#include "tnl/t_vertex.h"
#include "tnl/t_pipeline.h"
-#include "radeon_mipmap_tree.h"
#include "r600_context.h"
#include "r600_cmdbuf.h"
#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);
-GLboolean r700SyncSurf(context_t *context);
-unsigned int r700PrimitiveType(int prim);
+static unsigned int r700PrimitiveType(int prim);
void r600UpdateTextureState(GLcontext * ctx);
+GLboolean r700SyncSurf(context_t *context,
+ struct radeon_bo *pbo,
+ uint32_t read_domain,
+ uint32_t write_domain,
+ uint32_t sync_type);
void r700WaitForIdle(context_t *context)
{
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;
-
- return GL_TRUE;
-}
-
-GLboolean r700SendTextureState(context_t *context)
-{
- unsigned int i;
- R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw);
- offset_modifiers offset_mod = {NO_SHIFT, 0, 0xFFFFFFFF};
- struct radeon_bo *bo = NULL;
- BATCH_LOCALS(&context->radeon);
-
- for (i=0; i<R700_TEXTURE_NUMBERUNITS; i++) {
- radeonTexObj *t = r700->textures[i];
- if (t) {
- if (!t->image_override)
- bo = t->mt->bo;
- else
- bo = t->bo;
- if (bo) {
- BEGIN_BATCH_NO_AUTOSTATE(9);
- R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_RESOURCE, 7));
- R600_OUT_BATCH(i * 7);
- R600_OUT_BATCH(r700->textures[i]->SQ_TEX_RESOURCE0);
- R600_OUT_BATCH(r700->textures[i]->SQ_TEX_RESOURCE1);
- R600_OUT_BATCH_RELOC(r700->textures[i]->SQ_TEX_RESOURCE2,
- bo,
- 0,
- RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0, &offset_mod);
- R600_OUT_BATCH_RELOC(r700->textures[i]->SQ_TEX_RESOURCE3,
- bo,
- 0,
- RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0, &offset_mod);
- R600_OUT_BATCH(r700->textures[i]->SQ_TEX_RESOURCE4);
- R600_OUT_BATCH(r700->textures[i]->SQ_TEX_RESOURCE5);
- R600_OUT_BATCH(r700->textures[i]->SQ_TEX_RESOURCE6);
- END_BATCH();
-
- BEGIN_BATCH_NO_AUTOSTATE(5);
- R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_SAMPLER, 3));
- R600_OUT_BATCH(i * 3);
- R600_OUT_BATCH(r700->textures[i]->SQ_TEX_SAMPLER0);
- R600_OUT_BATCH(r700->textures[i]->SQ_TEX_SAMPLER1);
- R600_OUT_BATCH(r700->textures[i]->SQ_TEX_SAMPLER2);
- END_BATCH();
- COMMIT_BATCH();
- }
- }
- }
- return GL_TRUE;
-}
-
-GLboolean r700SyncSurf(context_t *context)
+GLboolean r700SyncSurf(context_t *context,
+ struct radeon_bo *pbo,
+ uint32_t read_domain,
+ uint32_t write_domain,
+ uint32_t sync_type)
{
BATCH_LOCALS(&context->radeon);
+ uint32_t cp_coher_size;
- /* TODO : too heavy? */
- unsigned int CP_COHER_CNTL = 0;
+ if (!pbo)
+ return GL_FALSE;
- CP_COHER_CNTL |= (TC_ACTION_ENA_bit
- | VC_ACTION_ENA_bit
- | CB_ACTION_ENA_bit
- | DB_ACTION_ENA_bit
- | SH_ACTION_ENA_bit
- | SMX_ACTION_ENA_bit);
+ if (pbo->size == 0xffffffff)
+ cp_coher_size = 0xffffffff;
+ 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(CP_COHER_CNTL);
- R600_OUT_BATCH(0xFFFFFFFF);
- R600_OUT_BATCH(0x00000000);
+ 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); // ???
END_BATCH();
COMMIT_BATCH();
return GL_TRUE;
}
-unsigned int r700PrimitiveType(int prim)
+static unsigned int r700PrimitiveType(int prim)
{
switch (prim & PRIM_MODE_MASK)
{
}
}
-static GLboolean r700RunRender(GLcontext * ctx,
- struct tnl_pipeline_stage *stage)
+static int r700NumVerts(int num_verts, int prim)
{
- context_t *context = R700_CONTEXT(ctx);
- R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw);
- int lastIndex = 0;
-#if 1
- BATCH_LOCALS(&context->radeon);
-
- unsigned int i, j;
- TNLcontext *tnl = TNL_CONTEXT(ctx);
- struct vertex_buffer *vb = &tnl->vb;
-
- struct r700_fragment_program *fp = (struct r700_fragment_program *)
- (ctx->FragmentProgram._Current);
- if (context->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV770)
- {
- fp->r700AsmCode.bR6xx = 1;
- }
+ 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;
+}
- r700Start3D(context); /* TODO : this is too much. */
+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;
+ 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;
+
+ total_emit = 3 /* VGT_PRIMITIVE_TYPE */
+ + 2 /* VGT_INDEX_TYPE */
+ + 2 /* NUM_INSTANCES */
+ + num_indices + 3; /* DRAW_INDEX_IMMD */
+
+ BEGIN_BATCH_NO_AUTOSTATE(total_emit);
+ // prim
+ SETfield(vgt_primitive_type, type,
+ VGT_PRIMITIVE_TYPE__PRIM_TYPE_shift, VGT_PRIMITIVE_TYPE__PRIM_TYPE_mask);
+ R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_CONFIG_REG, 1));
+ R600_OUT_BATCH(mmVGT_PRIMITIVE_TYPE - ASIC_CONFIG_BASE_INDEX);
+ R600_OUT_BATCH(vgt_primitive_type);
- r700SyncSurf(context); /* TODO : make it light. */
+ // index type
+ SETfield(vgt_index_type, DI_INDEX_SIZE_32_BIT, INDEX_TYPE_shift, INDEX_TYPE_mask);
+ R600_OUT_BATCH(CP_PACKET3(R600_IT_INDEX_TYPE, 0));
+ R600_OUT_BATCH(vgt_index_type);
- r700SendSQConfig(context);
+ // num instances
+ R600_OUT_BATCH(CP_PACKET3(R600_IT_NUM_INSTANCES, 0));
+ R600_OUT_BATCH(1);
- r700UpdateShaders(ctx);
+ // draw packet
+ vgt_num_indices = num_indices;
+ SETfield(vgt_draw_initiator, DI_SRC_SEL_IMMEDIATE, SOURCE_SELECT_shift, SOURCE_SELECT_mask);
+ SETfield(vgt_draw_initiator, DI_MAJOR_MODE_0, MAJOR_MODE_shift, MAJOR_MODE_mask);
- r700SetScissor(context);
- r700SetRenderTarget(context, 0);
- r700SetDepthTarget(context);
+ R600_OUT_BATCH(CP_PACKET3(R600_IT_DRAW_INDEX_IMMD, (num_indices + 1)));
+ R600_OUT_BATCH(vgt_num_indices);
+ R600_OUT_BATCH(vgt_draw_initiator);
- if(r700SetupStreams(ctx))
- {
- return GL_TRUE;
- }
+ for (i = start; i < (start + num_indices); i++) {
+ R600_OUT_BATCH(i);
+ }
+ END_BATCH();
+ COMMIT_BATCH();
- /* flush TX */
- //r700SyncSurf(context); /* */
+}
- r600UpdateTextureState(ctx);
- r700SendTextureState(context);
+/* start 3d, idle, cb/db flush */
+#define PRE_EMIT_STATE_BUFSZ 10 + 5 + 14
- if(GL_FALSE == fp->translated)
- {
- if( GL_FALSE == r700TranslateFragmentShader(fp, &(fp->mesa_program)) )
- {
- return GL_TRUE;
- }
- }
+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;
+}
- /* flush SQ */
- //r700SyncSurf(context); /* */
- //r700SyncSurf(context); /* */
+static GLboolean r700RunRender(GLcontext * ctx,
+ struct tnl_pipeline_stage *stage)
+{
+ context_t *context = R700_CONTEXT(ctx);
+ radeonContextPtr radeon = &context->radeon;
+ unsigned int i, id = 0;
+ TNLcontext *tnl = TNL_CONTEXT(ctx);
+ struct vertex_buffer *vb = &tnl->vb;
+ struct radeon_renderbuffer *rrb;
- r700SetupShaders(ctx);
+ if (RADEON_DEBUG & DEBUG_PRIMS)
+ fprintf(stderr, "%s: cs begin at %d\n",
+ __func__, context->radeon.cmdbuf.cs->cdw);
- r700SendFSState(context); // FIXME just a place holder for now
- r700SendPSState(context);
- r700SendVSState(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);
- /* flush vtx */
- //r700SyncSurf(context); /* */
+ r700UpdateShaders(ctx);
+ r700SetScissor(context);
+ r700SetupVertexProgram(ctx);
+ r700SetupFragmentProgram(ctx);
+ r600UpdateTextureState(ctx);
- r700SendContextStates(context);
- r700SendViewportState(context, 0);
- r700SendRenderTargetState(context, 0);
+ GLuint emit_end = r700PredictRenderSize(ctx)
+ + context->radeon.cmdbuf.cs->cdw;
+ r700SetupStreams(ctx);
+ radeonEmitState(radeon);
/* richard test code */
- for (i = 0; i < vb->PrimitiveCount; i++)
- {
+ 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;
- GLuint numIndices = vb->Primitive[i].count;
- GLuint numEntires;
-
- unsigned int VGT_DRAW_INITIATOR = 0;
- unsigned int VGT_INDEX_TYPE = 0;
- unsigned int VGT_PRIMITIVE_TYPE = 0;
- unsigned int VGT_NUM_INDICES = 0;
-
- numEntires = 2 /* VGT_INDEX_TYPE */
- + 3 /* VGT_PRIMITIVE_TYPE */
- + numIndices + 3; /* DRAW_INDEX_IMMD */
-
- BEGIN_BATCH_NO_AUTOSTATE(numEntires);
-
- VGT_INDEX_TYPE |= DI_INDEX_SIZE_32_BIT << INDEX_TYPE_shift;
-
- R600_OUT_BATCH(CP_PACKET3(R600_IT_INDEX_TYPE, 0));
- R600_OUT_BATCH(VGT_INDEX_TYPE);
-
- VGT_NUM_INDICES = numIndices;
-
- VGT_PRIMITIVE_TYPE |= r700PrimitiveType(prim) << VGT_PRIMITIVE_TYPE__PRIM_TYPE_shift;
- R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_CONFIG_REG, 1));
- R600_OUT_BATCH(mmVGT_PRIMITIVE_TYPE - ASIC_CONFIG_BASE_INDEX);
- R600_OUT_BATCH(VGT_PRIMITIVE_TYPE);
-
- VGT_DRAW_INITIATOR |= DI_SRC_SEL_IMMEDIATE << SOURCE_SELECT_shift;
- VGT_DRAW_INITIATOR |= DI_MAJOR_MODE_0 << MAJOR_MODE_shift;
-
- R600_OUT_BATCH(CP_PACKET3(R600_IT_DRAW_INDEX_IMMD, (numIndices + 1)));
- R600_OUT_BATCH(VGT_NUM_INDICES);
- R600_OUT_BATCH(VGT_DRAW_INITIATOR);
-
- for (j = lastIndex; j < lastIndex + numIndices; j++)
- {
- R600_OUT_BATCH(j);
- }
- lastIndex += numIndices;
-
- END_BATCH();
- COMMIT_BATCH();
+ r700RunRenderPrimitive(ctx, start, end, prim);
}
/* Flush render op cached for last several quads. */
r700WaitForIdleClean(context);
- /* flush dst */
- //r700SyncSurf(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)));
- radeonReleaseArrays(ctx, 0);
+ 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);
- //richard test
- /* test stamp, write a number to mmSCRATCH4 */
-#if 0
- BEGIN_BATCH_NO_AUTOSTATE(3);
- R600_OUT_BATCH_REGVAL((0x2144 << 2), 0x56785678);
- END_BATCH();
- COMMIT_BATCH();
-#endif
+ radeonReleaseArrays(ctx, ~0);
-#endif //0
- rcommonFlushCmdBuf( &context->radeon, __FUNCTION__ );
+ 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;
}
return GL_TRUE;
}
- context_t *context = R700_CONTEXT(ctx);
-
- r700UpdateShaders(ctx);
-
bRet = r700RunRender(ctx, stage);
return bRet;
projects / mesa.git / blobdiff