Skip to content

[cc] update docs #476

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
jgarzik merged 1 commit into from
Dec 19, 2025
Merged

[cc] update docs #476

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
24 changes: 1 addition & 23 deletions cc/doc/TODO.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,7 +4,6 @@

- [Technical Debt](#technical-debt)
- [R10 reserved globally for division scratch](#r10-reserved-globally-for-division-scratch)
- [XMM register save area for variadic functions](#xmm-register-save-area-for-variadic-functions-x86-64)
- [Future Features](#future-features)
- [C11 Atomics (`_Atomic`)](#c11-atomics-_atomic-implementation)
- [C11 Alignment Specifiers (`_Alignas`, `_Alignof`)](#c11-alignment-specifiers-_alignas-_alignof)
Expand All @@ -25,7 +24,7 @@

### R10 reserved globally for division scratch

**Location**: `arch/x86_64/codegen.rs` lines 160-179
**Location**: `arch/x86_64/regalloc.rs` lines 187-208

**Issue**: R10 is permanently excluded from the allocatable register pool because x86-64 `div`/`idiv` instructions clobber RAX (quotient) and RDX (remainder). When the divisor operand happens to be in RAX or RDX, we need a scratch register to hold it during the division.

Expand All @@ -45,27 +44,6 @@

Production compilers (GCC, LLVM) use approaches 2-4. The current approach is simple and correct but sacrifices some optimization potential.

### XMM register save area for variadic functions (x86-64)

**Location**: `arch/x86_64/codegen.rs` lines 1012-1015

**Issue**: On x86-64 SysV ABI, variadic functions that receive floating-point arguments via XMM0-XMM7 should save those registers in the register save area during the prologue. This allows `va_arg(ap, double)` to retrieve FP arguments passed in registers.

**Current state**: The compiler correctly:
- Passes FP arguments to variadic functions (caller side) in XMM registers
- Sets AL to the count of XMM registers used for variadic calls
- Saves GP registers (RDI, RSI, RDX, RCX, R8, R9) in the register save area

However, XMM registers are not saved, so user-defined variadic functions cannot use `va_arg` to retrieve `double` arguments.

**Impact**: Low. Most POSIX variadic functions (printf, sprintf, scanf, etc.) are provided by libc and work correctly. This only affects user-defined variadic functions that accept floating-point variadic arguments.

**Solution**: In the function prologue for variadic functions:
1. Allocate additional space in the register save area (8 XMM regs × 16 bytes = 128 bytes)
2. Use `movaps` to save XMM0-XMM7 at their ABI-specified offsets
3. Update `va_start` to initialize the FP register save area pointer correctly
4. Update `va_arg` to read from XMM save area when extracting FP types

---

## Future Features
Expand Down