diff options
Diffstat (limited to 'src/mesa')
26 files changed, 1202 insertions, 594 deletions
diff --git a/src/mesa/drivers/dri/common/dri_util.c b/src/mesa/drivers/dri/common/dri_util.c index e48e10d7c06..d7bcd565d73 100644 --- a/src/mesa/drivers/dri/common/dri_util.c +++ b/src/mesa/drivers/dri/common/dri_util.c @@ -82,6 +82,46 @@ driIntersectArea( drm_clip_rect_t rect1, drm_clip_rect_t rect2 ) return (rect1.x2 - rect1.x1) * (rect1.y2 - rect1.y1); } +static int driFreeDrawable(__DRIcontext *pcp) +{ + __DRIdrawable *pdp; + __DRIdrawable *prp; + + if (pcp == NULL) + return GL_FALSE; + + pdp = pcp->driDrawablePriv; + prp = pcp->driReadablePriv; + + /* already unbound */ + if (!pdp && !prp) + return GL_TRUE; + + if (pdp->refcount == 0) { + /* ERROR!!! */ + return GL_FALSE; + } + + dri_put_drawable(pdp); + + if (prp != pdp) { + if (prp->refcount == 0) { + /* ERROR!!! */ + return GL_FALSE; + } + + dri_put_drawable(prp); + } + + + /* XXX this is disabled so that if we call SwapBuffers on an unbound + * window we can determine the last context bound to the window and + * use that context's lock. (BrianP, 2-Dec-2000) + */ + pcp->driDrawablePriv = pcp->driReadablePriv = NULL; + return GL_TRUE; +} + /*****************************************************************/ /** \name Context (un)binding functions */ /*****************************************************************/ @@ -106,8 +146,6 @@ driIntersectArea( drm_clip_rect_t rect1, drm_clip_rect_t rect2 ) static int driUnbindContext(__DRIcontext *pcp) { __DRIscreen *psp; - __DRIdrawable *pdp; - __DRIdrawable *prp; /* ** Assume error checking is done properly in glXMakeCurrent before @@ -118,38 +156,10 @@ static int driUnbindContext(__DRIcontext *pcp) return GL_FALSE; psp = pcp->driScreenPriv; - pdp = pcp->driDrawablePriv; - prp = pcp->driReadablePriv; - /* already unbound */ - if (!pdp && !prp) - return GL_TRUE; - /* Let driver unbind drawable from context */ + /* Let driver unbind drawable from context */ (*psp->DriverAPI.UnbindContext)(pcp); - if (pdp->refcount == 0) { - /* ERROR!!! */ - return GL_FALSE; - } - - dri_put_drawable(pdp); - - if (prp != pdp) { - if (prp->refcount == 0) { - /* ERROR!!! */ - return GL_FALSE; - } - - dri_put_drawable(prp); - } - - - /* XXX this is disabled so that if we call SwapBuffers on an unbound - * window we can determine the last context bound to the window and - * use that context's lock. (BrianP, 2-Dec-2000) - */ - pcp->driDrawablePriv = pcp->driReadablePriv = NULL; - #if 0 /* Unbind the drawable */ pdp->driContextPriv = &psp->dummyContextPriv; @@ -171,17 +181,44 @@ static int driBindContext(__DRIcontext *pcp, /* Bind the drawable to the context */ - if (pcp) { - pcp->driDrawablePriv = pdp; - pcp->driReadablePriv = prp; - if (pdp) { - pdp->driContextPriv = pcp; - dri_get_drawable(pdp); - } - if ( prp && pdp != prp ) { - dri_get_drawable(prp); + if (pcp) { + + if (pcp->driDrawablePriv != pdp + || pcp->driReadablePriv != prp) + { + /* first increment ref count for new drawables */ + + if (pdp) + { + pdp->driContextPriv = pcp; + dri_get_drawable(pdp); + } + + if (prp && prp != pdp) + { + dri_get_drawable(prp); + } + + /* free old drawables */ + + if (pcp->driReadablePriv + && pcp->driReadablePriv != pcp->driDrawablePriv) + { + dri_put_drawable(pcp->driReadablePriv); + } + + if (pcp->driDrawablePriv) + { + dri_put_drawable(pcp->driDrawablePriv); + } + + /* assign new drawables to context */ + + pcp->driDrawablePriv = pdp; + pcp->driReadablePriv = prp; + + } } - } /* ** Now that we have a context associated with this drawable, we can @@ -542,6 +579,7 @@ static void driDestroyContext(__DRIcontext *pcp) { if (pcp) { + driFreeDrawable(pcp); (*pcp->driScreenPriv->DriverAPI.DestroyContext)(pcp); _mesa_free(pcp); } @@ -579,6 +617,7 @@ driCreateNewContext(__DRIscreen *psp, const __DRIconfig *config, pcp->driScreenPriv = psp; pcp->driDrawablePriv = NULL; + pcp->driReadablePriv = NULL; /* When the first context is created for a screen, initialize a "dummy" * context. diff --git a/src/mesa/drivers/dri/r200/Makefile b/src/mesa/drivers/dri/r200/Makefile index 2114ce55c14..25ecbef6091 100644 --- a/src/mesa/drivers/dri/r200/Makefile +++ b/src/mesa/drivers/dri/r200/Makefile @@ -54,7 +54,8 @@ C_SOURCES = $(COMMON_SOURCES) $(DRIVER_SOURCES) X86_SOURCES = -DRIVER_DEFINES = -DRADEON_COMMON=1 -DRADEON_COMMON_FOR_R200 +DRIVER_DEFINES = -DRADEON_COMMON=1 -DRADEON_COMMON_FOR_R200 \ + -Wall DRI_LIB_DEPS += $(RADEON_LDFLAGS) diff --git a/src/mesa/drivers/dri/r200/r200_context.h b/src/mesa/drivers/dri/r200/r200_context.h index 6267293817d..18360890ebd 100644 --- a/src/mesa/drivers/dri/r200/r200_context.h +++ b/src/mesa/drivers/dri/r200/r200_context.h @@ -526,8 +526,6 @@ struct r200_state { struct r200_tcl_info { GLuint hw_primitive; - GLuint *Elts; - int elt_used; }; diff --git a/src/mesa/drivers/dri/r200/r200_ioctl.h b/src/mesa/drivers/dri/r200/r200_ioctl.h index 2a4b8a11f4c..f6419f5a2cc 100644 --- a/src/mesa/drivers/dri/r200/r200_ioctl.h +++ b/src/mesa/drivers/dri/r200/r200_ioctl.h @@ -125,10 +125,12 @@ static INLINE int R200_DB_STATECHANGE( * are available, you will also be adding an rmesa->state.max_state_size because * r200EmitState is called from within r200EmitVbufPrim and r200FlushElts. */ -#define AOS_BUFSZ(nr) ((3 + ((nr / 2) * 3) + ((nr & 1) * 2)) * sizeof(int)) -#define VERT_AOS_BUFSZ (5 * sizeof(int)) +#define AOS_BUFSZ(nr) ((3 + ((nr / 2) * 3) + ((nr & 1) * 2) + nr*2)) +#define VERT_AOS_BUFSZ (5) #define ELTS_BUFSZ(nr) (12 + nr * 2) -#define VBUF_BUFSZ (3 * sizeof(int)) +#define VBUF_BUFSZ (3) +#define SCISSOR_BUFSZ (8) +#define INDEX_BUFSZ (8+2) static inline uint32_t cmdpacket3(int cmd_type) { diff --git a/src/mesa/drivers/dri/r200/r200_state_init.c b/src/mesa/drivers/dri/r200/r200_state_init.c index 78ad5baebb3..9b443bd0ea3 100644 --- a/src/mesa/drivers/dri/r200/r200_state_init.c +++ b/src/mesa/drivers/dri/r200/r200_state_init.c @@ -226,57 +226,73 @@ static int cmdscl2( int offset, int stride, int count ) return h.i; } -#define CHECK( NM, FLAG ) \ +/** + * Check functions are used to check if state is active. + * If it is active check function returns maximum emit size. + */ +#define CHECK( NM, FLAG, ADD ) \ static int check_##NM( GLcontext *ctx, struct radeon_state_atom *atom) \ { \ r200ContextPtr rmesa = R200_CONTEXT(ctx); \ (void) rmesa; \ - return (FLAG) ? atom->cmd_size : 0; \ + return (FLAG) ? atom->cmd_size + (ADD) : 0; \ } -#define TCL_CHECK( NM, FLAG ) \ +#define TCL_CHECK( NM, FLAG, ADD ) \ static int check_##NM( GLcontext *ctx, struct radeon_state_atom *atom) \ { \ r200ContextPtr rmesa = R200_CONTEXT(ctx); \ - return (!rmesa->radeon.TclFallback && !ctx->VertexProgram._Enabled && (FLAG)) ? atom->cmd_size : 0; \ + return (!rmesa->radeon.TclFallback && !ctx->VertexProgram._Enabled && (FLAG)) ? atom->cmd_size + (ADD) : 0; \ } -#define TCL_OR_VP_CHECK( NM, FLAG ) \ +#define TCL_OR_VP_CHECK( NM, FLAG, ADD ) \ static int check_##NM( GLcontext *ctx, struct radeon_state_atom *atom ) \ { \ r200ContextPtr rmesa = R200_CONTEXT(ctx); \ - return (!rmesa->radeon.TclFallback && (FLAG)) ? atom->cmd_size : 0; \ + return (!rmesa->radeon.TclFallback && (FLAG)) ? atom->cmd_size + (ADD) : 0; \ } -#define VP_CHECK( NM, FLAG ) \ +#define VP_CHECK( NM, FLAG, ADD ) \ static int check_##NM( GLcontext *ctx, struct radeon_state_atom *atom ) \ { \ r200ContextPtr rmesa = R200_CONTEXT(ctx); \ (void) atom; \ - return (!rmesa->radeon.TclFallback && ctx->VertexProgram._Enabled && (FLAG)) ? atom->cmd_size : 0; \ + return (!rmesa->radeon.TclFallback && ctx->VertexProgram._Enabled && (FLAG)) ? atom->cmd_size + (ADD) : 0; \ } -CHECK( always, GL_TRUE ) -CHECK( never, GL_FALSE ) -CHECK( tex_any, ctx->Texture._EnabledUnits ) -CHECK( tf, (ctx->Texture._EnabledUnits && !ctx->ATIFragmentShader._Enabled) ); -CHECK( tex_pair, (rmesa->state.texture.unit[atom->idx].unitneeded | rmesa->state.texture.unit[atom->idx & ~1].unitneeded) ) -CHECK( tex, rmesa->state.texture.unit[atom->idx].unitneeded ) -CHECK( pix_zero, !ctx->ATIFragmentShader._Enabled ) - CHECK( texenv, (rmesa->state.envneeded & (1 << (atom->idx)) && !ctx->ATIFragmentShader._Enabled) ) -CHECK( afs_pass1, (ctx->ATIFragmentShader._Enabled && (ctx->ATIFragmentShader.Current->NumPasses > 1)) ) -CHECK( afs, ctx->ATIFragmentShader._Enabled ) -CHECK( tex_cube, rmesa->state.texture.unit[atom->idx].unitneeded & TEXTURE_CUBE_BIT ) -TCL_CHECK( tcl_fog, ctx->Fog.Enabled ) -TCL_CHECK( tcl, GL_TRUE ) -TCL_CHECK( tcl_tex, rmesa->state.texture.unit[atom->idx].unitneeded ) -TCL_CHECK( tcl_lighting, ctx->Light.Enabled ) -TCL_CHECK( tcl_light, ctx->Light.Enabled && ctx->Light.Light[atom->idx].Enabled ) -TCL_OR_VP_CHECK( tcl_ucp, (ctx->Transform.ClipPlanesEnabled & (1 << (atom->idx))) ) -TCL_OR_VP_CHECK( tcl_or_vp, GL_TRUE ) -VP_CHECK( tcl_vp, GL_TRUE ) -VP_CHECK( tcl_vp_size, ctx->VertexProgram.Current->Base.NumNativeInstructions > 64 ) -VP_CHECK( tcl_vpp_size, ctx->VertexProgram.Current->Base.NumNativeParameters > 96 ) +CHECK( always, GL_TRUE, 0 ) +CHECK( always_add4, GL_TRUE, 4 ) +CHECK( never, GL_FALSE, 0 ) +CHECK( tex_any, ctx->Texture._EnabledUnits, 0 ) +CHECK( tf, (ctx->Texture._EnabledUnits && !ctx->ATIFragmentShader._Enabled), 0 ); +CHECK( pix_zero, !ctx->ATIFragmentShader._Enabled, 0 ) + CHECK( texenv, (rmesa->state.envneeded & (1 << (atom->idx)) && !ctx->ATIFragmentShader._Enabled), 0 ) +CHECK( afs_pass1, (ctx->ATIFragmentShader._Enabled && (ctx->ATIFragmentShader.Current->NumPasses > 1)), 0 ) +CHECK( afs, ctx->ATIFragmentShader._Enabled, 0 ) +CHECK( tex_cube, rmesa->state.texture.unit[atom->idx].unitneeded & TEXTURE_CUBE_BIT, 3 + 3*5 - CUBE_STATE_SIZE ) +CHECK( tex_cube_cs, rmesa->state.texture.unit[atom->idx].unitneeded & TEXTURE_CUBE_BIT, 2 + 2*5 - CUBE_STATE_SIZE ) +TCL_CHECK( tcl_fog, ctx->Fog.Enabled, 0 ) +TCL_CHECK( tcl_fog_add4, ctx->Fog.Enabled, 4 ) +TCL_CHECK( tcl, GL_TRUE, 0 ) +TCL_CHECK( tcl_add8, GL_TRUE, 8 ) +TCL_CHECK( tcl_add4, GL_TRUE, 4 ) +TCL_CHECK( tcl_tex, rmesa->state.texture.unit[atom->idx].unitneeded, 0 ) +TCL_CHECK( tcl_lighting, ctx->Light.Enabled, 0 ) +TCL_CHECK( tcl_light, ctx->Light.Enabled && ctx->Light.Light[atom->idx].Enabled, 0 ) +TCL_CHECK( tcl_tex_add4, rmesa->state.texture.unit[atom->idx].unitneeded, 4 ) +TCL_CHECK( tcl_lighting_add4, ctx->Light.Enabled, 4 ) +TCL_CHECK( tcl_lighting_add6, ctx->Light.Enabled, 6 ) +TCL_CHECK( tcl_light_add8, ctx->Light.Enabled && ctx->Light.Light[atom->idx].Enabled, 8 ) +TCL_OR_VP_CHECK( tcl_ucp, (ctx->Transform.ClipPlanesEnabled & (1 << (atom->idx))), 0 ) +TCL_OR_VP_CHECK( tcl_ucp_add4, (ctx->Transform.ClipPlanesEnabled & (1 << (atom->idx))), 4 ) +TCL_OR_VP_CHECK( tcl_or_vp, GL_TRUE, 0 ) +TCL_OR_VP_CHECK( tcl_or_vp_add2, GL_TRUE, 2 ) +VP_CHECK( tcl_vp, GL_TRUE, 0 ) +VP_CHECK( tcl_vp_add4, GL_TRUE, 4 ) +VP_CHECK( tcl_vp_size, ctx->VertexProgram.Current->Base.NumNativeInstructions > 64, 0 ) +VP_CHECK( tcl_vpp_size, ctx->VertexProgram.Current->Base.NumNativeParameters > 96, 0 ) +VP_CHECK( tcl_vp_size_add4, ctx->VertexProgram.Current->Base.NumNativeInstructions > 64, 4 ) +VP_CHECK( tcl_vpp_size_add4, ctx->VertexProgram.Current->Base.NumNativeParameters > 96, 4 ) #define OUT_VEC(hdr, data) do { \ drm_radeon_cmd_header_t h; \ @@ -329,9 +345,8 @@ static void mtl_emit(GLcontext *ctx, struct radeon_state_atom *atom) { r200ContextPtr r200 = R200_CONTEXT(ctx); BATCH_LOCALS(&r200->radeon); - uint32_t dwords = atom->cmd_size; + uint32_t dwords = atom->check(ctx, atom); - dwords += 6; BEGIN_BATCH_NO_AUTOSTATE(dwords); OUT_VEC(atom->cmd[MTL_CMD_0], (atom->cmd+1)); OUT_SCL2(atom->cmd[MTL_CMD_1], (atom->cmd + 18)); @@ -342,9 +357,8 @@ static void lit_emit(GLcontext *ctx, struct radeon_state_atom *atom) { r200ContextPtr r200 = R200_CONTEXT(ctx); BATCH_LOCALS(&r200->radeon); - uint32_t dwords = atom->cmd_size; + uint32_t dwords = atom->check(ctx, atom); - dwords += 8; BEGIN_BATCH_NO_AUTOSTATE(dwords); OUT_VEC(atom->cmd[LIT_CMD_0], atom->cmd+1); OUT_VEC(atom->cmd[LIT_CMD_1], atom->cmd+LIT_CMD_1+1); @@ -355,9 +369,8 @@ static void ptp_emit(GLcontext *ctx, struct radeon_state_atom *atom) { r200ContextPtr r200 = R200_CONTEXT(ctx); BATCH_LOCALS(&r200->radeon); - uint32_t dwords = atom->cmd_size; + uint32_t dwords = atom->check(ctx, atom); - dwords += 8; BEGIN_BATCH_NO_AUTOSTATE(dwords); OUT_VEC(atom->cmd[PTP_CMD_0], atom->cmd+1); OUT_VEC(atom->cmd[PTP_CMD_1], atom->cmd+PTP_CMD_1+1); @@ -368,9 +381,8 @@ static void veclinear_emit(GLcontext *ctx, struct radeon_state_atom *atom) { r200ContextPtr r200 = R200_CONTEXT(ctx); BATCH_LOCALS(&r200->radeon); - uint32_t dwords = atom->cmd_size; + uint32_t dwords = atom->check(ctx, atom); - dwords += 4; OUT_VECLINEAR(atom->cmd[0], atom->cmd+1); } @@ -378,9 +390,8 @@ static void scl_emit(GLcontext *ctx, struct radeon_state_atom *atom) { r200ContextPtr r200 = R200_CONTEXT(ctx); BATCH_LOCALS(&r200->radeon); - uint32_t dwords = atom->cmd_size; + uint32_t dwords = atom->check(ctx, atom); - dwords += 2; BEGIN_BATCH_NO_AUTOSTATE(dwords); OUT_SCL(atom->cmd[0], atom->cmd+1); END_BATCH(); @@ -391,9 +402,8 @@ static void vec_emit(GLcontext *ctx, struct radeon_state_atom *atom) { r200ContextPtr r200 = R200_CONTEXT(ctx); BATCH_LOCALS(&r200->radeon); - uint32_t dwords = atom->cmd_size; + uint32_t dwords = atom->check(ctx, atom); - dwords += 4; BEGIN_BATCH_NO_AUTOSTATE(dwords); OUT_VEC(atom->cmd[0], atom->cmd+1); END_BATCH(); @@ -406,10 +416,10 @@ static void ctx_emit(GLcontext *ctx, struct radeon_state_atom *atom) struct radeon_renderbuffer *rrb; uint32_t cbpitch; uint32_t zbpitch, depth_fmt; - uint32_t dwords = atom->cmd_size; + uint32_t dwords = atom->check(ctx, atom); /* output the first 7 bytes of context */ - BEGIN_BATCH_NO_AUTOSTATE(dwords+2+2); + BEGIN_BATCH_NO_AUTOSTATE(dwords); OUT_BATCH_TABLE(atom->cmd, 5); rrb = radeon_get_depthbuffer(&r200->radeon); @@ -466,6 +476,31 @@ static void ctx_emit(GLcontext *ctx, struct radeon_state_atom *atom) END_BATCH(); } +static int check_always_ctx( GLcontext *ctx, struct radeon_state_atom *atom) +{ + r200ContextPtr r200 = R200_CONTEXT(ctx); + struct radeon_renderbuffer *rrb, *drb; + uint32_t dwords; + + rrb = radeon_get_colorbuffer(&r200->radeon); + if (!rrb || !rrb->bo) { + return 0; + } + + drb = radeon_get_depthbuffer(&r200->radeon); + + dwords = 10; + if (drb) + dwords += 6; + if (rrb) + dwords += 8; + if (atom->cmd_size == CTX_STATE_SIZE_NEWDRM) + dwords += 4; + + + return dwords; +} + static void ctx_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom) { r200ContextPtr r200 = R200_CONTEXT(ctx); @@ -473,7 +508,7 @@ static void ctx_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom) struct radeon_renderbuffer *rrb, *drb; uint32_t cbpitch = 0; uint32_t zbpitch = 0; - uint32_t dwords = atom->cmd_size; + uint32_t dwords = atom->check(ctx, atom); uint32_t depth_fmt; rrb = radeon_get_colorbuffer(&r200->radeon); @@ -511,14 +546,6 @@ static void ctx_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom) atom->cmd[CTX_RB3D_ZSTENCILCNTL] |= depth_fmt; } - dwords = 10; - if (drb) - dwords += 6; - if (rrb) - dwords += 8; - if (atom->cmd_size == CTX_STATE_SIZE_NEWDRM) - dwords += 4; - /* output the first 7 bytes of context */ BEGIN_BATCH_NO_AUTOSTATE(dwords); @@ -556,16 +583,46 @@ static void ctx_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom) END_BATCH(); } +static int get_tex_size(GLcontext* ctx, struct radeon_state_atom *atom) +{ + r200ContextPtr r200 = R200_CONTEXT(ctx); + uint32_t dwords = atom->cmd_size + 2; + int i = atom->idx; + radeonTexObj *t = r200->state.texture.unit[i].texobj; + if (!(t && t->mt && !t->image_override)) + dwords -= 2; + + return dwords; +} + +static int check_tex_pair(GLcontext* ctx, struct radeon_state_atom *atom) +{ + r200ContextPtr r200 = R200_CONTEXT(ctx); + /** XOR is bit flip operation so use it for finding pair */ + if (!(r200->state.texture.unit[atom->idx].unitneeded | r200->state.texture.unit[atom->idx ^ 1].unitneeded)) + return 0; + + return get_tex_size(ctx, atom); +} + +static int check_tex(GLcontext* ctx, struct radeon_state_atom *atom) +{ + r200ContextPtr r200 = R200_CONTEXT(ctx); + if (!(r200->state.texture.unit[atom->idx].unitneeded)) + return 0; + + return get_tex_size(ctx, atom); +} + + static void tex_emit(GLcontext *ctx, struct radeon_state_atom *atom) { r200ContextPtr r200 = R200_CONTEXT(ctx); BATCH_LOCALS(&r200->radeon); - uint32_t dwords = atom->cmd_size; + uint32_t dwords = atom->check(ctx, atom); int i = atom->idx; radeonTexObj *t = r200->state.texture.unit[i].texobj; - if (t && t->mt && !t->image_override) - dwords += 2; BEGIN_BATCH_NO_AUTOSTATE(dwords); /* is this ok even with drm older than 1.18? */ OUT_BATCH_TABLE(atom->cmd, 10); @@ -583,17 +640,13 @@ static void tex_emit(GLcontext *ctx, struct radeon_state_atom *atom) END_BATCH(); } -static void tex_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom) +static int get_tex_mm_size(GLcontext* ctx, struct radeon_state_atom *atom) { r200ContextPtr r200 = R200_CONTEXT(ctx); - BATCH_LOCALS(&r200->radeon); - uint32_t dwords = atom->cmd_size; + uint32_t dwords = atom->cmd_size + 2; + int hastexture = 1; int i = atom->idx; radeonTexObj *t = r200->state.texture.unit[i].texobj; - int hastexture = 1; - - if (!r200->state.texture.unit[i].unitneeded) - hastexture = 0; if (!t) hastexture = 0; else { @@ -601,16 +654,46 @@ static void tex_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom) hastexture = 0; } - if (hastexture) - dwords += 2; - else - dwords -= 2; + if (!hastexture) + dwords -= 4; + return dwords; +} + +static int check_tex_pair_mm(GLcontext* ctx, struct radeon_state_atom *atom) +{ + r200ContextPtr r200 = R200_CONTEXT(ctx); + /** XOR is bit flip operation so use it for finding pair */ + if (!(r200->state.texture.unit[atom->idx].unitneeded | r200->state.texture.unit[atom->idx ^ 1].unitneeded)) + return 0; + + return get_tex_mm_size(ctx, atom); +} + +static int check_tex_mm(GLcontext* ctx, struct radeon_state_atom *atom) +{ + r200ContextPtr r200 = R200_CONTEXT(ctx); + if (!(r200->state.texture.unit[atom->idx].unitneeded)) + return 0; + + return get_tex_mm_size(ctx, atom); +} + + +static void tex_emit_mm(GLcontext *ctx, struct radeon_state_atom *atom) +{ + r200ContextPtr r200 = R200_CONTEXT(ctx); + BATCH_LOCALS(&r200->radeon); + uint32_t dwords = atom->check(ctx, atom); + int i = atom->idx; + radeonTexObj *t = r200->state.texture.unit[i].texobj; + if (!r200->state.texture.unit[i].unitneeded) + dwords -= 4; BEGIN_BATCH_NO_AUTOSTATE(dwords); OUT_BATCH(CP_PACKET0(R200_PP_TXFILTER_0 + (32 * i), 7)); OUT_BATCH_TABLE((atom->cmd + 1), 8); - if (hastexture) { + if (dwords > atom->cmd_size) { OUT_BATCH(CP_PACKET0(R200_PP_TXOFFSET_0 + (24 * i), 0)); if (t->mt && !t->image_override) { OUT_BATCH_RELOC(t->tile_bits, t->mt->bo, 0, @@ -629,12 +712,15 @@ static void cube_emit(GLcontext *ctx, struct radeon_state_atom *atom) { r200ContextPtr r200 = R200_CONTEXT(ctx); BATCH_LOCALS(&r200->radeon); - uint32_t dwords = 3; + uint32_t dwords = atom->check(ctx, atom); int i = atom->idx, j; radeonTexObj *t = r200->state.texture.unit[i].texobj; radeon_mipmap_level *lvl; - BEGIN_BATCH_NO_AUTOSTATE(dwords + (3 * 5)); + if (!(t && !t->image_override)) + dwords = 2; + + BEGIN_BATCH_NO_AUTOSTATE(dwords); /* XXX that size won't really match with image_override... */ OUT_BATCH_TABLE(atom->cmd, 2); @@ -653,12 +739,14 @@ static void cube_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom) { r200ContextPtr r200 = R200_CONTEXT(ctx); BATCH_LOCALS(&r200->radeon); - uint32_t dwords = 2; + uint32_t dwords = atom->check(ctx, atom); int i = atom->idx, j; radeonTexObj *t = r200->state.texture.unit[i].texobj; radeon_mipmap_level *lvl; + if (!(t && !t->image_override)) + dwords = 2; - BEGIN_BATCH_NO_AUTOSTATE(dwords + (4 * 5)); + BEGIN_BATCH_NO_AUTOSTATE(dwords); OUT_BATCH_TABLE(atom->cmd, 2); if (t && !t->image_override) { @@ -713,14 +801,19 @@ void r200InitState( r200ContextPtr rmesa ) /* Allocate state buffers: */ if (rmesa->radeon.radeonScreen->drmSupportsBlendColor) - ALLOC_STATE( ctx, always, CTX_STATE_SIZE_NEWDRM, "CTX/context", 0 ); + ALLOC_STATE( ctx, always_add4, CTX_STATE_SIZE_NEWDRM, "CTX/context", 0 ); else - ALLOC_STATE( ctx, always, CTX_STATE_SIZE_OLDDRM, "CTX/context", 0 ); + ALLOC_STATE( ctx, always_add4, CTX_STATE_SIZE_OLDDRM, "CTX/context", 0 ); if (rmesa->radeon.radeonScreen->kernel_mm) + { rmesa->hw.ctx.emit = ctx_emit_cs; + rmesa->hw.ctx.check = check_always_ctx; + } else + { rmesa->hw.ctx.emit = ctx_emit; + } ALLOC_STATE( set, always, SET_STATE_SIZE, "SET/setup", 0 ); ALLOC_STATE( lin, always, LIN_STATE_SIZE, "LIN/line", 0 ); ALLOC_STATE( msk, always, MSK_STATE_SIZE, "MSK/mask", 0 ); @@ -732,49 +825,56 @@ void r200InitState( r200ContextPtr rmesa ) ALLOC_STATE( cst, always, CST_STATE_SIZE, "CST/constant", 0 ); ALLOC_STATE( zbs, always, ZBS_STATE_SIZE, "ZBS/zbias", 0 ); ALLOC_STATE( tf, tf, TF_STATE_SIZE, "TF/tfactor", 0 ); - if (rmesa->radeon.radeonScreen->drmSupportsFragShader) { - if (rmesa->radeon.radeonScreen->chip_family == CHIP_FAMILY_R200) { - /* make sure texture units 0/1 are emitted pair-wise for r200 t0 hang workaround */ - ALLOC_STATE( tex[0], tex_pair, TEX_STATE_SIZE_NEWDRM, "TEX/tex-0", 0 ); - ALLOC_STATE( tex[1], tex_pair, TEX_STATE_SIZE_NEWDRM, "TEX/tex-1", 1 ); - ALLOC_STATE( tam, tex_any, TAM_STATE_SIZE, "TAM/tam", 0 ); - } - else { - ALLOC_STATE( tex[0], tex, TEX_STATE_SIZE_NEWDRM, "TEX/tex-0", 0 ); - ALLOC_STATE( tex[1], tex, TEX_STATE_SIZE_NEWDRM, "TEX/tex-1", 1 ); - ALLOC_STATE( tam, never, TAM_STATE_SIZE, "TAM/tam", 0 ); + { + int state_size = TEX_STATE_SIZE_NEWDRM; + if (!rmesa->radeon.radeonScreen->drmSupportsFragShader) { + state_size = TEX_STATE_SIZE_OLDDRM; } - ALLOC_STATE( tex[2], tex, TEX_STATE_SIZE_NEWDRM, "TEX/tex-2", 2 ); - ALLOC_STATE( tex[3], tex, TEX_STATE_SIZE_NEWDRM, "TEX/tex-3", 3 ); - ALLOC_STATE( tex[4], tex, TEX_STATE_SIZE_NEWDRM, "TEX/tex-4", 4 ); - ALLOC_STATE( tex[5], tex, TEX_STATE_SIZE_NEWDRM, "TEX/tex-5", 5 ); - ALLOC_STATE( atf, afs, ATF_STATE_SIZE, "ATF/tfactor", 0 ); - ALLOC_STATE( afs[0], afs_pass1, AFS_STATE_SIZE, "AFS/afsinst-0", 0 ); - ALLOC_STATE( afs[1], afs, AFS_STATE_SIZE, "AFS/afsinst-1", 1 ); - } - else { - if (rmesa->radeon.radeonScreen->chip_family == CHIP_FAMILY_R200) { - ALLOC_STATE( tex[0], tex_pair, TEX_STATE_SIZE_OLDDRM, "TEX/tex-0", 0 ); - ALLOC_STATE( tex[1], tex_pair, TEX_STATE_SIZE_OLDDRM, "TEX/tex-1", 1 ); - ALLOC_STATE( tam, tex_any, TAM_STATE_SIZE, "TAM/tam", 0 ); - } - else { - ALLOC_STATE( tex[0], tex, TEX_STATE_SIZE_OLDDRM, "TEX/tex-0", 0 ); - ALLOC_STATE( tex[1], tex, TEX_STATE_SIZE_OLDDRM, "TEX/tex-1", 1 ); - ALLOC_STATE( tam, never, TAM_STATE_SIZE, "TAM/tam", 0 ); + if (rmesa->radeon.radeonScreen->drmSupportsFragShader) { + if (rmesa->radeon.radeonScreen->chip_family == CHIP_FAMILY_R200) { + /* make sure texture units 0/1 are emitted pair-wise for r200 t0 hang workaround */ + ALLOC_STATE( tex[0], tex_pair_mm, state_size, "TEX/tex-0", 0 ); + ALLOC_STATE( tex[1], tex_pair_mm, state_size, "TEX/tex-1", 1 ); + ALLOC_STATE( tam, tex_any, TAM_STATE_SIZE, "TAM/tam", 0 ); + } + else { + ALLOC_STATE( tex[0], tex_mm, state_size, "TEX/tex-0", 0 ); + ALLOC_STATE( tex[1], tex_mm, state_size, "TEX/tex-1", 1 ); + ALLOC_STATE( tam, never, TAM_STATE_SIZE, "TAM/tam", 0 ); + } + ALLOC_STATE( tex[2], tex_mm, state_size, "TEX/tex-2", 2 ); + ALLOC_STATE( tex[3], tex_mm, state_size, "TEX/tex-3", 3 ); + ALLOC_STATE( tex[4], tex_mm, state_size, "TEX/tex-4", 4 ); + ALLOC_STATE( tex[5], tex_mm, state_size, "TEX/tex-5", 5 ); + if (!rmesa->radeon.radeonScreen->kernel_mm) + { + if (rmesa->radeon.radeonScreen->chip_family == CHIP_FAMILY_R200) { + rmesa->hw.tex[0].check = check_tex_pair; + rmesa->hw.tex[1].check = check_tex_pair; + } else { + rmesa->hw.tex[0].check = check_tex; + rmesa->hw.tex[1].check = check_tex; + } + rmesa->hw.tex[2].check = check_tex; + rmesa->hw.tex[3].check = check_tex; + rmesa->hw.tex[4].check = check_tex; + rmesa->hw.tex[5].check = check_tex; + } + if (rmesa->radeon.radeonScreen->drmSupportsFragShader) { + ALLOC_STATE( atf, afs, ATF_STATE_SIZE, "ATF/tfactor", 0 ); + ALLOC_STATE( afs[0], afs_pass1, AFS_STATE_SIZE, "AFS/afsinst-0", 0 ); + ALLOC_STATE( afs[1], afs, AFS_STATE_SIZE, "AFS/afsinst-1", 1 ); + } else { + ALLOC_STATE( atf, never, ATF_STATE_SIZE, "ATF/tfactor", 0 ); + ALLOC_STATE( afs[0], never, AFS_STATE_SIZE, "AFS/afsinst-0", 0 ); + ALLOC_STATE( afs[1], never, AFS_STATE_SIZE, "AFS/afsinst-1", 1 ); + } } - ALLOC_STATE( tex[2], tex, TEX_STATE_SIZE_OLDDRM, "TEX/tex-2", 2 ); - ALLOC_STATE( tex[3], tex, TEX_STATE_SIZE_OLDDRM, "TEX/tex-3", 3 ); - ALLOC_STATE( tex[4], tex, TEX_STATE_SIZE_OLDDRM, "TEX/tex-4", 4 ); - ALLOC_STATE( tex[5], tex, TEX_STATE_SIZE_OLDDRM, "TEX/tex-5", 5 ); - ALLOC_STATE( atf, never, ATF_STATE_SIZE, "TF/tfactor", 0 ); - ALLOC_STATE( afs[0], never, AFS_STATE_SIZE, "AFS/afsinst-0", 0 ); - ALLOC_STATE( afs[1], never, AFS_STATE_SIZE, "AFS/afsinst-1", 1 ); } for (i = 0; i < 6; i++) if (rmesa->radeon.radeonScreen->kernel_mm) - rmesa->hw.tex[i].emit = tex_emit_cs; + rmesa->hw.tex[i].emit = tex_emit_mm; else rmesa->hw.tex[i].emit = tex_emit; if (rmesa->radeon.radeonScreen->drmSupportsCubeMapsR200) { @@ -785,9 +885,10 @@ void r200InitState( r200ContextPtr rmesa ) ALLOC_STATE( cube[4], tex_cube, CUBE_STATE_SIZE, "CUBE/tex-4", 4 ); ALLOC_STATE( cube[5], tex_cube, CUBE_STATE_SIZE, "CUBE/tex-5", 5 ); for (i = 0; i < 6; i++) - if (rmesa->radeon.radeonScreen->kernel_mm) + if (rmesa->radeon.radeonScreen->kernel_mm) { rmesa->hw.cube[i].emit = cube_emit_cs; - else + rmesa->hw.cube[i].check = check_tex_cube_cs; + } else rmesa->hw.cube[i].emit = cube_emit; } else { @@ -801,10 +902,17 @@ void r200InitState( r200ContextPtr rmesa ) if (rmesa->radeon.radeonScreen->drmSupportsVertexProgram) { ALLOC_STATE( pvs, tcl_vp, PVS_STATE_SIZE, "PVS/pvscntl", 0 ); - ALLOC_STATE( vpi[0], tcl_vp, VPI_STATE_SIZE, "VP/vertexprog-0", 0 ); - ALLOC_STATE( vpi[1], tcl_vp_size, VPI_STATE_SIZE, "VP/vertexprog-1", 1 ); - ALLOC_STATE( vpp[0], tcl_vp, VPP_STATE_SIZE, "VPP/vertexparam-0", 0 ); - ALLOC_STATE( vpp[1], tcl_vpp_size, VPP_STATE_SIZE, "VPP/vertexparam-1", 1 ); + if (rmesa->radeon.radeonScreen->kernel_mm) { + ALLOC_STATE( vpi[0], tcl_vp_add4, VPI_STATE_SIZE, "VP/vertexprog-0", 0 ); + ALLOC_STATE( vpi[1], tcl_vp_size_add4, VPI_STATE_SIZE, "VP/vertexprog-1", 1 ); + ALLOC_STATE( vpp[0], tcl_vp_add4, VPP_STATE_SIZE, "VPP/vertexparam-0", 0 ); + ALLOC_STATE( vpp[1], tcl_vpp_size_add4, VPP_STATE_SIZE, "VPP/vertexparam-1", 1 ); + } else { + ALLOC_STATE( vpi[0], tcl_vp, VPI_STATE_SIZE, "VP/vertexprog-0", 0 ); + ALLOC_STATE( vpi[1], tcl_vp_size, VPI_STATE_SIZE, "VP/vertexprog-1", 1 ); + ALLOC_STATE( vpp[0], tcl_vp, VPP_STATE_SIZE, "VPP/vertexparam-0", 0 ); + ALLOC_STATE( vpp[1], tcl_vpp_size, VPP_STATE_SIZE, "VPP/vertexparam-1", 1 ); + } } else { ALLOC_STATE( pvs, never, PVS_STATE_SIZE, "PVS/pvscntl", 0 ); @@ -817,35 +925,67 @@ void r200InitState( r200ContextPtr rmesa ) ALLOC_STATE( tcl, tcl_or_vp, TCL_STATE_SIZE, "TCL/tcl", 0 ); ALLOC_STATE( msl, tcl, MSL_STATE_SIZE, "MSL/matrix-select", 0 ); ALLOC_STATE( tcg, tcl, TCG_STATE_SIZE, "TCG/texcoordgen", 0 ); - ALLOC_STATE( mtl[0], tcl_lighting, MTL_STATE_SIZE, "MTL0/material0", 0 ); - ALLOC_STATE( mtl[1], tcl_lighting, MTL_STATE_SIZE, "MTL1/material1", 1 ); - ALLOC_STATE( grd, tcl_or_vp, GRD_STATE_SIZE, "GRD/guard-band", 0 ); - ALLOC_STATE( fog, tcl_fog, FOG_STATE_SIZE, "FOG/fog", 0 ); - ALLOC_STATE( glt, tcl_lighting, GLT_STATE_SIZE, "GLT/light-global", 0 ); - ALLOC_STATE( eye, tcl_lighting, EYE_STATE_SIZE, "EYE/eye-vector", 0 ); - ALLOC_STATE( mat[R200_MTX_MV], tcl, MAT_STATE_SIZE, "MAT/modelview", 0 ); - ALLOC_STATE( mat[R200_MTX_IMV], tcl, MAT_STATE_SIZE, "MAT/it-modelview", 0 ); - ALLOC_STATE( mat[R200_MTX_MVP], tcl, MAT_STATE_SIZE, "MAT/modelproject", 0 ); - ALLOC_STATE( mat[R200_MTX_TEX0], tcl_tex, MAT_STATE_SIZE, "MAT/texmat0", 0 ); - ALLOC_STATE( mat[R200_MTX_TEX1], tcl_tex, MAT_STATE_SIZE, "MAT/texmat1", 1 ); - ALLOC_STATE( mat[R200_MTX_TEX2], tcl_tex, MAT_STATE_SIZE, "MAT/texmat2", 2 ); - ALLOC_STATE( mat[R200_MTX_TEX3], tcl_tex, MAT_STATE_SIZE, "MAT/texmat3", 3 ); - ALLOC_STATE( mat[R200_MTX_TEX4], tcl_tex, MAT_STATE_SIZE, "MAT/texmat4", 4 ); - ALLOC_STATE( mat[R200_MTX_TEX5], tcl_tex, MAT_STATE_SIZE, "MAT/texmat5", 5 ); - ALLOC_STATE( ucp[0], tcl_ucp, UCP_STATE_SIZE, "UCP/userclip-0", 0 ); - ALLOC_STATE( ucp[1], tcl_ucp, UCP_STATE_SIZE, "UCP/userclip-1", 1 ); - ALLOC_STATE( ucp[2], tcl_ucp, UCP_STATE_SIZE, "UCP/userclip-2", 2 ); - ALLOC_STATE( ucp[3], tcl_ucp, UCP_STATE_SIZE, "UCP/userclip-3", 3 ); - ALLOC_STATE( ucp[4], tcl_ucp, UCP_STATE_SIZE, "UCP/userclip-4", 4 ); - ALLOC_STATE( ucp[5], tcl_ucp, UCP_STATE_SIZE, "UCP/userclip-5", 5 ); - ALLOC_STATE( lit[0], tcl_light, LIT_STATE_SIZE, "LIT/light-0", 0 ); - ALLOC_STATE( lit[1], tcl_light, LIT_STATE_SIZE, "LIT/light-1", 1 ); - ALLOC_STATE( lit[2], tcl_light, LIT_STATE_SIZE, "LIT/light-2", 2 ); - ALLOC_STATE( lit[3], tcl_light, LIT_STATE_SIZE, "LIT/light-3", 3 ); - ALLOC_STATE( lit[4], tcl_light, LIT_STATE_SIZE, "LIT/light-4", 4 ); - ALLOC_STATE( lit[5], tcl_light, LIT_STATE_SIZE, "LIT/light-5", 5 ); - ALLOC_STATE( lit[6], tcl_light, LIT_STATE_SIZE, "LIT/light-6", 6 ); - ALLOC_STATE( lit[7], tcl_light, LIT_STATE_SIZE, "LIT/light-7", 7 ); + if (rmesa->radeon.radeonScreen->kernel_mm) { + ALLOC_STATE( mtl[0], tcl_lighting_add6, MTL_STATE_SIZE, "MTL0/material0", 0 ); + ALLOC_STATE( mtl[1], tcl_lighting_add6, MTL_STATE_SIZE, "MTL1/material1", 1 ); + ALLOC_STATE( grd, tcl_or_vp_add2, GRD_STATE_SIZE, "GRD/guard-band", 0 ); + ALLOC_STATE( fog, tcl_fog_add4, FOG_STATE_SIZE, "FOG/fog", 0 ); + ALLOC_STATE( glt, tcl_lighting_add4, GLT_STATE_SIZE, "GLT/light-global", 0 ); + ALLOC_STATE( eye, tcl_lighting_add4, EYE_STATE_SIZE, "EYE/eye-vector", 0 ); + ALLOC_STATE( mat[R200_MTX_MV], tcl_add4, MAT_STATE_SIZE, "MAT/modelview", 0 ); + ALLOC_STATE( mat[R200_MTX_IMV], tcl_add4, MAT_STATE_SIZE, "MAT/it-modelview", 0 ); + ALLOC_STATE( mat[R200_MTX_MVP], tcl_add4, MAT_STATE_SIZE, "MAT/modelproject", 0 ); + ALLOC_STATE( mat[R200_MTX_TEX0], tcl_tex_add4, MAT_STATE_SIZE, "MAT/texmat0", 0 ); + ALLOC_STATE( mat[R200_MTX_TEX1], tcl_tex_add4, MAT_STATE_SIZE, "MAT/texmat1", 1 ); + ALLOC_STATE( mat[R200_MTX_TEX2], tcl_tex_add4, MAT_STATE_SIZE, "MAT/texmat2", 2 ); + ALLOC_STATE( mat[R200_MTX_TEX3], tcl_tex_add4, MAT_STATE_SIZE, "MAT/texmat3", 3 ); + ALLOC_STATE( mat[R200_MTX_TEX4], tcl_tex_add4, MAT_STATE_SIZE, "MAT/texmat4", 4 ); + ALLOC_STATE( mat[R200_MTX_TEX5], tcl_tex_add4, MAT_STATE_SIZE, "MAT/texmat5", 5 ); + ALLOC_STATE( ucp[0], tcl_ucp_add4, UCP_STATE_SIZE, "UCP/userclip-0", 0 ); + ALLOC_STATE( ucp[1], tcl_ucp_add4, UCP_STATE_SIZE, "UCP/userclip-1", 1 ); + ALLOC_STATE( ucp[2], tcl_ucp_add4, UCP_STATE_SIZE, "UCP/userclip-2", 2 ); + ALLOC_STATE( ucp[3], tcl_ucp_add4, UCP_STATE_SIZE, "UCP/userclip-3", 3 ); + ALLOC_STATE( ucp[4], tcl_ucp_add4, UCP_STATE_SIZE, "UCP/userclip-4", 4 ); + ALLOC_STATE( ucp[5], tcl_ucp_add4, UCP_STATE_SIZE, "UCP/userclip-5", 5 ); + ALLOC_STATE( lit[0], tcl_light_add8, LIT_STATE_SIZE, "LIT/light-0", 0 ); + ALLOC_STATE( lit[1], tcl_light_add8, LIT_STATE_SIZE, "LIT/light-1", 1 ); + ALLOC_STATE( lit[2], tcl_light_add8, LIT_STATE_SIZE, "LIT/light-2", 2 ); + ALLOC_STATE( lit[3], tcl_light_add8, LIT_STATE_SIZE, "LIT/light-3", 3 ); + ALLOC_STATE( lit[4], tcl_light_add8, LIT_STATE_SIZE, "LIT/light-4", 4 ); + ALLOC_STATE( lit[5], tcl_light_add8, LIT_STATE_SIZE, "LIT/light-5", 5 ); + ALLOC_STATE( lit[6], tcl_light_add8, LIT_STATE_SIZE, "LIT/light-6", 6 ); + ALLOC_STATE( lit[7], tcl_light_add8, LIT_STATE_SIZE, "LIT/light-7", 7 ); + } else { + ALLOC_STATE( mtl[0], tcl_lighting, MTL_STATE_SIZE, "MTL0/material0", 0 ); + ALLOC_STATE( mtl[1], tcl_lighting, MTL_STATE_SIZE, "MTL1/material1", 1 ); + ALLOC_STATE( grd, tcl_or_vp, GRD_STATE_SIZE, "GRD/guard-band", 0 ); + ALLOC_STATE( fog, tcl_fog, FOG_STATE_SIZE, "FOG/fog", 0 ); + ALLOC_STATE( glt, tcl_lighting, GLT_STATE_SIZE, "GLT/light-global", 0 ); + ALLOC_STATE( eye, tcl_lighting, EYE_STATE_SIZE, "EYE/eye-vector", 0 ); + ALLOC_STATE( mat[R200_MTX_MV], tcl, MAT_STATE_SIZE, "MAT/modelview", 0 ); + ALLOC_STATE( mat[R200_MTX_IMV], tcl, MAT_STATE_SIZE, "MAT/it-modelview", 0 ); + ALLOC_STATE( mat[R200_MTX_MVP], tcl, MAT_STATE_SIZE, "MAT/modelproject", 0 ); + ALLOC_STATE( mat[R200_MTX_TEX0], tcl_tex, MAT_STATE_SIZE, "MAT/texmat0", 0 ); + ALLOC_STATE( mat[R200_MTX_TEX1], tcl_tex, MAT_STATE_SIZE, "MAT/texmat1", 1 ); + ALLOC_STATE( mat[R200_MTX_TEX2], tcl_tex, MAT_STATE_SIZE, "MAT/texmat2", 2 ); + ALLOC_STATE( mat[R200_MTX_TEX3], tcl_tex, MAT_STATE_SIZE, "MAT/texmat3", 3 ); + ALLOC_STATE( mat[R200_MTX_TEX4], tcl_tex, MAT_STATE_SIZE, "MAT/texmat4", 4 ); + ALLOC_STATE( mat[R200_MTX_TEX5], tcl_tex, MAT_STATE_SIZE, "MAT/texmat5", 5 ); + ALLOC_STATE( ucp[0], tcl_ucp, UCP_STATE_SIZE, "UCP/userclip-0", 0 ); + ALLOC_STATE( ucp[1], tcl_ucp, UCP_STATE_SIZE, "UCP/userclip-1", 1 ); + ALLOC_STATE( ucp[2], tcl_ucp, UCP_STATE_SIZE, "UCP/userclip-2", 2 ); + ALLOC_STATE( ucp[3], tcl_ucp, UCP_STATE_SIZE, "UCP/userclip-3", 3 ); + ALLOC_STATE( ucp[4], tcl_ucp, UCP_STATE_SIZE, "UCP/userclip-4", 4 ); + ALLOC_STATE( ucp[5], tcl_ucp, UCP_STATE_SIZE, "UCP/userclip-5", 5 ); + ALLOC_STATE( lit[0], tcl_light, LIT_STATE_SIZE, "LIT/light-0", 0 ); + ALLOC_STATE( lit[1], tcl_light, LIT_STATE_SIZE, "LIT/light-1", 1 ); + ALLOC_STATE( lit[2], tcl_light, LIT_STATE_SIZE, "LIT/light-2", 2 ); + ALLOC_STATE( lit[3], tcl_light, LIT_STATE_SIZE, "LIT/light-3", 3 ); + ALLOC_STATE( lit[4], tcl_light, LIT_STATE_SIZE, "LIT/light-4", 4 ); + ALLOC_STATE( lit[5], tcl_light, LIT_STATE_SIZE, "LIT/light-5", 5 ); + ALLOC_STATE( lit[6], tcl_light, LIT_STATE_SIZE, "LIT/light-6", 6 ); + ALLOC_STATE( lit[7], tcl_light, LIT_STATE_SIZE, "LIT/light-7", 7 ); + } ALLOC_STATE( pix[0], pix_zero, PIX_STATE_SIZE, "PIX/pixstage-0", 0 ); ALLOC_STATE( pix[1], texenv, PIX_STATE_SIZE, "PIX/pixstage-1", 1 ); ALLOC_STATE( pix[2], texenv, PIX_STATE_SIZE, "PIX/pixstage-2", 2 ); @@ -860,7 +1000,10 @@ void r200InitState( r200ContextPtr rmesa ) } if (rmesa->radeon.radeonScreen->drmSupportsPointSprites) { ALLOC_STATE( spr, always, SPR_STATE_SIZE, "SPR/pointsprite", 0 ); - ALLOC_STATE( ptp, tcl, PTP_STATE_SIZE, "PTP/pointparams", 0 ); + if (rmesa->radeon.radeonScreen->kernel_mm) + ALLOC_STATE( ptp, tcl_add8, PTP_STATE_SIZE, "PTP/pointparams", 0 ); + else + ALLOC_STATE( ptp, tcl, PTP_STATE_SIZE, "PTP/pointparams", 0 ); } else { ALLOC_STATE (spr, never, SPR_STATE_SIZE, "SPR/pointsprite", 0 ); @@ -953,18 +1096,18 @@ void r200InitState( r200ContextPtr rmesa ) rmesa->hw.prf.cmd[PRF_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_TRI_PERF_CNTL); rmesa->hw.spr.cmd[SPR_CMD_0] = cmdpkt(rmesa, R200_EMIT_TCL_POINT_SPRITE_CNTL); if (rmesa->radeon.radeonScreen->kernel_mm) { - rmesa->hw.mtl[0].emit = mtl_emit; - rmesa->hw.mtl[1].emit = mtl_emit; - - rmesa->hw.vpi[0].emit = veclinear_emit; - rmesa->hw.vpi[1].emit = veclinear_emit; - rmesa->hw.vpp[0].emit = veclinear_emit; - rmesa->hw.vpp[1].emit = veclinear_emit; - - rmesa->hw.grd.emit = scl_emit; - rmesa->hw.fog.emit = vec_emit; - rmesa->hw.glt.emit = vec_emit; - rmesa->hw.eye.emit = vec_emit; + rmesa->hw.mtl[0].emit = mtl_emit; + rmesa->hw.mtl[1].emit = mtl_emit; + + rmesa->hw.vpi[0].emit = veclinear_emit; + rmesa->hw.vpi[1].emit = veclinear_emit; + rmesa->hw.vpp[0].emit = veclinear_emit; + rmesa->hw.vpp[1].emit = veclinear_emit; + + rmesa->hw.grd.emit = scl_emit; + rmesa->hw.fog.emit = vec_emit; + rmesa->hw.glt.emit = vec_emit; + rmesa->hw.eye.emit = vec_emit; for (i = R200_MTX_MV; i <= R200_MTX_TEX5; i++) rmesa->hw.mat[i].emit = vec_emit; diff --git a/src/mesa/drivers/dri/r200/r200_swtcl.c b/src/mesa/drivers/dri/r200/r200_swtcl.c index 1b238911408..56930c7863a 100644 --- a/src/mesa/drivers/dri/r200/r200_swtcl.c +++ b/src/mesa/drivers/dri/r200/r200_swtcl.c @@ -269,7 +269,7 @@ void r200_swtcl_flush(GLcontext *ctx, uint32_t current_offset) { r200ContextPtr rmesa = R200_CONTEXT(ctx); rcommonEnsureCmdBufSpace(&rmesa->radeon, - rmesa->radeon.hw.max_state_size + (12*sizeof(int)), + radeonCountStateEmitSize(&rmesa->radeon) + (12*sizeof(int)), __FUNCTION__); diff --git a/src/mesa/drivers/dri/r200/r200_tcl.c b/src/mesa/drivers/dri/r200/r200_tcl.c index ca9a8dbf8c2..7b3a9fc859d 100644 --- a/src/mesa/drivers/dri/r200/r200_tcl.c +++ b/src/mesa/drivers/dri/r200/r200_tcl.c @@ -51,6 +51,8 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "r200_swtcl.h" #include "r200_maos.h" +#include "radeon_common_context.h" + #define HAVE_POINTS 1 @@ -109,7 +111,7 @@ static GLboolean discrete_prim[0x10] = { #define ELT_INIT(prim, hw_prim) \ r200TclPrimitive( ctx, prim, hw_prim | R200_VF_PRIM_WALK_IND ) -#define GET_MESA_ELTS() rmesa->tcl.Elts +#define GET_MESA_ELTS() TNL_CONTEXT(ctx)->vb.Elts /* Don't really know how many elts will fit in what's left of cmdbuf, @@ -156,8 +158,6 @@ static GLushort *r200AllocElts( r200ContextPtr rmesa, GLuint nr ) if (rmesa->radeon.dma.flush) rmesa->radeon.dma.flush( rmesa->radeon.glCtx ); - rcommonEnsureCmdBufSpace(&rmesa->radeon, AOS_BUFSZ(rmesa->radeon.tcl.aos_count), __FUNCTION__); - r200EmitAOS( rmesa, rmesa->radeon.tcl.aos_count, 0 ); @@ -187,9 +187,6 @@ static void r200EmitPrim( GLcontext *ctx, r200TclPrimitive( ctx, prim, hwprim ); // fprintf(stderr,"Emit prim %d\n", rmesa->radeon.tcl.aos_count); - rcommonEnsureCmdBufSpace( &rmesa->radeon, - AOS_BUFSZ(rmesa->radeon.tcl.aos_count) + - rmesa->radeon.hw.max_state_size + VBUF_BUFSZ, __FUNCTION__ ); r200EmitAOS( rmesa, rmesa->radeon.tcl.aos_count, @@ -206,6 +203,7 @@ static void r200EmitPrim( GLcontext *ctx, r200EmitPrim( ctx, prim, hwprim, start, count ); \ (void) rmesa; } while (0) +#define MAX_CONVERSION_SIZE 40 /* Try & join small primitives */ #if 0 @@ -368,6 +366,63 @@ r200ComputeFogBlendFactor( GLcontext *ctx, GLfloat fogcoord ) } } +/** + * Predict total emit size for next rendering operation so there is no flush in middle of rendering + * Prediction has to aim towards the best possible value that is worse than worst case scenario + */ +static GLuint r200EnsureEmitSize( GLcontext * ctx , GLubyte* vimap_rev ) +{ + r200ContextPtr rmesa = R200_CONTEXT(ctx); + TNLcontext *tnl = TNL_CONTEXT(ctx); + struct vertex_buffer *VB = &tnl->vb; + GLuint space_required; + GLuint state_size; + GLuint nr_aos = 0; + int i; + /* predict number of aos to emit */ + for (i = 0; i < 15; ++i) + { + if (vimap_rev[i] != 255) + { + ++nr_aos; + } + } + + { + /* count the prediction for state size */ + space_required = 0; + state_size = radeonCountStateEmitSize( &rmesa->radeon ); + /* vtx may be changed in r200EmitArrays so account for it if not dirty */ + if (!rmesa->hw.vtx.dirty) + state_size += rmesa->hw.vtx.check(rmesa->radeon.glCtx, &rmesa->hw.vtx); + /* predict size for elements */ + for (i = 0; i < VB->PrimitiveCount; ++i) + { + if (!VB->Primitive[i].count) + continue; + /* If primitive.count is less than MAX_CONVERSION_SIZE + rendering code may decide convert to elts. + In that case we have to make pessimistic prediction. + and use larger of 2 paths. */ + const GLuint elts = ELTS_BUFSZ(nr_aos); + const GLuint index = INDEX_BUFSZ; + const GLuint vbuf = VBUF_BUFSZ; + if ( (!VB->Elts && VB->Primitive[i].count >= MAX_CONVERSION_SIZE) + || vbuf > index + elts) + space_required += vbuf; + else + space_required += index + elts; + space_required += AOS_BUFSZ(nr_aos); + } + space_required += SCISSOR_BUFSZ; + } + /* flush the buffer in case we need more than is left. */ + if (rcommonEnsureCmdBufSpace(&rmesa->radeon, space_required + state_size, __FUNCTION__)) + return space_required + radeonCountStateEmitSize( &rmesa->radeon ); + else + return space_required + state_size; +} + /**********************************************************************/ /* Render pipeline stage */ @@ -482,10 +537,10 @@ static GLboolean r200_run_tcl_render( GLcontext *ctx, /* Do the actual work: */ radeonReleaseArrays( ctx, ~0 /* stage->changed_inputs */ ); + GLuint emit_end = r200EnsureEmitSize( ctx, vimap_rev ) + + rmesa->radeon.cmdbuf.cs->cdw; r200EmitArrays( ctx, vimap_rev ); - rmesa->tcl.Elts = VB->Elts; - for (i = 0 ; i < VB->PrimitiveCount ; i++) { GLuint prim = _tnl_translate_prim(&VB->Primitive[i]); @@ -495,11 +550,14 @@ static GLboolean r200_run_tcl_render( GLcontext *ctx, if (!length) continue; - if (rmesa->tcl.Elts) + if (VB->Elts) r200EmitEltPrimitive( ctx, start, start+length, prim ); else r200EmitPrimitive( ctx, start, start+length, prim ); } + if ( emit_end < rmesa->radeon.cmdbuf.cs->cdw ) + WARN_ONCE("Rendering was %d commands larger than predicted size." + " We might overflow command buffer.\n", rmesa->radeon.cmdbuf.cs->cdw - emit_end); return GL_FALSE; /* finished the pipe */ } diff --git a/src/mesa/drivers/dri/r300/r300_cmdbuf.c b/src/mesa/drivers/dri/r300/r300_cmdbuf.c index 6b33f48885b..87651716fef 100644 --- a/src/mesa/drivers/dri/r300/r300_cmdbuf.c +++ b/src/mesa/drivers/dri/r300/r300_cmdbuf.c @@ -74,6 +74,21 @@ static unsigned packet0_count(r300ContextPtr r300, uint32_t *pkt) #define vpu_count(ptr) (((drm_r300_cmd_header_t*)(ptr))->vpu.count) #define r500fp_count(ptr) (((drm_r300_cmd_header_t*)(ptr))->r500fp.count) +int check_vpu(GLcontext *ctx, struct radeon_state_atom *atom) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + int cnt; + int extra = 1; + cnt = vpu_count(atom->cmd); + + if (r300->radeon.radeonScreen->kernel_mm) { + extra = 5; + } + + return cnt ? (cnt * 4) + extra : 0; +} + + void emit_vpu(GLcontext *ctx, struct radeon_state_atom * atom) { r300ContextPtr r300 = R300_CONTEXT(ctx); @@ -81,39 +96,18 @@ void emit_vpu(GLcontext *ctx, struct radeon_state_atom * atom) drm_r300_cmd_header_t cmd; uint32_t addr, ndw; - if (!r300->radeon.radeonScreen->kernel_mm) { - uint32_t dwords; - dwords = (*atom->check) (ctx, atom); - BEGIN_BATCH_NO_AUTOSTATE(dwords); - OUT_BATCH_TABLE(atom->cmd, dwords); - END_BATCH(); - return; - } - cmd.u = atom->cmd[0]; addr = (cmd.vpu.adrhi << 8) | cmd.vpu.adrlo; - ndw = cmd.vpu.count * 4; - if (ndw) { + ndw = atom->check(ctx, atom); - if (r300->vap_flush_needed) { - BEGIN_BATCH_NO_AUTOSTATE(15 + ndw); + BEGIN_BATCH_NO_AUTOSTATE(ndw); - /* flush processing vertices */ - OUT_BATCH_REGVAL(R300_SC_SCREENDOOR, 0); - OUT_BATCH_REGVAL(R300_RB3D_DSTCACHE_CTLSTAT, R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D); - OUT_BATCH_REGVAL(RADEON_WAIT_UNTIL, RADEON_WAIT_3D_IDLECLEAN); - OUT_BATCH_REGVAL(R300_SC_SCREENDOOR, 0xffffff); - OUT_BATCH_REGVAL(R300_VAP_PVS_STATE_FLUSH_REG, 0); - r300->vap_flush_needed = GL_FALSE; - } else { - BEGIN_BATCH_NO_AUTOSTATE(5 + ndw); - } - OUT_BATCH_REGVAL(R300_VAP_PVS_VECTOR_INDX_REG, addr); - OUT_BATCH(CP_PACKET0(R300_VAP_PVS_UPLOAD_DATA, ndw-1) | RADEON_ONE_REG_WR); - OUT_BATCH_TABLE(&atom->cmd[1], ndw); - OUT_BATCH_REGVAL(R300_VAP_PVS_STATE_FLUSH_REG, 0); - END_BATCH(); - } + ndw -= 5; + OUT_BATCH_REGVAL(R300_VAP_PVS_VECTOR_INDX_REG, addr); + OUT_BATCH(CP_PACKET0(R300_VAP_PVS_UPLOAD_DATA, ndw-1) | RADEON_ONE_REG_WR); + OUT_BATCH_TABLE(&atom->cmd[1], ndw); + OUT_BATCH_REGVAL(R300_VAP_PVS_STATE_FLUSH_REG, 0); + END_BATCH(); } void emit_r500fp(GLcontext *ctx, struct radeon_state_atom * atom) @@ -122,16 +116,9 @@ void emit_r500fp(GLcontext *ctx, struct radeon_state_atom * atom) BATCH_LOCALS(&r300->radeon); drm_r300_cmd_header_t cmd; uint32_t addr, ndw, sz; - int type, clamp, stride; + int type, clamp; - if (!r300->radeon.radeonScreen->kernel_mm) { - uint32_t dwords; - dwords = (*atom->check) (ctx, atom); - BEGIN_BATCH_NO_AUTOSTATE(dwords); - OUT_BATCH_TABLE(atom->cmd, dwords); - END_BATCH(); - return; - } + ndw = atom->check(ctx, atom); cmd.u = atom->cmd[0]; sz = cmd.r500fp.count; @@ -142,18 +129,34 @@ void emit_r500fp(GLcontext *ctx, struct radeon_state_atom * atom) addr |= (type << 16); addr |= (clamp << 17); - stride = type ? 4 : 6; - - ndw = sz * stride; - if (ndw) { + BEGIN_BATCH_NO_AUTOSTATE(ndw); + OUT_BATCH(CP_PACKET0(R500_GA_US_VECTOR_INDEX, 0)); + OUT_BATCH(addr); + ndw-=3; + OUT_BATCH(CP_PACKET0(R500_GA_US_VECTOR_DATA, ndw-1) | RADEON_ONE_REG_WR); + OUT_BATCH_TABLE(&atom->cmd[1], ndw); + END_BATCH(); +} - BEGIN_BATCH_NO_AUTOSTATE(3 + ndw); - OUT_BATCH(CP_PACKET0(R500_GA_US_VECTOR_INDEX, 0)); - OUT_BATCH(addr); - OUT_BATCH(CP_PACKET0(R500_GA_US_VECTOR_DATA, ndw-1) | RADEON_ONE_REG_WR); - OUT_BATCH_TABLE(&atom->cmd[1], ndw); - END_BATCH(); +static int check_tex_offsets(GLcontext *ctx, struct radeon_state_atom * atom) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + int numtmus = packet0_count(r300, r300->hw.tex.offset.cmd); + int dw = 0, i; + if (atom->cmd[0] == CP_PACKET2) { + return dw; + } + for(i = 0; i < numtmus; ++i) { + radeonTexObj *t = r300->hw.textures[i]; + if (!t && !r300->radeon.radeonScreen->kernel_mm) { + dw += 0; + } else if (t && t->image_override && !t->bo) { + if (!r300->radeon.radeonScreen->kernel_mm) + dw += 2; + } else + dw += 4; } + return dw; } static void emit_tex_offsets(GLcontext *ctx, struct radeon_state_atom * atom) @@ -243,6 +246,17 @@ void r300_emit_scissor(GLcontext *ctx) OUT_BATCH((x2 << R300_SCISSORS_X_SHIFT)|(y2 << R300_SCISSORS_Y_SHIFT)); END_BATCH(); } +static int check_cb_offset(GLcontext *ctx, struct radeon_state_atom * atom) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + uint32_t dw = 6 + 3 + 16; + if (r300->radeon.radeonScreen->kernel_mm) + dw += 2; + if (!r300->radeon.radeonScreen->driScreen->dri2.enabled) { + dw -= 3 + 16; + } + return dw; +} static void emit_cb_offset(GLcontext *ctx, struct radeon_state_atom * atom) { @@ -252,7 +266,7 @@ static void emit_cb_offset(GLcontext *ctx, struct radeon_state_atom * atom) uint32_t cbpitch; uint32_t offset = r300->radeon.state.color.draw_offset; uint32_t dw = 6; - int i; + int i; rrb = radeon_get_colorbuffer(&r300->radeon); if (!rrb || !rrb->bo) { @@ -334,13 +348,23 @@ static void emit_cb_offset(GLcontext *ctx, struct radeon_state_atom * atom) } } +static int check_zb_offset(GLcontext *ctx, struct radeon_state_atom * atom) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + uint32_t dw; + dw = 6; + if (r300->radeon.radeonScreen->kernel_mm) + dw += 2; + return dw; +} + static void emit_zb_offset(GLcontext *ctx, struct radeon_state_atom * atom) { r300ContextPtr r300 = R300_CONTEXT(ctx); BATCH_LOCALS(&r300->radeon); struct radeon_renderbuffer *rrb; uint32_t zbpitch; - uint32_t dw; + uint32_t dw = atom->check(ctx, atom); rrb = radeon_get_depthbuffer(&r300->radeon); if (!rrb) @@ -356,9 +380,6 @@ static void emit_zb_offset(GLcontext *ctx, struct radeon_state_atom * atom) } } - dw = 6; - if (r300->radeon.radeonScreen->kernel_mm) - dw += 2; BEGIN_BATCH_NO_AUTOSTATE(dw); OUT_BATCH_REGSEQ(R300_ZB_DEPTHOFFSET, 1); OUT_BATCH_RELOC(0, rrb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0); @@ -370,46 +391,6 @@ static void emit_zb_offset(GLcontext *ctx, struct radeon_state_atom * atom) END_BATCH(); } -static void emit_gb_misc(GLcontext *ctx, struct radeon_state_atom * atom) -{ - r300ContextPtr r300 = R300_CONTEXT(ctx); - BATCH_LOCALS(&r300->radeon); - if (!r300->radeon.radeonScreen->driScreen->dri2.enabled) { - BEGIN_BATCH_NO_AUTOSTATE(4); - OUT_BATCH(atom->cmd[0]); - OUT_BATCH(atom->cmd[1]); - OUT_BATCH(atom->cmd[2]); - OUT_BATCH(atom->cmd[3]); - END_BATCH(); - } -} - -static void emit_threshold_misc(GLcontext *ctx, struct radeon_state_atom * atom) -{ - r300ContextPtr r300 = R300_CONTEXT(ctx); - BATCH_LOCALS(&r300->radeon); - if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) { - BEGIN_BATCH_NO_AUTOSTATE(3); - OUT_BATCH(atom->cmd[0]); - OUT_BATCH(atom->cmd[1]); - OUT_BATCH(atom->cmd[2]); - END_BATCH(); - } -} - -static void emit_shade_misc(GLcontext *ctx, struct radeon_state_atom * atom) -{ - r300ContextPtr r300 = R300_CONTEXT(ctx); - BATCH_LOCALS(&r300->radeon); - - if (!r300->radeon.radeonScreen->driScreen->dri2.enabled) { - BEGIN_BATCH_NO_AUTOSTATE(2); - OUT_BATCH(atom->cmd[0]); - OUT_BATCH(atom->cmd[1]); - END_BATCH(); - } -} - static void emit_zstencil_format(GLcontext *ctx, struct radeon_state_atom * atom) { r300ContextPtr r300 = R300_CONTEXT(ctx); @@ -427,7 +408,7 @@ static void emit_zstencil_format(GLcontext *ctx, struct radeon_state_atom * atom format = R300_DEPTHFORMAT_24BIT_INT_Z_8BIT_STENCIL; } - BEGIN_BATCH_NO_AUTOSTATE(5); + BEGIN_BATCH_NO_AUTOSTATE(atom->cmd_size); OUT_BATCH(atom->cmd[0]); atom->cmd[1] &= ~0xf; atom->cmd[1] |= format; @@ -438,6 +419,11 @@ static void emit_zstencil_format(GLcontext *ctx, struct radeon_state_atom * atom END_BATCH(); } +static int check_never(GLcontext *ctx, struct radeon_state_atom *atom) +{ + return 0; +} + static int check_always(GLcontext *ctx, struct radeon_state_atom *atom) { return atom->cmd_size; @@ -454,28 +440,29 @@ static int check_variable(GLcontext *ctx, struct radeon_state_atom *atom) return cnt ? cnt + 1 : 0; } -int check_vpu(GLcontext *ctx, struct radeon_state_atom *atom) -{ - int cnt; - - cnt = vpu_count(atom->cmd); - return cnt ? (cnt * 4) + 1 : 0; -} - int check_r500fp(GLcontext *ctx, struct radeon_state_atom *atom) { int cnt; - + r300ContextPtr r300 = R300_CONTEXT(ctx); + int extra = 1; cnt = r500fp_count(atom->cmd); - return cnt ? (cnt * 6) + 1 : 0; + if (r300->radeon.radeonScreen->kernel_mm) + extra = 3; + + return cnt ? (cnt * 6) + extra : 0; } int check_r500fp_const(GLcontext *ctx, struct radeon_state_atom *atom) { int cnt; + r300ContextPtr r300 = R300_CONTEXT(ctx); + int extra = 1; + cnt = r500fp_count(atom->cmd); + if (r300->radeon.radeonScreen->kernel_mm) + extra = 3; cnt = r500fp_count(atom->cmd); - return cnt ? (cnt * 4) + 1 : 0; + return cnt ? (cnt * 4) + extra : 0; } #define ALLOC_STATE( ATOM, CHK, SZ, IDX ) \ @@ -565,11 +552,14 @@ void r300InitCmdBuf(r300ContextPtr r300) ALLOC_STATE(gb_enable, always, 2, 0); r300->hw.gb_enable.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GB_ENABLE, 1); - ALLOC_STATE(gb_misc, always, R300_GB_MISC_CMDSIZE, 0); + if (!r300->radeon.radeonScreen->driScreen->dri2.enabled) { + ALLOC_STATE(gb_misc, always, R300_GB_MISC_CMDSIZE, 0); + } else { + ALLOC_STATE(gb_misc, never, R300_GB_MISC_CMDSIZE, 0); + } r300->hw.gb_misc.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GB_MSPOS0, 3); - r300->hw.gb_misc.emit = emit_gb_misc; ALLOC_STATE(gb_misc2, always, R300_GB_MISC2_CMDSIZE, 0); - r300->hw.gb_misc2.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, 0x401C, 2); + r300->hw.gb_misc2.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, 0x401C, 2); ALLOC_STATE(txe, always, R300_TXE_CMDSIZE, 0); r300->hw.txe.cmd[R300_TXE_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_TX_ENABLE, 1); ALLOC_STATE(ga_point_s0, always, 5, 0); @@ -584,9 +574,12 @@ void r300InitCmdBuf(r300ContextPtr r300) r300->hw.lcntl.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GA_LINE_CNTL, 1); ALLOC_STATE(ga_line_stipple, always, 4, 0); r300->hw.ga_line_stipple.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GA_LINE_STIPPLE_VALUE, 3); - ALLOC_STATE(shade, always, 2, 0); + if (!r300->radeon.radeonScreen->driScreen->dri2.enabled) { + ALLOC_STATE(shade, always, 2, 0); + } else { + ALLOC_STATE(shade, never, 2, 0); + } r300->hw.shade.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GA_ENHANCE, 1); - r300->hw.shade.emit = emit_shade_misc; ALLOC_STATE(shade2, always, 4, 0); r300->hw.shade2.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, 0x4278, 3); ALLOC_STATE(polygon_mode, always, 4, 0); @@ -635,11 +628,14 @@ void r300InitCmdBuf(r300ContextPtr r300) ALLOC_STATE(r500fp, r500fp, R500_FPI_CMDSIZE, 0); r300->hw.r500fp.cmd[R300_FPI_CMD_0] = cmdr500fp(r300->radeon.radeonScreen, 0, 0, 0, 0); - r300->hw.r500fp.emit = emit_r500fp; + if (r300->radeon.radeonScreen->kernel_mm) + r300->hw.r500fp.emit = emit_r500fp; + ALLOC_STATE(r500fp_const, r500fp_const, R500_FPP_CMDSIZE, 0); r300->hw.r500fp_const.cmd[R300_FPI_CMD_0] = cmdr500fp(r300->radeon.radeonScreen, 0, 0, 1, 0); - r300->hw.r500fp_const.emit = emit_r500fp; + if (r300->radeon.radeonScreen->kernel_mm) + r300->hw.r500fp_const.emit = emit_r500fp; } else { ALLOC_STATE(fp, always, R300_FP_CMDSIZE, 0); r300->hw.fp.cmd[R300_FP_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_US_CONFIG, 3); @@ -682,15 +678,18 @@ void r300InitCmdBuf(r300ContextPtr r300) } ALLOC_STATE(rop, always, 2, 0); r300->hw.rop.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_RB3D_ROPCNTL, 1); - ALLOC_STATE(cb, always, R300_CB_CMDSIZE, 0); + ALLOC_STATE(cb, cb_offset, R300_CB_CMDSIZE, 0); r300->hw.cb.emit = &emit_cb_offset; ALLOC_STATE(rb3d_dither_ctl, always, 10, 0); r300->hw.rb3d_dither_ctl.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_RB3D_DITHER_CTL, 9); ALLOC_STATE(rb3d_aaresolve_ctl, always, 2, 0); r300->hw.rb3d_aaresolve_ctl.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_RB3D_AARESOLVE_CTL, 1); - ALLOC_STATE(rb3d_discard_src_pixel_lte_threshold, always, 3, 0); - r300->hw.rb3d_discard_src_pixel_lte_threshold.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 2); - r300->hw.rb3d_discard_src_pixel_lte_threshold.emit = emit_threshold_misc; + if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) { + ALLOC_STATE(rb3d_discard_src_pixel_lte_threshold, always, 3, 0); + } else { + ALLOC_STATE(rb3d_discard_src_pixel_lte_threshold, never, 3, 0); + } + r300->hw.rb3d_discard_src_pixel_lte_threshold.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 2); ALLOC_STATE(zs, always, R300_ZS_CMDSIZE, 0); r300->hw.zs.cmd[R300_ZS_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_ZB_CNTL, 3); @@ -700,7 +699,7 @@ void r300InitCmdBuf(r300ContextPtr r300) cmdpacket0(r300->radeon.radeonScreen, R300_ZB_FORMAT, 4); r300->hw.zstencil_format.emit = emit_zstencil_format; - ALLOC_STATE(zb, always, R300_ZB_CMDSIZE, 0); + ALLOC_STATE(zb, zb_offset, R300_ZB_CMDSIZE, 0); r300->hw.zb.emit = emit_zb_offset; ALLOC_STATE(zb_depthclearvalue, always, 2, 0); r300->hw.zb_depthclearvalue.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_ZB_DEPTHCLEARVALUE, 1); @@ -713,47 +712,72 @@ void r300InitCmdBuf(r300ContextPtr r300) /* VPU only on TCL */ if (has_tcl) { - int i; + int i; + if (r300->radeon.radeonScreen->kernel_mm) { + ALLOC_STATE(vap_flush, always, 10, 0); + /* flush processing vertices */ + r300->hw.vap_flush.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_SC_SCREENDOOR, 1); + r300->hw.vap_flush.cmd[1] = 0; + r300->hw.vap_flush.cmd[2] = cmdpacket0(r300->radeon.radeonScreen, R300_RB3D_DSTCACHE_CTLSTAT, 1); + r300->hw.vap_flush.cmd[3] = R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D; + r300->hw.vap_flush.cmd[4] = cmdpacket0(r300->radeon.radeonScreen, RADEON_WAIT_UNTIL, 1); + r300->hw.vap_flush.cmd[5] = RADEON_WAIT_3D_IDLECLEAN; + r300->hw.vap_flush.cmd[6] = cmdpacket0(r300->radeon.radeonScreen, R300_SC_SCREENDOOR, 1); + r300->hw.vap_flush.cmd[7] = 0xffffff; + r300->hw.vap_flush.cmd[8] = cmdpacket0(r300->radeon.radeonScreen, R300_VAP_PVS_STATE_FLUSH_REG, 1); + r300->hw.vap_flush.cmd[9] = 0; + } else { + ALLOC_STATE(vap_flush, never, 10, 0); + } + + ALLOC_STATE(vpi, vpu, R300_VPI_CMDSIZE, 0); r300->hw.vpi.cmd[0] = - cmdvpu(r300->radeon.radeonScreen, R300_PVS_CODE_START, 0); - r300->hw.vpi.emit = emit_vpu; + cmdvpu(r300->radeon.radeonScreen, R300_PVS_CODE_START, 0); + if (r300->radeon.radeonScreen->kernel_mm) + r300->hw.vpi.emit = emit_vpu; if (is_r500) { - ALLOC_STATE(vpp, vpu, R300_VPP_CMDSIZE, 0); - r300->hw.vpp.cmd[0] = - cmdvpu(r300->radeon.radeonScreen, R500_PVS_CONST_START, 0); - r300->hw.vpp.emit = emit_vpu; - - ALLOC_STATE(vps, vpu, R300_VPS_CMDSIZE, 0); - r300->hw.vps.cmd[0] = - cmdvpu(r300->radeon.radeonScreen, R500_POINT_VPORT_SCALE_OFFSET, 1); - r300->hw.vps.emit = emit_vpu; + ALLOC_STATE(vpp, vpu, R300_VPP_CMDSIZE, 0); + r300->hw.vpp.cmd[0] = + cmdvpu(r300->radeon.radeonScreen, R500_PVS_CONST_START, 0); + if (r300->radeon.radeonScreen->kernel_mm) + r300->hw.vpp.emit = emit_vpu; + + ALLOC_STATE(vps, vpu, R300_VPS_CMDSIZE, 0); + r300->hw.vps.cmd[0] = + cmdvpu(r300->radeon.radeonScreen, R500_POINT_VPORT_SCALE_OFFSET, 1); + if (r300->radeon.radeonScreen->kernel_mm) + r300->hw.vps.emit = emit_vpu; for (i = 0; i < 6; i++) { - ALLOC_STATE(vpucp[i], vpu, R300_VPUCP_CMDSIZE, 0); - r300->hw.vpucp[i].cmd[0] = - cmdvpu(r300->radeon.radeonScreen, - R500_PVS_UCP_START + i, 1); - r300->hw.vpucp[i].emit = emit_vpu; + ALLOC_STATE(vpucp[i], vpu, R300_VPUCP_CMDSIZE, 0); + r300->hw.vpucp[i].cmd[0] = + cmdvpu(r300->radeon.radeonScreen, + R500_PVS_UCP_START + i, 1); + if (r300->radeon.radeonScreen->kernel_mm) + r300->hw.vpucp[i].emit = emit_vpu; } } else { - ALLOC_STATE(vpp, vpu, R300_VPP_CMDSIZE, 0); - r300->hw.vpp.cmd[0] = - cmdvpu(r300->radeon.radeonScreen, R300_PVS_CONST_START, 0); - r300->hw.vpp.emit = emit_vpu; - - ALLOC_STATE(vps, vpu, R300_VPS_CMDSIZE, 0); - r300->hw.vps.cmd[0] = - cmdvpu(r300->radeon.radeonScreen, R300_POINT_VPORT_SCALE_OFFSET, 1); - r300->hw.vps.emit = emit_vpu; + ALLOC_STATE(vpp, vpu, R300_VPP_CMDSIZE, 0); + r300->hw.vpp.cmd[0] = + cmdvpu(r300->radeon.radeonScreen, R300_PVS_CONST_START, 0); + if (r300->radeon.radeonScreen->kernel_mm) + r300->hw.vpp.emit = emit_vpu; + + ALLOC_STATE(vps, vpu, R300_VPS_CMDSIZE, 0); + r300->hw.vps.cmd[0] = + cmdvpu(r300->radeon.radeonScreen, R300_POINT_VPORT_SCALE_OFFSET, 1); + if (r300->radeon.radeonScreen->kernel_mm) + r300->hw.vps.emit = emit_vpu; for (i = 0; i < 6; i++) { ALLOC_STATE(vpucp[i], vpu, R300_VPUCP_CMDSIZE, 0); r300->hw.vpucp[i].cmd[0] = cmdvpu(r300->radeon.radeonScreen, - R300_PVS_UCP_START + i, 1); - r300->hw.vpucp[i].emit = emit_vpu; + R300_PVS_UCP_START + i, 1); + if (r300->radeon.radeonScreen->kernel_mm) + r300->hw.vpucp[i].emit = emit_vpu; } } } @@ -777,7 +801,7 @@ void r300InitCmdBuf(r300ContextPtr r300) ALLOC_STATE(tex.pitch, variable, mtu + 1, 0); r300->hw.tex.pitch.cmd[R300_TEX_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_TX_FORMAT2_0, 0); - ALLOC_STATE(tex.offset, variable, 1, 0); + ALLOC_STATE(tex.offset, tex_offsets, 1, 0); r300->hw.tex.offset.cmd[R300_TEX_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_TX_OFFSET_0, 0); r300->hw.tex.offset.emit = &emit_tex_offsets; diff --git a/src/mesa/drivers/dri/r300/r300_cmdbuf.h b/src/mesa/drivers/dri/r300/r300_cmdbuf.h index 53bcc0eeb49..1b703e518a0 100644 --- a/src/mesa/drivers/dri/r300/r300_cmdbuf.h +++ b/src/mesa/drivers/dri/r300/r300_cmdbuf.h @@ -38,6 +38,12 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "r300_context.h" +#define CACHE_FLUSH_BUFSZ (4*2) +#define PRE_EMIT_STATE_BUFSZ (2+2) +#define AOS_BUFSZ(nr) (3+(nr >>1)*3 + (nr&1)*2 + (nr*2)) +#define FIREAOS_BUFSZ (3) +#define SCISSORS_BUFSZ (3) + extern void r300InitCmdBuf(r300ContextPtr r300); void r300_emit_scissor(GLcontext *ctx); diff --git a/src/mesa/drivers/dri/r300/r300_context.c b/src/mesa/drivers/dri/r300/r300_context.c index f701c434a45..0bf002caeab 100644 --- a/src/mesa/drivers/dri/r300/r300_context.c +++ b/src/mesa/drivers/dri/r300/r300_context.c @@ -214,11 +214,8 @@ static void r300_vtbl_emit_cs_header(struct radeon_cs *cs, radeonContextPtr rmes static void r300_vtbl_pre_emit_atoms(radeonContextPtr radeon) { - r300ContextPtr r300 = (r300ContextPtr)radeon; BATCH_LOCALS(radeon); - r300->vap_flush_needed = GL_TRUE; - cp_wait(radeon, R300_WAIT_3D | R300_WAIT_3D_CLEAN); BEGIN_BATCH_NO_AUTOSTATE(2); OUT_BATCH_REGVAL(R300_TX_INVALTAGS, R300_TX_FLUSH); diff --git a/src/mesa/drivers/dri/r300/r300_context.h b/src/mesa/drivers/dri/r300/r300_context.h index a8fe508c4a5..1dadcc0a697 100644 --- a/src/mesa/drivers/dri/r300/r300_context.h +++ b/src/mesa/drivers/dri/r300/r300_context.h @@ -350,6 +350,7 @@ struct r300_hw_state { struct radeon_state_atom zb_hiz_offset; /* (4F44) */ struct radeon_state_atom zb_hiz_pitch; /* (4F54) */ + struct radeon_state_atom vap_flush; struct radeon_state_atom vpi; /* vp instructions */ struct radeon_state_atom vpp; /* vp parameters */ struct radeon_state_atom vps; /* vertex point size (?) */ @@ -524,7 +525,6 @@ struct r300_context { struct r300_swtcl_info swtcl; struct r300_vertex_buffer vbuf; struct r300_index_buffer ind_buf; - GLboolean vap_flush_needed; uint32_t fallback; diff --git a/src/mesa/drivers/dri/r300/r300_draw.c b/src/mesa/drivers/dri/r300/r300_draw.c index e98dc335183..56680516c83 100644 --- a/src/mesa/drivers/dri/r300/r300_draw.c +++ b/src/mesa/drivers/dri/r300/r300_draw.c @@ -39,8 +39,10 @@ #include "r300_render.h" #include "r300_state.h" #include "r300_tex.h" +#include "r300_cmdbuf.h" #include "radeon_buffer_objects.h" +#include "radeon_common_context.h" #include "tnl/tnl.h" #include "tnl/t_vp_build.h" @@ -328,7 +330,6 @@ static void r300TranslateAttrib(GLcontext *ctx, GLuint attr, int count, const st type = GL_FLOAT; - r300ConvertAttrib(ctx, count, input, &r300_attr); if (input->StrideB == 0) { r300_attr.stride = 0; } else { @@ -339,41 +340,14 @@ static void r300TranslateAttrib(GLcontext *ctx, GLuint attr, int count, const st } else { type = input->Type; r300_attr.dwords = (getTypeSize(type) * input->Size + 3)/ 4; - if (input->BufferObj->Name) { - if (stride % 4 != 0) { - assert(((int) input->Ptr) % input->StrideB == 0); - r300AlignDataToDword(ctx, input, count, &r300_attr); - r300_attr.is_named_bo = GL_FALSE; - } else { - r300_attr.stride = input->StrideB; - r300_attr.bo_offset = (GLuint) input->Ptr; - r300_attr.bo = get_radeon_buffer_object(input->BufferObj)->bo; - r300_attr.is_named_bo = GL_TRUE; - } - } else { - int size; - uint32_t *dst; + if (!input->BufferObj->Name) { if (input->StrideB == 0) { - size = getTypeSize(input->Type) * input->Size; - count = 1; r300_attr.stride = 0; } else { - size = getTypeSize(input->Type) * input->Size * count; r300_attr.stride = (getTypeSize(type) * input->Size + 3) & ~3; } - radeonAllocDmaRegion(&r300->radeon, &r300_attr.bo, &r300_attr.bo_offset, size, 32); - assert(r300_attr.bo->ptr != NULL); - dst = (uint32_t *)ADD_POINTERS(r300_attr.bo->ptr, r300_attr.bo_offset); - switch (r300_attr.dwords) { - case 1: radeonEmitVec4(dst, input->Ptr, input->StrideB, count); break; - case 2: radeonEmitVec8(dst, input->Ptr, input->StrideB, count); break; - case 3: radeonEmitVec12(dst, input->Ptr, input->StrideB, count); break; - case 4: radeonEmitVec16(dst, input->Ptr, input->StrideB, count); break; - default: assert(0); break; - } - r300_attr.is_named_bo = GL_FALSE; } } @@ -468,7 +442,6 @@ static void r300SetVertexFormat(GLcontext *ctx, const struct gl_client_array *ar { r300ContextPtr r300 = R300_CONTEXT(ctx); struct r300_vertex_buffer *vbuf = &r300->vbuf; - int ret; { int i, tmp; @@ -492,29 +465,83 @@ static void r300SetVertexFormat(GLcontext *ctx, const struct gl_client_array *ar r300SwitchFallback(ctx, R300_FALLBACK_AOS_LIMIT, vbuf->num_attribs > R300_MAX_AOS_ARRAYS); if (r300->fallback) return; +} - { - int i; +static void r300AllocDmaRegions(GLcontext *ctx, const struct gl_client_array *input[], int count) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + struct r300_vertex_buffer *vbuf = &r300->vbuf; + GLuint stride; + int ret; + int i, index; + + for (index = 0; index < vbuf->num_attribs; index++) { + struct radeon_aos *aos = &r300->radeon.tcl.aos[index]; + i = vbuf->attribs[index].element; - for (i = 0; i < vbuf->num_attribs; i++) { - struct radeon_aos *aos = &r300->radeon.tcl.aos[i]; + stride = (input[i]->StrideB == 0) ? getTypeSize(input[i]->Type) * input[i]->Size : input[i]->StrideB; - aos->count = vbuf->attribs[i].stride == 0 ? 1 : count; - aos->stride = vbuf->attribs[i].stride / sizeof(float); - aos->offset = vbuf->attribs[i].bo_offset; - aos->components = vbuf->attribs[i].dwords; - aos->bo = vbuf->attribs[i].bo; + if (input[i]->Type == GL_DOUBLE || input[i]->Type == GL_UNSIGNED_INT || input[i]->Type == GL_INT || +#if MESA_BIG_ENDIAN + getTypeSize(input[i]->Type) != 4 || +#endif + stride < 4) { + + r300ConvertAttrib(ctx, count, input[i], &vbuf->attribs[index]); + } else { + if (input[i]->BufferObj->Name) { + if (stride % 4 != 0) { + assert(((intptr_t) input[i]->Ptr) % input[i]->StrideB == 0); + r300AlignDataToDword(ctx, input[i], count, &vbuf->attribs[index]); + vbuf->attribs[index].is_named_bo = GL_FALSE; + } else { + vbuf->attribs[index].stride = input[i]->StrideB; + vbuf->attribs[index].bo_offset = (intptr_t) input[i]->Ptr; + vbuf->attribs[index].bo = get_radeon_buffer_object(input[i]->BufferObj)->bo; + vbuf->attribs[index].is_named_bo = GL_TRUE; + } + } else { + + int size; + int local_count = count; + uint32_t *dst; + + if (input[i]->StrideB == 0) { + size = getTypeSize(input[i]->Type) * input[i]->Size; + local_count = 1; + } else { + size = getTypeSize(input[i]->Type) * input[i]->Size * local_count; + } + + radeonAllocDmaRegion(&r300->radeon, &vbuf->attribs[index].bo, &vbuf->attribs[index].bo_offset, size, 32); + assert(vbuf->attribs[index].bo->ptr != NULL); + dst = (uint32_t *)ADD_POINTERS(vbuf->attribs[index].bo->ptr, vbuf->attribs[index].bo_offset); + switch (vbuf->attribs[index].dwords) { + case 1: radeonEmitVec4(dst, input[i]->Ptr, input[i]->StrideB, local_count); break; + case 2: radeonEmitVec8(dst, input[i]->Ptr, input[i]->StrideB, local_count); break; + case 3: radeonEmitVec12(dst, input[i]->Ptr, input[i]->StrideB, local_count); break; + case 4: radeonEmitVec16(dst, input[i]->Ptr, input[i]->StrideB, local_count); break; + default: assert(0); break; + } - if (vbuf->attribs[i].is_named_bo) { - radeon_cs_space_add_persistent_bo(r300->radeon.cmdbuf.cs, r300->vbuf.attribs[i].bo, RADEON_GEM_DOMAIN_GTT, 0); } } - r300->radeon.tcl.aos_count = vbuf->num_attribs; - ret = radeon_cs_space_check_with_bo(r300->radeon.cmdbuf.cs, first_elem(&r300->radeon.dma.reserved)->bo, RADEON_GEM_DOMAIN_GTT, 0); - if (ret) - r300SwitchFallback(ctx, R300_FALLBACK_INVALID_BUFFERS, GL_TRUE); + aos->count = vbuf->attribs[index].stride == 0 ? 1 : count; + aos->stride = vbuf->attribs[index].stride / sizeof(float); + aos->components = vbuf->attribs[index].dwords; + aos->bo = vbuf->attribs[index].bo; + aos->offset = vbuf->attribs[index].bo_offset; + + if (vbuf->attribs[index].is_named_bo) { + radeon_cs_space_add_persistent_bo(r300->radeon.cmdbuf.cs, r300->vbuf.attribs[index].bo, RADEON_GEM_DOMAIN_GTT, 0); + } } + + r300->radeon.tcl.aos_count = vbuf->num_attribs; + ret = radeon_cs_space_check_with_bo(r300->radeon.cmdbuf.cs, first_elem(&r300->radeon.dma.reserved)->bo, RADEON_GEM_DOMAIN_GTT, 0); + r300SwitchFallback(ctx, R300_FALLBACK_INVALID_BUFFERS, ret); + } static void r300FreeData(GLcontext *ctx) @@ -542,6 +569,34 @@ static void r300FreeData(GLcontext *ctx) } } +static GLuint r300PredictTryDrawPrimsSize(GLcontext *ctx, GLuint nr_prims) +{ + struct r300_context *r300 = R300_CONTEXT(ctx); + struct r300_vertex_buffer *vbuf = &r300->vbuf; + GLboolean flushed; + GLuint dwords; + GLuint state_size; + + dwords = 2*CACHE_FLUSH_BUFSZ; + dwords += PRE_EMIT_STATE_BUFSZ; + dwords += (AOS_BUFSZ(vbuf->num_attribs) + + SCISSORS_BUFSZ + + FIREAOS_BUFSZ )*nr_prims; + + state_size = radeonCountStateEmitSize(&r300->radeon); + flushed = rcommonEnsureCmdBufSpace(&r300->radeon, + dwords + state_size, + __FUNCTION__); + if (flushed) + dwords += radeonCountStateEmitSize(&r300->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 r300TryDrawPrims(GLcontext *ctx, const struct gl_client_array *arrays[], const struct _mesa_prim *prim, @@ -553,6 +608,10 @@ static GLboolean r300TryDrawPrims(GLcontext *ctx, struct r300_context *r300 = R300_CONTEXT(ctx); GLuint i; + if (RADEON_DEBUG & DEBUG_PRIMS) + fprintf(stderr, "%s: %u (%d-%d) cs begin at %d\n", + __FUNCTION__, nr_prims, min_index, max_index, r300->radeon.cmdbuf.cs->cdw ); + if (ctx->NewState) _mesa_update_state( ctx ); @@ -563,14 +622,6 @@ static GLboolean r300TryDrawPrims(GLcontext *ctx, r300SwitchFallback(ctx, R300_FALLBACK_INVALID_BUFFERS, !r300ValidateBuffers(ctx)); - /* ensure we have the cmd buf space in advance to cover - * the state + DMA AOS pointers */ - rcommonEnsureCmdBufSpace(&r300->radeon, - r300->radeon.hw.max_state_size + (60*sizeof(int)), - __FUNCTION__); - - r300SetupIndexBuffer(ctx, ib); - r300SetVertexFormat(ctx, arrays, max_index + 1); if (r300->fallback) @@ -580,6 +631,18 @@ static GLboolean r300TryDrawPrims(GLcontext *ctx, r300UpdateShaderStates(r300); + /* ensure we have the cmd buf space in advance to cover + * the state + DMA AOS pointers */ + GLuint emit_end = r300PredictTryDrawPrimsSize(ctx, nr_prims) + + r300->radeon.cmdbuf.cs->cdw; + + r300SetupIndexBuffer(ctx, ib); + + r300AllocDmaRegions(ctx, arrays, max_index + 1); + + if (r300->fallback) + return GL_FALSE; + r300EmitCacheFlush(r300); radeonEmitState(&r300->radeon); @@ -591,6 +654,14 @@ static GLboolean r300TryDrawPrims(GLcontext *ctx, r300FreeData(ctx); + if (RADEON_DEBUG & DEBUG_PRIMS) + fprintf(stderr, "%s: %u (%d-%d) cs ending at %d\n", + __FUNCTION__, nr_prims, min_index, max_index, r300->radeon.cmdbuf.cs->cdw ); + + if (emit_end < r300->radeon.cmdbuf.cs->cdw) + WARN_ONCE("Rendering was %d commands larger than predicted size." + " We might overflow command buffer.\n", r300->radeon.cmdbuf.cs->cdw - emit_end); + return GL_TRUE; } diff --git a/src/mesa/drivers/dri/r300/r300_ioctl.c b/src/mesa/drivers/dri/r300/r300_ioctl.c index 7ab6928247a..3303078e39a 100644 --- a/src/mesa/drivers/dri/r300/r300_ioctl.c +++ b/src/mesa/drivers/dri/r300/r300_ioctl.c @@ -507,7 +507,15 @@ static void r300EmitClearState(GLcontext * ctx) R500_ALU_RGBA_A_SWIZ_0; r500fp.cmd[7] = 0; - emit_r500fp(ctx, &r500fp); + if (r300->radeon.radeonScreen->kernel_mm) { + emit_r500fp(ctx, &r500fp); + } else { + int dwords = r500fp.check(ctx,&r500fp); + BEGIN_BATCH_NO_AUTOSTATE(dwords); + OUT_BATCH_TABLE(r500fp.cmd, dwords); + END_BATCH(); + } + } BEGIN_BATCH(2); @@ -551,6 +559,7 @@ static void r300EmitClearState(GLcontext * ctx) struct radeon_state_atom vpu; uint32_t _cmd[10]; R300_STATECHANGE(r300, pvs); + R300_STATECHANGE(r300, vap_flush); R300_STATECHANGE(r300, vpi); BEGIN_BATCH(4); @@ -592,8 +601,19 @@ static void r300EmitClearState(GLcontext * ctx) PVS_SRC_REG_INPUT, NEGATE_NONE); vpu.cmd[8] = 0x0; - r300->vap_flush_needed = GL_TRUE; - emit_vpu(ctx, &vpu); + if (r300->radeon.radeonScreen->kernel_mm) { + int dwords = r300->hw.vap_flush.check(ctx,&r300->hw.vap_flush); + BEGIN_BATCH_NO_AUTOSTATE(dwords); + OUT_BATCH_TABLE(r300->hw.vap_flush.cmd, dwords); + END_BATCH(); + emit_vpu(ctx, &vpu); + } else { + int dwords = vpu.check(ctx,&vpu); + BEGIN_BATCH_NO_AUTOSTATE(dwords); + OUT_BATCH_TABLE(vpu.cmd, dwords); + END_BATCH(); + } + } } diff --git a/src/mesa/drivers/dri/r300/r300_render.c b/src/mesa/drivers/dri/r300/r300_render.c index 37a40f6c36a..446cf40131e 100644 --- a/src/mesa/drivers/dri/r300/r300_render.c +++ b/src/mesa/drivers/dri/r300/r300_render.c @@ -341,12 +341,6 @@ void r300RunRenderPrimitive(GLcontext * ctx, int start, int end, int prim) if (type < 0 || num_verts <= 0) return; - /* Make space for at least 128 dwords. - * This is supposed to ensure that we can get all rendering - * commands into a single command buffer. - */ - rcommonEnsureCmdBufSpace(&rmesa->radeon, 128, __FUNCTION__); - if (rmesa->ind_buf.bo) { GLuint first, incr, offset = 0; diff --git a/src/mesa/drivers/dri/r300/r300_state.c b/src/mesa/drivers/dri/r300/r300_state.c index ce0666b9019..4fe9175b61d 100644 --- a/src/mesa/drivers/dri/r300/r300_state.c +++ b/src/mesa/drivers/dri/r300/r300_state.c @@ -371,6 +371,7 @@ static void r300ClipPlane( GLcontext *ctx, GLenum plane, const GLfloat *eq ) p = (GLint) plane - (GLint) GL_CLIP_PLANE0; ip = (GLint *)ctx->Transform._ClipUserPlane[p]; + R300_STATECHANGE( rmesa, vap_flush ); R300_STATECHANGE( rmesa, vpucp[p] ); rmesa->hw.vpucp[p].cmd[R300_VPUCP_X] = ip[0]; rmesa->hw.vpucp[p].cmd[R300_VPUCP_Y] = ip[1]; diff --git a/src/mesa/drivers/dri/r300/r300_vertprog.c b/src/mesa/drivers/dri/r300/r300_vertprog.c index c5edbd0052b..862f2120850 100644 --- a/src/mesa/drivers/dri/r300/r300_vertprog.c +++ b/src/mesa/drivers/dri/r300/r300_vertprog.c @@ -298,6 +298,8 @@ static void r300EmitVertexProgram(r300ContextPtr r300, int dest, struct r300_ver assert((code->length > 0) && (code->length % 4 == 0)); + R300_STATECHANGE( r300, vap_flush ); + switch ((dest >> 8) & 0xf) { case 0: R300_STATECHANGE(r300, vpi); @@ -335,6 +337,7 @@ void r300SetupVertexProgram(r300ContextPtr rmesa) ((drm_r300_cmd_header_t *) rmesa->hw.vpi.cmd)->vpu.count = 0; ((drm_r300_cmd_header_t *) rmesa->hw.vps.cmd)->vpu.count = 0; + R300_STATECHANGE(rmesa, vap_flush); R300_STATECHANGE(rmesa, vpp); param_count = r300VertexProgUpdateParams(ctx, prog, (float *)&rmesa->hw.vpp.cmd[R300_VPP_PARAM_0]); bump_vpu_count(rmesa->hw.vpp.cmd, param_count); diff --git a/src/mesa/drivers/dri/r600/r700_render.c b/src/mesa/drivers/dri/r600/r700_render.c index cdf3220a7f3..d64e921bdad 100644 --- a/src/mesa/drivers/dri/r600/r700_render.c +++ b/src/mesa/drivers/dri/r600/r700_render.c @@ -53,6 +53,8 @@ #include "r700_fragprog.h" #include "r700_state.h" +#include "radeon_common_context.h" + void r700WaitForIdle(context_t *context); void r700WaitForIdleClean(context_t *context); GLboolean r700SendTextureState(context_t *context); @@ -298,22 +300,52 @@ static void r700RunRenderPrimitive(GLcontext * ctx, int start, int end, int prim } +/* start 3d, idle, cb/db flush */ +#define PRE_EMIT_STATE_BUFSZ 10 + 5 + 14 + +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) { context_t *context = R700_CONTEXT(ctx); radeonContextPtr radeon = &context->radeon; - unsigned int i, ind_count = 0, id = 0; + unsigned int i, id = 0; TNLcontext *tnl = TNL_CONTEXT(ctx); struct vertex_buffer *vb = &tnl->vb; struct radeon_renderbuffer *rrb; - for (i = 0; i < vb->PrimitiveCount; i++) - ind_count += vb->Primitive[i].count + 10; - - /* just an estimate, need to properly calculate this */ - rcommonEnsureCmdBufSpace(&context->radeon, - radeon->hw.max_state_size + ind_count, __FUNCTION__); + if (RADEON_DEBUG & DEBUG_PRIMS) + fprintf(stderr, "%s: cs begin at %d\n", + __func__, context->radeon.cmdbuf.cs->cdw); /* always emit CB base to prevent * lock ups on some chips. @@ -327,6 +359,9 @@ static GLboolean r700RunRender(GLcontext * ctx, r700SetupVertexProgram(ctx); r700SetupFragmentProgram(ctx); r600UpdateTextureState(ctx); + + GLuint emit_end = r700PredictRenderSize(ctx) + + context->radeon.cmdbuf.cs->cdw; r700SetupStreams(ctx); radeonEmitState(radeon); @@ -336,7 +371,7 @@ static GLboolean r700RunRender(GLcontext * ctx, 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. */ @@ -354,6 +389,14 @@ static GLboolean r700RunRender(GLcontext * ctx, 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; } diff --git a/src/mesa/drivers/dri/radeon/Makefile b/src/mesa/drivers/dri/radeon/Makefile index 6bf67d2ea59..8fe9d98a0b2 100644 --- a/src/mesa/drivers/dri/radeon/Makefile +++ b/src/mesa/drivers/dri/radeon/Makefile @@ -46,7 +46,8 @@ C_SOURCES = \ $(DRIVER_SOURCES) \ $(CS_SOURCES) -DRIVER_DEFINES = -DRADEON_COMMON=0 +DRIVER_DEFINES = -DRADEON_COMMON=0 \ + -Wall DRI_LIB_DEPS += $(RADEON_LDFLAGS) diff --git a/src/mesa/drivers/dri/radeon/radeon_cmdbuf.h b/src/mesa/drivers/dri/radeon/radeon_cmdbuf.h index c31421c253a..410df4d665c 100644 --- a/src/mesa/drivers/dri/radeon/radeon_cmdbuf.h +++ b/src/mesa/drivers/dri/radeon/radeon_cmdbuf.h @@ -3,7 +3,7 @@ #include "radeon_bocs_wrapper.h" -void rcommonEnsureCmdBufSpace(radeonContextPtr rmesa, int dwords, const char *caller); +GLboolean rcommonEnsureCmdBufSpace(radeonContextPtr rmesa, int dwords, const char *caller); int rcommonFlushCmdBuf(radeonContextPtr rmesa, const char *caller); int rcommonFlushCmdBufLocked(radeonContextPtr rmesa, const char *caller); void rcommonInitCmdBuf(radeonContextPtr rmesa); @@ -71,7 +71,7 @@ void rcommonBeginBatch(radeonContextPtr rmesa, */ #define OUT_BATCH_TABLE(ptr,n) \ do { \ - radeon_cs_write_table(b_l_rmesa->cmdbuf.cs, ptr, n); \ + radeon_cs_write_table(b_l_rmesa->cmdbuf.cs, (ptr), (n));\ } while(0) /** diff --git a/src/mesa/drivers/dri/radeon/radeon_common.c b/src/mesa/drivers/dri/radeon/radeon_common.c index b5b4fed8fa8..8f34fbf6bc4 100644 --- a/src/mesa/drivers/dri/radeon/radeon_common.c +++ b/src/mesa/drivers/dri/radeon/radeon_common.c @@ -895,7 +895,7 @@ void radeon_viewport(GLcontext *ctx, GLint x, GLint y, GLsizei width, GLsizei he ctx->Driver.Viewport = old_viewport; } -static void radeon_print_state_atom(radeonContextPtr radeon, struct radeon_state_atom *state) +static void radeon_print_state_atom_prekmm(radeonContextPtr radeon, struct radeon_state_atom *state) { int i, j, reg; int dwords = (*state->check) (radeon->glCtx, state); @@ -904,6 +904,9 @@ static void radeon_print_state_atom(radeonContextPtr radeon, struct radeon_state fprintf(stderr, " emit %s %d/%d\n", state->name, dwords, state->cmd_size); if (RADEON_DEBUG & DEBUG_VERBOSE) { + if (dwords > state->cmd_size) + dwords = state->cmd_size; + for (i = 0; i < dwords;) { cmd = *((drm_r300_cmd_header_t *) &state->cmd[i]); reg = (cmd.packet0.reghi << 8) | cmd.packet0.reglo; @@ -920,16 +923,27 @@ static void radeon_print_state_atom(radeonContextPtr radeon, struct radeon_state } } -static void radeon_print_state_atom_kmm(radeonContextPtr radeon, struct radeon_state_atom *state) +static void radeon_print_state_atom(radeonContextPtr radeon, struct radeon_state_atom *state) { int i, j, reg, count; - int dwords = (*state->check) (radeon->glCtx, state); + int dwords; uint32_t packet0; + if (! (DEBUG_CMDBUF && RADEON_DEBUG & DEBUG_STATE) ) + return; + + if (!radeon->radeonScreen->kernel_mm) { + radeon_print_state_atom_prekmm(radeon, state); + return; + } + + dwords = (*state->check) (radeon->glCtx, state); fprintf(stderr, " emit %s %d/%d\n", state->name, dwords, state->cmd_size); if (RADEON_DEBUG & DEBUG_VERBOSE) { - for (i = 0; i < dwords;) { + if (dwords > state->cmd_size) + dwords = state->cmd_size; + for (i = 0; i < state->cmd_size;) { packet0 = state->cmd[i]; reg = (packet0 & 0x1FFF) << 2; count = ((packet0 & 0x3FFF0000) >> 16) + 1; @@ -946,40 +960,84 @@ static void radeon_print_state_atom_kmm(radeonContextPtr radeon, struct radeon_s } } -static INLINE void radeonEmitAtoms(radeonContextPtr radeon, GLboolean dirty) +/** + * Count total size for next state emit. + **/ +GLuint radeonCountStateEmitSize(radeonContextPtr radeon) { - BATCH_LOCALS(radeon); struct radeon_state_atom *atom; + GLuint dwords = 0; + /* check if we are going to emit full state */ + if (DEBUG_CMDBUF && RADEON_DEBUG & DEBUG_VERBOSE) + fprintf(stderr, "%s\n", __func__); + + if (radeon->cmdbuf.cs->cdw && !radeon->hw.all_dirty) { + if (!radeon->hw.is_dirty) + return dwords; + foreach(atom, &radeon->hw.atomlist) { + if (atom->dirty) { + const GLuint atom_size = atom->check(radeon->glCtx, atom); + dwords += atom_size; + if (DEBUG_CMDBUF && atom_size) { + radeon_print_state_atom(radeon, atom); + } + } + } + } else { + foreach(atom, &radeon->hw.atomlist) { + const GLuint atom_size = atom->check(radeon->glCtx, atom); + dwords += atom_size; + if (DEBUG_CMDBUF && atom_size) { + radeon_print_state_atom(radeon, atom); + } + + } + } + return dwords; +} + +static INLINE void radeon_emit_atom(radeonContextPtr radeon, struct radeon_state_atom *atom) +{ + BATCH_LOCALS(radeon); int dwords; + dwords = (*atom->check) (radeon->glCtx, atom); + if (dwords) { + + radeon_print_state_atom(radeon, atom); + + if (atom->emit) { + (*atom->emit)(radeon->glCtx, atom); + } else { + BEGIN_BATCH_NO_AUTOSTATE(dwords); + OUT_BATCH_TABLE(atom->cmd, dwords); + END_BATCH(); + } + } else { + if (DEBUG_CMDBUF && RADEON_DEBUG & DEBUG_STATE) { + fprintf(stderr, " skip state %s\n", + atom->name); + } + } + atom->dirty = GL_FALSE; + +} + +static INLINE void radeonEmitAtoms(radeonContextPtr radeon, GLboolean emitAll) +{ + struct radeon_state_atom *atom; + if (radeon->vtbl.pre_emit_atoms) radeon->vtbl.pre_emit_atoms(radeon); /* Emit actual atoms */ - foreach(atom, &radeon->hw.atomlist) { - if ((atom->dirty || radeon->hw.all_dirty) == dirty) { - dwords = (*atom->check) (radeon->glCtx, atom); - if (dwords) { - if (DEBUG_CMDBUF && RADEON_DEBUG & DEBUG_STATE) { - if (radeon->radeonScreen->kernel_mm) - radeon_print_state_atom_kmm(radeon, atom); - else - radeon_print_state_atom(radeon, atom); - } - if (atom->emit) { - (*atom->emit)(radeon->glCtx, atom); - } else { - BEGIN_BATCH_NO_AUTOSTATE(dwords); - OUT_BATCH_TABLE(atom->cmd, dwords); - END_BATCH(); - } - atom->dirty = GL_FALSE; - } else { - if (DEBUG_CMDBUF && RADEON_DEBUG & DEBUG_STATE) { - fprintf(stderr, " skip state %s\n", - atom->name); - } - } + if (radeon->hw.all_dirty || emitAll) { + foreach(atom, &radeon->hw.atomlist) + radeon_emit_atom( radeon, atom ); + } else { + foreach(atom, &radeon->hw.atomlist) { + if ( atom->dirty ) + radeon_emit_atom( radeon, atom ); } } @@ -1009,26 +1067,21 @@ void radeonEmitState(radeonContextPtr radeon) if (radeon->cmdbuf.cs->cdw && !radeon->hw.is_dirty && !radeon->hw.all_dirty) return; - /* To avoid going across the entire set of states multiple times, just check - * for enough space for the case of emitting all state, and inline the - * radeonAllocCmdBuf code here without all the checks. - */ - rcommonEnsureCmdBufSpace(radeon, radeon->hw.max_state_size, __FUNCTION__); - if (!radeon->cmdbuf.cs->cdw) { if (RADEON_DEBUG & DEBUG_STATE) fprintf(stderr, "Begin reemit state\n"); + radeonEmitAtoms(radeon, GL_TRUE); + } else { + + if (RADEON_DEBUG & DEBUG_STATE) + fprintf(stderr, "Begin dirty state\n"); + radeonEmitAtoms(radeon, GL_FALSE); } - if (RADEON_DEBUG & DEBUG_STATE) - fprintf(stderr, "Begin dirty state\n"); - - radeonEmitAtoms(radeon, GL_TRUE); radeon->hw.is_dirty = GL_FALSE; radeon->hw.all_dirty = GL_FALSE; - } @@ -1172,12 +1225,16 @@ int rcommonFlushCmdBuf(radeonContextPtr rmesa, const char *caller) * * \param dwords The number of dwords we need to be free on the command buffer */ -void rcommonEnsureCmdBufSpace(radeonContextPtr rmesa, int dwords, const char *caller) +GLboolean rcommonEnsureCmdBufSpace(radeonContextPtr rmesa, int dwords, const char *caller) { - if ((rmesa->cmdbuf.cs->cdw + dwords + 128) > rmesa->cmdbuf.size || - radeon_cs_need_flush(rmesa->cmdbuf.cs)) { - rcommonFlushCmdBuf(rmesa, caller); - } + if ((rmesa->cmdbuf.cs->cdw + dwords + 128) > rmesa->cmdbuf.size + || radeon_cs_need_flush(rmesa->cmdbuf.cs)) { + /* If we try to flush empty buffer there is too big rendering operation. */ + assert(rmesa->cmdbuf.cs->cdw); + rcommonFlushCmdBuf(rmesa, caller); + return GL_TRUE; + } + return GL_FALSE; } void rcommonInitCmdBuf(radeonContextPtr rmesa) @@ -1252,7 +1309,6 @@ void rcommonBeginBatch(radeonContextPtr rmesa, int n, const char *function, int line) { - rcommonEnsureCmdBufSpace(rmesa, n, function); if (!rmesa->cmdbuf.cs->cdw && dostate) { if (RADEON_DEBUG & DEBUG_IOCTL) fprintf(stderr, "Reemit state after flush (from %s)\n", function); diff --git a/src/mesa/drivers/dri/radeon/radeon_common.h b/src/mesa/drivers/dri/radeon/radeon_common.h index cebae18b2d6..a9e1ca49eb9 100644 --- a/src/mesa/drivers/dri/radeon/radeon_common.h +++ b/src/mesa/drivers/dri/radeon/radeon_common.h @@ -24,6 +24,7 @@ void radeonUpdatePageFlipping(radeonContextPtr rmesa); void radeonFlush(GLcontext *ctx); void radeonFinish(GLcontext * ctx); void radeonEmitState(radeonContextPtr radeon); +GLuint radeonCountStateEmitSize(radeonContextPtr radeon); void radeon_clear_tris(GLcontext *ctx, GLbitfield mask); diff --git a/src/mesa/drivers/dri/radeon/radeon_dma.c b/src/mesa/drivers/dri/radeon/radeon_dma.c index a1835427f19..f306befec47 100644 --- a/src/mesa/drivers/dri/radeon/radeon_dma.c +++ b/src/mesa/drivers/dri/radeon/radeon_dma.c @@ -438,7 +438,7 @@ restart: if (!rmesa->dma.flush) { /* make sure we have enough space to use this in cmdbuf */ rcommonEnsureCmdBufSpace(rmesa, - rmesa->hw.max_state_size + (20*sizeof(int)), + radeonCountStateEmitSize( rmesa ) + (20*sizeof(int)), __FUNCTION__); /* if cmdbuf flushed DMA restart */ if (is_empty_list(&rmesa->dma.reserved)) diff --git a/src/mesa/drivers/dri/radeon/radeon_ioctl.h b/src/mesa/drivers/dri/radeon/radeon_ioctl.h index 18805d4c571..deb53ae3138 100644 --- a/src/mesa/drivers/dri/radeon/radeon_ioctl.h +++ b/src/mesa/drivers/dri/radeon/radeon_ioctl.h @@ -132,16 +132,18 @@ static INLINE int RADEON_DB_STATECHANGE(r100ContextPtr rmesa, * r200EmitState is called from within r200EmitVbufPrim and r200FlushElts. */ #if RADEON_OLD_PACKETS -#define AOS_BUFSZ(nr) ((3 + ((nr / 2) * 3) + ((nr & 1) * 2)) * sizeof(int)) +#define AOS_BUFSZ(nr) ((3 + ((nr / 2) * 3) + ((nr & 1) * 2))+nr*2) #define VERT_AOS_BUFSZ (0) #define ELTS_BUFSZ(nr) (24 + nr * 2) -#define VBUF_BUFSZ (6 * sizeof(int)) +#define VBUF_BUFSZ (8) #else -#define AOS_BUFSZ(nr) ((3 + ((nr / 2) * 3) + ((nr & 1) * 2)) * sizeof(int)) -#define VERT_AOS_BUFSZ (5 * sizeof(int)) +#define AOS_BUFSZ(nr) ((3 + ((nr / 2) * 3) + ((nr & 1) * 2) + nr*2)) +#define VERT_AOS_BUFSZ (5) #define ELTS_BUFSZ(nr) (16 + nr * 2) -#define VBUF_BUFSZ (4 * sizeof(int)) +#define VBUF_BUFSZ (4) #endif +#define SCISSOR_BUFSZ (8) +#define INDEX_BUFSZ (7) static inline uint32_t cmdpacket3(int cmd_type) diff --git a/src/mesa/drivers/dri/radeon/radeon_state_init.c b/src/mesa/drivers/dri/radeon/radeon_state_init.c index 501ea0b66b1..b34a4e803d8 100644 --- a/src/mesa/drivers/dri/radeon/radeon_state_init.c +++ b/src/mesa/drivers/dri/radeon/radeon_state_init.c @@ -197,55 +197,85 @@ static int cmdscl( int offset, int stride, int count ) return h.i; } -#define CHECK( NM, FLAG ) \ +#define CHECK( NM, FLAG, ADD ) \ static int check_##NM( GLcontext *ctx, struct radeon_state_atom *atom ) \ { \ - return FLAG ? atom->cmd_size : 0; \ + return FLAG ? atom->cmd_size + (ADD) : 0; \ } -#define TCL_CHECK( NM, FLAG ) \ +#define TCL_CHECK( NM, FLAG, ADD ) \ static int check_##NM( GLcontext *ctx, struct radeon_state_atom *atom ) \ { \ r100ContextPtr rmesa = R100_CONTEXT(ctx); \ - return (!rmesa->radeon.TclFallback && (FLAG)) ? atom->cmd_size : 0; \ + return (!rmesa->radeon.TclFallback && (FLAG)) ? atom->cmd_size + (ADD) : 0; \ } -CHECK( always, GL_TRUE ) -CHECK( never, GL_FALSE ) -CHECK( tex0, ctx->Texture.Unit[0]._ReallyEnabled ) -CHECK( tex1, ctx->Texture.Unit[1]._ReallyEnabled ) +CHECK( always, GL_TRUE, 0 ) +CHECK( always_add2, GL_TRUE, 2 ) +CHECK( always_add4, GL_TRUE, 4 ) +CHECK( never, GL_FALSE, 0 ) +CHECK( tex0_mm, ctx->Texture.Unit[0]._ReallyEnabled, 3 ) +CHECK( tex1_mm, ctx->Texture.Unit[1]._ReallyEnabled, 3 ) /* need this for the cubic_map on disabled unit 2 bug, maybe r100 only? */ -CHECK( tex2, ctx->Texture._EnabledUnits ) -CHECK( cube0, (ctx->Texture.Unit[0]._ReallyEnabled & TEXTURE_CUBE_BIT)) -CHECK( cube1, (ctx->Texture.Unit[1]._ReallyEnabled & TEXTURE_CUBE_BIT)) -CHECK( cube2, (ctx->Texture.Unit[2]._ReallyEnabled & TEXTURE_CUBE_BIT)) -CHECK( fog, ctx->Fog.Enabled ) -TCL_CHECK( tcl, GL_TRUE ) -TCL_CHECK( tcl_tex0, ctx->Texture.Unit[0]._ReallyEnabled ) -TCL_CHECK( tcl_tex1, ctx->Texture.Unit[1]._ReallyEnabled ) -TCL_CHECK( tcl_tex2, ctx->Texture.Unit[2]._ReallyEnabled ) -TCL_CHECK( tcl_lighting, ctx->Light.Enabled ) -TCL_CHECK( tcl_eyespace_or_lighting, ctx->_NeedEyeCoords || ctx->Light.Enabled ) -TCL_CHECK( tcl_lit0, ctx->Light.Enabled && ctx->Light.Light[0].Enabled ) -TCL_CHECK( tcl_lit1, ctx->Light.Enabled && ctx->Light.Light[1].Enabled ) -TCL_CHECK( tcl_lit2, ctx->Light.Enabled && ctx->Light.Light[2].Enabled ) -TCL_CHECK( tcl_lit3, ctx->Light.Enabled && ctx->Light.Light[3].Enabled ) -TCL_CHECK( tcl_lit4, ctx->Light.Enabled && ctx->Light.Light[4].Enabled ) -TCL_CHECK( tcl_lit5, ctx->Light.Enabled && ctx->Light.Light[5].Enabled ) -TCL_CHECK( tcl_lit6, ctx->Light.Enabled && ctx->Light.Light[6].Enabled ) -TCL_CHECK( tcl_lit7, ctx->Light.Enabled && ctx->Light.Light[7].Enabled ) -TCL_CHECK( tcl_ucp0, (ctx->Transform.ClipPlanesEnabled & 0x1) ) -TCL_CHECK( tcl_ucp1, (ctx->Transform.ClipPlanesEnabled & 0x2) ) -TCL_CHECK( tcl_ucp2, (ctx->Transform.ClipPlanesEnabled & 0x4) ) -TCL_CHECK( tcl_ucp3, (ctx->Transform.ClipPlanesEnabled & 0x8) ) -TCL_CHECK( tcl_ucp4, (ctx->Transform.ClipPlanesEnabled & 0x10) ) -TCL_CHECK( tcl_ucp5, (ctx->Transform.ClipPlanesEnabled & 0x20) ) -TCL_CHECK( tcl_eyespace_or_fog, ctx->_NeedEyeCoords || ctx->Fog.Enabled ) - -CHECK( txr0, (ctx->Texture.Unit[0]._ReallyEnabled & TEXTURE_RECT_BIT)) -CHECK( txr1, (ctx->Texture.Unit[1]._ReallyEnabled & TEXTURE_RECT_BIT)) -CHECK( txr2, (ctx->Texture.Unit[2]._ReallyEnabled & TEXTURE_RECT_BIT)) +CHECK( tex2_mm, ctx->Texture._EnabledUnits, 3 ) +CHECK( tex0, ctx->Texture.Unit[0]._ReallyEnabled, 2 ) +CHECK( tex1, ctx->Texture.Unit[1]._ReallyEnabled, 2 ) +CHECK( tex2, ctx->Texture._EnabledUnits, 2 ) +CHECK( cube0, (ctx->Texture.Unit[0]._ReallyEnabled & TEXTURE_CUBE_BIT), 3 + 3*5 - CUBE_STATE_SIZE ) +CHECK( cube1, (ctx->Texture.Unit[1]._ReallyEnabled & TEXTURE_CUBE_BIT), 3 + 3*5 - CUBE_STATE_SIZE ) +CHECK( cube2, (ctx->Texture.Unit[2]._ReallyEnabled & TEXTURE_CUBE_BIT), 3 + 3*5 - CUBE_STATE_SIZE ) +CHECK( cube0_mm, (ctx->Texture.Unit[0]._ReallyEnabled & TEXTURE_CUBE_BIT), 2 + 4*5 - CUBE_STATE_SIZE ) +CHECK( cube1_mm, (ctx->Texture.Unit[1]._ReallyEnabled & TEXTURE_CUBE_BIT), 2 + 4*5 - CUBE_STATE_SIZE ) +CHECK( cube2_mm, (ctx->Texture.Unit[2]._ReallyEnabled & TEXTURE_CUBE_BIT), 2 + 4*5 - CUBE_STATE_SIZE ) +CHECK( fog, ctx->Fog.Enabled, 0 ) +CHECK( fog_add4, ctx->Fog.Enabled, 4 ) +TCL_CHECK( tcl, GL_TRUE, 0 ) +TCL_CHECK( tcl_add4, GL_TRUE, 4 ) +TCL_CHECK( tcl_tex0, ctx->Texture.Unit[0]._ReallyEnabled, 0 ) +TCL_CHECK( tcl_tex1, ctx->Texture.Unit[1]._ReallyEnabled, 0 ) +TCL_CHECK( tcl_tex2, ctx->Texture.Unit[2]._ReallyEnabled, 0 ) +TCL_CHECK( tcl_tex0_add4, ctx->Texture.Unit[0]._ReallyEnabled, 4 ) +TCL_CHECK( tcl_tex1_add4, ctx->Texture.Unit[1]._ReallyEnabled, 4 ) +TCL_CHECK( tcl_tex2_add4, ctx->Texture.Unit[2]._ReallyEnabled, 4 ) +TCL_CHECK( tcl_lighting, ctx->Light.Enabled, 0 ) +TCL_CHECK( tcl_lighting_add4, ctx->Light.Enabled, 4 ) +TCL_CHECK( tcl_eyespace_or_lighting, ctx->_NeedEyeCoords || ctx->Light.Enabled, 0 ) +TCL_CHECK( tcl_eyespace_or_lighting_add4, ctx->_NeedEyeCoords || ctx->Light.Enabled, 4 ) +TCL_CHECK( tcl_lit0, ctx->Light.Enabled && ctx->Light.Light[0].Enabled, 0 ) +TCL_CHECK( tcl_lit1, ctx->Light.Enabled && ctx->Light.Light[1].Enabled, 0 ) +TCL_CHECK( tcl_lit2, ctx->Light.Enabled && ctx->Light.Light[2].Enabled, 0 ) +TCL_CHECK( tcl_lit3, ctx->Light.Enabled && ctx->Light.Light[3].Enabled, 0 ) +TCL_CHECK( tcl_lit4, ctx->Light.Enabled && ctx->Light.Light[4].Enabled, 0 ) +TCL_CHECK( tcl_lit5, ctx->Light.Enabled && ctx->Light.Light[5].Enabled, 0 ) +TCL_CHECK( tcl_lit6, ctx->Light.Enabled && ctx->Light.Light[6].Enabled, 0 ) +TCL_CHECK( tcl_lit7, ctx->Light.Enabled && ctx->Light.Light[7].Enabled, 0 ) +TCL_CHECK( tcl_lit0_add6, ctx->Light.Enabled && ctx->Light.Light[0].Enabled, 6 ) +TCL_CHECK( tcl_lit1_add6, ctx->Light.Enabled && ctx->Light.Light[1].Enabled, 6 ) +TCL_CHECK( tcl_lit2_add6, ctx->Light.Enabled && ctx->Light.Light[2].Enabled, 6 ) +TCL_CHECK( tcl_lit3_add6, ctx->Light.Enabled && ctx->Light.Light[3].Enabled, 6 ) +TCL_CHECK( tcl_lit4_add6, ctx->Light.Enabled && ctx->Light.Light[4].Enabled, 6 ) +TCL_CHECK( tcl_lit5_add6, ctx->Light.Enabled && ctx->Light.Light[5].Enabled, 6 ) +TCL_CHECK( tcl_lit6_add6, ctx->Light.Enabled && ctx->Light.Light[6].Enabled, 6 ) +TCL_CHECK( tcl_lit7_add6, ctx->Light.Enabled && ctx->Light.Light[7].Enabled, 6 ) +TCL_CHECK( tcl_ucp0, (ctx->Transform.ClipPlanesEnabled & 0x1), 0 ) +TCL_CHECK( tcl_ucp1, (ctx->Transform.ClipPlanesEnabled & 0x2), 0 ) +TCL_CHECK( tcl_ucp2, (ctx->Transform.ClipPlanesEnabled & 0x4), 0 ) +TCL_CHECK( tcl_ucp3, (ctx->Transform.ClipPlanesEnabled & 0x8), 0 ) +TCL_CHECK( tcl_ucp4, (ctx->Transform.ClipPlanesEnabled & 0x10), 0 ) +TCL_CHECK( tcl_ucp5, (ctx->Transform.ClipPlanesEnabled & 0x20), 0 ) +TCL_CHECK( tcl_ucp0_add4, (ctx->Transform.ClipPlanesEnabled & 0x1), 4 ) +TCL_CHECK( tcl_ucp1_add4, (ctx->Transform.ClipPlanesEnabled & 0x2), 4 ) +TCL_CHECK( tcl_ucp2_add4, (ctx->Transform.ClipPlanesEnabled & 0x4), 4 ) +TCL_CHECK( tcl_ucp3_add4, (ctx->Transform.ClipPlanesEnabled & 0x8), 4 ) +TCL_CHECK( tcl_ucp4_add4, (ctx->Transform.ClipPlanesEnabled & 0x10), 4 ) +TCL_CHECK( tcl_ucp5_add4, (ctx->Transform.ClipPlanesEnabled & 0x20), 4 ) +TCL_CHECK( tcl_eyespace_or_fog, ctx->_NeedEyeCoords || ctx->Fog.Enabled, 0 ) +TCL_CHECK( tcl_eyespace_or_fog_add4, ctx->_NeedEyeCoords || ctx->Fog.Enabled, 4 ) + +CHECK( txr0, (ctx->Texture.Unit[0]._ReallyEnabled & TEXTURE_RECT_BIT), 0 ) +CHECK( txr1, (ctx->Texture.Unit[1]._ReallyEnabled & TEXTURE_RECT_BIT), 0 ) +CHECK( txr2, (ctx->Texture.Unit[2]._ReallyEnabled & TEXTURE_RECT_BIT), 0 ) #define OUT_VEC(hdr, data) do { \ drm_radeon_cmd_header_t h; \ @@ -271,9 +301,8 @@ static void scl_emit(GLcontext *ctx, struct radeon_state_atom *atom) { r100ContextPtr r100 = R100_CONTEXT(ctx); BATCH_LOCALS(&r100->radeon); - uint32_t dwords = atom->cmd_size; + uint32_t dwords = atom->check(ctx, atom); - dwords += 2; BEGIN_BATCH_NO_AUTOSTATE(dwords); OUT_SCL(atom->cmd[0], atom->cmd+1); END_BATCH(); @@ -284,9 +313,8 @@ static void vec_emit(GLcontext *ctx, struct radeon_state_atom *atom) { r100ContextPtr r100 = R100_CONTEXT(ctx); BATCH_LOCALS(&r100->radeon); - uint32_t dwords = atom->cmd_size; + uint32_t dwords = atom->check(ctx, atom); - dwords += 4; BEGIN_BATCH_NO_AUTOSTATE(dwords); OUT_VEC(atom->cmd[0], atom->cmd+1); END_BATCH(); @@ -297,9 +325,8 @@ static void lit_emit(GLcontext *ctx, struct radeon_state_atom *atom) { r100ContextPtr r100 = R100_CONTEXT(ctx); BATCH_LOCALS(&r100->radeon); - uint32_t dwords = atom->cmd_size; + uint32_t dwords = atom->check(ctx, atom); - dwords += 6; BEGIN_BATCH_NO_AUTOSTATE(dwords); OUT_VEC(atom->cmd[LIT_CMD_0], atom->cmd+1); OUT_SCL(atom->cmd[LIT_CMD_1], atom->cmd+LIT_CMD_1+1); @@ -313,10 +340,10 @@ static void ctx_emit(GLcontext *ctx, struct radeon_state_atom *atom) struct radeon_renderbuffer *rrb; uint32_t cbpitch; uint32_t zbpitch, depth_fmt; - uint32_t dwords = atom->cmd_size; + uint32_t dwords = atom->check(ctx, atom); /* output the first 7 bytes of context */ - BEGIN_BATCH_NO_AUTOSTATE(dwords + 4); + BEGIN_BATCH_NO_AUTOSTATE(dwords); OUT_BATCH_TABLE(atom->cmd, 5); rrb = radeon_get_depthbuffer(&r100->radeon); @@ -371,6 +398,28 @@ static void ctx_emit(GLcontext *ctx, struct radeon_state_atom *atom) END_BATCH(); } +static int check_always_ctx( GLcontext *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 ctx_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom) { r100ContextPtr r100 = R100_CONTEXT(ctx); @@ -378,7 +427,7 @@ static void ctx_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom) struct radeon_renderbuffer *rrb, *drb; uint32_t cbpitch = 0; uint32_t zbpitch = 0; - uint32_t dwords = atom->cmd_size; + uint32_t dwords = atom->check(ctx, atom); uint32_t depth_fmt; rrb = radeon_get_colorbuffer(&r100->radeon); @@ -418,12 +467,6 @@ static void ctx_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom) } - /* output the first 7 bytes of context */ - dwords = 10; - if (drb) - dwords += 6; - if (rrb) - dwords += 8; BEGIN_BATCH_NO_AUTOSTATE(dwords); /* In the CS case we need to split this up */ @@ -474,7 +517,7 @@ static void cube_emit(GLcontext *ctx, struct radeon_state_atom *atom) { r100ContextPtr r100 = R100_CONTEXT(ctx); BATCH_LOCALS(&r100->radeon); - uint32_t dwords = 3; + uint32_t dwords = atom->check(ctx, atom); int i = atom->idx, j; radeonTexObj *t = r100->state.texture.unit[i].texobj; radeon_mipmap_level *lvl; @@ -488,7 +531,7 @@ static void cube_emit(GLcontext *ctx, struct radeon_state_atom *atom) if (!t->mt) return; - BEGIN_BATCH_NO_AUTOSTATE(dwords + (5 * 3)); + BEGIN_BATCH_NO_AUTOSTATE(dwords); OUT_BATCH_TABLE(atom->cmd, 3); lvl = &t->mt->levels[0]; for (j = 0; j < 5; j++) { @@ -502,7 +545,7 @@ static void cube_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom) { r100ContextPtr r100 = R100_CONTEXT(ctx); BATCH_LOCALS(&r100->radeon); - uint32_t dwords = 2; + uint32_t dwords = atom->check(ctx, atom); int i = atom->idx, j; radeonTexObj *t = r100->state.texture.unit[i].texobj; radeon_mipmap_level *lvl; @@ -523,7 +566,7 @@ static void cube_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom) default: case 0: base_reg = RADEON_PP_CUBIC_OFFSET_T0_0; break; }; - BEGIN_BATCH_NO_AUTOSTATE(dwords + (5 * 4)); + BEGIN_BATCH_NO_AUTOSTATE(dwords); OUT_BATCH_TABLE(atom->cmd, 2); lvl = &t->mt->levels[0]; for (j = 0; j < 5; j++) { @@ -665,10 +708,11 @@ void radeonInitState( r100ContextPtr rmesa ) /* Allocate state buffers: */ - ALLOC_STATE( ctx, always, CTX_STATE_SIZE, "CTX/context", 0 ); - if (rmesa->radeon.radeonScreen->kernel_mm) + ALLOC_STATE( ctx, always_add4, CTX_STATE_SIZE, "CTX/context", 0 ); + if (rmesa->radeon.radeonScreen->kernel_mm) { rmesa->hw.ctx.emit = ctx_emit_cs; - else + rmesa->hw.ctx.check = check_always_ctx; + } else rmesa->hw.ctx.emit = ctx_emit; ALLOC_STATE( lin, always, LIN_STATE_SIZE, "LIN/line", 0 ); ALLOC_STATE( msk, always, MSK_STATE_SIZE, "MSK/mask", 0 ); @@ -678,13 +722,63 @@ void radeonInitState( r100ContextPtr rmesa ) ALLOC_STATE( zbs, always, ZBS_STATE_SIZE, "ZBS/zbias", 0 ); ALLOC_STATE( tcl, always, TCL_STATE_SIZE, "TCL/tcl", 1 ); ALLOC_STATE( mtl, tcl_lighting, MTL_STATE_SIZE, "MTL/material", 1 ); - ALLOC_STATE( grd, always, GRD_STATE_SIZE, "GRD/guard-band", 1 ); - ALLOC_STATE( fog, fog, FOG_STATE_SIZE, "FOG/fog", 1 ); - ALLOC_STATE( glt, tcl_lighting, GLT_STATE_SIZE, "GLT/light-global", 1 ); - ALLOC_STATE( eye, tcl_lighting, EYE_STATE_SIZE, "EYE/eye-vector", 1 ); - ALLOC_STATE_IDX( tex[0], tex0, TEX_STATE_SIZE, "TEX/tex-0", 0, 0); - ALLOC_STATE_IDX( tex[1], tex1, TEX_STATE_SIZE, "TEX/tex-1", 0, 1); - ALLOC_STATE_IDX( tex[2], tex2, TEX_STATE_SIZE, "TEX/tex-2", 0, 2); + if (rmesa->radeon.radeonScreen->kernel_mm) { + ALLOC_STATE( grd, always_add2, GRD_STATE_SIZE, "GRD/guard-band", 1 ); + ALLOC_STATE( fog, fog_add4, FOG_STATE_SIZE, "FOG/fog", 1 ); + ALLOC_STATE( glt, tcl_lighting_add4, GLT_STATE_SIZE, "GLT/light-global", 1 ); + ALLOC_STATE( eye, tcl_lighting_add4, EYE_STATE_SIZE, "EYE/eye-vector", 1 ); + ALLOC_STATE_IDX( tex[0], tex0_mm, TEX_STATE_SIZE, "TEX/tex-0", 0, 0); + ALLOC_STATE_IDX( tex[1], tex1_mm, TEX_STATE_SIZE, "TEX/tex-1", 0, 1); + ALLOC_STATE_IDX( tex[2], tex2_mm, TEX_STATE_SIZE, "TEX/tex-2", 0, 2); + ALLOC_STATE( mat[0], tcl_add4, MAT_STATE_SIZE, "MAT/modelproject", 1 ); + ALLOC_STATE( mat[1], tcl_eyespace_or_fog_add4, MAT_STATE_SIZE, "MAT/modelview", 1 ); + ALLOC_STATE( mat[2], tcl_eyespace_or_lighting_add4, MAT_STATE_SIZE, "MAT/it-modelview", 1 ); + ALLOC_STATE( mat[3], tcl_tex0_add4, MAT_STATE_SIZE, "MAT/texmat0", 1 ); + ALLOC_STATE( mat[4], tcl_tex1_add4, MAT_STATE_SIZE, "MAT/texmat1", 1 ); + ALLOC_STATE( mat[5], tcl_tex2_add4, MAT_STATE_SIZE, "MAT/texmat2", 1 ); + ALLOC_STATE( lit[0], tcl_lit0_add6, LIT_STATE_SIZE, "LIT/light-0", 1 ); + ALLOC_STATE( lit[1], tcl_lit1_add6, LIT_STATE_SIZE, "LIT/light-1", 1 ); + ALLOC_STATE( lit[2], tcl_lit2_add6, LIT_STATE_SIZE, "LIT/light-2", 1 ); + ALLOC_STATE( lit[3], tcl_lit3_add6, LIT_STATE_SIZE, "LIT/light-3", 1 ); + ALLOC_STATE( lit[4], tcl_lit4_add6, LIT_STATE_SIZE, "LIT/light-4", 1 ); + ALLOC_STATE( lit[5], tcl_lit5_add6, LIT_STATE_SIZE, "LIT/light-5", 1 ); + ALLOC_STATE( lit[6], tcl_lit6_add6, LIT_STATE_SIZE, "LIT/light-6", 1 ); + ALLOC_STATE( lit[7], tcl_lit7_add6, LIT_STATE_SIZE, "LIT/light-7", 1 ); + ALLOC_STATE( ucp[0], tcl_ucp0_add4, UCP_STATE_SIZE, "UCP/userclip-0", 1 ); + ALLOC_STATE( ucp[1], tcl_ucp1_add4, UCP_STATE_SIZE, "UCP/userclip-1", 1 ); + ALLOC_STATE( ucp[2], tcl_ucp2_add4, UCP_STATE_SIZE, "UCP/userclip-2", 1 ); + ALLOC_STATE( ucp[3], tcl_ucp3_add4, UCP_STATE_SIZE, "UCP/userclip-3", 1 ); + ALLOC_STATE( ucp[4], tcl_ucp4_add4, UCP_STATE_SIZE, "UCP/userclip-4", 1 ); + ALLOC_STATE( ucp[5], tcl_ucp5_add4, UCP_STATE_SIZE, "UCP/userclip-5", 1 ); + } else { + ALLOC_STATE( grd, always, GRD_STATE_SIZE, "GRD/guard-band", 1 ); + ALLOC_STATE( fog, fog, FOG_STATE_SIZE, "FOG/fog", 1 ); + ALLOC_STATE( glt, tcl_lighting, GLT_STATE_SIZE, "GLT/light-global", 1 ); + ALLOC_STATE( eye, tcl_lighting, EYE_STATE_SIZE, "EYE/eye-vector", 1 ); + ALLOC_STATE_IDX( tex[0], tex0, TEX_STATE_SIZE, "TEX/tex-0", 0, 0); + ALLOC_STATE_IDX( tex[1], tex1, TEX_STATE_SIZE, "TEX/tex-1", 0, 1); + ALLOC_STATE_IDX( tex[2], tex2, TEX_STATE_SIZE, "TEX/tex-2", 0, 2); + ALLOC_STATE( mat[0], tcl, MAT_STATE_SIZE, "MAT/modelproject", 1 ); + ALLOC_STATE( mat[1], tcl_eyespace_or_fog, MAT_STATE_SIZE, "MAT/modelview", 1 ); + ALLOC_STATE( mat[2], tcl_eyespace_or_lighting, MAT_STATE_SIZE, "MAT/it-modelview", 1 ); + ALLOC_STATE( mat[3], tcl_tex0, MAT_STATE_SIZE, "MAT/texmat0", 1 ); + ALLOC_STATE( mat[4], tcl_tex1, MAT_STATE_SIZE, "MAT/texmat1", 1 ); + ALLOC_STATE( mat[5], tcl_tex2, MAT_STATE_SIZE, "MAT/texmat2", 1 ); + ALLOC_STATE( lit[0], tcl_lit0, LIT_STATE_SIZE, "LIT/light-0", 1 ); + ALLOC_STATE( lit[1], tcl_lit1, LIT_STATE_SIZE, "LIT/light-1", 1 ); + ALLOC_STATE( lit[2], tcl_lit2, LIT_STATE_SIZE, "LIT/light-2", 1 ); + ALLOC_STATE( lit[3], tcl_lit3, LIT_STATE_SIZE, "LIT/light-3", 1 ); + ALLOC_STATE( lit[4], tcl_lit4, LIT_STATE_SIZE, "LIT/light-4", 1 ); + ALLOC_STATE( lit[5], tcl_lit5, LIT_STATE_SIZE, "LIT/light-5", 1 ); + ALLOC_STATE( lit[6], tcl_lit6, LIT_STATE_SIZE, "LIT/light-6", 1 ); + ALLOC_STATE( lit[7], tcl_lit7, LIT_STATE_SIZE, "LIT/light-7", 1 ); + ALLOC_STATE( ucp[0], tcl_ucp0, UCP_STATE_SIZE, "UCP/userclip-0", 1 ); + ALLOC_STATE( ucp[1], tcl_ucp1, UCP_STATE_SIZE, "UCP/userclip-1", 1 ); + ALLOC_STATE( ucp[2], tcl_ucp2, UCP_STATE_SIZE, "UCP/userclip-2", 1 ); + ALLOC_STATE( ucp[3], tcl_ucp3, UCP_STATE_SIZE, "UCP/userclip-3", 1 ); + ALLOC_STATE( ucp[4], tcl_ucp4, UCP_STATE_SIZE, "UCP/userclip-4", 1 ); + ALLOC_STATE( ucp[5], tcl_ucp5, UCP_STATE_SIZE, "UCP/userclip-5", 1 ); + } for (i = 0; i < 3; i++) { if (rmesa->radeon.radeonScreen->kernel_mm) @@ -694,14 +788,19 @@ void radeonInitState( r100ContextPtr rmesa ) } if (rmesa->radeon.radeonScreen->drmSupportsCubeMapsR100) { - ALLOC_STATE_IDX( cube[0], cube0, CUBE_STATE_SIZE, "CUBE/cube-0", 0, 0 ); - ALLOC_STATE_IDX( cube[1], cube1, CUBE_STATE_SIZE, "CUBE/cube-1", 0, 1 ); - ALLOC_STATE_IDX( cube[2], cube2, CUBE_STATE_SIZE, "CUBE/cube-2", 0, 2 ); - for (i = 0; i < 3; i++) - if (rmesa->radeon.radeonScreen->kernel_mm) - rmesa->hw.cube[i].emit = cube_emit_cs; - else - rmesa->hw.cube[i].emit = cube_emit; + if (rmesa->radeon.radeonScreen->kernel_mm) { + ALLOC_STATE_IDX( cube[0], cube0_mm, CUBE_STATE_SIZE, "CUBE/cube-0", 0, 0 ); + ALLOC_STATE_IDX( cube[1], cube1_mm, CUBE_STATE_SIZE, "CUBE/cube-1", 0, 1 ); + ALLOC_STATE_IDX( cube[2], cube2_mm, CUBE_STATE_SIZE, "CUBE/cube-2", 0, 2 ); + for (i = 0; i < 3; i++) + rmesa->hw.cube[i].emit = cube_emit_cs; + } else { + ALLOC_STATE_IDX( cube[0], cube0, CUBE_STATE_SIZE, "CUBE/cube-0", 0, 0 ); + ALLOC_STATE_IDX( cube[1], cube1, CUBE_STATE_SIZE, "CUBE/cube-1", 0, 1 ); + ALLOC_STATE_IDX( cube[2], cube2, CUBE_STATE_SIZE, "CUBE/cube-2", 0, 2 ); + for (i = 0; i < 3; i++) + rmesa->hw.cube[i].emit = cube_emit; + } } else { @@ -709,26 +808,6 @@ void radeonInitState( r100ContextPtr rmesa ) ALLOC_STATE_IDX( cube[1], never, CUBE_STATE_SIZE, "CUBE/cube-1", 0, 1 ); ALLOC_STATE_IDX( cube[2], never, CUBE_STATE_SIZE, "CUBE/cube-2", 0, 2 ); } - ALLOC_STATE( mat[0], tcl, MAT_STATE_SIZE, "MAT/modelproject", 1 ); - ALLOC_STATE( mat[1], tcl_eyespace_or_fog, MAT_STATE_SIZE, "MAT/modelview", 1 ); - ALLOC_STATE( mat[2], tcl_eyespace_or_lighting, MAT_STATE_SIZE, "MAT/it-modelview", 1 ); - ALLOC_STATE( mat[3], tcl_tex0, MAT_STATE_SIZE, "MAT/texmat0", 1 ); - ALLOC_STATE( mat[4], tcl_tex1, MAT_STATE_SIZE, "MAT/texmat1", 1 ); - ALLOC_STATE( mat[5], tcl_tex2, MAT_STATE_SIZE, "MAT/texmat2", 1 ); - ALLOC_STATE( ucp[0], tcl_ucp0, UCP_STATE_SIZE, "UCP/userclip-0", 1 ); - ALLOC_STATE( ucp[1], tcl_ucp1, UCP_STATE_SIZE, "UCP/userclip-1", 1 ); - ALLOC_STATE( ucp[2], tcl_ucp2, UCP_STATE_SIZE, "UCP/userclip-2", 1 ); - ALLOC_STATE( ucp[3], tcl_ucp3, UCP_STATE_SIZE, "UCP/userclip-3", 1 ); - ALLOC_STATE( ucp[4], tcl_ucp4, UCP_STATE_SIZE, "UCP/userclip-4", 1 ); - ALLOC_STATE( ucp[5], tcl_ucp5, UCP_STATE_SIZE, "UCP/userclip-5", 1 ); - ALLOC_STATE( lit[0], tcl_lit0, LIT_STATE_SIZE, "LIT/light-0", 1 ); - ALLOC_STATE( lit[1], tcl_lit1, LIT_STATE_SIZE, "LIT/light-1", 1 ); - ALLOC_STATE( lit[2], tcl_lit2, LIT_STATE_SIZE, "LIT/light-2", 1 ); - ALLOC_STATE( lit[3], tcl_lit3, LIT_STATE_SIZE, "LIT/light-3", 1 ); - ALLOC_STATE( lit[4], tcl_lit4, LIT_STATE_SIZE, "LIT/light-4", 1 ); - ALLOC_STATE( lit[5], tcl_lit5, LIT_STATE_SIZE, "LIT/light-5", 1 ); - ALLOC_STATE( lit[6], tcl_lit6, LIT_STATE_SIZE, "LIT/light-6", 1 ); - ALLOC_STATE( lit[7], tcl_lit7, LIT_STATE_SIZE, "LIT/light-7", 1 ); ALLOC_STATE_IDX( txr[0], txr0, TXR_STATE_SIZE, "TXR/txr-0", 0, 0 ); ALLOC_STATE_IDX( txr[1], txr1, TXR_STATE_SIZE, "TXR/txr-1", 0, 1 ); ALLOC_STATE_IDX( txr[2], txr2, TXR_STATE_SIZE, "TXR/txr-2", 0, 2 ); diff --git a/src/mesa/drivers/dri/radeon/radeon_swtcl.c b/src/mesa/drivers/dri/radeon/radeon_swtcl.c index 58b3be9391b..20ce6c470b4 100644 --- a/src/mesa/drivers/dri/radeon/radeon_swtcl.c +++ b/src/mesa/drivers/dri/radeon/radeon_swtcl.c @@ -285,7 +285,7 @@ void r100_swtcl_flush(GLcontext *ctx, uint32_t current_offset) r100ContextPtr rmesa = R100_CONTEXT(ctx); rcommonEnsureCmdBufSpace(&rmesa->radeon, - rmesa->radeon.hw.max_state_size + (12*sizeof(int)), + radeonCountStateEmitSize( &rmesa->radeon ) + (12*sizeof(int)), __FUNCTION__); diff --git a/src/mesa/drivers/dri/radeon/radeon_tcl.c b/src/mesa/drivers/dri/radeon/radeon_tcl.c index df6708f05e3..2404f284505 100644 --- a/src/mesa/drivers/dri/radeon/radeon_tcl.c +++ b/src/mesa/drivers/dri/radeon/radeon_tcl.c @@ -50,6 +50,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "radeon_tcl.h" #include "radeon_swtcl.h" #include "radeon_maos.h" +#include "radeon_common_context.h" @@ -149,9 +150,6 @@ static GLushort *radeonAllocElts( r100ContextPtr rmesa, GLuint nr ) if (rmesa->radeon.dma.flush) rmesa->radeon.dma.flush( rmesa->radeon.glCtx ); - rcommonEnsureCmdBufSpace(&rmesa->radeon, rmesa->radeon.hw.max_state_size + ELTS_BUFSZ(nr) + - AOS_BUFSZ(rmesa->radeon.tcl.aos_count), __FUNCTION__); - radeonEmitAOS( rmesa, rmesa->radeon.tcl.aos_count, 0 ); @@ -176,10 +174,6 @@ static void radeonEmitPrim( GLcontext *ctx, r100ContextPtr rmesa = R100_CONTEXT( ctx ); radeonTclPrimitive( ctx, prim, hwprim ); - rcommonEnsureCmdBufSpace( &rmesa->radeon, - AOS_BUFSZ(rmesa->radeon.tcl.aos_count) + - rmesa->radeon.hw.max_state_size + VBUF_BUFSZ, __FUNCTION__ ); - radeonEmitAOS( rmesa, rmesa->radeon.tcl.aos_count, start ); @@ -196,6 +190,8 @@ static void radeonEmitPrim( GLcontext *ctx, radeonEmitPrim( ctx, prim, hwprim, start, count ); \ (void) rmesa; } while (0) +#define MAX_CONVERSION_SIZE 40 + /* Try & join small primitives */ #if 0 @@ -360,6 +356,73 @@ radeonComputeFogBlendFactor( GLcontext *ctx, GLfloat fogcoord ) } } +/** + * Predict total emit size for next rendering operation so there is no flush in middle of rendering + * Prediction has to aim towards the best possible value that is worse than worst case scenario + */ +static GLuint radeonEnsureEmitSize( GLcontext * ctx , GLuint inputs ) +{ + r100ContextPtr rmesa = R100_CONTEXT(ctx); + TNLcontext *tnl = TNL_CONTEXT(ctx); + struct vertex_buffer *VB = &tnl->vb; + GLuint space_required; + GLuint state_size; + GLuint nr_aos = 1; /* radeonEmitArrays does always emit one */ + int i; + /* list of flags that are allocating aos object */ + const GLuint flags_to_check[] = { + VERT_BIT_NORMAL, + VERT_BIT_COLOR0, + VERT_BIT_COLOR1, + VERT_BIT_FOG + }; + /* predict number of aos to emit */ + for (i=0; i < sizeof(flags_to_check)/sizeof(flags_to_check[0]); ++i) + { + if (inputs & flags_to_check[i]) + ++nr_aos; + } + for (i = 0; i < ctx->Const.MaxTextureUnits; ++i) + { + if (inputs & VERT_BIT_TEX(i)) + ++nr_aos; + } + + { + /* count the prediction for state size */ + space_required = 0; + state_size = radeonCountStateEmitSize( &rmesa->radeon ); + /* tcl may be changed in radeonEmitArrays so account for it if not dirty */ + if (!rmesa->hw.tcl.dirty) + state_size += rmesa->hw.tcl.check( rmesa->radeon.glCtx, &rmesa->hw.tcl ); + /* predict size for elements */ + for (i = 0; i < VB->PrimitiveCount; ++i) + { + if (!VB->Primitive[i].count) + continue; + /* If primitive.count is less than MAX_CONVERSION_SIZE + rendering code may decide convert to elts. + In that case we have to make pessimistic prediction. + and use larger of 2 paths. */ + const GLuint elts = ELTS_BUFSZ(nr_aos); + const GLuint index = INDEX_BUFSZ; + const GLuint vbuf = VBUF_BUFSZ; + if ( (!VB->Elts && VB->Primitive[i].count >= MAX_CONVERSION_SIZE) + || vbuf > index + elts) + space_required += vbuf; + else + space_required += index + elts; + space_required += AOS_BUFSZ(nr_aos); + } + space_required += SCISSOR_BUFSZ; + } + /* flush the buffer in case we need more than is left. */ + if (rcommonEnsureCmdBufSpace(&rmesa->radeon, space_required, __FUNCTION__)) + return space_required + radeonCountStateEmitSize( &rmesa->radeon ); + else + return space_required + state_size; +} + /**********************************************************************/ /* Render pipeline stage */ /**********************************************************************/ @@ -410,6 +473,8 @@ static GLboolean radeon_run_tcl_render( GLcontext *ctx, } radeonReleaseArrays( ctx, ~0 ); + GLuint emit_end = radeonEnsureEmitSize( ctx, inputs ) + + rmesa->radeon.cmdbuf.cs->cdw; radeonEmitArrays( ctx, inputs ); rmesa->tcl.Elts = VB->Elts; @@ -429,6 +494,10 @@ static GLboolean radeon_run_tcl_render( GLcontext *ctx, radeonEmitPrimitive( ctx, start, start+length, prim ); } + if (emit_end < rmesa->radeon.cmdbuf.cs->cdw) + WARN_ONCE("Rendering was %d commands larger than predicted size." + " We might overflow command buffer.\n", rmesa->radeon.cmdbuf.cs->cdw - emit_end); + return GL_FALSE; /* finished the pipe */ } |