Tensor type assertion failure when switching off the matrix scalarization

[turbo0628]

Script:

import taichi as ti
import taichi.math as tm

ti.init(arch=ti.cuda, print_kernel_llvm_ir=True, print_ir=True, real_matrix_scalarize=False)

f = ti.field(ti.f32, 2)

@ti.kernel
def vec_test()->ti.f32:
    a = 1.0
    v1 = ti.math.vec4([a, a, a, a * 2])
    return v1[3]

vec_test()

Log and trace

[I 01/05/23 16:01:21.445 36690] [compile_to_offloads.cpp:operator()@22] [vec_test_c76_0] Matrix ptr lowered:
kernel {
  <f32> $0 = const 1.0
  <i32> $1 = const 2
  <f32> $2 = mul $0 $1
  <[Tensor (4) f32]> $3 = [$0, $0, $0, $2]
  <[Tensor (4) f32]> $4 = cast_value<f32> $3
  <[Tensor (4) f32]> $5 = alloca
  $6 : local store [$5 <- $4]
  <i32> $7 = const 3
  <*f32> $8 = shift ptr [$5 + $7]
  <f32> $9 = local load [$8]
  <f32> $10 = cast_value<f32> $9
  $11 : return tmp10
}
[E 01/05/23 16:01:21.445 36690] [type_factory.cpp:promoted_type@222] Assertion failure: a->is<TensorType>() && b->is<TensorType>()


Traceback (most recent call last):
  File "/home/taichigraphics/dev/taichi-fork/gemm.py", line 14, in <module>
    vec_test()
  File "/home/taichigraphics/miniconda3/envs/taichi-dev/lib/python3.9/site-packages/taichi-1.4.0-py3.9-linux-x86_64.egg/taichi/lang/kernel_impl.py", line 955, in wrapped
    return primal(*args, **kwargs)
  File "/home/taichigraphics/miniconda3/envs/taichi-dev/lib/python3.9/site-packages/taichi-1.4.0-py3.9-linux-x86_64.egg/taichi/lang/kernel_impl.py", line 882, in __call__
    return self.runtime.compiled_functions[key](*args)
  File "/home/taichigraphics/miniconda3/envs/taichi-dev/lib/python3.9/site-packages/taichi-1.4.0-py3.9-linux-x86_64.egg/taichi/lang/kernel_impl.py", line 807, in func__
    raise e from None
  File "/home/taichigraphics/miniconda3/envs/taichi-dev/lib/python3.9/site-packages/taichi-1.4.0-py3.9-linux-x86_64.egg/taichi/lang/kernel_impl.py", line 804, in func__
    t_kernel(launch_ctx)
RuntimeError: [type_factory.cpp:promoted_type@222] Assertion failure: a->is<TensorType>() && b->is<TensorType>()

[jim19930609]

Seems to be working with latest master commit, and I believe this is the PR that fixed it: #7124. Assigning back to @turbo0628 for further verification.