);
}
const struct brw_tracked_state brw_indices = {
if (index_buffer == NULL)
return;
- /* Emit the indexbuffer packet:
- */
- {
- struct brw_indexbuffer ib;
-
- memset(&ib, 0, sizeof(ib));
-
- ib.header.bits.opcode = CMD_INDEX_BUFFER;
- ib.header.bits.length = sizeof(ib)/4 - 2;
- ib.header.bits.index_format = get_index_type(index_buffer->type);
- ib.header.bits.cut_index_enable = 0;
-
- BEGIN_BATCH(4, IGNORE_CLIPRECTS);
- OUT_BATCH( ib.header.dword );
- OUT_RELOC(brw->ib.bo,
- I915_GEM_DOMAIN_VERTEX, 0,
- brw->ib.offset);
- OUT_RELOC(brw->ib.bo,
- I915_GEM_DOMAIN_VERTEX, 0,
- brw->ib.offset + brw->ib.size - 1);
- OUT_BATCH( 0 );
- ADVANCE_BATCH();
- }
+ BEGIN_BATCH(3);
+ OUT_BATCH(CMD_INDEX_BUFFER << 16 |
+ /* cut index enable << 10 */
+ get_index_type(index_buffer->type) << 8 |
+ 1);
+ OUT_RELOC(brw->ib.bo,
+ I915_GEM_DOMAIN_VERTEX, 0,
+ brw->ib.offset);
+ OUT_RELOC(brw->ib.bo,
+ I915_GEM_DOMAIN_VERTEX, 0,
+ brw->ib.bo->size - 1);
+ ADVANCE_BATCH();
}
const struct brw_tracked_state brw_index_buffer = {
projects / mesa.git / blobdiff