Skip to content

riscv: keep tiny FMA RNE/RMM results on the host rounded path #313

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

Open
carlosqwqqwq wants to merge 1 commit into
base: staging
Choose a base branch
from
Open

riscv: keep tiny FMA RNE/RMM results on the host rounded path #313

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
13 changes: 12 additions & 1 deletion src/util/fpu_lib.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1099,8 +1099,19 @@ static forceinline fpu_f64_t fpu_fma64_raw(fpu_f64_t a, fpu_f64_t b, fpu_f64_t c
return fpu_wrap_f64(__builtin_fma(fpu_raw_f64(a), fpu_raw_f64(b), fpu_raw_f64(c)));
#else
fpu_f64_t mul = fpu_mul64(a, b);
fpu_f64_t e_m = fpu_mul_error64(mul, a, b);
fpu_f64_t sum = fpu_add64(mul, c);
if (unlikely((fpu_bit_f64_to_u64(sum) & FPU_LIB_FP64_NOSIGNED_MASK) < 0x0350000000000000ULL &&
(fpu_get_rounding_mode() == FPU_LIB_ROUND_NE || fpu_get_rounding_mode() == FPU_LIB_ROUND_MM))) {
const uint64_t mul_bits = fpu_bit_f64_to_u64(mul);
const uint64_t c_bits = fpu_bit_f64_to_u64(c);
const uint64_t sum_mag = fpu_bit_f64_to_u64(sum) & FPU_LIB_FP64_NOSIGNED_MASK;
const uint64_t mul_mag = mul_bits & FPU_LIB_FP64_NOSIGNED_MASK;
const uint64_t c_mag = c_bits & FPU_LIB_FP64_NOSIGNED_MASK;
const bool cancellation = ((mul_bits ^ c_bits) >> 63) &&
sum_mag < (mul_mag < c_mag ? mul_mag : c_mag);
if (!cancellation) return sum;
}
fpu_f64_t e_m = fpu_mul_error64(mul, a, b);
fpu_f64_t e_s = fpu_add_error64(sum, mul, c);
fpu_f64_t e_f = fpu_add64(e_m, e_s);
fpu_f64_t err = fpu_odd_round64(e_f, fpu_add_error64(e_f, e_s, e_m));
Expand Down
Loading