/**
* Ensure all enabled and complete textures are uploaded along with any buffers being used.
*/
GLboolean r300ValidateBuffers(GLcontext * ctx)
{
r300ContextPtr rmesa = R300_CONTEXT(ctx);
struct radeon_renderbuffer *rrb;
int i;
int ret;
radeon_cs_space_reset_bos(rmesa->radeon.cmdbuf.cs);
rrb = radeon_get_colorbuffer(&rmesa->radeon);
/* color buffer */
if (rrb && rrb->bo) {
radeon_cs_space_add_persistent_bo(rmesa->radeon.cmdbuf.cs,
rrb->bo, 0,
RADEON_GEM_DOMAIN_VRAM);
}
/* depth buffer */
rrb = radeon_get_depthbuffer(&rmesa->radeon);
if (rrb && rrb->bo) {
radeon_cs_space_add_persistent_bo(rmesa->radeon.cmdbuf.cs,
rrb->bo, 0,
RADEON_GEM_DOMAIN_VRAM);
}
for (i = 0; i < ctx->Const.MaxTextureImageUnits; ++i) {
radeonTexObj *t;
if (!ctx->Texture.Unit[i]._ReallyEnabled)
continue;
if (!r300_validate_texture(ctx, ctx->Texture.Unit[i]._Current)) {
_mesa_warning(ctx,
"failed to validate texture for unit %d.\n",
i);
}
t = radeon_tex_obj(ctx->Texture.Unit[i]._Current);
if (t->image_override && t->bo)
radeon_cs_space_add_persistent_bo(rmesa->radeon.cmdbuf.cs,
t->bo,
RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0);
else if (t->mt->bo)
radeon_cs_space_add_persistent_bo(rmesa->radeon.cmdbuf.cs,
t->mt->bo,
RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0);
}
ret = radeon_cs_space_check_with_bo(rmesa->radeon.cmdbuf.cs, first_elem(&rmesa->radeon.dma.reserved)->bo, RADEON_GEM_DOMAIN_GTT, 0);
if (ret)
return GL_FALSE;
return GL_TRUE;
}
static int check_always_ctx( struct gl_context *ctx, struct radeon_state_atom *atom)
{
r100ContextPtr r100 = R100_CONTEXT(ctx);
struct radeon_renderbuffer *rrb, *drb;
uint32_t dwords;
rrb = radeon_get_colorbuffer(&r100->radeon);
if (!rrb || !rrb->bo) {
return 0;
}
drb = radeon_get_depthbuffer(&r100->radeon);
dwords = 10;
if (drb)
dwords += 6;
if (rrb)
dwords += 8;
return dwords;
}
static void r700SetDepthTarget(context_t *context)
{
R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw);
struct radeon_renderbuffer *rrb;
unsigned int nPitchInPixel;
rrb = radeon_get_depthbuffer(&context->radeon);
if (!rrb)
return;
R600_STATECHANGE(context, db_target);
/* depth buf */
r700->DB_DEPTH_SIZE.u32All = 0;
r700->DB_DEPTH_BASE.u32All = 0;
r700->DB_DEPTH_INFO.u32All = 0;
r700->DB_DEPTH_VIEW.u32All = 0;
nPitchInPixel = rrb->pitch/rrb->cpp;
SETfield(r700->DB_DEPTH_SIZE.u32All, (nPitchInPixel/8)-1,
PITCH_TILE_MAX_shift, PITCH_TILE_MAX_mask);
SETfield(r700->DB_DEPTH_SIZE.u32All, ( (nPitchInPixel * context->radeon.radeonScreen->driScreen->fbHeight)/64 )-1,
SLICE_TILE_MAX_shift, SLICE_TILE_MAX_mask); /* size in pixel / 64 - 1 */
if(4 == rrb->cpp)
{
SETfield(r700->DB_DEPTH_INFO.u32All, DEPTH_8_24,
DB_DEPTH_INFO__FORMAT_shift, DB_DEPTH_INFO__FORMAT_mask);
}
else
{
SETfield(r700->DB_DEPTH_INFO.u32All, DEPTH_16,
DB_DEPTH_INFO__FORMAT_shift, DB_DEPTH_INFO__FORMAT_mask);
}
SETfield(r700->DB_DEPTH_INFO.u32All, ARRAY_1D_TILED_THIN1,
DB_DEPTH_INFO__ARRAY_MODE_shift, DB_DEPTH_INFO__ARRAY_MODE_mask);
/* r700->DB_PREFETCH_LIMIT.bits.DEPTH_HEIGHT_TILE_MAX = (context->currentDraw->h >> 3) - 1; */ /* z buffer sie may much bigger than what need, so use actual used h. */
}
static void r700SendDepthTargetState(GLcontext *ctx, struct radeon_state_atom *atom)
{
context_t *context = R700_CONTEXT(ctx);
R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context);
struct radeon_renderbuffer *rrb;
BATCH_LOCALS(&context->radeon);
radeon_print(RADEON_STATE, RADEON_VERBOSE, "%s\n", __func__);
rrb = radeon_get_depthbuffer(&context->radeon);
if (!rrb || !rrb->bo) {
return;
}
r700SetDepthTarget(context);
BEGIN_BATCH_NO_AUTOSTATE(8 + 2);
R600_OUT_BATCH_REGSEQ(DB_DEPTH_SIZE, 2);
R600_OUT_BATCH(r700->DB_DEPTH_SIZE.u32All);
R600_OUT_BATCH(r700->DB_DEPTH_VIEW.u32All);
R600_OUT_BATCH_REGSEQ(DB_DEPTH_BASE, 2);
R600_OUT_BATCH(r700->DB_DEPTH_BASE.u32All);
R600_OUT_BATCH(r700->DB_DEPTH_INFO.u32All);
R600_OUT_BATCH_RELOC(r700->DB_DEPTH_BASE.u32All,
rrb->bo,
r700->DB_DEPTH_BASE.u32All,
0, RADEON_GEM_DOMAIN_VRAM, 0);
END_BATCH();
if ((context->radeon.radeonScreen->chip_family > CHIP_FAMILY_R600) &&
(context->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV770)) {
BEGIN_BATCH_NO_AUTOSTATE(2);
R600_OUT_BATCH(CP_PACKET3(R600_IT_SURFACE_BASE_UPDATE, 0));
R600_OUT_BATCH(1 << 0);
END_BATCH();
}
COMMIT_BATCH();
}
static void ctx_emit_cs(struct gl_context *ctx, struct radeon_state_atom *atom)
{
r100ContextPtr r100 = R100_CONTEXT(ctx);
BATCH_LOCALS(&r100->radeon);
struct radeon_renderbuffer *rrb, *drb;
uint32_t cbpitch = 0;
uint32_t zbpitch = 0;
uint32_t dwords = atom->check(ctx, atom);
uint32_t depth_fmt;
rrb = radeon_get_colorbuffer(&r100->radeon);
if (!rrb || !rrb->bo) {
fprintf(stderr, "no rrb\n");
return;
}
atom->cmd[CTX_RB3D_CNTL] &= ~(0xf << 10);
if (rrb->cpp == 4)
atom->cmd[CTX_RB3D_CNTL] |= RADEON_COLOR_FORMAT_ARGB8888;
else switch (rrb->base.Base.Format) {
case MESA_FORMAT_RGB565:
atom->cmd[CTX_RB3D_CNTL] |= RADEON_COLOR_FORMAT_RGB565;
break;
case MESA_FORMAT_ARGB4444:
atom->cmd[CTX_RB3D_CNTL] |= RADEON_COLOR_FORMAT_ARGB4444;
break;
case MESA_FORMAT_ARGB1555:
atom->cmd[CTX_RB3D_CNTL] |= RADEON_COLOR_FORMAT_ARGB1555;
break;
default:
_mesa_problem(ctx, "unexpected format in ctx_emit_cs()");
}
cbpitch = (rrb->pitch / rrb->cpp);
if (rrb->bo->flags & RADEON_BO_FLAGS_MACRO_TILE)
cbpitch |= R200_COLOR_TILE_ENABLE;
if (rrb->bo->flags & RADEON_BO_FLAGS_MICRO_TILE)
cbpitch |= RADEON_COLOR_MICROTILE_ENABLE;
drb = radeon_get_depthbuffer(&r100->radeon);
if (drb) {
zbpitch = (drb->pitch / drb->cpp);
if (drb->cpp == 4)
depth_fmt = RADEON_DEPTH_FORMAT_24BIT_INT_Z;
else
depth_fmt = RADEON_DEPTH_FORMAT_16BIT_INT_Z;
atom->cmd[CTX_RB3D_ZSTENCILCNTL] &= ~RADEON_DEPTH_FORMAT_MASK;
atom->cmd[CTX_RB3D_ZSTENCILCNTL] |= depth_fmt;
}
BEGIN_BATCH_NO_AUTOSTATE(dwords);
/* In the CS case we need to split this up */
OUT_BATCH(CP_PACKET0(packet[0].start, 3));
OUT_BATCH_TABLE((atom->cmd + 1), 4);
if (drb) {
OUT_BATCH(CP_PACKET0(RADEON_RB3D_DEPTHOFFSET, 0));
OUT_BATCH_RELOC(0, drb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
OUT_BATCH(CP_PACKET0(RADEON_RB3D_DEPTHPITCH, 0));
OUT_BATCH(zbpitch);
}
OUT_BATCH(CP_PACKET0(RADEON_RB3D_ZSTENCILCNTL, 0));
OUT_BATCH(atom->cmd[CTX_RB3D_ZSTENCILCNTL]);
OUT_BATCH(CP_PACKET0(RADEON_PP_CNTL, 1));
OUT_BATCH(atom->cmd[CTX_PP_CNTL]);
OUT_BATCH(atom->cmd[CTX_RB3D_CNTL]);
if (rrb) {
OUT_BATCH(CP_PACKET0(RADEON_RB3D_COLOROFFSET, 0));
OUT_BATCH_RELOC(rrb->draw_offset, rrb->bo, rrb->draw_offset, 0, RADEON_GEM_DOMAIN_VRAM, 0);
OUT_BATCH(CP_PACKET0(RADEON_RB3D_COLORPITCH, 0));
OUT_BATCH_RELOC(cbpitch, rrb->bo, cbpitch, 0, RADEON_GEM_DOMAIN_VRAM, 0);
}
// if (atom->cmd_size == CTX_STATE_SIZE_NEWDRM) {
// OUT_BATCH_TABLE((atom->cmd + 14), 4);
// }
END_BATCH();
BEGIN_BATCH_NO_AUTOSTATE(4);
OUT_BATCH(CP_PACKET0(RADEON_RE_TOP_LEFT, 0));
OUT_BATCH(0);
OUT_BATCH(CP_PACKET0(RADEON_RE_WIDTH_HEIGHT, 0));
if (rrb) {
OUT_BATCH(((rrb->base.Base.Width - 1) << RADEON_RE_WIDTH_SHIFT) |
((rrb->base.Base.Height - 1) << RADEON_RE_HEIGHT_SHIFT));
} else {
OUT_BATCH(0);
}
END_BATCH();
}
static void ctx_emit(GLcontext *ctx, struct radeon_state_atom *atom)
{
r100ContextPtr r100 = R100_CONTEXT(ctx);
BATCH_LOCALS(&r100->radeon);
struct radeon_renderbuffer *rrb;
uint32_t cbpitch;
uint32_t zbpitch, depth_fmt;
uint32_t dwords = atom->check(ctx, atom);
/* output the first 7 bytes of context */
BEGIN_BATCH_NO_AUTOSTATE(dwords);
OUT_BATCH_TABLE(atom->cmd, 5);
rrb = radeon_get_depthbuffer(&r100->radeon);
if (!rrb) {
OUT_BATCH(0);
OUT_BATCH(0);
} else {
zbpitch = (rrb->pitch / rrb->cpp);
if (r100->using_hyperz)
zbpitch |= RADEON_DEPTH_HYPERZ;
OUT_BATCH_RELOC(0, rrb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
OUT_BATCH(zbpitch);
if (rrb->cpp == 4)
depth_fmt = RADEON_DEPTH_FORMAT_24BIT_INT_Z;
else
depth_fmt = RADEON_DEPTH_FORMAT_16BIT_INT_Z;
atom->cmd[CTX_RB3D_ZSTENCILCNTL] &= ~RADEON_DEPTH_FORMAT_MASK;
atom->cmd[CTX_RB3D_ZSTENCILCNTL] |= depth_fmt;
}
OUT_BATCH(atom->cmd[CTX_RB3D_ZSTENCILCNTL]);
OUT_BATCH(atom->cmd[CTX_CMD_1]);
OUT_BATCH(atom->cmd[CTX_PP_CNTL]);
rrb = radeon_get_colorbuffer(&r100->radeon);
if (!rrb || !rrb->bo) {
OUT_BATCH(atom->cmd[CTX_RB3D_CNTL]);
OUT_BATCH(atom->cmd[CTX_RB3D_COLOROFFSET]);
} else {
atom->cmd[CTX_RB3D_CNTL] &= ~(0xf << 10);
if (rrb->cpp == 4)
atom->cmd[CTX_RB3D_CNTL] |= RADEON_COLOR_FORMAT_ARGB8888;
else
atom->cmd[CTX_RB3D_CNTL] |= RADEON_COLOR_FORMAT_RGB565;
OUT_BATCH(atom->cmd[CTX_RB3D_CNTL]);
OUT_BATCH_RELOC(0, rrb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
}
OUT_BATCH(atom->cmd[CTX_CMD_2]);
if (!rrb || !rrb->bo) {
OUT_BATCH(atom->cmd[CTX_RB3D_COLORPITCH]);
} else {
cbpitch = (rrb->pitch / rrb->cpp);
if (rrb->bo->flags & RADEON_BO_FLAGS_MACRO_TILE)
cbpitch |= RADEON_COLOR_TILE_ENABLE;
OUT_BATCH(cbpitch);
}
END_BATCH();
}
static GLboolean evergreenTryDrawPrims(GLcontext *ctx,
const struct gl_client_array *arrays[],
const struct _mesa_prim *prim,
GLuint nr_prims,
const struct _mesa_index_buffer *ib,
GLuint min_index,
GLuint max_index )
{
context_t *context = EVERGREEN_CONTEXT(ctx);
radeonContextPtr radeon = &context->radeon;
GLuint i, id = 0;
struct radeon_renderbuffer *rrb;
if (ctx->NewState)
_mesa_update_state( ctx );
if (evergreen_check_fallbacks(ctx))
return GL_FALSE;
_tnl_UpdateFixedFunctionProgram(ctx);
evergreenSetVertexFormat(ctx, arrays, max_index + 1);
/* shaders need to be updated before buffers are validated */
evergreenUpdateShaders(ctx);
if (!evergreenValidateBuffers(ctx))
return GL_FALSE;
/* always emit CB base to prevent
* lock ups on some chips.
*/
EVERGREEN_STATECHANGE(context, cb);
/* mark vtx as dirty since it changes per-draw */
EVERGREEN_STATECHANGE(context, vtx);
evergreenSetScissor(context);
evergreenSetupVertexProgram(ctx);
evergreenSetupFragmentProgram(ctx);
evergreenUpdateShaderStates(ctx);
GLuint emit_end = evergreenPredictRenderSize(ctx, prim, ib, nr_prims)
+ context->radeon.cmdbuf.cs->cdw;
/* evergreenPredictRenderSize will call radeonReleaseDmaRegions, so update VP/FP const buf after it. */
evergreenSetupVPconstants(ctx);
evergreenSetupFPconstants(ctx);
evergreenSetupIndexBuffer(ctx, ib);
evergreenSetupStreams(ctx, arrays, max_index + 1);
radeonEmitState(radeon);
radeon_debug_add_indent();
for (i = 0; i < nr_prims; ++i)
{
if (context->ind_buf.bo)
evergreenRunRenderPrimitive(ctx,
prim[i].start,
prim[i].start + prim[i].count,
prim[i].mode,
prim[i].basevertex);
else
evergreenRunRenderPrimitiveImmediate(ctx,
prim[i].start,
prim[i].start + prim[i].count,
prim[i].mode);
}
radeon_debug_remove_indent();
/* Flush render op cached for last several quads. */
/* XXX drm should handle this in fence submit */
//evergreeWaitForIdleClean(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);
evergreenFreeData(ctx);
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_TRUE;
}