Skip to content

RISC-V __mulsf3: handle exponent increase due to rounding #2086

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
merged 1 commit into from
Nov 23, 2024
Merged

RISC-V __mulsf3: handle exponent increase due to rounding #2086

Show file tree
Hide file tree
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
11 changes: 8 additions & 3 deletions src/rp2_common/pico_float/float_single_hazard3.S
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -216,13 +216,10 @@ __mulsf3:
clz a0, a4
sll a4, a4, a0
sub a2, a2, a0
// After normalising we can calculate the final exponent, since rounding
// cannot increase the exponent for multiplication (unlike addition)
add a2, a2, a3
// Subtract redundant bias term (127), add 1 for normalisation correction
addi a2, a2, -126
blez a2, __mulsf_underflow
bge a2, t0, __mulsf_overflow

// Gather sticky bits from low fraction:
snez a1, a1
Expand All @@ -231,13 +228,21 @@ __mulsf3:
bexti a1, a4, 8
add a4, a4, a1
addi a4, a4, 127
// Check carry-out: exponent may increase due to rounding
bgez a4, 2f
1:
bge a2, t0, __mulsf_overflow
// Pack it and ship it
packh a2, a2, a6
slli a2, a2, 23
slli a4, a4, 1
srli a4, a4, 9
add a0, a4, a2
ret
2:
srli a4, a4, 1
addi a2, a2, 1
j 1b

__mulsf_underflow:
// Signed zero
Expand Down
12 changes: 12 additions & 0 deletions test/pico_float_test/pico_float_test_hazard3.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -149,6 +149,18 @@ test_t mul_directed_tests[] = {
// 1.25 x 2^-63 x 1.25 x 2^-64 = 0
// (normal inputs with subnormal output, and we claim to be FTZ)
{0x20200000u, 0x1fa00000u, 0x00000000u},
// 1.333333 (rounded down) x 1.5 = 2 - 1 ulp
{0x3faaaaaau, 0x3fc00000u, 0x3fffffffu},
// 1.333333 (rounded down) x (1.5 + 1 ulp) = 2 exactly
{0x3faaaaaau, 0x3fc00001u, 0x40000000u},
// (1.333333 (rounded down) + 1 ulp) x 1.5 = 2 exactly
{0x3faaaaabu, 0x3fc00000u, 0x40000000u},
// (1.25 - 1 ulp) x (0.8 + 1 ulp) = 1 exactly (exponent increases after rounding)
{0x3f9fffffu, 0x3f4cccceu, 0x3f800000u},
// as above, but overflow on exponent increase -> +inf
{0x3f9fffffu, 0x7f4cccceu, 0x7f800000u},
// subtract 1 ulp from rhs -> largest normal
{0x3f9fffffu, 0x7f4ccccdu, 0x7f7fffffu},
};

#define N_RANDOM_TESTS 1000
Expand Down