Improve SSA optimizations #268
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
jserv
requested review from
ChAoSUnItY,
DrXiao,
fennecJ,
nosba0957 and
vacantron
jserv
force-pushed
the
improve-ssa
branch
3 times, most recently
from
a651980 to
c6ebcf0
Compare
|
@cubic-dev-ai review this PR |
![cubic-dev-ai[bot]](https://avatars.githubusercontent.com/in/1082092?s=60&v=4){kind=small}
jserv
force-pushed
the
improve-ssa
branch
4 times, most recently
from
cdd200f to
8b2fed7
Compare
DrXiao
reviewed
Sep 4, 2025
Comment on lines
+1755
to
+1767
| if (rs1 && rs1->is_const && !rs1->ptr_level && !rs1->is_global) { | ||
| vd = require_var(parent); | ||
| gen_name_to(vd->var_name); | ||
| vd->is_const = true; | ||
| vd->init_val = !rs1->init_val; | ||
| opstack_push(vd); | ||
| add_insn(parent, *bb, OP_load_constant, vd, NULL, NULL, 0, NULL); | ||
| } else { | ||
| vd = require_var(parent); | ||
| gen_name_to(vd->var_name); | ||
| opstack_push(vd); | ||
| add_insn(parent, *bb, OP_log_not, vd, rs1, NULL, 0, NULL); | ||
| } |
There was a problem hiding this comment.
require_var() and gen_name_to() can be moved to before the if statement to improve code reusability. That is:
vd = require_var(parent);
gen_name_to(vd->var_name);
if (rs1 && rs1->is_const && !rs1->ptr_level && !rs1->is_global) {
vd->is_const = true;
...
} else {
opstack_push(vd);
add_insn(...);
}
Comment on lines
+1774
to
+1786
| if (rs1 && rs1->is_const && !rs1->ptr_level && !rs1->is_global) { | ||
| vd = require_var(parent); | ||
| gen_name_to(vd->var_name); | ||
| vd->is_const = true; | ||
| vd->init_val = ~rs1->init_val; | ||
| opstack_push(vd); | ||
| add_insn(parent, *bb, OP_load_constant, vd, NULL, NULL, 0, NULL); | ||
| } else { | ||
| vd = require_var(parent); | ||
| gen_name_to(vd->var_name); | ||
| opstack_push(vd); | ||
| add_insn(parent, *bb, OP_bit_not, vd, rs1, NULL, 0, NULL); | ||
| } |
Comment on lines
+2205
to
+2219
| /* Constant folding for negation */ | ||
| if (rs1 && rs1->is_const && !rs1->ptr_level && !rs1->is_global) { | ||
| vd = require_var(parent); | ||
| gen_name_to(vd->var_name); | ||
| vd->is_const = true; | ||
| vd->init_val = -rs1->init_val; | ||
| opstack_push(vd); | ||
| add_insn(parent, *bb, OP_load_constant, vd, NULL, NULL, 0, | ||
| NULL); | ||
| } else { | ||
| vd = require_var(parent); | ||
| gen_name_to(vd->var_name); | ||
| opstack_push(vd); | ||
| add_insn(parent, *bb, OP_negate, vd, rs1, NULL, 0, NULL); | ||
| } |
DrXiao
reviewed
Sep 4, 2025
There was a problem hiding this comment.
Most of the introduced optimization look good. However, I currently don't understand "Phi node optimization" because I lack knowledge about the phi function.
It would be better if the other collaborators who have the relevant knowledge could review it. Otherwise, I'll need to spend some time learning the phi function before coming back to review the phi node optimization.
The BX (Branch and Exchange) instruction is needed for proper function returns in ARM code generation. Unlike BLX which saves the return address, BX simply branches to the address in the register. This fixes undefined reference errors during compilation.
Optimize unary operations on constants at parse time: - Logical NOT (!x) when x is constant - Bitwise NOT (~x) when x is constant - Negation (-x) when x is constant This reduces the number of instructions generated and enables further optimizations in later passes.
Enhanced dead store elimination to detect stores overwritten within a small window (3 instructions). The optimization: - Checks for intervening uses before marking as dead - Stops at control flow boundaries for safety - Marks dead stores for removal by DCE sweep This catches common patterns like consecutive assignments while remaining conservative to avoid incorrect elimination.
Implement zero-check guards to prevent undefined behavior in: - x / x = 1 optimization (only when x is provably non-zero) - x % x = 0 optimization (only when x is provably non-zero) - x / 1 = x and x % 1 = 0 (always safe) These guards ensure correctness by only applying optimizations when operands are compile-time constants with non-zero values, addressing reviewer concerns about potential division by zero.
Add comprehensive algebraic optimization patterns: - Self-operations: x-x=0, x^x=0, x&x=x, x|x=x - Comparisons: x==x=1, x!=x=0, x<x=0, x<=x=1 - Identity operations: x+0=x, x*1=x, x&-1=x - Constant folding: x*0=0, x&0=0, x|-1=-1 - Special cases: x%1=0, x*-1=-x, x<<0=x These patterns improve code generation by eliminating redundant operations and simplifying expressions at compile time.
Implement trivial phi node elimination: - Remove phi nodes where all operands are the same variable - Replace with simple assignment (phi(x,x,x) = x) - Fold phi nodes with all same constant values - Convert to load_constant for compile-time evaluation This optimization reduces unnecessary phi operations in SSA form, improving both compile time and generated code quality by eliminating redundant merge points.
Add sophisticated cross-instruction optimizations: Store-to-Load Forwarding: - Forward stored values directly to subsequent loads - Eliminate unnecessary memory round-trips - Validate no intervening calls or branches Redundant Load Elimination: - Reuse values from previous loads to same location - Check for no intervening stores or calls - Convert redundant loads to simple assignments Strength Reduction: - Convert multiply by power-of-2 to left shift - Convert divide by power-of-2 to right shift - Convert modulo by power-of-2 to bitwise AND These patterns analyze instruction sequences to find optimization opportunities that single-instruction analysis would miss.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Optimizations Added:
x-x,x^x,x&x,x|x) to peepholeSSA handles constants, peephole handles registers.