: vIndices2 = GetSimdValid32bitIndices(indices2, pLastIndex);
#endif
break; // incoming type is already 32bit int
- default: SWR_INVALID("Unsupported index type"); vIndices = nullptr; break;
+ default:
+ SWR_INVALID("Unsupported index type");
+ vIndices = nullptr;
+#if USE_SIMD16_SHADERS
+ vIndices2 = nullptr;
+#endif
+ break;
}
if(fetchState.bForceSequentialAccessEnable)
}
else if (ied.InstanceStrideEnable)
{
+ // silence unused variable warnings
+ startOffset = C(0);
+ vCurIndices = vIndices;
+
SWR_ASSERT((0), "TODO: Fill out more once driver sends this down.");
}
else
// But, we know that elements must be aligned for FETCH. :)
// Right shift the offset by a bit and then scale by 2 to remove the sign extension.
Value *shiftedOffsets16 = LSHR(vOffsets16, 1);
- pVtxSrc2[currentVertexElement++] = GATHERPS_16(gatherSrc16, pStreamBase, shiftedOffsets16, vGatherMask16, 2);
+ pVtxSrc2[currentVertexElement++] = GATHERPS_16(gatherSrc16, pStreamBaseGFX, shiftedOffsets16, vGatherMask16, 2, GFX_MEM_CLIENT_FETCH);
}
else
{
currentVertexElement = 0;
}
}
-
- // offset base to the next component in the vertex to gather
- pStreamBase = GEP(pStreamBase, C((char)4));
#else
if (isComponentEnabled(compMask, i))
{
// But, we know that elements must be aligned for FETCH. :)
// Right shift the offset by a bit and then scale by 2 to remove the sign extension.
Value *vShiftedOffsets = LSHR(vOffsets, 1);
- vVertexElements[currentVertexElement++] = GATHERPS(gatherSrc, pStreamBaseGFX, vShiftedOffsets, vGatherMask, 2);
+ vVertexElements[currentVertexElement++] = GATHERPS(gatherSrc, pStreamBaseGFX, vShiftedOffsets, vGatherMask, 2, GFX_MEM_CLIENT_FETCH);
}
else
{
currentVertexElement = 0;
}
}
+#endif
// offset base to the next component in the vertex to gather
pStreamBase = GEP(pStreamBase, C((char)4));
pStreamBaseGFX = ADD(pStreamBaseGFX, C((int64_t)4));
-#endif
}
}
break;
// if valid, load the index. if not, load 0 from the stack
Value* pValid = SELECT(mask, pIndex, pZeroIndex);
- Value *index = LOAD(pValid, "valid index");
+ Value *index = LOAD(pValid, "valid index", GFX_MEM_CLIENT_FETCH);
// zero extended index to 32 bits and insert into the correct simd lane
index = Z_EXT(index, mInt32Ty);
// vIndexMask -1-1-1-1 0 0 0 0 : offsets < max pass
// vLoadedIndices 0 1 2 3 0 0 0 0 : offsets >= max masked to 0
Value* vMaxIndex = VBROADCAST(numIndicesLeft);
- Value* vIndexMask = VPCMPGTD(vMaxIndex, vIndexOffsets);
-
- // VMASKLOAD takes an *i8 src pointer
- pIndices = BITCAST(pIndices,PointerType::get(mInt8Ty,0));
+ Value* vIndexMask = ICMP_SGT(vMaxIndex, vIndexOffsets);
// Load the indices; OOB loads 0
- return MASKLOADD(pIndices,vIndexMask);
+ pIndices = BITCAST(pIndices, PointerType::get(mSimdInt32Ty, 0));
+ return MASKED_LOAD(pIndices, 4, vIndexMask, VIMMED1(0));
}
//////////////////////////////////////////////////////////////////////////