fastest way to convert int8 to int7 - c++

Question

I've a function that takes int8_t val and converts it to int7_t.
//Bit [7] reserved
//Bits [6:0] = signed -64 to +63 offset value
// user who calls this function will use it correctly (-64 to +63)
uint8_t func_int7_t(int8_t val){
uint8_t val_6 = val & 0b01111111;
if (val & 0x80)
val_6 |= 0x40;
//...
//do stuff...
return val_6;
}
What is the best and fastest way to manipulate the int8 to int7? Did I do it efficient and fast? or there is better way?
The target is ARM Cortex M0+ if that matters
UPDATE:
After reading different answers I can say the question was asked wrong? (or my code in the question is what gave wrong assumptions to others) I had the intension to make an int8 to int7
So it will be done by doing nothing because
8bit:
63 = 0011 1111
62 = 0011 1110
0 = 0000 0000
-1 = 1111 1111
-2 = 1111 1110
-63 = 1100 0001
-64 = 1100 0000
7bit:
63 = 011 1111
62 = 011 1110
0 = 000 0000
-1 = 111 1111
-2 = 111 1110
-63 = 100 0001
-64 = 100 0000

Answer 1

the faster way is probably :
uint8_t val_7 = (val & 0x3f) | ((val >> 1) & 0x40);
val & 0x3f get the 6 lower bits (truncate) and ((val >> 1) & 0x40) move the bit to sign from the position 8 to 7
The advantage to not use a if is to have a shorter code (even you can use arithmetic if) and to have a code without sequence break

Answer 2

To clear the reserved bit, just
return val & 0x7f;
To leave the reserved bit exactly like how it was from input, nothing needs to be done
return val;
and the low 7 bits will contain the values in [-64, 63]. Because in two's complement down casting is done by a simple truncation. The value remains the same. That's what happens for an assignment like (int8_t)some_int_value
There's no such thing as 0bX1100001. There's no undefined bit in machine language. That state only exists in hardware, like the high-Z state or undefined state in Verilog or other hardware description languages

Answer 3

Use bitfield to narrow the value and let compiler to choose what sequence of shifts and/or masks is most efficient for that on your platform.
inline uint8_t to7bit(int8_t x)
{
struct {uint8_t x:7;} s;
return s.x = x;
}

Answer 4

If you are not concerned about what happens to out-of-range values, then
return val & 0x7f;
is enough. This correctly handles values in the range -64 <= val <= 63.
You haven't said how you want to handle out-of-range values, so I have nothing to say about that.
Updated to add: The question has been updated so stipulate that the function will never be called with out-of-range values. So this method qualifies unambiguously as "best and fastest".

Answer 5

the user who calls this function he knows he should put data -64 to +63
So not considering any other values, the really fastest thing you can do is not doing anything at all!
You have a 7 bit value stored in eight bits. Any value within specified range will have both bit 7 and bit 6 the same value, and when you process the 7-bit value, you just ignore the MSB (of 8-bit value), no matter if set or not, e. g.:
for(unsigned int bit = 0x40; bit; bit >>= 1)
// NOT: 0x80!
std::cout << (value & bit);
The other way round is more critical: whenever you receive these seven bits via some communication channel, then you need to do manual sign extension for eight (or more) bits to be able to correctly use that value.