r8asm: make .loop consume the bit it just cleared
.endloop expanded to >-+]< : flb clears the LSB and rnz then tests that
very same bit, so it always sees 0 and the body ran exactly once no
matter how many 1s the counter cell held. Measured: a counter of 0b111
came out as 0b110 after a single pass.
Putting a ror after the flb pulls the next bit down before the test,
which is what SPEC describes -- consecutive 1s from the LSB, one per
iteration. Measured:
0b0000_0111 -> 3 passes, counter 0
0b1111_1111 -> 8 passes, counter 0 (a bare ror would spin forever)
Nothing under asm/ uses .loop, so no existing tape changes.
This commit is contained in:
1 parent
3f3e431d64
commit
1af92eb156
1 file changed
+5
@@ -281,8 +281,13 @@ r8asm::ast_node r8asm::ast::expand_node_to_ins(const ast_node &node) {
|
||||
// >[<
|
||||
} break;
|
||||
case builtin_t::ENDLOOP: {
|
||||
// Consume one '1' from the auxiliary cell and pull the next bit down.
|
||||
// Without the ror, the rnz could only ever see the bit that flb just
|
||||
// cleared, so a .loop over 0b111 ran once instead of three times
|
||||
// (SPEC: "consecutive 1s starting from the LSB").
|
||||
to_return.children.push_back(ROT8_NODE("stp"));
|
||||
to_return.children.push_back(ROT8_NODE("flb"));
|
||||
to_return.children.push_back(ROT8_NODE("ror"));
|
||||
to_return.children.push_back(ROT8_NODE("rnz"));
|
||||
to_return.children.push_back(ROT8_NODE("btp"));
|
||||
// >-+]<
|
||||
|
||||
Reference in new issue
Block a user