Attachment 136240 Details for Bug 104226 – 0001-intel-fs-bank_conflicts-Use-posix_memalign-instead-o.patch

[patch] 0001-intel-fs-bank_conflicts-Use-posix_memalign-instead-o.patch

0001-intel-fs-bank_conflicts-Use-posix_memalign-instead-o.patch (text/plain), 3.01 KB, created by Francisco Jerez on 2017-12-17 22:16:09 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: Francisco Jerez

Created: 2017-12-17 22:16:09 UTC

Size: 3.01 KB

patch

obsolete

>From 3dae9e585dee788bfde882c4499f8f0a97f47bca Mon Sep 17 00:00:00 2001
>From: Francisco Jerez <currojerez@riseup.net>
>Date: Sun, 17 Dec 2017 13:05:55 -0800
>Subject: [PATCH] intel/fs/bank_conflicts: Use posix_memalign() instead of
> overaligned new to obtain vector storage.
>MIME-Version: 1.0
>Content-Type: text/plain; charset=UTF-8
>Content-Transfer-Encoding: 8bit
>
>The weight_vector_type constructor was inadvertently assuming C++17
>semantics of the new operator applied on a type with alignment
>requirement greater than the largest fundamental alignment.
>Unfortunately on earlier C++ dialects the implementation was allowed
>to raise an allocation failure when the alignment requirement of the
>allocated type was unsupported, in an implementation-defined fashion.
>It's expected that a C++ implementation recent enough to implement
>P0035R4 would have honored allocation requests for such over-aligned
>types even if the C++17 dialect wasn't active, which is likely the
>reason why this problem wasn't caught by our CI system.
>
>A more elegant fix would involve wrapping the __SSE2__ block in a
>'__cpp_aligned_new >= 201606' preprocessor conditional, but that would
>yield lower compile-time performance on old compilers not implementing
>this language feature (e.g. GCC versions older than 7.0).
>
>Fixes: af2c320190f3c731 "intel/fs: Implement GRF bank conflict mitigation pass."
>Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104226
>Reported-by: JÃ³zef Kucia <joseph.kucia@gmail.com>
>---
> src/intel/compiler/brw_fs_bank_conflicts.cpp | 21 +++++++++++++++------
> 1 file changed, 15 insertions(+), 6 deletions(-)
>
>diff --git a/src/intel/compiler/brw_fs_bank_conflicts.cpp b/src/intel/compiler/brw_fs_bank_conflicts.cpp
>index 0cd880d44f2..783e672d564 100644
>--- a/src/intel/compiler/brw_fs_bank_conflicts.cpp
>+++ b/src/intel/compiler/brw_fs_bank_conflicts.cpp
>@@ -277,13 +277,10 @@ namespace {
>    struct weight_vector_type {
>       weight_vector_type() : v(NULL), size(0) {}
> 
>-      weight_vector_type(unsigned n) :
>-         v(new vector_type[DIV_ROUND_UP(n, vector_width)]()),
>-         size(n) {}
>+      weight_vector_type(unsigned n) : v(alloc(n)), size(n) {}
> 
>       weight_vector_type(const weight_vector_type &u) :
>-         v(new vector_type[DIV_ROUND_UP(u.size, vector_width)]()),
>-         size(u.size)
>+         v(alloc(u.size)), size(u.size)
>       {
>          memcpy(v, u.v,
>                 DIV_ROUND_UP(u.size, vector_width) * sizeof(vector_type));
>@@ -291,7 +288,7 @@ namespace {
> 
>       ~weight_vector_type()
>       {
>-         delete[] v;
>+         free(v);
>       }
> 
>       weight_vector_type &
>@@ -304,6 +301,18 @@ namespace {
> 
>       vector_type *v;
>       unsigned size;
>+
>+   private:
>+      static vector_type *
>+      alloc(unsigned n)
>+      {
>+         const unsigned sz = DIV_ROUND_UP(n, vector_width) * sizeof(vector_type);
>+         void *p;
>+         if (posix_memalign(&p, __alignof__(vector_type), sz))
>+            return NULL;
>+         memset(p, 0, sz);
>+         return reinterpret_cast<vector_type *>(p);
>+      }
>    };
> 
>    /**
>-- 
>2.14.2
>

Actions: View | Splinter Review

Attachments on bug 104226: 136195 | 136196 | 136215 | 136237 | 136238 | 136240

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.