* obviously this does work... Further investigation is needed.
*
* \author Nicolai Haehnle <prefect_@gmx.net>
+ *
+ * \todo Add immediate implementation back? Perhaps this is useful if there are
+ * no bugs...
*/
-#include "glheader.h"
-#include "state.h"
-#include "imports.h"
-#include "enums.h"
-#include "macros.h"
-#include "context.h"
-#include "dd.h"
-#include "simple_list.h"
-#include "api_arrayelt.h"
+#include "main/glheader.h"
+#include "main/state.h"
+#include "main/imports.h"
+#include "main/enums.h"
+#include "main/macros.h"
+#include "main/context.h"
+#include "main/dd.h"
+#include "main/simple_list.h"
+#include "main/api_arrayelt.h"
#include "swrast/swrast.h"
#include "swrast_setup/swrast_setup.h"
#include "vbo/vbo.h"
#include "r300_state.h"
#include "r300_reg.h"
#include "r300_tex.h"
-#include "r300_maos.h"
#include "r300_emit.h"
+#include "r300_fragprog.h"
extern int future_hw_tcl_on;
/**
* \brief Convert a OpenGL primitive type into a R300 primitive type.
*/
-static int r300PrimitiveType(r300ContextPtr rmesa, GLcontext * ctx, int prim)
+int r300PrimitiveType(r300ContextPtr rmesa, int prim)
{
- int type = -1;
-
switch (prim & PRIM_MODE_MASK) {
case GL_POINTS:
- type = R300_VAP_VF_CNTL__PRIM_POINTS;
+ return R300_VAP_VF_CNTL__PRIM_POINTS;
break;
case GL_LINES:
- type = R300_VAP_VF_CNTL__PRIM_LINES;
+ return R300_VAP_VF_CNTL__PRIM_LINES;
break;
case GL_LINE_STRIP:
- type = R300_VAP_VF_CNTL__PRIM_LINE_STRIP;
+ return R300_VAP_VF_CNTL__PRIM_LINE_STRIP;
break;
case GL_LINE_LOOP:
- type = R300_VAP_VF_CNTL__PRIM_LINE_LOOP;
+ return R300_VAP_VF_CNTL__PRIM_LINE_LOOP;
break;
case GL_TRIANGLES:
- type = R300_VAP_VF_CNTL__PRIM_TRIANGLES;
+ return R300_VAP_VF_CNTL__PRIM_TRIANGLES;
break;
case GL_TRIANGLE_STRIP:
- type = R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP;
+ return R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP;
break;
case GL_TRIANGLE_FAN:
- type = R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN;
+ return R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN;
break;
case GL_QUADS:
- type = R300_VAP_VF_CNTL__PRIM_QUADS;
+ return R300_VAP_VF_CNTL__PRIM_QUADS;
break;
case GL_QUAD_STRIP:
- type = R300_VAP_VF_CNTL__PRIM_QUAD_STRIP;
+ return R300_VAP_VF_CNTL__PRIM_QUAD_STRIP;
break;
case GL_POLYGON:
- type = R300_VAP_VF_CNTL__PRIM_POLYGON;
+ return R300_VAP_VF_CNTL__PRIM_POLYGON;
break;
default:
- fprintf(stderr,
- "%s:%s Do not know how to handle primitive 0x%04x - help me !\n",
- __FILE__, __FUNCTION__, prim & PRIM_MODE_MASK);
+ assert(0);
return -1;
break;
}
- return type;
}
-static int r300NumVerts(r300ContextPtr rmesa, int num_verts, int prim)
+int r300NumVerts(r300ContextPtr rmesa, int num_verts, int prim)
{
int verts_off = 0;
verts_off = num_verts;
break;
default:
- fprintf(stderr,
- "%s:%s Do not know how to handle primitive 0x%04x - help me !\n",
- __FILE__, __FUNCTION__, prim & PRIM_MODE_MASK);
+ assert(0);
return -1;
break;
}
- if (num_verts - verts_off == 0) {
- WARN_ONCE
- ("user error: Need more than %d vertices to draw primitive 0x%04x !\n",
- num_verts, prim & PRIM_MODE_MASK);
- return -1;
- }
-
- if (verts_off > 0) {
- WARN_ONCE
- ("user error: %d is not a valid number of vertices for primitive 0x%04x !\n",
- num_verts, prim & PRIM_MODE_MASK);
- return -1;
- }
-
return num_verts - verts_off;
}
-static void inline r300FireEB(r300ContextPtr rmesa, unsigned long addr,
- int vertex_count, int type, int elt_size)
+static void r300EmitElts(GLcontext * ctx, void *elts, unsigned long n_elts)
{
- int cmd_reserved = 0;
- int cmd_written = 0;
- drm_radeon_cmd_header_t *cmd = NULL;
- unsigned long t_addr;
- unsigned long magic_1, magic_2;
-
- assert(elt_size == 2 || elt_size == 4);
+ r300ContextPtr rmesa = R300_CONTEXT(ctx);
+ void *out;
+
+ rmesa->state.elt_dma_bo = radeon_bo_open(rmesa->radeon.radeonScreen->bom,
+ 0, n_elts * 4, 4,
+ RADEON_GEM_DOMAIN_GTT, 0);
+ rmesa->state.elt_dma_offset = 0;
+ radeon_bo_map(rmesa->state.elt_dma_bo, 1);
+ out = rmesa->state.elt_dma_bo->ptr + rmesa->state.elt_dma_offset;
+ memcpy(out, elts, n_elts * 4);
+ radeon_bo_unmap(rmesa->state.elt_dma_bo);
+}
- if (addr & (elt_size - 1)) {
- WARN_ONCE("Badly aligned buffer\n");
- return;
- }
+static void r300FireEB(r300ContextPtr rmesa, int vertex_count, int type)
+{
+ BATCH_LOCALS(rmesa);
+
+ if (vertex_count > 0) {
+ BEGIN_BATCH(8);
+ OUT_BATCH_PACKET3(R300_PACKET3_3D_DRAW_INDX_2, 0);
+ OUT_BATCH(R300_VAP_VF_CNTL__PRIM_WALK_INDICES |
+ ((vertex_count + 0) << 16) |
+ type |
+ R300_VAP_VF_CNTL__INDEX_SIZE_32bit);
+
+ if (!rmesa->radeon.radeonScreen->driScreen->dri2.enabled) {
+ OUT_BATCH_PACKET3(R300_PACKET3_INDX_BUFFER, 2);
+ OUT_BATCH(R300_EB_UNK1 | (0 << 16) | R300_EB_UNK2);
+ OUT_BATCH_RELOC(rmesa->state.elt_dma_offset,
+ rmesa->state.elt_dma_bo,
+ rmesa->state.elt_dma_offset,
+ RADEON_GEM_DOMAIN_GTT, 0, 0);
+ OUT_BATCH(vertex_count);
+ } else {
+ OUT_BATCH_PACKET3(R300_PACKET3_INDX_BUFFER, 2);
+ OUT_BATCH(R300_EB_UNK1 | (0 << 16) | R300_EB_UNK2);
+ OUT_BATCH(rmesa->state.elt_dma_offset);
+ OUT_BATCH(vertex_count);
+ radeon_cs_write_reloc(rmesa->cmdbuf.cs,
+ rmesa->state.elt_dma_bo,
+ 0,
+ rmesa->state.elt_dma_bo->size,
+ RADEON_GEM_DOMAIN_GTT, 0, 0);
+ }
+ END_BATCH();
+ }
+}
- magic_1 = (addr % 32) / 4;
- t_addr = addr & ~0x1d;
- magic_2 = (vertex_count + 1 + (t_addr & 0x2)) / 2 + magic_1;
+static void r300EmitAOS(r300ContextPtr rmesa, GLuint nr, GLuint offset)
+{
+ BATCH_LOCALS(rmesa);
+ uint32_t voffset;
+ int sz = 1 + (nr >> 1) * 3 + (nr & 1) * 2;
+ int i;
- start_packet3(RADEON_CP_PACKET3_3D_DRAW_INDX_2, 0);
- if (elt_size == 4) {
- e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES |
- (vertex_count << 16) | type |
- R300_VAP_VF_CNTL__INDEX_SIZE_32bit);
- } else {
- e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES |
- (vertex_count << 16) | type);
+ if (RADEON_DEBUG & DEBUG_VERTS)
+ fprintf(stderr, "%s: nr=%d, ofs=0x%08x\n", __FUNCTION__, nr,
+ offset);
+
+ BEGIN_BATCH(sz+2);
+ OUT_BATCH_PACKET3(R300_PACKET3_3D_LOAD_VBPNTR, sz - 1);
+ OUT_BATCH(nr);
+
+
+ if (!rmesa->radeon.radeonScreen->driScreen->dri2.enabled) {
+ for (i = 0; i + 1 < nr; i += 2) {
+ OUT_BATCH((rmesa->state.aos[i].components << 0) |
+ (rmesa->state.aos[i].stride << 8) |
+ (rmesa->state.aos[i + 1].components << 16) |
+ (rmesa->state.aos[i + 1].stride << 24));
+
+ voffset = rmesa->state.aos[i + 0].offset +
+ offset * 4 * rmesa->state.aos[i + 0].stride;
+ OUT_BATCH_RELOC(voffset,
+ rmesa->state.aos[i].bo,
+ voffset,
+ RADEON_GEM_DOMAIN_GTT,
+ 0, 0);
+ voffset = rmesa->state.aos[i + 1].offset +
+ offset * 4 * rmesa->state.aos[i + 1].stride;
+ OUT_BATCH_RELOC(voffset,
+ rmesa->state.aos[i+1].bo,
+ voffset,
+ RADEON_GEM_DOMAIN_GTT,
+ 0, 0);
}
- start_packet3(RADEON_CP_PACKET3_INDX_BUFFER, 2);
-#ifdef OPTIMIZE_ELTS
- if (elt_size == 4) {
- e32(R300_EB_UNK1 | (0 << 16) | R300_EB_UNK2);
- e32(addr);
- } else {
- e32(R300_EB_UNK1 | (magic_1 << 16) | R300_EB_UNK2);
- e32(t_addr);
+ if (nr & 1) {
+ OUT_BATCH((rmesa->state.aos[nr - 1].components << 0) |
+ (rmesa->state.aos[nr - 1].stride << 8));
+ voffset = rmesa->state.aos[nr - 1].offset +
+ offset * 4 * rmesa->state.aos[nr - 1].stride;
+ OUT_BATCH_RELOC(voffset,
+ rmesa->state.aos[nr - 1].bo,
+ voffset,
+ RADEON_GEM_DOMAIN_GTT,
+ 0, 0);
+ }
+ } else {
+ for (i = 0; i + 1 < nr; i += 2) {
+ OUT_BATCH((rmesa->state.aos[i].components << 0) |
+ (rmesa->state.aos[i].stride << 8) |
+ (rmesa->state.aos[i + 1].components << 16) |
+ (rmesa->state.aos[i + 1].stride << 24));
+
+ voffset = rmesa->state.aos[i + 0].offset +
+ offset * 4 * rmesa->state.aos[i + 0].stride;
+ OUT_BATCH(voffset);
+ voffset = rmesa->state.aos[i + 1].offset +
+ offset * 4 * rmesa->state.aos[i + 1].stride;
+ OUT_BATCH(voffset);
}
-#else
- e32(R300_EB_UNK1 | (0 << 16) | R300_EB_UNK2);
- e32(addr);
-#endif
- if (elt_size == 4) {
- e32(vertex_count);
- } else {
-#ifdef OPTIMIZE_ELTS
- e32(magic_2);
-#else
- e32((vertex_count + 1) / 2);
-#endif
+ if (nr & 1) {
+ OUT_BATCH((rmesa->state.aos[nr - 1].components << 0) |
+ (rmesa->state.aos[nr - 1].stride << 8));
+ voffset = rmesa->state.aos[nr - 1].offset +
+ offset * 4 * rmesa->state.aos[nr - 1].stride;
+ OUT_BATCH(voffset);
+ }
+ for (i = 0; i + 1 < nr; i += 2) {
+ voffset = rmesa->state.aos[i + 0].offset +
+ offset * 4 * rmesa->state.aos[i + 0].stride;
+ radeon_cs_write_reloc(rmesa->cmdbuf.cs,
+ rmesa->state.aos[i+0].bo,
+ voffset,
+ rmesa->state.aos[i+0].bo->size,
+ RADEON_GEM_DOMAIN_GTT,
+ 0, 0);
+ voffset = rmesa->state.aos[i + 1].offset +
+ offset * 4 * rmesa->state.aos[i + 1].stride;
+ radeon_cs_write_reloc(rmesa->cmdbuf.cs,
+ rmesa->state.aos[i+1].bo,
+ voffset,
+ rmesa->state.aos[i+1].bo->size,
+ RADEON_GEM_DOMAIN_GTT,
+ 0, 0);
+ }
+ if (nr & 1) {
+ voffset = rmesa->state.aos[nr - 1].offset +
+ offset * 4 * rmesa->state.aos[nr - 1].stride;
+ radeon_cs_write_reloc(rmesa->cmdbuf.cs,
+ rmesa->state.aos[nr-1].bo,
+ voffset,
+ rmesa->state.aos[nr-1].bo->size,
+ RADEON_GEM_DOMAIN_GTT,
+ 0, 0);
}
+ }
+ END_BATCH();
+}
+
+static void r300FireAOS(r300ContextPtr rmesa, int vertex_count, int type)
+{
+ BATCH_LOCALS(rmesa);
+
+ BEGIN_BATCH(3);
+ OUT_BATCH_PACKET3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
+ OUT_BATCH(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (vertex_count << 16) | type);
+ END_BATCH();
}
static void r300RunRenderPrimitive(r300ContextPtr rmesa, GLcontext * ctx,
int start, int end, int prim)
{
+ BATCH_LOCALS(rmesa);
int type, num_verts;
+ TNLcontext *tnl = TNL_CONTEXT(ctx);
+ struct vertex_buffer *vb = &tnl->vb;
- type = r300PrimitiveType(rmesa, ctx, prim);
+ type = r300PrimitiveType(rmesa, prim);
num_verts = r300NumVerts(rmesa, end - start, prim);
if (type < 0 || num_verts <= 0)
return;
- if (rmesa->state.VB.Elts) {
- r300EmitAOS(rmesa, rmesa->state.aos_count, start);
+ /* Make space for at least 64 dwords.
+ * This is supposed to ensure that we can get all rendering
+ * commands into a single command buffer.
+ */
+ r300EnsureCmdBufSpace(rmesa, 64, __FUNCTION__);
+
+ if (vb->Elts) {
if (num_verts > 65535) {
/* not implemented yet */
WARN_ONCE("Too many elts\n");
return;
}
- r300EmitElts(ctx, rmesa->state.VB.Elts, num_verts,
- rmesa->state.VB.elt_size);
- r300FireEB(rmesa, rmesa->state.elt_dma.aos_offset,
- num_verts, type, rmesa->state.VB.elt_size);
- } else {
+ /* Note: The following is incorrect, but it's the best I can do
+ * without a major refactoring of how DMA memory is handled.
+ * The problem: Ensuring that both vertex arrays *and* index
+ * arrays are at the right position, and then ensuring that
+ * the LOAD_VBPNTR, DRAW_INDX and INDX_BUFFER packets are emitted
+ * at once.
+ *
+ * So why is the following incorrect? Well, it seems like
+ * allocating the index array might actually evict the vertex
+ * arrays. *sigh*
+ */
+ r300EmitElts(ctx, vb->Elts, num_verts);
r300EmitAOS(rmesa, rmesa->state.aos_count, start);
- fire_AOS(rmesa, num_verts, type);
- }
-}
-
-#define CONV_VB(a, b) rvb->AttribPtr[(a)].size = vb->b->size, \
- rvb->AttribPtr[(a)].type = GL_FLOAT, \
- rvb->AttribPtr[(a)].stride = vb->b->stride, \
- rvb->AttribPtr[(a)].data = vb->b->data
-
-static void radeon_vb_to_rvb(r300ContextPtr rmesa,
- struct radeon_vertex_buffer *rvb,
- struct vertex_buffer *vb)
-{
- int i;
- GLcontext *ctx;
- ctx = rmesa->radeon.glCtx;
-
- memset(rvb, 0, sizeof(*rvb));
-
- rvb->Elts = vb->Elts;
- rvb->elt_size = 4;
- rvb->elt_min = 0;
- rvb->elt_max = vb->Count;
-
- rvb->Count = vb->Count;
-
- if (hw_tcl_on) {
- CONV_VB(VERT_ATTRIB_POS, ObjPtr);
+ r300FireEB(rmesa, num_verts, type);
} else {
- assert(vb->ClipPtr);
- CONV_VB(VERT_ATTRIB_POS, ClipPtr);
+ r300EmitAOS(rmesa, rmesa->state.aos_count, start);
+ r300FireAOS(rmesa, num_verts, type);
}
-
- CONV_VB(VERT_ATTRIB_NORMAL, NormalPtr);
- CONV_VB(VERT_ATTRIB_COLOR0, ColorPtr[0]);
- CONV_VB(VERT_ATTRIB_COLOR1, SecondaryColorPtr[0]);
- CONV_VB(VERT_ATTRIB_FOG, FogCoordPtr);
-
- for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++)
- CONV_VB(VERT_ATTRIB_TEX0 + i, TexCoordPtr[i]);
-
- for (i = 0; i < MAX_VERTEX_PROGRAM_ATTRIBS; i++)
- CONV_VB(VERT_ATTRIB_GENERIC0 + i,
- AttribPtr[VERT_ATTRIB_GENERIC0 + i]);
-
- rvb->Primitive = vb->Primitive;
- rvb->PrimitiveCount = vb->PrimitiveCount;
- rvb->LockFirst = rvb->LockCount = 0;
- rvb->lock_uptodate = GL_FALSE;
+ COMMIT_BATCH();
}
static GLboolean r300RunRender(GLcontext * ctx,
struct tnl_pipeline_stage *stage)
{
r300ContextPtr rmesa = R300_CONTEXT(ctx);
- struct radeon_vertex_buffer *VB = &rmesa->state.VB;
int i;
- int cmd_reserved = 0;
- int cmd_written = 0;
- drm_radeon_cmd_header_t *cmd = NULL;
+ TNLcontext *tnl = TNL_CONTEXT(ctx);
+ struct vertex_buffer *vb = &tnl->vb;
if (RADEON_DEBUG & DEBUG_PRIMS)
fprintf(stderr, "%s\n", __FUNCTION__);
- if (stage) {
- TNLcontext *tnl = TNL_CONTEXT(ctx);
- radeon_vb_to_rvb(rmesa, VB, &tnl->vb);
- }
-
r300UpdateShaders(rmesa);
if (r300EmitArrays(ctx))
return GL_TRUE;
r300UpdateShaderStates(rmesa);
- reg_start(R300_RB3D_DSTCACHE_CTLSTAT, 0);
- e32(R300_RB3D_DSTCACHE_UNKNOWN_0A);
-
- reg_start(R300_RB3D_ZCACHE_CTLSTAT, 0);
- e32(R300_RB3D_ZCACHE_UNKNOWN_03);
-
+ r300EmitCacheFlush(rmesa);
r300EmitState(rmesa);
- for (i = 0; i < VB->PrimitiveCount; i++) {
- GLuint prim = _tnl_translate_prim(&VB->Primitive[i]);
- GLuint start = VB->Primitive[i].start;
- GLuint length = VB->Primitive[i].count;
- GLuint end = VB->Primitive[i].start + VB->Primitive[i].count;
+ for (i = 0; i < vb->PrimitiveCount; i++) {
+ GLuint prim = _tnl_translate_prim(&vb->Primitive[i]);
+ GLuint start = vb->Primitive[i].start;
+ GLuint end = vb->Primitive[i].start + vb->Primitive[i].count;
r300RunRenderPrimitive(rmesa, ctx, start, end, prim);
}
- reg_start(R300_RB3D_DSTCACHE_CTLSTAT, 0);
- e32(R300_RB3D_DSTCACHE_UNKNOWN_0A);
-
- reg_start(R300_RB3D_ZCACHE_CTLSTAT, 0);
- e32(R300_RB3D_ZCACHE_UNKNOWN_03);
-
-#ifdef USER_BUFFERS
- r300UseArrays(ctx);
-#endif
+ r300EmitCacheFlush(rmesa);
r300ReleaseArrays(ctx);
static int r300Fallback(GLcontext * ctx)
{
r300ContextPtr r300 = R300_CONTEXT(ctx);
- struct r300_fragment_program *fp = (struct r300_fragment_program *)
+ /* Do we need to use new-style shaders?
+ * Also is there a better way to do this? */
+ if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) {
+ struct r500_fragment_program *fp = (struct r500_fragment_program *)
(char *)ctx->FragmentProgram._Current;
-
- if (fp) {
- if (!fp->translated)
- r300TranslateFragmentShader(r300, fp);
- FALLBACK_IF(!fp->translated);
+ if (fp) {
+ if (!fp->translated) {
+ r500TranslateFragmentShader(r300, fp);
+ FALLBACK_IF(!fp->translated);
+ }
+ }
+ } else {
+ struct r300_fragment_program *fp = (struct r300_fragment_program *)
+ (char *)ctx->FragmentProgram._Current;
+ if (fp) {
+ if (!fp->translated) {
+ r300TranslateFragmentShader(r300, fp);
+ FALLBACK_IF(!fp->translated);
+ }
+ }
}
FALLBACK_IF(ctx->RenderMode != GL_RENDER);
|| ctx->Stencil.WriteMask[0] !=
ctx->Stencil.WriteMask[1]));
- FALLBACK_IF(ctx->Color.ColorLogicOpEnabled);
-
if (ctx->Extensions.NV_point_sprite || ctx->Extensions.ARB_point_sprite)
FALLBACK_IF(ctx->Point.PointSprite);
if (!r300->disable_lowimpact_fallback) {
- FALLBACK_IF(ctx->Polygon.OffsetPoint);
- FALLBACK_IF(ctx->Polygon.OffsetLine);
FALLBACK_IF(ctx->Polygon.StippleFlag);
- FALLBACK_IF(ctx->Multisample.Enabled);
+ FALLBACK_IF(ctx->Multisample._Enabled);
FALLBACK_IF(ctx->Line.StippleFlag);
FALLBACK_IF(ctx->Line.SmoothFlag);
FALLBACK_IF(ctx->Point.SmoothFlag);
static GLboolean r300RunNonTCLRender(GLcontext * ctx,
struct tnl_pipeline_stage *stage)
{
+ r300ContextPtr rmesa = R300_CONTEXT(ctx);
+
if (RADEON_DEBUG & DEBUG_PRIMS)
fprintf(stderr, "%s\n", __FUNCTION__);
if (r300Fallback(ctx) >= R300_FALLBACK_RAST)
return GL_TRUE;
+ if (!(rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL))
+ return GL_TRUE;
+
return r300RunRender(ctx, stage);
}
projects / mesa.git / blobdiff