* 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_reg.h"
#include "r300_tex.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)
{
switch (prim & PRIM_MODE_MASK) {
case GL_POINTS:
}
}
-static int r300NumVerts(r300ContextPtr rmesa, int num_verts, int prim)
+int r300NumVerts(r300ContextPtr rmesa, int num_verts, int prim)
{
int verts_off = 0;
return num_verts - verts_off;
}
-static void r300EmitElts(GLcontext * ctx, void *elts, unsigned long n_elts,
- int elt_size)
+static void r300EmitElts(GLcontext * ctx, void *elts, unsigned long n_elts)
{
r300ContextPtr rmesa = R300_CONTEXT(ctx);
- struct r300_dma_region *rvb = &rmesa->state.elt_dma;
void *out;
- assert(elt_size == 2 || elt_size == 4);
-
- if (r300IsGartMemory(rmesa, elts, n_elts * elt_size)) {
- rvb->address = rmesa->radeon.radeonScreen->gartTextures.map;
- rvb->start = ((char *)elts) - rvb->address;
- rvb->aos_offset =
- rmesa->radeon.radeonScreen->gart_texture_offset +
- rvb->start;
- return;
- } else if (r300IsGartMemory(rmesa, elts, 1)) {
- WARN_ONCE("Pointer not within GART memory!\n");
- _mesa_exit(-1);
- }
-
- r300AllocDmaRegion(rmesa, rvb, n_elts * elt_size, elt_size);
- rvb->aos_offset = GET_START(rvb);
-
- out = rvb->address + rvb->start;
- memcpy(out, elts, n_elts * elt_size);
+ 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);
}
-static void r300FireEB(r300ContextPtr rmesa, unsigned long addr,
- int vertex_count, int type, int elt_size)
+static void r300FireEB(r300ContextPtr rmesa, int vertex_count, int type)
{
- 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);
-
- if (addr & (elt_size - 1)) {
- WARN_ONCE("Badly aligned buffer\n");
- return;
- }
-
- magic_1 = (addr % 32) / 4;
- t_addr = addr & ~0x1d;
- magic_2 = (vertex_count + 1 + (t_addr & 0x2)) / 2 + magic_1;
-
- 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);
- }
-
- 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);
- }
-#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
- }
+ 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();
+ }
}
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;
- int cmd_reserved = 0;
- int cmd_written = 0;
- drm_radeon_cmd_header_t *cmd = NULL;
if (RADEON_DEBUG & DEBUG_VERTS)
fprintf(stderr, "%s: nr=%d, ofs=0x%08x\n", __FUNCTION__, nr,
offset);
- start_packet3(RADEON_CP_PACKET3_3D_LOAD_VBPNTR, sz - 1);
- e32(nr);
+ 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) {
- e32((rmesa->state.aos[i].aos_size << 0) |
- (rmesa->state.aos[i].aos_stride << 8) |
- (rmesa->state.aos[i + 1].aos_size << 16) |
- (rmesa->state.aos[i + 1].aos_stride << 24));
+ 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);
+ }
- e32(rmesa->state.aos[i].aos_offset + offset * 4 * rmesa->state.aos[i].aos_stride);
- e32(rmesa->state.aos[i + 1].aos_offset + offset * 4 * rmesa->state.aos[i + 1].aos_stride);
+ 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);
}
if (nr & 1) {
- e32((rmesa->state.aos[nr - 1].aos_size << 0) |
- (rmesa->state.aos[nr - 1].aos_stride << 8));
- e32(rmesa->state.aos[nr - 1].aos_offset + offset * 4 * rmesa->state.aos[nr - 1].aos_stride);
+ 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)
{
- int cmd_reserved = 0;
- int cmd_written = 0;
- drm_radeon_cmd_header_t *cmd = NULL;
+ BATCH_LOCALS(rmesa);
- start_packet3(RADEON_CP_PACKET3_3D_DRAW_VBUF_2, 0);
- e32(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (vertex_count << 16) | type);
+ 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;
+ /* 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) {
- r300EmitAOS(rmesa, rmesa->state.aos_count, start);
if (num_verts > 65535) {
/* not implemented yet */
WARN_ONCE("Too many elts\n");
return;
}
- r300EmitElts(ctx, vb->Elts, num_verts, 4);
- r300FireEB(rmesa, rmesa->state.elt_dma.aos_offset,
- num_verts, type, 4);
+ /* 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);
+ r300FireEB(rmesa, num_verts, type);
} else {
r300EmitAOS(rmesa, rmesa->state.aos_count, start);
r300FireAOS(rmesa, num_verts, type);
}
+ COMMIT_BATCH();
}
static GLboolean r300RunRender(GLcontext * ctx,
{
r300ContextPtr rmesa = R300_CONTEXT(ctx);
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;
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++) {
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);
}