static void
VTX_FETCH(State &state, const ControlFlowInst &cf, const VertexFetchInst &inst)
{
// FETCH R4.xyzw, R0.y, b131 NO_INDEX_OFFSET FMT_FROM_FETCH_CONSTANT MEGA(16) OFFSET(0)
auto id = inst.word0.BUFFER_ID() + SQ_VS_RESOURCE_BASE;
// For now we only support reading from vertex buffers (uniform blocks)
decaf_assert(id >= SQ_VS_BUF_RESOURCE_0 && id < SQ_VS_GSOUT_RESOURCE, fmt::format("Unsupported VTX_FETCH buffer id {}", id));
// Let's only support a very expected set of values
decaf_check(inst.word0.FETCH_TYPE() == SQ_VTX_FETCH_NO_INDEX_OFFSET);
decaf_check(inst.word1.USE_CONST_FIELDS() == 1);
decaf_check(inst.word2.OFFSET() == 0);
decaf_check(inst.word2.MEGA_FETCH() && (inst.word0.MEGA_FETCH_COUNT() + 1) == 16);
auto dstSelX = inst.word1.DST_SEL_X();
auto dstSelY = inst.word1.DST_SEL_Y();
auto dstSelZ = inst.word1.DST_SEL_Z();
auto dstSelW = inst.word1.DST_SEL_W();
auto numDstSels = 4u;
auto dstSelMask = condenseSelections(dstSelX, dstSelY, dstSelZ, dstSelW, numDstSels);
if (numDstSels > 0) {
auto dst = getExportRegister(inst.gpr.DST_GPR(), inst.gpr.DST_REL());
auto src = getExportRegister(inst.word0.SRC_GPR(), inst.word0.SRC_REL());
inst.word0.SRC_SEL_X();
auto blockID = id - SQ_VS_BUF_RESOURCE_0;
if (state.shader) {
state.shader->usedUniformBlocks[blockID] = true;
}
fmt::MemoryWriter tmp;
tmp << "UB_" << blockID << ".values[floatBitsToInt(";
insertSelectValue(tmp, src, inst.word0.SRC_SEL_X());
tmp << ")]";
insertLineStart(state);
state.out << dst << "." << dstSelMask << " = ";
insertSelectVector(state.out, tmp.str(), dstSelX, dstSelY, dstSelZ, dstSelW, numDstSels);
state.out << ";";
insertLineEnd(state);
}
}
static void
GET_TEXTURE_INFO(State &state, const latte::ControlFlowInst &cf, const latte::TextureFetchInst &inst)
{
auto dstSelX = inst.word1.DST_SEL_X();
auto dstSelY = inst.word1.DST_SEL_Y();
auto dstSelZ = inst.word1.DST_SEL_Z();
auto dstSelW = inst.word1.DST_SEL_W();
auto srcSelX = inst.word2.SRC_SEL_X();
auto srcSelY = inst.word2.SRC_SEL_Y();
auto srcSelZ = inst.word2.SRC_SEL_Z();
auto srcSelW = inst.word2.SRC_SEL_W();
auto resourceID = inst.word0.RESOURCE_ID();
// SAMPLER_ID is a don't care in this instruction, but we ensure that
// textures and samplers use the same IDs, so we can safely use
// RESOURCE_ID as the sampler ID.
auto samplerID = resourceID;
auto samplerDim = state.shader->samplerDim[samplerID];
registerSamplerID(state, samplerID, false);
auto dst = getExportRegister(inst.word1.DST_GPR(), inst.word1.DST_REL());
auto src = getExportRegister(inst.word0.SRC_GPR(), inst.word0.SRC_REL());
// TODO: Which source component is used to select the LoD? Xenoblade has:
// GET_TEXTURE_INFO R6.xy__, R4.xx0x, t4, s0 (with R4.x = 0)
auto srcSelLod = srcSelX;
// GET_TEXTURE_INFO returns {width, height, depth, mipmap count}, but GLSL
// has separate functions for W/H/D and mipmap count, so we need to split
// this up into two operations.
auto numDstSels = 3u;
SQ_SEL dummy = SQ_SEL::SEL_MASK;
auto dstSelMask = condenseSelections(dstSelX, dstSelY, dstSelZ, dummy, numDstSels);
if (numDstSels > 0) {
auto samplerElements = getSamplerArgCount(samplerDim, false);
insertLineStart(state);
state.out << "texTmp.xyz = intBitsToFloat(ivec3(textureSize(sampler_" << samplerID;
if (!getSamplerIsMsaa(samplerDim)) {
state.out << ", floatBitsToInt(";
insertSelectValue(state.out, src, srcSelLod);
state.out << ")";
}
state.out << ")";
for (auto i = samplerElements; i < 3; ++i) {
state.out << ", 1";
}
state.out << "));";
insertLineEnd(state);
insertLineStart(state);
state.out << dst << "." << dstSelMask;
state.out << " = ";
insertSelectVector(state.out, "texTmp", dstSelX, dstSelY, dstSelZ, SQ_SEL::SEL_MASK, numDstSels);
state.out << ";";
insertLineEnd(state);
}
if (dstSelW != SQ_SEL::SEL_MASK) {
insertLineStart(state);
insertSelectValue(state.out, dst, dstSelW);
state.out << " = intBitsToFloat(textureQueryLevels(sampler_" << samplerID << "));";
insertLineEnd(state);
}
}
static void
sampleFunc(State &state,
const latte::ControlFlowInst &cf,
const latte::TextureFetchInst &inst,
const std::string &func,
const std::string &offsetFunc,
bool isShadowOp = false,
latte::SQ_SEL extraArg = latte::SQ_SEL::SEL_MASK,
bool asInts = false)
{
auto dstSelX = inst.word1.DST_SEL_X();
auto dstSelY = inst.word1.DST_SEL_Y();
auto dstSelZ = inst.word1.DST_SEL_Z();
auto dstSelW = inst.word1.DST_SEL_W();
auto srcSelX = inst.word2.SRC_SEL_X();
auto srcSelY = inst.word2.SRC_SEL_Y();
auto srcSelZ = inst.word2.SRC_SEL_Z();
auto srcSelW = inst.word2.SRC_SEL_W();
int32_t offsetX = sign_extend<5>(inst.word2.OFFSET_X());
int32_t offsetY = sign_extend<5>(inst.word2.OFFSET_Y());
int32_t offsetZ = sign_extend<5>(inst.word2.OFFSET_Z());
auto resourceID = inst.word0.RESOURCE_ID();
auto samplerID = inst.word2.SAMPLER_ID();
auto samplerDim = state.shader->samplerDim[samplerID];
auto samplerUsage = registerSamplerID(state, samplerID, isShadowOp);
if (resourceID != samplerID) {
throw translate_exception("Unsupported sample with RESOURCE_ID != SAMPLER_ID");
}
auto dst = getExportRegister(inst.word1.DST_GPR(), inst.word1.DST_REL());
auto src = getExportRegister(inst.word0.SRC_GPR(), inst.word0.SRC_REL());
auto numDstSels = 4u;
auto dstSelMask = condenseSelections(dstSelX, dstSelY, dstSelZ, dstSelW, numDstSels);
if (numDstSels > 0) {
insertLineStart(state);
auto samplerElements = getSamplerArgCount(samplerDim, isShadowOp);
if (!isShadowOp) {
state.out << "texTmp";
} else {
state.out << "texTmp.x";
}
state.out << " = ";
bool writeOffsets = false;
if (offsetX != 0 || offsetY != 0 || offsetZ != 0) {
decaf_check(offsetFunc.size());
state.out << offsetFunc;
writeOffsets = true;
} else {
decaf_check(func.size());
state.out << func;
}
state.out << "(sampler_" << samplerID << ", ";
if (isShadowOp) {
/* In r600 the .w channel holds the compare value whereas OpenGL
* shadow samplers just expect it to be the last texture coordinate
* so we must set the last channel to SQ_SEL::SEL_W
*/
if (samplerElements == 2) {
srcSelY = srcSelW;
} else if (samplerElements == 3) {
srcSelZ = srcSelW;
} else if (samplerElements == 4) {
// The value will already be in place
} else {
decaf_abort(fmt::format("Unexpected samplerElements {} for shadow sampler", samplerElements));
}
}
if (asInts) {
state.out << "floatBitsToInt(";
}
insertSelectVector(state.out, src, srcSelX, srcSelY, srcSelZ, srcSelW, samplerElements);
if (asInts) {
state.out << ")";
}
switch (extraArg) {
case latte::SQ_SEL::SEL_X:
state.out << ", ";
insertSelectValue(state.out, src, srcSelX);
break;
case latte::SQ_SEL::SEL_Y:
state.out << ", ";
insertSelectValue(state.out, src, srcSelY);
break;
case latte::SQ_SEL::SEL_Z:
state.out << ", ";
insertSelectValue(state.out, src, srcSelZ);
break;
case latte::SQ_SEL::SEL_W:
state.out << ", ";
insertSelectValue(state.out, src, srcSelW);
break;
case latte::SQ_SEL::SEL_0:
state.out << ", 0";
break;
case latte::SQ_SEL::SEL_1:
state.out << ", 1";
break;
}
if (writeOffsets) {
switch (samplerDim) {
case latte::SQ_TEX_DIM::DIM_1D:
case latte::SQ_TEX_DIM::DIM_1D_ARRAY:
state.out << ", " << offsetX;
break;
case latte::SQ_TEX_DIM::DIM_2D:
case latte::SQ_TEX_DIM::DIM_2D_ARRAY:
case latte::SQ_TEX_DIM::DIM_2D_MSAA:
case latte::SQ_TEX_DIM::DIM_2D_ARRAY_MSAA:
state.out << ", ivec2(" << offsetX << ", " << offsetY << ")";
break;
case latte::SQ_TEX_DIM::DIM_3D:
state.out << ", ivec3(" << offsetX << ", " << offsetY << ", " << offsetZ << ")";
break;
case latte::SQ_TEX_DIM::DIM_CUBEMAP:
default:
throw translate_exception(fmt::format("Unsupported sampler dim {}", static_cast<unsigned>(samplerDim)));
}
}
if (getSamplerIsMsaa(samplerDim)) {
// Write the sample number if this is an MSAA sampler
state.out << ", 0";
}
state.out << ");";
insertLineEnd(state);
insertLineStart(state);
state.out << dst << "." << dstSelMask;
state.out << " = ";
insertSelectVector(state.out, "texTmp", dstSelX, dstSelY, dstSelZ, dstSelW, numDstSels);
state.out << ";";
insertLineEnd(state);
}
}