#include "r700_fragprog.h"
#include "r700_state.h"
+#include "radeon_common_context.h"
+
void r700WaitForIdle(context_t *context);
void r700WaitForIdleClean(context_t *context);
-void r700Start3D(context_t *context);
GLboolean r700SendTextureState(context_t *context);
static unsigned int r700PrimitiveType(int prim);
void r600UpdateTextureState(GLcontext * ctx);
r700WaitForIdleClean(context);
}
-static GLboolean r700SetupShaders(GLcontext * ctx)
-{
- context_t *context = R700_CONTEXT(ctx);
-
- R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw);
-
- GLuint exportCount;
-
- r700->ps.SQ_PGM_RESOURCES_PS.u32All = 0;
- r700->vs.SQ_PGM_RESOURCES_VS.u32All = 0;
-
- SETbit(r700->ps.SQ_PGM_RESOURCES_PS.u32All, PGM_RESOURCES__PRIME_CACHE_ON_DRAW_bit);
- SETbit(r700->vs.SQ_PGM_RESOURCES_VS.u32All, PGM_RESOURCES__PRIME_CACHE_ON_DRAW_bit);
-
- r700SetupVertexProgram(ctx);
-
- r700SetupFragmentProgram(ctx);
-
- exportCount = (r700->ps.SQ_PGM_EXPORTS_PS.u32All & EXPORT_MODE_mask) / (1 << EXPORT_MODE_shift);
- r700->CB_SHADER_CONTROL.u32All = (1 << exportCount) - 1;
-
- r600UpdateTextureState(ctx);
-
- r700SendFSState(context); // FIXME just a place holder for now
- r700SendPSState(context);
- r700SendVSState(context);
-
- r700SendTextureState(context);
- r700SetupStreams(ctx);
-
- return GL_TRUE;
-}
-
GLboolean r700SyncSurf(context_t *context,
struct radeon_bo *pbo,
uint32_t read_domain,
else
cp_coher_size = ((pbo->size + 255) >> 8);
- BEGIN_BATCH_NO_AUTOSTATE(5);
+ BEGIN_BATCH_NO_AUTOSTATE(5 + 2);
R600_OUT_BATCH(CP_PACKET3(R600_IT_SURFACE_SYNC, 3));
R600_OUT_BATCH(sync_type);
R600_OUT_BATCH(cp_coher_size);
+ R600_OUT_BATCH(0);
+ R600_OUT_BATCH(10);
R600_OUT_BATCH_RELOC(0,
pbo,
0,
read_domain, write_domain, 0); // ???
- R600_OUT_BATCH(10);
END_BATCH();
COMMIT_BATCH();
}
}
+static int r700NumVerts(int num_verts, int prim)
+{
+ int verts_off = 0;
+
+ switch (prim & PRIM_MODE_MASK) {
+ case GL_POINTS:
+ verts_off = 0;
+ break;
+ case GL_LINES:
+ verts_off = num_verts % 2;
+ break;
+ case GL_LINE_STRIP:
+ if (num_verts < 2)
+ verts_off = num_verts;
+ break;
+ case GL_LINE_LOOP:
+ if (num_verts < 2)
+ verts_off = num_verts;
+ break;
+ case GL_TRIANGLES:
+ verts_off = num_verts % 3;
+ break;
+ case GL_TRIANGLE_STRIP:
+ if (num_verts < 3)
+ verts_off = num_verts;
+ break;
+ case GL_TRIANGLE_FAN:
+ if (num_verts < 3)
+ verts_off = num_verts;
+ break;
+ case GL_QUADS:
+ verts_off = num_verts % 4;
+ break;
+ case GL_QUAD_STRIP:
+ if (num_verts < 4)
+ verts_off = num_verts;
+ else
+ verts_off = num_verts % 2;
+ break;
+ case GL_POLYGON:
+ if (num_verts < 3)
+ verts_off = num_verts;
+ break;
+ default:
+ assert(0);
+ return -1;
+ break;
+ }
+
+ return num_verts - verts_off;
+}
+
static void r700RunRenderPrimitive(GLcontext * ctx, int start, int end, int prim)
{
context_t *context = R700_CONTEXT(ctx);
BATCH_LOCALS(&context->radeon);
int type, i, total_emit;
- int num_indices = end - start;
+ int num_indices;
uint32_t vgt_draw_initiator = 0;
uint32_t vgt_index_type = 0;
uint32_t vgt_primitive_type = 0;
uint32_t vgt_num_indices = 0;
type = r700PrimitiveType(prim);
+ num_indices = r700NumVerts(end - start, prim);
if (type < 0 || num_indices <= 0)
return;
R600_OUT_BATCH(vgt_num_indices);
R600_OUT_BATCH(vgt_draw_initiator);
- for (i = start; i < end; i++) {
+ for (i = start; i < (start + num_indices); i++) {
R600_OUT_BATCH(i);
}
END_BATCH();
}
-void r700EmitState(GLcontext * ctx)
-{
- context_t *context = R700_CONTEXT(ctx);
- radeonContextPtr radeon = &context->radeon;
-
- if (radeon->cmdbuf.cs->cdw && !radeon->hw.is_dirty && !radeon->hw.all_dirty)
- return;
-
- rcommonEnsureCmdBufSpace(&context->radeon,
- context->radeon.hw.max_state_size, __FUNCTION__);
-
- r700SendSQConfig(context);
+/* start 3d, idle, cb/db flush */
+#define PRE_EMIT_STATE_BUFSZ 10 + 5 + 14
- r700SendUCPState(context);
- r700SendContextStates(context);
- r700SendViewportState(context, 0);
- r700SendRenderTargetState(context, 0);
- r700SendDepthTargetState(context);
+static GLuint r700PredictRenderSize(GLcontext* ctx)
+{
+ context_t *context = R700_CONTEXT(ctx);
+ TNLcontext *tnl = TNL_CONTEXT(ctx);
+ struct r700_vertex_program *vpc
+ = (struct r700_vertex_program *)ctx->VertexProgram._Current;
+ struct vertex_buffer *vb = &tnl->vb;
+ GLboolean flushed;
+ GLuint dwords, i;
+ GLuint state_size;
+ /* pre calculate aos count so state prediction works */
+ context->radeon.tcl.aos_count = _mesa_bitcount(vpc->mesa_program.Base.InputsRead);
+
+ dwords = PRE_EMIT_STATE_BUFSZ;
+ for (i = 0; i < vb->PrimitiveCount; i++)
+ dwords += vb->Primitive[i].count + 10;
+ state_size = radeonCountStateEmitSize(&context->radeon);
+ flushed = rcommonEnsureCmdBufSpace(&context->radeon,
+ dwords + state_size, __FUNCTION__);
+
+ if (flushed)
+ dwords += radeonCountStateEmitSize(&context->radeon);
+ else
+ dwords += state_size;
+ if (RADEON_DEBUG & DEBUG_PRIMS)
+ fprintf(stderr, "%s: total prediction size is %d.\n", __FUNCTION__, dwords);
+ return dwords;
}
static GLboolean r700RunRender(GLcontext * ctx,
- struct tnl_pipeline_stage *stage)
+ struct tnl_pipeline_stage *stage)
{
context_t *context = R700_CONTEXT(ctx);
- unsigned int i;
+ radeonContextPtr radeon = &context->radeon;
+ unsigned int i, id = 0;
TNLcontext *tnl = TNL_CONTEXT(ctx);
struct vertex_buffer *vb = &tnl->vb;
+ struct radeon_renderbuffer *rrb;
+
+ if (RADEON_DEBUG & DEBUG_PRIMS)
+ fprintf(stderr, "%s: cs begin at %d\n",
+ __func__, context->radeon.cmdbuf.cs->cdw);
- r700Start3D(context);
+ /* always emit CB base to prevent
+ * lock ups on some chips.
+ */
+ R600_STATECHANGE(context, cb_target);
+ /* mark vtx as dirty since it changes per-draw */
+ R600_STATECHANGE(context, vtx);
r700UpdateShaders(ctx);
r700SetScissor(context);
- r700SetupShaders(ctx);
+ r700SetupVertexProgram(ctx);
+ r700SetupFragmentProgram(ctx);
+ r600UpdateTextureState(ctx);
- r700EmitState(ctx);
+ GLuint emit_end = r700PredictRenderSize(ctx)
+ + context->radeon.cmdbuf.cs->cdw;
+ r700SetupStreams(ctx);
+
+ radeonEmitState(radeon);
/* richard test code */
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;
- r700RunRenderPrimitive(ctx, start, end, prim);
+ r700RunRenderPrimitive(ctx, start, end, prim);
}
/* Flush render op cached for last several quads. */
r700WaitForIdleClean(context);
+ rrb = radeon_get_colorbuffer(&context->radeon);
+ if (rrb && rrb->bo)
+ r700SyncSurf(context, rrb->bo, 0, RADEON_GEM_DOMAIN_VRAM,
+ CB_ACTION_ENA_bit | (1 << (id + 6)));
+
+ rrb = radeon_get_depthbuffer(&context->radeon);
+ if (rrb && rrb->bo)
+ r700SyncSurf(context, rrb->bo, 0, RADEON_GEM_DOMAIN_VRAM,
+ DB_ACTION_ENA_bit | DB_DEST_BASE_ENA_bit);
+
radeonReleaseArrays(ctx, ~0);
+ if (RADEON_DEBUG & DEBUG_PRIMS)
+ fprintf(stderr, "%s: cs end at %d\n",
+ __func__, context->radeon.cmdbuf.cs->cdw);
+
+ if ( emit_end < context->radeon.cmdbuf.cs->cdw )
+ WARN_ONCE("Rendering was %d commands larger than predicted size."
+ " We might overflow command buffer.\n", context->radeon.cmdbuf.cs->cdw - emit_end);
+
return GL_FALSE;
}
projects / mesa.git / blobdiff