/**
* Scan program instructions to update the program's InputsRead and
* OutputsWritten fields.
*/
static void
_slang_update_inputs_outputs(struct gl_program *prog)
{
GLuint i, j;
GLuint maxAddrReg = 0;
prog->InputsRead = 0x0;
prog->OutputsWritten = 0x0;
for (i = 0; i < prog->NumInstructions; i++) {
const struct prog_instruction *inst = prog->Instructions + i;
const GLuint numSrc = _mesa_num_inst_src_regs(inst->Opcode);
for (j = 0; j < numSrc; j++) {
if (inst->SrcReg[j].File == PROGRAM_INPUT) {
prog->InputsRead |= 1 << inst->SrcReg[j].Index;
}
else if (inst->SrcReg[j].File == PROGRAM_ADDRESS) {
maxAddrReg = MAX2(maxAddrReg, (GLuint) (inst->SrcReg[j].Index + 1));
}
}
if (inst->DstReg.File == PROGRAM_OUTPUT) {
prog->OutputsWritten |= BITFIELD64_BIT(inst->DstReg.Index);
if (inst->DstReg.RelAddr) {
/* If the output attribute is indexed with relative addressing
* we know that it must be a varying or texcoord such as
* gl_TexCoord[i] = v; In this case, mark all the texcoords
* or varying outputs as being written. It's not an error if
* a vertex shader writes varying vars that aren't used by the
* fragment shader. But it is an error for a fragment shader
* to use varyings that are not written by the vertex shader.
*/
if (prog->Target == GL_VERTEX_PROGRAM_ARB) {
if (inst->DstReg.Index == VERT_RESULT_TEX0) {
/* mark all texcoord outputs as written */
const GLbitfield64 mask =
BITFIELD64_RANGE(VERT_RESULT_TEX0,
(VERT_RESULT_TEX0
+ MAX_TEXTURE_COORD_UNITS - 1));
prog->OutputsWritten |= mask;
}
else if (inst->DstReg.Index == VERT_RESULT_VAR0) {
/* mark all generic varying outputs as written */
const GLbitfield64 mask =
BITFIELD64_RANGE(VERT_RESULT_VAR0,
(VERT_RESULT_VAR0 + MAX_VARYING - 1));
prog->OutputsWritten |= mask;
}
}
}
}
else if (inst->DstReg.File == PROGRAM_ADDRESS) {
maxAddrReg = MAX2(maxAddrReg, inst->DstReg.Index + 1);
}
}
prog->NumAddressRegs = maxAddrReg;
}
/**
* Set vertex state for SW TCL. The primary purpose of this function is to
* determine in advance whether or not the hardware can / should do the
* projection divide or Mesa should do it.
*/
void radeonChooseVertexState( struct gl_context *ctx )
{
r100ContextPtr rmesa = R100_CONTEXT( ctx );
TNLcontext *tnl = TNL_CONTEXT(ctx);
GLuint se_coord_fmt = rmesa->hw.set.cmd[SET_SE_COORDFMT];
GLboolean unfilled = (ctx->Polygon.FrontMode != GL_FILL ||
ctx->Polygon.BackMode != GL_FILL);
GLboolean twosided = ctx->Light.Enabled && ctx->Light.Model.TwoSide;
se_coord_fmt &= ~(RADEON_VTX_XY_PRE_MULT_1_OVER_W0 |
RADEON_VTX_Z_PRE_MULT_1_OVER_W0 |
RADEON_VTX_W0_IS_NOT_1_OVER_W0);
/* We must ensure that we don't do _tnl_need_projected_coords while in a
* rasterization fallback. As this function will be called again when we
* leave a rasterization fallback, we can just skip it for now.
*/
if (rmesa->radeon.Fallback != 0)
return;
/* HW perspective divide is a win, but tiny vertex formats are a
* bigger one.
*/
if ((0 == (tnl->render_inputs_bitset &
(BITFIELD64_RANGE(_TNL_ATTRIB_TEX0, _TNL_NUM_TEX)
| BITFIELD64_BIT(_TNL_ATTRIB_COLOR1))))
|| twosided
|| unfilled) {
rmesa->swtcl.needproj = GL_TRUE;
se_coord_fmt |= (RADEON_VTX_XY_PRE_MULT_1_OVER_W0 |
RADEON_VTX_Z_PRE_MULT_1_OVER_W0);
}
else {
rmesa->swtcl.needproj = GL_FALSE;
se_coord_fmt |= (RADEON_VTX_W0_IS_NOT_1_OVER_W0);
}
_tnl_need_projected_coords( ctx, rmesa->swtcl.needproj );
if ( se_coord_fmt != rmesa->hw.set.cmd[SET_SE_COORDFMT] ) {
RADEON_STATECHANGE( rmesa, set );
rmesa->hw.set.cmd[SET_SE_COORDFMT] = se_coord_fmt;
}
}
/**
* Tell the tnl module how to build SWvertex objects for swrast.
* We'll build the map[] array with that info and pass it to
* _tnl_install_attrs().
*/
static void
setup_vertex_format(struct gl_context *ctx)
{
TNLcontext *tnl = TNL_CONTEXT(ctx);
SScontext *swsetup = SWSETUP_CONTEXT(ctx);
GLboolean intColors = !ctx->FragmentProgram._Current
&& !ctx->ATIFragmentShader._Enabled
&& ctx->RenderMode == GL_RENDER
&& CHAN_TYPE != GL_FLOAT;
if (intColors != swsetup->intColors ||
tnl->render_inputs_bitset != swsetup->last_index_bitset) {
GLbitfield64 index_bitset = tnl->render_inputs_bitset;
struct tnl_attr_map map[_TNL_ATTRIB_MAX];
unsigned int i, e = 0;
swsetup->intColors = intColors;
EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, attrib[VARYING_SLOT_POS] );
if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_COLOR0)) {
if (swsetup->intColors)
EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4CHAN_4F_RGBA, color );
else
EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4F, attrib[VARYING_SLOT_COL0]);
}
if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_COLOR1)) {
EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_4F, attrib[VARYING_SLOT_COL1]);
}
if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_FOG)) {
const GLint emit = ctx->FragmentProgram._Current ? EMIT_4F : EMIT_1F;
EMIT_ATTR( _TNL_ATTRIB_FOG, emit, attrib[VARYING_SLOT_FOGC]);
}
if (index_bitset & BITFIELD64_RANGE(_TNL_ATTRIB_TEX0, _TNL_NUM_TEX))
{
for (i = 0; i < MAX_TEXTURE_COORD_UNITS; i++) {
if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_TEX(i))) {
EMIT_ATTR( _TNL_ATTRIB_TEX(i), EMIT_4F,
attrib[VARYING_SLOT_TEX0 + i] );
}
}
}
/* shader varying vars */
if (index_bitset & BITFIELD64_RANGE(_TNL_ATTRIB_GENERIC0, _TNL_NUM_GENERIC)) {
for (i = 0; i < ctx->Const.MaxVarying; i++) {
if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_GENERIC(i))) {
EMIT_ATTR( _TNL_ATTRIB_GENERIC(i), VARYING_EMIT_STYLE,
attrib[VARYING_SLOT_VAR0 + i] );
}
}
}
if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_POINTSIZE))
EMIT_ATTR( _TNL_ATTRIB_POINTSIZE, EMIT_1F, pointSize );
_tnl_install_attrs( ctx, map, e,
ctx->Viewport._WindowMap.m,
sizeof(SWvertex) );
swsetup->last_index_bitset = index_bitset;
}
}
static void radeonSetVertexFormat( struct gl_context *ctx )
{
r100ContextPtr rmesa = R100_CONTEXT( ctx );
TNLcontext *tnl = TNL_CONTEXT(ctx);
struct vertex_buffer *VB = &tnl->vb;
GLbitfield64 index_bitset = tnl->render_inputs_bitset;
int fmt_0 = 0;
int offset = 0;
/* Important:
*/
if ( VB->NdcPtr != NULL ) {
VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
}
else {
VB->AttribPtr[VERT_ATTRIB_POS] = VB->ClipPtr;
}
assert( VB->AttribPtr[VERT_ATTRIB_POS] != NULL );
rmesa->radeon.swtcl.vertex_attr_count = 0;
/* EMIT_ATTR's must be in order as they tell t_vertex.c how to
* build up a hardware vertex.
*/
if ( !rmesa->swtcl.needproj ||
(index_bitset & BITFIELD64_RANGE(_TNL_ATTRIB_TEX0, _TNL_NUM_TEX))) {
/* for projtex */
EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_4F,
RADEON_CP_VC_FRMT_XY | RADEON_CP_VC_FRMT_Z | RADEON_CP_VC_FRMT_W0 );
offset = 4;
}
else {
EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_3F,
RADEON_CP_VC_FRMT_XY | RADEON_CP_VC_FRMT_Z );
offset = 3;
}
rmesa->swtcl.coloroffset = offset;
#if MESA_LITTLE_ENDIAN
EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4UB_4F_RGBA,
RADEON_CP_VC_FRMT_PKCOLOR );
#else
EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4UB_4F_ABGR,
RADEON_CP_VC_FRMT_PKCOLOR );
#endif
offset += 1;
rmesa->swtcl.specoffset = 0;
if (index_bitset &
(BITFIELD64_BIT(_TNL_ATTRIB_COLOR1) | BITFIELD64_BIT(_TNL_ATTRIB_FOG))) {
#if MESA_LITTLE_ENDIAN
if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_COLOR1)) {
rmesa->swtcl.specoffset = offset;
EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_3UB_3F_RGB,
RADEON_CP_VC_FRMT_PKSPEC );
}
else {
EMIT_PAD( 3 );
}
if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_FOG)) {
EMIT_ATTR( _TNL_ATTRIB_FOG, EMIT_1UB_1F,
RADEON_CP_VC_FRMT_PKSPEC );
}
else {
EMIT_PAD( 1 );
}
#else
if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_FOG)) {
EMIT_ATTR( _TNL_ATTRIB_FOG, EMIT_1UB_1F,
RADEON_CP_VC_FRMT_PKSPEC );
}
else {
EMIT_PAD( 1 );
}
if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_COLOR1)) {
rmesa->swtcl.specoffset = offset;
EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_3UB_3F_BGR,
RADEON_CP_VC_FRMT_PKSPEC );
}
else {
EMIT_PAD( 3 );
}
#endif
}
if (index_bitset & BITFIELD64_RANGE(_TNL_ATTRIB_TEX0, _TNL_NUM_TEX)) {
int i;
for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_TEX(i))) {
GLuint sz = VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]->size;
switch (sz) {
case 1:
case 2:
EMIT_ATTR( _TNL_ATTRIB_TEX0+i, EMIT_2F,
radeon_cp_vc_frmts[i][0] );
break;
case 3:
if (ctx->Texture.Unit[i]._Current &&
ctx->Texture.Unit[i]._Current->Target == GL_TEXTURE_CUBE_MAP) {
EMIT_ATTR( _TNL_ATTRIB_TEX0+i, EMIT_3F,
radeon_cp_vc_frmts[i][1] );
} else {
EMIT_ATTR( _TNL_ATTRIB_TEX0+i, EMIT_2F,
radeon_cp_vc_frmts[i][0] );
}
break;
case 4:
if (ctx->Texture.Unit[i]._Current &&
ctx->Texture.Unit[i]._Current->Target == GL_TEXTURE_CUBE_MAP) {
EMIT_ATTR( _TNL_ATTRIB_TEX0+i, EMIT_3F,
radeon_cp_vc_frmts[i][1] );
} else {
EMIT_ATTR( _TNL_ATTRIB_TEX0+i, EMIT_3F_XYW,
radeon_cp_vc_frmts[i][1] );
}
break;
default:
continue;
};
}
}
}
if (rmesa->radeon.tnl_index_bitset != index_bitset ||
fmt_0 != rmesa->swtcl.vertex_format) {
RADEON_NEWPRIM(rmesa);
rmesa->swtcl.vertex_format = fmt_0;
rmesa->radeon.swtcl.vertex_size =
_tnl_install_attrs( ctx,
rmesa->radeon.swtcl.vertex_attrs,
rmesa->radeon.swtcl.vertex_attr_count,
NULL, 0 );
rmesa->radeon.swtcl.vertex_size /= 4;
rmesa->radeon.tnl_index_bitset = index_bitset;
radeon_print(RADEON_SWRENDER, RADEON_VERBOSE,
"%s: vertex_size= %d floats\n", __func__, rmesa->radeon.swtcl.vertex_size);
}
}
static void
i830_render_start(struct intel_context *intel)
{
struct gl_context *ctx = &intel->ctx;
struct i830_context *i830 = i830_context(ctx);
TNLcontext *tnl = TNL_CONTEXT(ctx);
struct vertex_buffer *VB = &tnl->vb;
GLbitfield64 index_bitset = tnl->render_inputs_bitset;
GLuint v0 = _3DSTATE_VFT0_CMD;
GLuint v2 = _3DSTATE_VFT1_CMD;
GLuint mcsb1 = 0;
/* Important:
*/
VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
intel->vertex_attr_count = 0;
/* EMIT_ATTR's must be in order as they tell t_vertex.c how to
* build up a hardware vertex.
*/
if (index_bitset & BITFIELD64_RANGE(_TNL_ATTRIB_TEX0, _TNL_NUM_TEX)) {
EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, VFT0_XYZW);
intel->coloroffset = 4;
}
else {
EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, VFT0_XYZ);
intel->coloroffset = 3;
}
if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_POINTSIZE)) {
EMIT_ATTR(_TNL_ATTRIB_POINTSIZE, EMIT_1F, VFT0_POINT_WIDTH);
}
EMIT_ATTR(_TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, VFT0_DIFFUSE);
intel->specoffset = 0;
if (index_bitset & (BITFIELD64_BIT(_TNL_ATTRIB_COLOR1) |
BITFIELD64_BIT(_TNL_ATTRIB_FOG))) {
if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_COLOR1)) {
intel->specoffset = intel->coloroffset + 1;
EMIT_ATTR(_TNL_ATTRIB_COLOR1, EMIT_3UB_3F_BGR, VFT0_SPEC);
}
else
EMIT_PAD(3);
if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_FOG))
EMIT_ATTR(_TNL_ATTRIB_FOG, EMIT_1UB_1F, VFT0_SPEC);
else
EMIT_PAD(1);
}
if (index_bitset & BITFIELD64_RANGE(_TNL_ATTRIB_TEX0, _TNL_NUM_TEX)) {
int i, count = 0;
for (i = 0; i < I830_TEX_UNITS; i++) {
if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_TEX(i))) {
GLuint sz = VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]->size;
GLuint emit;
GLuint mcs = (i830->state.Tex[i][I830_TEXREG_MCS] &
~TEXCOORDTYPE_MASK);
switch (sz) {
case 1:
case 2:
emit = EMIT_2F;
sz = 2;
mcs |= TEXCOORDTYPE_CARTESIAN;
break;
case 3:
emit = EMIT_3F;
sz = 3;
mcs |= TEXCOORDTYPE_VECTOR;
break;
case 4:
emit = EMIT_3F_XYW;
sz = 3;
mcs |= TEXCOORDTYPE_HOMOGENEOUS;
break;
default:
continue;
};
EMIT_ATTR(_TNL_ATTRIB_TEX0 + i, emit, 0);
v2 |= VRTX_TEX_SET_FMT(count, SZ_TO_HW(sz));
mcsb1 |= (count + 8) << (i * 4);
if (mcs != i830->state.Tex[i][I830_TEXREG_MCS]) {
I830_STATECHANGE(i830, I830_UPLOAD_TEX(i));
i830->state.Tex[i][I830_TEXREG_MCS] = mcs;
}
count++;
}
}
v0 |= VFT0_TEX_COUNT(count);
}
/* Only need to change the vertex emit code if there has been a
* statechange to a new hardware vertex format:
*/
if (v0 != i830->state.Ctx[I830_CTXREG_VF] ||
v2 != i830->state.Ctx[I830_CTXREG_VF2] ||
mcsb1 != i830->state.Ctx[I830_CTXREG_MCSB1] ||
index_bitset != i830->last_index_bitset) {
I830_STATECHANGE(i830, I830_UPLOAD_CTX);
/* Must do this *after* statechange, so as not to affect
* buffered vertices reliant on the old state:
*/
intel->vertex_size =
_tnl_install_attrs(ctx,
intel->vertex_attrs,
intel->vertex_attr_count,
intel->ViewportMatrix.m, 0);
intel->vertex_size >>= 2;
i830->state.Ctx[I830_CTXREG_VF] = v0;
i830->state.Ctx[I830_CTXREG_VF2] = v2;
i830->state.Ctx[I830_CTXREG_MCSB1] = mcsb1;
i830->last_index_bitset = index_bitset;
assert(i830_check_vertex_size(intel, intel->vertex_size));
}