From Bytes to Registers

This page answers a question that the Bit Lab’s signed-or-unsigned section raises: why does a compiled if use one load instruction for a uint8_t and a different one for an int8_t, when both are a single byte in memory? It is part 8 of From Gates to a Compiled If, after the compare-and-branch page.

What it covers¶

1. Why a byte has to widen. The processor computes in registers, ARM’s are 32 bits wide, and there is no 8-bit compare, so a byte must be loaded into a full register and 24 bits that were never in memory have to be filled with something.

2. One byte, both loads. 0x80 through LDRB (zeros, giving 128) and LDRSB (copies of bit 7, giving −128), then 0x05, on which the two loads agree. The top bit decides.

3. Your pair, live. Two tappable bytes and a uint8_t/int8_t toggle. The panel widens both bytes, runs the 32-bit compare as A + NOT B + 1 across 32 columns with the carries shown, reads N Z C V off the result, and says when and why the 32-bit flags differ from the lab’s 8-bit ones while the branch lands the same way.

Next¶

• The Bit Lab: the loads inside a compiled if, beside the C that produced it.

• The flag drills: widen a byte by hand and check it.