[PATCH] D69870: [builtins] Implement rounding mode support for i386/x86_64

[Nick Desaulniers via Phabricator via llvm-commits]

nickdesaulniers added inline comments.


================
Comment at: compiler-rt/lib/builtins/i386/fp_mode.c:19
+
+#define X87_INEXACT    0x20
+
----------------
This could be a local `const uint32_t` (local to `__fe_raise_inexact`).


================
Comment at: compiler-rt/lib/builtins/i386/fp_mode.c:43-45
+  cw &= X87_RMODE_MASK;
+
+  switch (cw) {
----------------
`switch (cw & X87_RMODE_MASK)`. LLVM will optimize such expressions in a switch.


================
Comment at: compiler-rt/lib/builtins/i386/fp_mode.c:66
+#ifdef __SSE_MATH__
+  uint32_t mxcsr;
+  __asm__ __volatile__ ("stmxcsr %0": "=m" (mxcsr));
----------------
Should you use `env.__mxcsr` here? If so, then the preprocessor guard differs from the above.


================
Comment at: compiler-rt/lib/builtins/i386/fp_mode.c:69
+  mxcsr |= X87_INEXACT;
+  __asm__ __volatile__ ("ldmxcsr %0": : "m" (mxcsr));
+#endif
----------------
Do *any* of these `asm` statements in this patch require `volatile` qualification?


Repository:
  rCRT Compiler Runtime

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D69870/new/

https://reviews.llvm.org/D69870