Lines Matching full:token
147 Value token) {
159 return builder.create<gpu::AllocOp>(loc, TypeRange({memTp, token.getType()}),
160 token, dynamicSizes, ValueRange());
172 Value size, Value token) {
174 return builder.create<gpu::AllocOp>(loc, TypeRange({memTp, token.getType()}),
175 token, size, ValueRange());
180 Value token) {
181 return genAllocBuffer(builder, loc, builder.getI8Type(), size, token);
186 Value token) {
187 return builder.create<gpu::DeallocOp>(loc, token.getType(), token, mem)
193 Value src, Value token) {
194 return builder.create<gpu::MemcpyOp>(loc, token.getType(), token, dst, src)
246 /// on the kernel token and then deallocated. All other buffers are simply
519 Value token, Value sz1, Value sz2, Value nseA,
526 return builder.create<gpu::CreateCooOp>(loc, handleTp, tokenTp, token,
531 return builder.create<gpu::CreateCooAoSOp>(loc, handleTp, tokenTp, token,
539 return builder.create<gpu::CreateCsrOp>(loc, handleTp, tokenTp, token, sz1,
542 return builder.create<gpu::CreateCscOp>(loc, handleTp, tokenTp, token, sz1,
556 return builder.create<gpu::CreateBsrOp>(loc, handleTp, tokenTp, token, bRows,
603 Value token = genFirstWait(rewriter, loc);
605 genSpMat(rewriter, loc, aTp, spmatHandleTp, tokenTp, token, szY, szX,
608 token = spGenA->getResult(1);
610 loc, dnTensorHandleTp, tokenTp, token, vecX, szX);
612 token = dvecX.getAsyncToken();
614 loc, dnTensorHandleTp, tokenTp, token, vecY, szY);
616 token = dvecY.getAsyncToken();
621 loc, indexTp, tokenTp, token, spMatA, dnX, dnY,
624 token = bufferComp.getAsyncToken();
625 auto buf = genAllocBuffer(rewriter, loc, bufferSz, token);
627 token = buf.getAsyncToken();
631 loc, tokenTp, token, spMatA, dnX, dnY, /*computeType=*/dnYType, buffer);
632 token = spmvComp.getAsyncToken();
635 token = rewriter.create<gpu::DestroySpMatOp>(loc, tokenTp, token, spMatA)
637 token = rewriter.create<gpu::DestroyDnTensorOp>(loc, tokenTp, token, dnX)
639 token = rewriter.create<gpu::DestroyDnTensorOp>(loc, tokenTp, token, dnY)
641 token = genDeallocMemRef(rewriter, loc, rowA, token);
643 token = genDeallocMemRef(rewriter, loc, colA, token);
644 token = genDeallocMemRef(rewriter, loc, valA, token);
645 token = genDeallocMemRef(rewriter, loc, buffer, token);
646 token = genDeallocMemRef(rewriter, loc, vecX, token);
647 token = genCopyMemRef(rewriter, loc, memY, vecY, token);
648 token = genDeallocMemRef(rewriter, loc, vecY, token);
649 tokens.push_back(token);
701 Value token = genFirstWait(rewriter, loc);
703 genSpMat(rewriter, loc, aTp, spMatHandleTp, tokenTp, token, szm, szk,
706 token = spGenA->getResult(1);
708 loc, dnTensorHandleTp, tokenTp, token, matB,
711 token = dmatB.getAsyncToken();
713 loc, dnTensorHandleTp, tokenTp, token, matC,
716 token = dmatC.getAsyncToken();
721 loc, indexTp, tokenTp, token, spMatA, dnB, dnC,
724 token = bufferComp.getAsyncToken();
725 auto buf = genAllocBuffer(rewriter, loc, bufferSz, token);
727 token = buf.getAsyncToken();
732 loc, tokenTp, token, spMatA, dnB, dnC, /*computeType=*/dnCType, buffer);
733 token = spmmComp.getAsyncToken();
736 token = rewriter.create<gpu::DestroySpMatOp>(loc, tokenTp, token, spMatA)
738 token = rewriter.create<gpu::DestroyDnTensorOp>(loc, tokenTp, token, dnB)
740 token = rewriter.create<gpu::DestroyDnTensorOp>(loc, tokenTp, token, dnC)
742 token = genDeallocMemRef(rewriter, loc, rowA, token);
744 token = genDeallocMemRef(rewriter, loc, colA, token);
745 token = genDeallocMemRef(rewriter, loc, valA, token);
746 token = genDeallocMemRef(rewriter, loc, buffer, token);
747 token = genDeallocMemRef(rewriter, loc, matB, token);
748 token = genCopyMemRef(rewriter, loc, bufC, matC, token);
749 token = genDeallocMemRef(rewriter, loc, matC, token);
750 tokens.push_back(token);
806 Value token = genFirstWait(rewriter, loc);
808 genSpMat(rewriter, loc, aTp, spmatHandleTp, tokenTp, token, szm, szk,
811 token = spGenA->getResult(1);
813 genSpMat(rewriter, loc, bTp, spmatHandleTp, tokenTp, token, szk, szn,
816 token = spGenB->getResult(1);
822 auto e1 = genAllocBuffer(rewriter, loc, cTp.getPosType(), mplus1, token);
824 token = e1.getAsyncToken();
825 auto e2 = genAllocBuffer(rewriter, loc, cTp.getCrdType(), zero, token);
827 token = e2.getAsyncToken();
828 auto e3 = genAllocBuffer(rewriter, loc, dnCType, zero, token);
830 token = e3.getAsyncToken();
832 genSpMat(rewriter, loc, cTp, spmatHandleTp, tokenTp, token, szm, szn,
835 token = spGenC->getResult(1);
839 rewriter.create<gpu::SpGEMMCreateDescrOp>(loc, descTp, tokenTp, token);
841 token = descOp->getResult(1);
843 loc, indexTp, tokenTp, token, desc, gpu::TransposeMode::NON_TRANSPOSE,
847 token = work1->getResult(1);
848 auto buf1 = genAllocBuffer(rewriter, loc, bufferSz1, token);
850 token = buf1.getAsyncToken();
852 loc, indexTp, tokenTp, token, desc, gpu::TransposeMode::NON_TRANSPOSE,
856 token = work2->getResult(1);
860 loc, indexTp, tokenTp, token, desc, gpu::TransposeMode::NON_TRANSPOSE,
864 token = compute1->getResult(1);
865 auto buf2 = genAllocBuffer(rewriter, loc, bufferSz2, token);
867 token = buf2.getAsyncToken();
869 loc, indexTp, tokenTp, token, desc, gpu::TransposeMode::NON_TRANSPOSE,
872 token = compute2->getResult(1);
876 loc, indexTp, indexTp, indexTp, tokenTp, token, spMatC);
878 token = sizes->getResult(3);
879 auto a2 = genAllocBuffer(rewriter, loc, cTp.getCrdType(), nnz, token);
881 token = a2.getAsyncToken();
882 auto a3 = genAllocBuffer(rewriter, loc, dnCType, nnz, token);
884 token = a3.getAsyncToken();
888 loc, tokenTp, token, spMatC, rowC, colC, valC);
889 token = update->getResult(0);
891 loc, tokenTp, token, desc, gpu::TransposeMode::NON_TRANSPOSE,
893 token = copy->getResult(0);
901 token = rewriter.create<gpu::SpGEMMDestroyDescrOp>(loc, tokenTp, token, desc)
903 token = rewriter.create<gpu::DestroySpMatOp>(loc, tokenTp, token, spMatA)
905 token = rewriter.create<gpu::DestroySpMatOp>(loc, tokenTp, token, spMatB)
907 token = rewriter.create<gpu::DestroySpMatOp>(loc, tokenTp, token, spMatC)
909 token = genCopyMemRef(rewriter, loc, rowH, rowC, token);
910 token = genCopyMemRef(rewriter, loc, colH, colC, token);
911 token = genCopyMemRef(rewriter, loc, valH, valC, token);
912 token = genDeallocMemRef(rewriter, loc, rowA, token);
913 token = genDeallocMemRef(rewriter, loc, colA, token);
914 token = genDeallocMemRef(rewriter, loc, valA, token);
915 token = genDeallocMemRef(rewriter, loc, rowB, token);
916 token = genDeallocMemRef(rewriter, loc, colB, token);
917 token = genDeallocMemRef(rewriter, loc, valB, token);
918 token = genDeallocMemRef(rewriter, loc, rowC, token);
919 token = genDeallocMemRef(rewriter, loc, colC, token);
920 token = genDeallocMemRef(rewriter, loc, valC, token);
921 token = genDeallocMemRef(rewriter, loc, buffer1, token);
922 token = genDeallocMemRef(rewriter, loc, buffer2, token);
923 tokens.push_back(token);
979 Value token = genFirstWait(rewriter, loc);
981 loc, spMatHandleTp, tokenTp, token, szm, szk,
984 token = spGenA->getResult(1);
986 loc, dnTensorHandleTp, tokenTp, token, matB,
989 token = dmatB.getAsyncToken();
991 loc, dnTensorHandleTp, tokenTp, token, matC,
994 token = dmatC.getAsyncToken();
1001 loc, bufferTypes, tokenTp, token, gpu::TransposeMode::NON_TRANSPOSE,
1004 token = bufferComp.getAsyncToken();
1008 auto buf1 = genAllocBuffer(rewriter, loc, bufferSz1, token);
1010 token = buf1.getAsyncToken();
1012 auto buf2 = genAllocBuffer(rewriter, loc, bufferSz2, token);
1014 token = buf2.getAsyncToken();
1016 auto buf3 = genAllocBuffer(rewriter, loc, bufferSz3, token);
1018 token = buf3.getAsyncToken();
1023 loc, tokenTp, token, spMatA, dnB, dnC, /*computeType=*/dnCType,
1025 token = spmmComp.getAsyncToken();
1028 token = rewriter.create<gpu::DestroySpMatOp>(loc, tokenTp, token, spMatA)
1030 token = rewriter.create<gpu::DestroyDnTensorOp>(loc, tokenTp, token, dnB)
1032 token = rewriter.create<gpu::DestroyDnTensorOp>(loc, tokenTp, token, dnC)
1035 token = genDeallocMemRef(rewriter, loc, buffer1, token);
1036 token = genDeallocMemRef(rewriter, loc, buffer2, token);
1037 token = genDeallocMemRef(rewriter, loc, buffer3, token);
1038 token = genDeallocMemRef(rewriter, loc, matA, token);
1039 token = genDeallocMemRef(rewriter, loc, matB, token);
1040 token = genCopyMemRef(rewriter, loc, bufC, matC, token);
1041 token = genDeallocMemRef(rewriter, loc, matC, token);
1042 tokens.push_back(token);
1096 Value token = genFirstWait(rewriter, loc);
1098 loc, dnMatHandleTp, tokenTp, token, matA, SmallVector<Value>{szm, szk});
1100 token = dmatA.getAsyncToken();
1102 loc, dnMatHandleTp, tokenTp, token, matB, SmallVector<Value>{szk, szn});
1104 token = dmatB.getAsyncToken();
1106 genSpMat(rewriter, loc, cTp, spMatHandleTp, tokenTp, token, szm, szn,
1109 token = spGenC->getResult(1);
1114 loc, indexTp, tokenTp, token, dnA, dnB, spMatC, dnCType);
1116 token = bufferComp.getAsyncToken();
1117 auto buf = genAllocBuffer(rewriter, loc, bufferSz, token);
1119 token = buf.getAsyncToken();
1122 auto sddmmComp = rewriter.create<gpu::SDDMMOp>(loc, tokenTp, token, dnA, dnB,
1124 token = sddmmComp.getAsyncToken();
1127 token = rewriter.create<gpu::DestroyDnTensorOp>(loc, tokenTp, token, dnA)
1129 token = rewriter.create<gpu::DestroyDnTensorOp>(loc, tokenTp, token, dnB)
1131 token = rewriter.create<gpu::DestroySpMatOp>(loc, tokenTp, token, spMatC)
1133 token = genDeallocMemRef(rewriter, loc, buffer, token);
1134 token = genDeallocMemRef(rewriter, loc, matA, token);
1135 token = genDeallocMemRef(rewriter, loc, matB, token);
1136 token = genDeallocMemRef(rewriter, loc, rowC, token);
1138 token = genDeallocMemRef(rewriter, loc, colC, token);
1139 token = genCopyMemRef(rewriter, loc, memV, valC, token);
1140 token = genDeallocMemRef(rewriter, loc, valC, token);
1141 tokens.push_back(token);
1223 // opens tokens and yielding a new token for the output.