[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH WIP] replace numpy with struct
From: |
Philippe Mathieu-Daudé |
Subject: |
Re: [Qemu-devel] [PATCH WIP] replace numpy with struct |
Date: |
Thu, 26 Oct 2017 10:50:26 -0300 |
User-agent: |
Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101 Thunderbird/52.3.0 |
Hi Joannah,
The common subject to use is [RFC PATCH] instead of [PATCH WIP]
On 10/26/2017 09:14 AM, Joannah Nanjekye wrote:
> This patch replaces the use of numpy with the standard Library struct module
> where possible.
This seems a good idea but why?
Do you have any performance improvements results to share?
> Signed-off-by: Joannah Nanjekye <address@hidden>
> ---
>
> scripts/analyze-migration.py | 19 ++++++++++++-------
> 1 file changed, 12 insertions(+), 7 deletions(-)
>
> diff --git a/scripts/analyze-migration.py b/scripts/analyze-migration.py
> index 1455387..f421012 100755
> --- a/scripts/analyze-migration.py
> +++ b/scripts/analyze-migration.py
> @@ -36,23 +36,28 @@ class MigrationFile(object):
> self.file = open(self.filename, "rb")
>
> def read64(self):
> - return np.asscalar(np.fromfile(self.file, count=1, dtype='>i8')[0])
> + buffer = file.read(64)
> + return struct.unpack('>i16', buffer)[0]
>
> def read32(self):
> - return np.asscalar(np.fromfile(self.file, count=1, dtype='>i4')[0])
> + buffer = file.read(32)
> + return struct.unpack('>i8', buffer)[0]
>
> def read16(self):
> - return np.asscalar(np.fromfile(self.file, count=1, dtype='>i2')[0])
> + buffer = file.read(16)
> + return struct.unpack('>i4', buffer)[0]
>
> def read8(self):
> - return np.asscalar(np.fromfile(self.file, count=1, dtype='>i1')[0])
> + buffer = file.read(8)
> + return struct.unpack('>i2', buffer)[0]
>
> def readstr(self, len = None):
> + buffer = file.read(8)
^ move this
> if len is None:
> len = self.read8()
> if len == 0:
> return ""
here to avoid a read() if len=0
> - return np.fromfile(self.file, count=1, dtype=('S%d' % len))[0]
> + return np.array(struct.unpack(str(len) + 'd', buffer)[0])
Well this is now a mix of np + struct ...
This looks a bit more messy.
>
> def readvar(self, size = None):
> if size is None:
> @@ -303,8 +308,8 @@ class VMSDFieldInt(VMSDFieldGeneric):
>
> def read(self):
> super(VMSDFieldInt, self).read()
> - self.sdata = np.fromstring(self.data, count=1,
> dtype=(self.sdtype))[0]
> - self.udata = np.fromstring(self.data, count=1,
> dtype=(self.udtype))[0]
> + self.sdata = np.array(struct.unpack(self.sdtype, self.data)[0])
> + self.udata = np.array(struct.unpack(self.udtype, self.data)[0])
ditto.
> self.data = self.sdata
> return self.data
Regards,
Phil.