From 7837eeb90de3dd955089cbd554d59c07e8c295eb Mon Sep 17 00:00:00 2001 From: Jakub Doka Date: Sun, 15 Dec 2024 19:37:37 +0100 Subject: [PATCH] implementing the loop iteration optimization the multiplication and addition to a pointer is replaced with simply incremrnting the pointer it self Signed-off-by: Jakub Doka --- lang/src/son.rs | 113 +++++++++++------- ..._tests_different_function_destinations.txt | 54 ++++----- .../tests/son_tests_tests_ptr_to_ptr_copy.txt | 37 +++--- 3 files changed, 115 insertions(+), 89 deletions(-) diff --git a/lang/src/son.rs b/lang/src/son.rs index b0cd8a2b..1e4be551 100644 --- a/lang/src/son.rs +++ b/lang/src/son.rs @@ -1068,47 +1068,72 @@ impl Nodes { return Some(self.new_node(ty, K::BinOp { op }, [ctrl, a, new_rhs], tys)); } - //if op == T::Add - // && self[rhs].kind == (K::BinOp { op: T::Mul }) - // && let &[_, mul_lhs, mul_rhs] = self[rhs].inputs.as_slice() - // && self[mul_lhs].kind == K::Phi - // && let &[phi_ctrl, phi_lhs, phi_rhs] = self[mul_lhs].inputs.as_slice() - // && phi_rhs != VOID - // && self[phi_ctrl].kind == K::Loop - // && self[phi_rhs].kind == (K::BinOp { op: T::Add }) - // && self[phi_rhs].inputs[1] == mul_lhs - //{ - // debug_assert_eq!( - // self[mul_lhs].outputs.len(), - // 1, - // "{:?}", - // self[mul_lhs] - // .outputs - // .iter() - // .map(|&n| ( - // &self[n].kind, - // self[n].inputs.iter().position(|&n| n == mul_lhs) - // )) - // .collect::>() - // ); + if op == T::Add + && self[rhs].kind == (K::BinOp { op: T::Mul }) + && let &[_, index, step] = self[rhs].inputs.as_slice() + && self[index].kind == K::Phi + && let &[iter_loop, index_init, new_index] = self[index].inputs.as_slice() + && new_index != VOID + && self[iter_loop].kind == K::Loop + && self[new_index].kind == (K::BinOp { op: T::Add }) + && self[new_index].inputs[1] == index + && let Some(&iter_cond) = self[index].outputs.iter().find( + |&&n| matches!(self[n].kind, Kind::BinOp { op } if op.is_compatison()), + ) + && self[index].outputs.iter().all(|n| [iter_cond, rhs, new_index].contains(n)) + { + // arr := @as([u32; 10], idk) + // + // i := 0 + // loop if i == 10 break else { + // arr[i] = 0 + // i += 1 + // } + // + // ||||| + // VVVVV + // + // cursor := &arr[0] + 0 + // end := &arr[0] + 10 + // loop if cursor == end else { + // *cursor = 0 + // i += 1 + // } - // let init_shift = self.new_node( - // self[rhs].ty, - // K::BinOp { op: T::Mul }, - // [ctrl, phi_lhs, mul_rhs], - // tys, - // ); - // let init = - // self.new_node(ty, K::BinOp { op: T::Add }, [ctrl, lhs, init_shift], tys); - // let new_value = self.new_node_nop(ty, K::Phi, [phi_ctrl, init, 0]); - // let next = self.new_node( - // ty, - // Kind::BinOp { op: T::Add }, - // [ctrl, new_value, mul_rhs], - // tys, - // ); - // return Some(self.modify_input(new_value, 2, next)); - //} + debug_assert!(self[iter_cond].inputs.contains(&index)); + let iter_bound_index = + self[iter_cond].inputs.iter().rposition(|&n| n != index).unwrap(); + debug_assert_ne!(iter_bound_index, 0); + let end_shift = self.new_node( + self[rhs].ty, + K::BinOp { op: T::Mul }, + [ctrl, self[iter_cond].inputs[iter_bound_index], step], + tys, + ); + let end = + self.new_node(ty, K::BinOp { op: T::Add }, [ctrl, lhs, end_shift], tys); + + let init_shift = self.new_node( + self[rhs].ty, + K::BinOp { op: T::Mul }, + [ctrl, index_init, step], + tys, + ); + let init = + self.new_node(ty, K::BinOp { op: T::Add }, [ctrl, lhs, init_shift], tys); + + let new_value = self.new_node_nop(ty, K::Phi, [iter_loop, init, 0]); + let next = + self.new_node(ty, Kind::BinOp { op: T::Add }, [ctrl, new_value, step], tys); + + let mut new_cond_inputs = self[iter_cond].inputs.clone(); + new_cond_inputs[iter_bound_index] = end; + new_cond_inputs[3 - iter_bound_index] = new_value; + let new_cond = self.new_node(ty, self[iter_cond].kind, new_cond_inputs, tys); + self.replace(iter_cond, new_cond); + + return Some(self.modify_input(new_value, 2, next)); + } if changed { return Some(self.new_node(ty, self[target].kind, [ctrl, lhs, rhs], tys)); @@ -1268,6 +1293,14 @@ impl Nodes { return Some(lhs); } + // TODO: travese the graph downward and chech if this phi is only consumed by it + // self + if self[target].outputs.as_slice() == [rhs] + && self[rhs].outputs.as_slice() == [target] + { + return Some(lhs); + } + if self[lhs].kind == Kind::Stre && self[rhs].kind == Kind::Stre && self[lhs].ty == self[rhs].ty diff --git a/lang/tests/son_tests_different_function_destinations.txt b/lang/tests/son_tests_different_function_destinations.txt index a15a2c05..d86d005b 100644 --- a/lang/tests/son_tests_different_function_destinations.txt +++ b/lang/tests/son_tests_different_function_destinations.txt @@ -1,36 +1,35 @@ main: - ADDI64 r254, r254, -104d - ST r31, r254, 48a, 56h + ADDI64 r254, r254, -88d + ST r31, r254, 48a, 40h LRA r32, r0, :glob_stru JAL r31, r0, :new_stru ST r1, r32, 0a, 16h - CP r33, r0 - LD r34, r32, 0a, 8h - JEQ r34, r33, :0 + LD r33, r32, 0a, 8h + JEQ r33, r0, :0 LI64 r32, 300d CP r1, r32 JMP :1 - 0: ST r33, r32, 0a, 8h - LD r34, r32, 0a, 8h - JEQ r34, r33, :2 + 0: ST r0, r32, 0a, 8h + LD r33, r32, 0a, 8h + JEQ r33, r0, :2 LI64 r32, 200d CP r1, r32 JMP :1 2: LI64 r34, 1d ST r34, r32, 0a, 8h ST r34, r32, 8a, 8h - ADDI64 r35, r254, 0d + ADDI64 r33, r254, 0d ST r34, r254, 0a, 8h ST r34, r254, 8a, 8h ST r34, r254, 16a, 8h ST r34, r254, 24a, 8h ST r34, r254, 32a, 8h ST r34, r254, 40a, 8h - LI64 r36, 3d + ADDI64 r35, r33, 48d CP r32, r33 - 8: JNE r32, r36, :3 + 8: JNE r35, r32, :3 LD r32, r254, 32a, 8h - JEQ r32, r33, :4 + JEQ r32, r0, :4 LI64 r32, 100d CP r1, r32 JMP :1 @@ -41,28 +40,25 @@ main: ST r34, r254, 32a, 8h ST r34, r254, 40a, 8h CP r32, r33 - 7: LD r37, r254, 32a, 8h - JNE r32, r36, :5 - JEQ r37, r33, :6 + 7: LD r34, r254, 32a, 8h + JNE r35, r32, :5 + JEQ r34, r0, :6 LI64 r32, 10d CP r1, r32 JMP :1 - 6: CP r1, r33 + 6: CP r1, r0 JMP :1 - 5: MULI64 r37, r32, 16d - ADD64 r37, r35, r37 - ST r33, r37, 0a, 8h - ST r33, r37, 8a, 8h - ADD64 r32, r32, r34 + 5: ADDI64 r34, r32, 16d + ST r0, r32, 0a, 8h + ST r0, r32, 8a, 8h + CP r32, r34 JMP :7 - 3: MULI64 r37, r32, 16d - ADD64 r37, r35, r37 - JAL r31, r0, :new_stru - ST r1, r37, 0a, 16h - ADD64 r32, r32, r34 + 3: JAL r31, r0, :new_stru + ST r1, r32, 0a, 16h + ADDI64 r32, r32, 16d JMP :8 - 1: LD r31, r254, 48a, 56h - ADDI64 r254, r254, 104d + 1: LD r31, r254, 48a, 40h + ADDI64 r254, r254, 88d JALA r0, r31, 0a new_stru: ADDI64 r254, r254, -16d @@ -71,6 +67,6 @@ new_stru: LD r1, r254, 0a, 16h ADDI64 r254, r254, 16d JALA r0, r31, 0a -code size: 673 +code size: 658 ret: 0 status: Ok(()) diff --git a/lang/tests/son_tests_tests_ptr_to_ptr_copy.txt b/lang/tests/son_tests_tests_ptr_to_ptr_copy.txt index 18482953..13510ef8 100644 --- a/lang/tests/son_tests_tests_ptr_to_ptr_copy.txt +++ b/lang/tests/son_tests_tests_ptr_to_ptr_copy.txt @@ -1,31 +1,28 @@ main: ADDI64 r254, r254, -10240d - LI64 r14, 1d - LI8 r16, 64b - LI64 r17, 1024d + LI8 r15, 64b + LI64 r16, 1024d CP r13, r0 - ADDI64 r18, r254, 0d - 4: JLTU r13, r17, :0 - LI64 r16, 10d - CP r13, r14 - 3: LD r15, r254, 2048a, 1h - JLTU r13, r16, :1 - ANDI r13, r15, 255d + ADDI64 r17, r254, 0d + 4: JLTU r13, r16, :0 + ADDI64 r13, r17, 1024d + ADDI64 r15, r17, 10240d + 3: LD r14, r254, 2048a, 1h + JLTU r13, r15, :1 + ANDI r13, r14, 255d CP r1, r13 JMP :2 - 1: ADD64 r15, r13, r14 - MULI64 r13, r13, 1024d - ADD64 r13, r18, r13 - BMC r18, r13, 1024h - CP r13, r15 + 1: ADDI64 r14, r13, 1024d + BMC r17, r13, 1024h + CP r13, r14 JMP :3 - 0: ADD64 r15, r13, r14 - ADD64 r13, r18, r13 - ST r16, r13, 0a, 1h - CP r13, r15 + 0: ADDI64 r14, r13, 1d + ADD64 r13, r17, r13 + ST r15, r13, 0a, 1h + CP r13, r14 JMP :4 2: ADDI64 r254, r254, 10240d JALA r0, r31, 0a -code size: 194 +code size: 192 ret: 64 status: Ok(())