[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Fri, 28 Jul 2023 14:50:34 -0400
From: Rich Felker <dalias@...c.org>
To: zhangfei <zhang_fei_0403@....com>
Cc: musl@...ts.openwall.com, zhangfei <zhangfei@...iscas.ac.cn>
Subject: Re: [PATCH 1/1] RISC-V: Add some mathematical functions to
riscv64
On Fri, Jul 28, 2023 at 02:19:55PM +0800, zhangfei wrote:
> From: zhangfei <zhangfei@...iscas.ac.cn>
>
> Add a series of function implementations such as lrint and lround.
>
> Signed-off-by: Zhang Fei<zhangfei@...iscas.ac.cn>
> ---
> src/math/riscv64/llrint.c | 16 ++++++++++++++++
> src/math/riscv64/llrintf.c | 16 ++++++++++++++++
> src/math/riscv64/llround.c | 16 ++++++++++++++++
> src/math/riscv64/llroundf.c | 16 ++++++++++++++++
> src/math/riscv64/lrint.c | 16 ++++++++++++++++
> src/math/riscv64/lrintf.c | 16 ++++++++++++++++
> src/math/riscv64/lround.c | 16 ++++++++++++++++
> src/math/riscv64/lroundf.c | 16 ++++++++++++++++
> 8 files changed, 128 insertions(+)
> create mode 100644 src/math/riscv64/llrint.c
> create mode 100644 src/math/riscv64/llrintf.c
> create mode 100644 src/math/riscv64/llround.c
> create mode 100644 src/math/riscv64/llroundf.c
> create mode 100644 src/math/riscv64/lrint.c
> create mode 100644 src/math/riscv64/lrintf.c
> create mode 100644 src/math/riscv64/lround.c
> create mode 100644 src/math/riscv64/lroundf.c
>
> diff --git a/src/math/riscv64/llrint.c b/src/math/riscv64/llrint.c
> new file mode 100644
> index 0000000..2b5ea25
> --- /dev/null
> +++ b/src/math/riscv64/llrint.c
> @@ -0,0 +1,16 @@
> +#include <math.h>
> +
> +#if __riscv_flen >= 64
> +
> +long long llrint (double x)
> +{
> + long long res;
> + __asm__ ("fcvt.l.d %0, %1" : "=r" (res) : "f" (x));
> + return res;
> +}
> +
> +#else
> +
> +#include "../llrint.c"
> +
> +#endif
> diff --git a/src/math/riscv64/llrintf.c b/src/math/riscv64/llrintf.c
> new file mode 100644
> index 0000000..d69566b
> --- /dev/null
> +++ b/src/math/riscv64/llrintf.c
> @@ -0,0 +1,16 @@
> +#include <math.h>
> +
> +#if __riscv_flen >= 32
> +
> +long long llrintf (float x)
> +{
> + long long res;
> + __asm__ ("fcvt.l.s %0, %1" : "=r" (res) : "f" (x));
> + return res;
> +}
> +
> +#else
> +
> +#include "../llrintf.c"
> +
> +#endif
> diff --git a/src/math/riscv64/llround.c b/src/math/riscv64/llround.c
> new file mode 100644
> index 0000000..d6d2619
> --- /dev/null
> +++ b/src/math/riscv64/llround.c
> @@ -0,0 +1,16 @@
> +#include <math.h>
> +
> +#if __riscv_flen >= 64
> +
> +long long llround (double x)
> +{
> + long long res;
> + __asm__ ("fcvt.l.d %0, %1, rmm" : "=r" (res) : "f" (x));
> + return res;
> +}
> +
> +#else
> +
> +#include "../llround.c"
> +
> +#endif
This and all of the other lround() variants are almost surely wrong.
AFAICT, as you've written them, they behave identically to the
corresponding lrint variants. However they have very different
rounding semantics.
Please check that these instructions are right for what you're trying
to use them for (for the lrint variants as well, tho I suspect they're
right for that).
Rich
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
