Fixing the boolean copy from data

[akahard2dj]

closes #2921

Hello,
I am very pleased with contributing to numba project.
This PR is first job for me.
Please reviewing my PR.

I fixed the bug for taking a numpy boolean array to memory.
In my opion, there is no problem for copying from disk by using 1bit size.

I have also tested in llvmlite using ir.
There is no problem when boolean copy to 1 bit size.

import numpy as np
from llvmlite import ir

a = np.bool(True)
print(a)

int8 = ir.IntType(8)
int1 = ir.IntType(1)

bit_bool = int1(a)
byte_bool = int8(a)
print(bit_bool)
print(byte_bool)

True
i1 true
i8 true

If there any problems in my PR, I welcome valuable comments.

[stuartarchibald]

Many thanks for the patch. Unfortunately this does not fix the problem presented and many unit tests fail because there are now multiple occurrences of invalid store of i8* to i1* that occur when trying to lower the Numba IR to LLVM IR. I think #2921 (comment) was alluding to something along the lines of:

    def from_data(self, builder, value):
        ty = self.get_value_type()
        resalloca = cgutils.alloca_once(builder, ty)
        cond = builder.icmp_unsigned('==', value, value.type(0))
        with builder.if_else(cond) as (then, otherwise):
            with then:
                builder.store(ty(0), resalloca)
            with otherwise:
                builder.store(ty(1), resalloca)
        return builder.load(resalloca)

in place of:

numba/numba/datamodel/models.py

Lines 192 to 193 in 2a2b772

	def from_data(self, builder, value):
	return builder.trunc(value, self.get_value_type())

[akahard2dj]

That too bad.
I will re-fix this feature.
Thanks! for your valuable comments!

[stuartarchibald]

Closing as #3862 fixed the issue.