GENERIC load not implemented

[fighterkin]

I build the gass with
"cmake -G Ninja ../llvm -DLLVM_ENABLE_PROJECTS="clang;clang-tools-extra" -DLLVM_TARGETS_TO_BUILD="GASS;NVPTX;X86" -DCMAKE_BUILD_TYPE=Debug -DLLVM_ENABLE_ASSERTIONS=ON -DLLVM_ENABLE_RTTI=ON -DCUDA_ARCH=70"
Then i create a .bc file with
"clang++ --cuda-path=/usr/local/cuda-11.3/ -std=c++11 --cuda-gpu-arch=sm_70 -emit-llvm test.cu -c "
test.cu looks like

Then i use llc to call gass backend
"llc --march=gass test-cuda-nvptx64-nvidia-cuda-sm_70.bc -o test.s"
I get the error report
"GENERIC load not implemented"
so, my question is what am doing is wrong and can you tell me how to test this program?
the error inst is
"t74: i32,ch = load<(dereferenceable load 4 from %ir.ix)> t73, FrameIndex:i64<5>, undef:i64"
the N->dumpr() is
"t74: i32,ch = load<(dereferenceable load 4 from %ir.ix)> t73, FrameIndex:i64<5>, undef:i64
t73: ch = store<(store 4 into %ir.idx)> t72, t70, FrameIndex:i64<7>, undef:i64
t72: ch = TokenFactor t66:1, t67:1, t68:1
t66: i32,ch = load<(dereferenceable load 4 from %ir.ix)> t65, FrameIndex:i64<5>, undef:i64
t65: ch = store<(store 4 into %ir.iy)> t52, t63, FrameIndex:i64<6>, undef:i64
t52: ch = store<(store 4 into %ir.ix)> t38, t50, FrameIndex:i64<5>, undef:i64
t38: ch = store<(store 4 into %ir.ny.addr)> t36, t10, FrameIndex:i64<4>, undef:i64
t36: ch = store<(store 4 into %ir.nx.addr)> t34, t8, FrameIndex:i64<3>, undef:i64
t34: ch = store<(store 8 into %ir.MatC.addr)> t32, t6, FrameIndex:i64<2>, undef:i64
t32: ch = store<(store 8 into %ir.MatB.addr)> t30, t4, FrameIndex:i64<1>, undef:i64
t30: ch = store<(store 8 into %ir.MatA.addr)> t0, t2, FrameIndex:i64<0>, undef:i64
t0: ch = EntryToken
t2: i64 = <<Unknown Node #338>> TargetConstant:i32<352>
t4: i64 = <<Unknown Node #338>> TargetConstant:i32<360>
t6: i64 = <<Unknown Node #338>> TargetConstant:i32<368>
t8: i32 = <<Unknown Node #338>> TargetConstant:i32<376>
t10: i32 = <<Unknown Node #338>> TargetConstant:i32<380>
t50: i32 = add t42, t49
t42: i32 = AssertZext t40, ValueType:ch:i10
t40: i32 = llvm.nvvm.read.ptx.sreg.tid.x TargetConstant:i64<5183>
t49: i32 = mul t44, t48
t44: i32 = llvm.nvvm.read.ptx.sreg.ntid.x TargetConstant:i64<5173>
t48: i32 = AssertZext t46, ValueType:ch:i31
t46: i32 = llvm.nvvm.read.ptx.sreg.ctaid.x TargetConstant:i64<5125>
t63: i32 = add t55, t62
t55: i32 = AssertZext t54, ValueType:ch:i10
t54: i32 = llvm.nvvm.read.ptx.sreg.tid.y TargetConstant:i64<5184>
t62: i32 = mul t57, t61
t57: i32 = llvm.nvvm.read.ptx.sreg.ntid.y TargetConstant:i64<5174>
t61: i32 = AssertZext t59, ValueType:ch:i16
t59: i32 = llvm.nvvm.read.ptx.sreg.ctaid.y TargetConstant:i64<5126>
t67: i32,ch = load<(dereferenceable load 4 from %ir.iy)> t65, FrameIndex:i64<6>, undef:i64
t68: i32,ch = load<(dereferenceable load 4 from %ir.ny.addr)> t65, FrameIndex:i64<4>, undef:i64
t70: i32 = add nsw t66, t69
t69: i32 = mul nsw t67, t68"

[fighterkin]

主要是想咨询一下，基于 sm_70 架构环境下，要怎么跑测试用例。