On 2017/1/23 9:08, David Wohlferd wrote:
> Hmm.
>
> It seems a bit backwards to have the function that takes a 'long double'
> calling the function that takes a 'double.' Yes, they are both the same
> size on ARM, but I think I would have gone the other way. Plus I kinda
> like having all the
Hmm.
It seems a bit backwards to have the function that takes a 'long double'
calling the function that takes a 'double.' Yes, they are both the same
size on ARM, but I think I would have gone the other way. Plus I kinda
like having all the implementations in one file (fmal.c).
Other than
The mail has been being rejected for spamming for a few hours.
Hope it wouldn't be this time.
--
Best regards,
lh_mouse
From 82fd24e992a402ff2f7c55780fd76945ef83e094 Mon Sep 17 00:00:00 2001
From: LH_Mouse
Date: Wed, 18 Jan 2017 19:35:43 +0800
Subject: [PATCH]
> So you have decided that __builtins can't be used then? That's too bad.
Yes it results in a call to `fma()` on x64. Can't test it on ARM though.
> I know almost nothing about the guts of floating point, so I'm prepared
> to defer to your judgement, but here's what I think:
>
> Let me propose
So you have decided that __builtins can't be used then? That's too bad.
I know almost nothing about the guts of floating point, so I'm prepared
to defer to your judgement, but here's what I think:
Let me propose an alternative for fma.c:
/**
* This file has no copyright assigned
New patch attached.
This patch fixes ARM functions and adds a check in `fpu_fma()` for potential
NaN or INF results.
--
Best regards,
lh_mouse
2017-01-19
From 3c55daec84dac190b9e3cb032371960e1acbc38f Mon Sep 17 00:00:00 2001
From: LH_Mouse
> I see that you have replaced the x86 parts for fma and fmaf with C
> code. That seems like a good thing. Is there some reason you can't do
> that with the ARM versions too?
ARM has hardware FMA and software emulation is not optimal.
> Reducing the amount of platform-specific code also seems
On 1/18/2017 5:14 AM, lhmouse wrote:
> Patch is attached.
I see that you have replaced the x86 parts for fma and fmaf with C
code. That seems like a good thing. Is there some reason you can't do
that with the ARM versions too?
Reducing the amount of platform-specific code also seems like a
The correctness of fma() function can be verified using the following program:
---
#include
#include
volatile double x = 0x1.3p52;
volatile double y = 0x1.5p52;
volatile double z = -0x1.8p104;
int main(){
Patch is attached.
This patch removes assembly files that implement FMA on ARM and merges
them into the corresponding C files with the same name using inline assembly.
A re-generation of Makefile.in is required.
I don't have any knowledge about ARM assembly. Those functions for ARM
were created
10 matches
Mail list logo