*
**************************************************************************/
-#undef NDEBUG
-
#include "main/glheader.h"
#include "main/bufferobj.h"
#include "main/context.h"
#include "main/enums.h"
+#include "main/macros.h"
+#include "main/glformats.h"
#include "brw_draw.h"
#include "brw_defines.h"
BRW_SURFACEFORMAT_R16G16B16A16_FLOAT
};
+static GLuint fixed_point_types[5] = {
+ 0,
+ BRW_SURFACEFORMAT_R32_SFIXED,
+ BRW_SURFACEFORMAT_R32G32_SFIXED,
+ BRW_SURFACEFORMAT_R32G32B32_SFIXED,
+ BRW_SURFACEFORMAT_R32G32B32A32_SFIXED,
+};
+
+static GLuint uint_types_direct[5] = {
+ 0,
+ BRW_SURFACEFORMAT_R32_UINT,
+ BRW_SURFACEFORMAT_R32G32_UINT,
+ BRW_SURFACEFORMAT_R32G32B32_UINT,
+ BRW_SURFACEFORMAT_R32G32B32A32_UINT
+};
+
static GLuint uint_types_norm[5] = {
0,
BRW_SURFACEFORMAT_R32_UNORM,
BRW_SURFACEFORMAT_R32G32B32A32_USCALED
};
+static GLuint int_types_direct[5] = {
+ 0,
+ BRW_SURFACEFORMAT_R32_SINT,
+ BRW_SURFACEFORMAT_R32G32_SINT,
+ BRW_SURFACEFORMAT_R32G32B32_SINT,
+ BRW_SURFACEFORMAT_R32G32B32A32_SINT
+};
+
static GLuint int_types_norm[5] = {
0,
BRW_SURFACEFORMAT_R32_SNORM,
BRW_SURFACEFORMAT_R32G32B32A32_SSCALED
};
+static GLuint ushort_types_direct[5] = {
+ 0,
+ BRW_SURFACEFORMAT_R16_UINT,
+ BRW_SURFACEFORMAT_R16G16_UINT,
+ BRW_SURFACEFORMAT_R16G16B16A16_UINT,
+ BRW_SURFACEFORMAT_R16G16B16A16_UINT
+};
+
static GLuint ushort_types_norm[5] = {
0,
BRW_SURFACEFORMAT_R16_UNORM,
BRW_SURFACEFORMAT_R16G16B16A16_USCALED
};
+static GLuint short_types_direct[5] = {
+ 0,
+ BRW_SURFACEFORMAT_R16_SINT,
+ BRW_SURFACEFORMAT_R16G16_SINT,
+ BRW_SURFACEFORMAT_R16G16B16A16_SINT,
+ BRW_SURFACEFORMAT_R16G16B16A16_SINT
+};
+
static GLuint short_types_norm[5] = {
0,
BRW_SURFACEFORMAT_R16_SNORM,
BRW_SURFACEFORMAT_R16G16B16A16_SSCALED
};
+static GLuint ubyte_types_direct[5] = {
+ 0,
+ BRW_SURFACEFORMAT_R8_UINT,
+ BRW_SURFACEFORMAT_R8G8_UINT,
+ BRW_SURFACEFORMAT_R8G8B8A8_UINT,
+ BRW_SURFACEFORMAT_R8G8B8A8_UINT
+};
+
static GLuint ubyte_types_norm[5] = {
0,
BRW_SURFACEFORMAT_R8_UNORM,
BRW_SURFACEFORMAT_R8G8B8A8_USCALED
};
+static GLuint byte_types_direct[5] = {
+ 0,
+ BRW_SURFACEFORMAT_R8_SINT,
+ BRW_SURFACEFORMAT_R8G8_SINT,
+ BRW_SURFACEFORMAT_R8G8B8A8_SINT,
+ BRW_SURFACEFORMAT_R8G8B8A8_SINT
+};
+
static GLuint byte_types_norm[5] = {
0,
BRW_SURFACEFORMAT_R8_SNORM,
* the appopriate hardware surface type.
* Format will be GL_RGBA or possibly GL_BGRA for GLubyte[4] color arrays.
*/
-static GLuint get_surface_type( GLenum type, GLuint size,
- GLenum format, GLboolean normalized )
+unsigned
+brw_get_vertex_surface_type(struct brw_context *brw,
+ const struct gl_client_array *glarray)
{
+ int size = glarray->Size;
+
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) {
+ _mesa_lookup_enum_by_nr(glarray->Type),
+ glarray->Size, glarray->Normalized);
+
+ if (glarray->Integer) {
+ assert(glarray->Format == GL_RGBA); /* sanity check */
+ switch (glarray->Type) {
+ case GL_INT: return int_types_direct[size];
+ case GL_SHORT: return short_types_direct[size];
+ case GL_BYTE: return byte_types_direct[size];
+ case GL_UNSIGNED_INT: return uint_types_direct[size];
+ case GL_UNSIGNED_SHORT: return ushort_types_direct[size];
+ case GL_UNSIGNED_BYTE: return ubyte_types_direct[size];
+ default: assert(0); return 0;
+ }
+ } else if (glarray->Normalized) {
+ switch (glarray->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_UNSIGNED_INT: return uint_types_norm[size];
case GL_UNSIGNED_SHORT: return ushort_types_norm[size];
case GL_UNSIGNED_BYTE:
- if (format == GL_BGRA) {
+ if (glarray->Format == GL_BGRA) {
/* See GL_EXT_vertex_array_bgra */
assert(size == 4);
return BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
else {
return ubyte_types_norm[size];
}
+ case GL_FIXED:
+ if (brw->gen >= 8 || brw->is_haswell)
+ return fixed_point_types[size];
+
+ /* This produces GL_FIXED inputs as values between INT32_MIN and
+ * INT32_MAX, which will be scaled down by 1/65536 by the VS.
+ */
+ return int_types_scale[size];
+ /* See GL_ARB_vertex_type_2_10_10_10_rev.
+ * W/A: Pre-Haswell, the hardware doesn't really support the formats we'd
+ * like to use here, so upload everything as UINT and fix
+ * it in the shader
+ */
+ case GL_INT_2_10_10_10_REV:
+ assert(size == 4);
+ if (brw->gen >= 8 || brw->is_haswell) {
+ return glarray->Format == GL_BGRA
+ ? BRW_SURFACEFORMAT_B10G10R10A2_SNORM
+ : BRW_SURFACEFORMAT_R10G10B10A2_SNORM;
+ }
+ return BRW_SURFACEFORMAT_R10G10B10A2_UINT;
+ case GL_UNSIGNED_INT_2_10_10_10_REV:
+ assert(size == 4);
+ if (brw->gen >= 8 || brw->is_haswell) {
+ return glarray->Format == GL_BGRA
+ ? BRW_SURFACEFORMAT_B10G10R10A2_UNORM
+ : BRW_SURFACEFORMAT_R10G10B10A2_UNORM;
+ }
+ return BRW_SURFACEFORMAT_R10G10B10A2_UINT;
default: assert(0); return 0;
- }
+ }
}
else {
- assert(format == GL_RGBA); /* sanity check */
- switch (type) {
+ /* See GL_ARB_vertex_type_2_10_10_10_rev.
+ * W/A: the hardware doesn't really support the formats we'd
+ * like to use here, so upload everything as UINT and fix
+ * it in the shader
+ */
+ if (glarray->Type == GL_INT_2_10_10_10_REV) {
+ assert(size == 4);
+ if (brw->gen >= 8 || brw->is_haswell) {
+ return glarray->Format == GL_BGRA
+ ? BRW_SURFACEFORMAT_B10G10R10A2_SSCALED
+ : BRW_SURFACEFORMAT_R10G10B10A2_SSCALED;
+ }
+ return BRW_SURFACEFORMAT_R10G10B10A2_UINT;
+ } else if (glarray->Type == GL_UNSIGNED_INT_2_10_10_10_REV) {
+ assert(size == 4);
+ if (brw->gen >= 8 || brw->is_haswell) {
+ return glarray->Format == GL_BGRA
+ ? BRW_SURFACEFORMAT_B10G10R10A2_USCALED
+ : BRW_SURFACEFORMAT_R10G10B10A2_USCALED;
+ }
+ return BRW_SURFACEFORMAT_R10G10B10A2_UINT;
+ }
+ assert(glarray->Format == GL_RGBA); /* sanity check */
+ switch (glarray->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_UNSIGNED_INT: return uint_types_scale[size];
case GL_UNSIGNED_SHORT: return ushort_types_scale[size];
case GL_UNSIGNED_BYTE: return ubyte_types_scale[size];
- /* This produces GL_FIXED inputs as values between INT32_MIN and
- * INT32_MAX, which will be scaled down by 1/65536 by the VS.
- */
- case GL_FIXED: return int_types_scale[size];
+ case GL_FIXED:
+ if (brw->gen >= 8 || brw->is_haswell)
+ return fixed_point_types[size];
+
+ /* This produces GL_FIXED inputs as values between INT32_MIN and
+ * INT32_MAX, which will be scaled down by 1/65536 by the VS.
+ */
+ return int_types_scale[size];
default: assert(0); return 0;
}
}
}
-
-static GLuint get_size( GLenum type )
-{
- 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);
- case GL_UNSIGNED_INT: return sizeof(GLuint);
- case GL_UNSIGNED_SHORT: return sizeof(GLushort);
- case GL_UNSIGNED_BYTE: return sizeof(GLubyte);
- case GL_FIXED: return sizeof(GLuint);
- default: assert(0); return 0;
- }
-}
-
-static GLuint get_index_type(GLenum type)
+unsigned
+brw_get_index_type(GLenum type)
{
switch (type) {
case GL_UNSIGNED_BYTE: return BRW_INDEX_BYTE;
struct brw_vertex_buffer *buffer,
GLuint dst_stride)
{
- int src_stride = element->glarray->StrideB;
+ const int src_stride = element->glarray->StrideB;
+
+ /* If the source stride is zero, we just want to upload the current
+ * attribute once and set the buffer's stride to 0. There's no need
+ * to replicate it out.
+ */
+ if (src_stride == 0) {
+ intel_upload_data(brw, element->glarray->Ptr,
+ element->glarray->_ElementSize,
+ element->glarray->_ElementSize,
+ &buffer->bo, &buffer->offset);
+
+ buffer->stride = 0;
+ return;
+ }
+
const unsigned char *src = element->glarray->Ptr + min * src_stride;
int count = max - min + 1;
GLuint size = count * dst_stride;
if (dst_stride == src_stride) {
- intel_upload_data(&brw->intel, src, size, dst_stride,
+ intel_upload_data(brw, src, size, dst_stride,
&buffer->bo, &buffer->offset);
} else {
- char * const map = intel_upload_map(&brw->intel, size, dst_stride);
+ char * const map = intel_upload_map(brw, size, dst_stride);
char *dst = map;
while (count--) {
src += src_stride;
dst += dst_stride;
}
- intel_upload_unmap(&brw->intel, map, size, dst_stride,
+ intel_upload_unmap(brw, map, size, dst_stride,
&buffer->bo, &buffer->offset);
}
buffer->stride = dst_stride;
static void brw_prepare_vertices(struct brw_context *brw)
{
- struct gl_context *ctx = &brw->intel.ctx;
- struct intel_context *intel = intel_context(ctx);
+ struct gl_context *ctx = &brw->ctx;
/* CACHE_NEW_VS_PROG */
- GLbitfield vs_inputs = brw->vs.prog_data->inputs_read;
+ GLbitfield64 vs_inputs = brw->vs.prog_data->inputs_read;
const unsigned char *ptr = NULL;
- GLuint interleaved = 0, total_size = 0;
- unsigned int min_index = brw->vb.min_index;
- unsigned int max_index = brw->vb.max_index;
+ GLuint interleaved = 0;
+ unsigned int min_index = brw->vb.min_index + brw->basevertex;
+ unsigned int max_index = brw->vb.max_index + brw->basevertex;
int delta, i, j;
struct brw_vertex_element *upload[VERT_ATTRIB_MAX];
GLuint nr_uploads = 0;
- /* First build an array of pointers to ve's in vb.inputs_read
+ /* _NEW_POLYGON
+ *
+ * On gen6+, edge flags don't end up in the VUE (either in or out of the
+ * VS). Instead, they're uploaded as the last vertex element, and the data
+ * is passed sideband through the fixed function units. So, we need to
+ * prepare the vertex buffer for it, but it's not present in inputs_read.
*/
+ if (brw->gen >= 6 && (ctx->Polygon.FrontMode != GL_FILL ||
+ ctx->Polygon.BackMode != GL_FILL)) {
+ vs_inputs |= VERT_BIT_EDGEFLAG;
+ }
+
if (0)
printf("%s %d..%d\n", __FUNCTION__, min_index, max_index);
/* Accumulate the list of enabled arrays. */
brw->vb.nr_enabled = 0;
while (vs_inputs) {
- GLuint i = ffs(vs_inputs) - 1;
+ GLuint i = ffsll(vs_inputs) - 1;
struct brw_vertex_element *input = &brw->vb.inputs[i];
- vs_inputs &= ~(1 << i);
- if (input->glarray->Size && get_size(input->glarray->Type))
- brw->vb.enabled[brw->vb.nr_enabled++] = input;
+ vs_inputs &= ~BITFIELD64_BIT(i);
+ brw->vb.enabled[brw->vb.nr_enabled++] = input;
}
if (brw->vb.nr_enabled == 0)
return;
if (brw->vb.nr_buffers)
- goto validate;
-
- /* 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
- * cases with > 17 vertex attributes enabled, so it probably
- * isn't an issue at this point.
- */
- if (brw->vb.nr_enabled >= BRW_VEP_MAX) {
- intel->Fallback = GL_TRUE; /* boolean, not bitfield */
return;
- }
for (i = j = 0; i < brw->vb.nr_enabled; i++) {
struct brw_vertex_element *input = brw->vb.enabled[i];
const struct gl_client_array *glarray = input->glarray;
- int type_size = get_size(glarray->Type);
-
- input->element_size = type_size * glarray->Size;
if (_mesa_is_bufferobj(glarray->BufferObj)) {
struct intel_buffer_object *intel_buffer =
intel_buffer_object(glarray->BufferObj);
int k;
+ /* If we have a VB set to be uploaded for this buffer object
+ * already, reuse that VB state so that we emit fewer
+ * relocations.
+ */
for (k = 0; k < i; k++) {
const struct gl_client_array *other = brw->vb.enabled[k]->glarray;
if (glarray->BufferObj == other->BufferObj &&
glarray->StrideB == other->StrideB &&
+ glarray->InstanceDivisor == other->InstanceDivisor &&
(uintptr_t)(glarray->Ptr - other->Ptr) < glarray->StrideB)
{
input->buffer = brw->vb.enabled[k]->buffer;
struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
/* Named buffer object: Just reference its contents directly. */
- buffer->bo = intel_bufferobj_source(intel,
- intel_buffer, type_size,
+ buffer->bo = intel_bufferobj_source(brw,
+ intel_buffer, 1,
&buffer->offset);
drm_intel_bo_reference(buffer->bo);
buffer->offset += (uintptr_t)glarray->Ptr;
buffer->stride = glarray->StrideB;
+ buffer->step_rate = glarray->InstanceDivisor;
input->buffer = j++;
input->offset = 0;
* when we've decided if we're doing interleaved or not.
*/
if (nr_uploads == 0) {
- /* Position array not properly enabled:
- */
- if (input->attrib == VERT_ATTRIB_POS && glarray->StrideB == 0) {
- intel->Fallback = GL_TRUE; /* boolean, not bitfield */
- return;
- }
-
interleaved = glarray->StrideB;
ptr = glarray->Ptr;
}
else if (interleaved != glarray->StrideB ||
- (uintptr_t)(glarray->Ptr - ptr) > interleaved)
- {
- interleaved = 0;
- }
- else if ((uintptr_t)(glarray->Ptr - ptr) & (type_size -1))
+ glarray->Ptr < ptr ||
+ (uintptr_t)(glarray->Ptr - ptr) + glarray->_ElementSize > interleaved)
{
- /* enforce natural alignment (for doubles) */
+ /* If our stride is different from the first attribute's stride,
+ * or if the first attribute's stride didn't cover our element,
+ * disable the interleaved upload optimization. The second case
+ * can most commonly occur in cases where there is a single vertex
+ * and, for example, the data is stored on the application's
+ * stack.
+ *
+ * NOTE: This will also disable the optimization in cases where
+ * the data is in a different order than the array indices.
+ * Something like:
+ *
+ * float data[...];
+ * glVertexAttribPointer(0, 4, GL_FLOAT, 32, &data[4]);
+ * glVertexAttribPointer(1, 4, GL_FLOAT, 32, &data[0]);
+ */
interleaved = 0;
}
upload[nr_uploads++] = input;
- total_size = ALIGN(total_size, type_size);
- total_size += input->element_size;
}
}
brw->vb.start_vertex_bias = -delta;
delta = 0;
}
- if (delta && !brw->intel.intelScreen->relaxed_relocations)
- min_index = delta = 0;
/* Handle any arrays to be uploaded. */
if (nr_uploads > 1) {
- if (interleaved && interleaved <= 2*total_size) {
+ if (interleaved) {
struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
/* All uploads are interleaved, so upload the arrays together as
* interleaved. First, upload the contents and set up upload[0].
}
j++;
- nr_uploads = 0;
- }
- else if (total_size < 2048) {
- /* Upload non-interleaved arrays into a single interleaved array */
- struct brw_vertex_buffer *buffer;
- int count = max_index - min_index + 1;
- int offset;
- char *map;
-
- map = intel_upload_map(&brw->intel, total_size * count, total_size);
- for (i = offset = 0; i < nr_uploads; i++) {
- const unsigned char *src = upload[i]->glarray->Ptr;
- int size = upload[i]->element_size;
- int stride = upload[i]->glarray->StrideB;
- char *dst;
- int n;
-
- offset = ALIGN(offset, get_size(upload[i]->glarray->Type));
- dst = map + offset;
- src += min_index * stride;
-
- for (n = 0; n < count; n++) {
- memcpy(dst, src, size);
- src += stride;
- dst += total_size;
- }
-
- upload[i]->offset = offset;
- upload[i]->buffer = j;
-
- offset += size;
- }
- assert(offset == total_size);
- buffer = &brw->vb.buffers[j++];
- intel_upload_unmap(&brw->intel, map, offset * count, offset,
- &buffer->bo, &buffer->offset);
- buffer->stride = offset;
- buffer->offset -= delta * offset;
-
nr_uploads = 0;
}
}
/* Upload non-interleaved arrays */
for (i = 0; i < nr_uploads; i++) {
struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
- copy_array_to_vbo_array(brw, upload[i], min_index, max_index,
- buffer, upload[i]->element_size);
+ if (upload[i]->glarray->InstanceDivisor == 0) {
+ copy_array_to_vbo_array(brw, upload[i], min_index, max_index,
+ buffer, upload[i]->glarray->_ElementSize);
+ } else {
+ /* This is an instanced attribute, since its InstanceDivisor
+ * is not zero. Therefore, its data will be stepped after the
+ * instanced draw has been run InstanceDivisor times.
+ */
+ uint32_t instanced_attr_max_index =
+ (brw->num_instances - 1) / upload[i]->glarray->InstanceDivisor;
+ copy_array_to_vbo_array(brw, upload[i], 0, instanced_attr_max_index,
+ buffer, upload[i]->glarray->_ElementSize);
+ }
buffer->offset -= delta * buffer->stride;
+ buffer->step_rate = upload[i]->glarray->InstanceDivisor;
upload[i]->buffer = j++;
upload[i]->offset = 0;
}
- /* can we simply extend the current vb? */
- if (j == brw->vb.nr_current_buffers) {
- int delta = 0;
- for (i = 0; i < j; i++) {
- int d;
-
- if (brw->vb.current_buffers[i].handle != brw->vb.buffers[i].bo->handle ||
- brw->vb.current_buffers[i].stride != brw->vb.buffers[i].stride)
- break;
-
- d = brw->vb.buffers[i].offset - brw->vb.current_buffers[i].offset;
- if (d < 0)
- break;
- if (i == 0)
- delta = d / brw->vb.current_buffers[i].stride;
- if (delta * brw->vb.current_buffers[i].stride != d)
- break;
- }
-
- if (i == j) {
- brw->vb.start_vertex_bias += delta;
- while (--j >= 0)
- drm_intel_bo_unreference(brw->vb.buffers[j].bo);
- j = 0;
- }
- }
-
brw->vb.nr_buffers = j;
-
-validate:
- brw_prepare_query_begin(brw);
- for (i = 0; i < brw->vb.nr_buffers; i++) {
- brw_add_validated_bo(brw, brw->vb.buffers[i].bo);
- }
}
static void brw_emit_vertices(struct brw_context *brw)
{
- struct gl_context *ctx = &brw->intel.ctx;
- struct intel_context *intel = intel_context(ctx);
- GLuint i;
+ GLuint i, nr_elements;
+
+ brw_prepare_vertices(brw);
brw_emit_query_begin(brw);
+ nr_elements = brw->vb.nr_enabled + brw->vs.prog_data->uses_vertexid;
+
/* If the VS doesn't read any inputs (calculating vertex position from
* a state variable for some reason, for example), emit a single pad
* VERTEX_ELEMENT struct and bail.
* The stale VB state stays in place, but they don't do anything unless
* a VE loads from them.
*/
- if (brw->vb.nr_enabled == 0) {
+ if (nr_elements == 0) {
BEGIN_BATCH(3);
OUT_BATCH((_3DSTATE_VERTEX_ELEMENTS << 16) | 1);
- if (intel->gen >= 6) {
+ if (brw->gen >= 6) {
OUT_BATCH((0 << GEN6_VE0_INDEX_SHIFT) |
GEN6_VE0_VALID |
(BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) |
*/
if (brw->vb.nr_buffers) {
+ if (brw->gen >= 6) {
+ assert(brw->vb.nr_buffers <= 33);
+ } else {
+ assert(brw->vb.nr_buffers <= 17);
+ }
+
BEGIN_BATCH(1 + 4*brw->vb.nr_buffers);
OUT_BATCH((_3DSTATE_VERTEX_BUFFERS << 16) | (4*brw->vb.nr_buffers - 1));
for (i = 0; i < brw->vb.nr_buffers; i++) {
struct brw_vertex_buffer *buffer = &brw->vb.buffers[i];
uint32_t dw0;
- if (intel->gen >= 6) {
- dw0 = GEN6_VB0_ACCESS_VERTEXDATA | (i << GEN6_VB0_INDEX_SHIFT);
+ if (brw->gen >= 6) {
+ dw0 = buffer->step_rate
+ ? GEN6_VB0_ACCESS_INSTANCEDATA
+ : GEN6_VB0_ACCESS_VERTEXDATA;
+ dw0 |= i << GEN6_VB0_INDEX_SHIFT;
} else {
- dw0 = BRW_VB0_ACCESS_VERTEXDATA | (i << BRW_VB0_INDEX_SHIFT);
+ dw0 = buffer->step_rate
+ ? BRW_VB0_ACCESS_INSTANCEDATA
+ : BRW_VB0_ACCESS_VERTEXDATA;
+ dw0 |= i << BRW_VB0_INDEX_SHIFT;
}
- if (intel->gen >= 7)
+ if (brw->gen >= 7)
dw0 |= GEN7_VB0_ADDRESS_MODIFYENABLE;
+ if (brw->gen == 7)
+ dw0 |= GEN7_MOCS_L3 << 16;
+
OUT_BATCH(dw0 | (buffer->stride << BRW_VB0_PITCH_SHIFT));
OUT_RELOC(buffer->bo, I915_GEM_DOMAIN_VERTEX, 0, buffer->offset);
- if (intel->gen >= 5) {
+ if (brw->gen >= 5) {
OUT_RELOC(buffer->bo, I915_GEM_DOMAIN_VERTEX, 0, buffer->bo->size - 1);
} else
OUT_BATCH(0);
- OUT_BATCH(0); /* Instance data step rate */
-
- brw->vb.current_buffers[i].handle = buffer->bo->handle;
- brw->vb.current_buffers[i].offset = buffer->offset;
- brw->vb.current_buffers[i].stride = buffer->stride;
+ OUT_BATCH(buffer->step_rate);
}
- brw->vb.nr_current_buffers = i;
ADVANCE_BATCH();
}
- BEGIN_BATCH(1 + brw->vb.nr_enabled * 2);
- OUT_BATCH((_3DSTATE_VERTEX_ELEMENTS << 16) | (2*brw->vb.nr_enabled - 1));
+ /* The hardware allows one more VERTEX_ELEMENTS than VERTEX_BUFFERS, presumably
+ * for VertexID/InstanceID.
+ */
+ if (brw->gen >= 6) {
+ assert(nr_elements <= 34);
+ } else {
+ assert(nr_elements <= 18);
+ }
+
+ struct brw_vertex_element *gen6_edgeflag_input = NULL;
+
+ BEGIN_BATCH(1 + nr_elements * 2);
+ OUT_BATCH((_3DSTATE_VERTEX_ELEMENTS << 16) | (2 * nr_elements - 1));
for (i = 0; i < brw->vb.nr_enabled; i++) {
struct brw_vertex_element *input = brw->vb.enabled[i];
- uint32_t format = get_surface_type(input->glarray->Type,
- input->glarray->Size,
- input->glarray->Format,
- input->glarray->Normalized);
+ uint32_t format = brw_get_vertex_surface_type(brw, input->glarray);
uint32_t comp0 = BRW_VE1_COMPONENT_STORE_SRC;
uint32_t comp1 = BRW_VE1_COMPONENT_STORE_SRC;
uint32_t comp2 = BRW_VE1_COMPONENT_STORE_SRC;
uint32_t comp3 = BRW_VE1_COMPONENT_STORE_SRC;
+ /* The gen4 driver expects edgeflag to come in as a float, and passes
+ * that float on to the tests in the clipper. Mesa's current vertex
+ * attribute value for EdgeFlag is stored as a float, which works out.
+ * glEdgeFlagPointer, on the other hand, gives us an unnormalized
+ * integer ubyte. Just rewrite that to convert to a float.
+ */
+ if (input->attrib == VERT_ATTRIB_EDGEFLAG) {
+ /* Gen6+ passes edgeflag as sideband along with the vertex, instead
+ * of in the VUE. We have to upload it sideband as the last vertex
+ * element according to the B-Spec.
+ */
+ if (brw->gen >= 6) {
+ gen6_edgeflag_input = input;
+ continue;
+ }
+
+ if (format == BRW_SURFACEFORMAT_R8_UINT)
+ format = BRW_SURFACEFORMAT_R8_SSCALED;
+ }
+
switch (input->glarray->Size) {
case 0: comp0 = BRW_VE1_COMPONENT_STORE_0;
case 1: comp1 = BRW_VE1_COMPONENT_STORE_0;
case 2: comp2 = BRW_VE1_COMPONENT_STORE_0;
- case 3: comp3 = BRW_VE1_COMPONENT_STORE_1_FLT;
+ case 3: comp3 = input->glarray->Integer ? BRW_VE1_COMPONENT_STORE_1_INT
+ : BRW_VE1_COMPONENT_STORE_1_FLT;
break;
}
- if (intel->gen >= 6) {
+ if (brw->gen >= 6) {
OUT_BATCH((input->buffer << GEN6_VE0_INDEX_SHIFT) |
GEN6_VE0_VALID |
(format << BRW_VE0_FORMAT_SHIFT) |
(input->offset << BRW_VE0_SRC_OFFSET_SHIFT));
}
- if (intel->gen >= 5)
+ if (brw->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));
}
+
+ if (brw->gen >= 6 && gen6_edgeflag_input) {
+ uint32_t format =
+ brw_get_vertex_surface_type(brw, gen6_edgeflag_input->glarray);
+
+ OUT_BATCH((gen6_edgeflag_input->buffer << GEN6_VE0_INDEX_SHIFT) |
+ GEN6_VE0_VALID |
+ GEN6_VE0_EDGE_FLAG_ENABLE |
+ (format << BRW_VE0_FORMAT_SHIFT) |
+ (gen6_edgeflag_input->offset << BRW_VE0_SRC_OFFSET_SHIFT));
+ OUT_BATCH((BRW_VE1_COMPONENT_STORE_SRC << 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_0 << BRW_VE1_COMPONENT_3_SHIFT));
+ }
+
+ if (brw->vs.prog_data->uses_vertexid) {
+ uint32_t dw0 = 0, dw1 = 0;
+
+ dw1 = ((BRW_VE1_COMPONENT_STORE_VID << BRW_VE1_COMPONENT_0_SHIFT) |
+ (BRW_VE1_COMPONENT_STORE_IID << BRW_VE1_COMPONENT_1_SHIFT) |
+ (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_2_SHIFT) |
+ (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_3_SHIFT));
+
+ if (brw->gen >= 6) {
+ dw0 |= GEN6_VE0_VALID;
+ } else {
+ dw0 |= BRW_VE0_VALID;
+ dw1 |= (i * 4) << BRW_VE1_DST_OFFSET_SHIFT;
+ }
+
+ /* Note that for gl_VertexID, gl_InstanceID, and gl_PrimitiveID values,
+ * the format is ignored and the value is always int.
+ */
+
+ OUT_BATCH(dw0);
+ OUT_BATCH(dw1);
+ }
+
CACHED_BATCH();
}
const struct brw_tracked_state brw_vertices = {
.dirty = {
- .mesa = 0,
+ .mesa = _NEW_POLYGON,
.brw = BRW_NEW_BATCH | BRW_NEW_VERTICES,
.cache = CACHE_NEW_VS_PROG,
},
- .prepare = brw_prepare_vertices,
.emit = brw_emit_vertices,
};
-static void brw_prepare_indices(struct brw_context *brw)
+static void brw_upload_indices(struct brw_context *brw)
{
- struct gl_context *ctx = &brw->intel.ctx;
- struct intel_context *intel = &brw->intel;
+ struct gl_context *ctx = &brw->ctx;
const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
GLuint ib_size;
drm_intel_bo *bo = NULL;
if (index_buffer == NULL)
return;
- ib_type_size = get_size(index_buffer->type);
+ ib_type_size = _mesa_sizeof_type(index_buffer->type);
ib_size = ib_type_size * index_buffer->count;
bufferobj = index_buffer->obj;
/* Get new bufferobj, offset:
*/
- intel_upload_data(&brw->intel, index_buffer->ptr, ib_size, ib_type_size,
+ intel_upload_data(brw, index_buffer->ptr, ib_size, ib_type_size,
&bo, &offset);
brw->ib.start_vertex_offset = offset / ib_type_size;
} else {
/* If the index buffer isn't aligned to its element size, we have to
* rebase it into a temporary.
*/
- if ((get_size(index_buffer->type) - 1) & offset) {
- GLubyte *map = ctx->Driver.MapBufferRange(ctx,
- offset,
- ib_size,
- GL_MAP_WRITE_BIT,
- bufferobj);
-
- intel_upload_data(&brw->intel, map, ib_size, ib_type_size,
- &bo, &offset);
- brw->ib.start_vertex_offset = offset / ib_type_size;
-
- ctx->Driver.UnmapBuffer(ctx, bufferobj);
+ if ((ib_type_size - 1) & offset) {
+ perf_debug("copying index buffer to a temporary to work around "
+ "misaligned offset %d\n", offset);
+
+ GLubyte *map = ctx->Driver.MapBufferRange(ctx,
+ offset,
+ ib_size,
+ GL_MAP_READ_BIT,
+ bufferobj);
+
+ intel_upload_data(brw, map, ib_size, ib_type_size, &bo, &offset);
+ brw->ib.start_vertex_offset = offset / ib_type_size;
+
+ ctx->Driver.UnmapBuffer(ctx, bufferobj);
} else {
/* Use CMD_3D_PRIM's start_vertex_offset to avoid re-uploading
* the index buffer state when we're just moving the start index
*/
brw->ib.start_vertex_offset = offset / ib_type_size;
- bo = intel_bufferobj_source(intel,
+ bo = intel_bufferobj_source(brw,
intel_buffer_object(bufferobj),
ib_type_size,
&offset);
drm_intel_bo_unreference(brw->ib.bo);
brw->ib.bo = bo;
- brw_add_validated_bo(brw, brw->ib.bo);
brw->state.dirty.brw |= BRW_NEW_INDEX_BUFFER;
} else {
drm_intel_bo_unreference(bo);
.brw = BRW_NEW_INDICES,
.cache = 0,
},
- .prepare = brw_prepare_indices,
+ .emit = brw_upload_indices,
};
static void brw_emit_index_buffer(struct brw_context *brw)
{
- struct intel_context *intel = &brw->intel;
const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
+ GLuint cut_index_setting;
if (index_buffer == NULL)
return;
+ if (brw->prim_restart.enable_cut_index && !brw->is_haswell) {
+ cut_index_setting = BRW_CUT_INDEX_ENABLE;
+ } else {
+ cut_index_setting = 0;
+ }
+
BEGIN_BATCH(3);
OUT_BATCH(CMD_INDEX_BUFFER << 16 |
- /* cut index enable << 10 */
- get_index_type(index_buffer->type) << 8 |
+ cut_index_setting |
+ brw_get_index_type(index_buffer->type) << 8 |
1);
OUT_RELOC(brw->ib.bo,
I915_GEM_DOMAIN_VERTEX, 0,