[arch-commits] Commit in tensorflow/trunk (10868.patch PKGBUILD)
Sven-Hendrik Haase
svenstaro at archlinux.org
Thu Jun 22 06:55:35 UTC 2017
Date: Thursday, June 22, 2017 @ 06:55:34
Author: svenstaro
Revision: 239452
upgpkg: tensorflow 1.2.0-1
Added:
tensorflow/trunk/10868.patch
Modified:
tensorflow/trunk/PKGBUILD
-------------+
10868.patch | 89 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
PKGBUILD | 15 ++++++---
2 files changed, 99 insertions(+), 5 deletions(-)
Added: 10868.patch
===================================================================
--- 10868.patch (rev 0)
+++ 10868.patch 2017-06-22 06:55:34 UTC (rev 239452)
@@ -0,0 +1,89 @@
+From eb197c45eda55b1047b00f1136a4fec2b8adb2d5 Mon Sep 17 00:00:00 2001
+From: Todd Wang <toddwang at gmail.com>
+Date: Tue, 20 Jun 2017 14:16:26 -0700
+Subject: [PATCH 1/3] [XLA] Explicitly instantiate Permute template func
+
+This is an attempted fix for older compilers that can't deduce the template args.
+---
+ tensorflow/compiler/xla/service/llvm_ir/ir_array.cc | 2 +-
+ 1 file changed, 1 insertion(+), 1 deletion(-)
+
+diff --git a/tensorflow/compiler/xla/service/llvm_ir/ir_array.cc b/tensorflow/compiler/xla/service/llvm_ir/ir_array.cc
+index e401305ae73..b14104be6ad 100644
+--- a/tensorflow/compiler/xla/service/llvm_ir/ir_array.cc
++++ b/tensorflow/compiler/xla/service/llvm_ir/ir_array.cc
+@@ -158,7 +158,7 @@ IrArray::Index IrArray::Index::SourceIndexOfTranspose(
+ tensorflow::gtl::ArraySlice<int64> dimension_mapping,
+ llvm::IRBuilder<>* builder) const {
+ std::vector<llvm::Value*> operand_multidim_index =
+- Permute(dimension_mapping, multidim());
++ Permute<std::vector, llvm::Value*>(dimension_mapping, multidim());
+ if (linear() != nullptr &&
+ ShapeUtil::TransposeIsBitcast(operand_shape, shape, dimension_mapping)) {
+ return Index(operand_multidim_index, linear(), operand_shape);
+
+From 2df6cd3acd71a6dcf459e92c26ac426c90a57be4 Mon Sep 17 00:00:00 2001
+From: Todd Wang <toddwang at gmail.com>
+Date: Wed, 21 Jun 2017 17:51:55 -0700
+Subject: [PATCH 2/3] Another try
+
+---
+ tensorflow/compiler/xla/util.h | 17 ++++++++++++-----
+ 1 file changed, 12 insertions(+), 5 deletions(-)
+
+diff --git a/tensorflow/compiler/xla/util.h b/tensorflow/compiler/xla/util.h
+index 42d5c1d1550..32aa81d1f42 100644
+--- a/tensorflow/compiler/xla/util.h
++++ b/tensorflow/compiler/xla/util.h
+@@ -195,16 +195,23 @@ bool IsPermutation(tensorflow::gtl::ArraySlice<int64> permutation, int64 rank);
+ // 2. permutation.size() == input.size().
+ template <template <typename...> class C, typename T>
+ std::vector<T> Permute(tensorflow::gtl::ArraySlice<int64> permutation,
+- C<T> input_) {
+- tensorflow::gtl::ArraySlice<T> input(input_);
+- CHECK(IsPermutation(permutation, input.size()));
+- std::vector<T> output(input.size());
++ C<T> input) {
++ tensorflow::gtl::ArraySlice<T> data(input);
++ CHECK(IsPermutation(permutation, data.size()));
++ std::vector<T> output(data.size());
+ for (size_t i = 0; i < permutation.size(); ++i) {
+- output[permutation[i]] = input[i];
++ output[permutation[i]] = data[i];
+ }
+ return output;
+ }
+
++// Override of the above that works around compile failures with vectors.
++template <typename T>
++std::vector<T> Permute(tensorflow::gtl::ArraySlice<int64> permutation,
++ const std::vector<T>& input) {
++ return Permute<std::vector, T>(permutation, input);
++}
++
+ // Inverts a permutation, i.e., output_permutation[input_permutation[i]] = i.
+ std::vector<int64> InversePermutation(
+ tensorflow::gtl::ArraySlice<int64> input_permutation);
+
+From 03da6113468d6b256a755d0521b9b239eaf8b6a9 Mon Sep 17 00:00:00 2001
+From: Todd Wang <toddwang at gmail.com>
+Date: Wed, 21 Jun 2017 17:54:15 -0700
+Subject: [PATCH 3/3] Update ir_array.cc
+
+---
+ tensorflow/compiler/xla/service/llvm_ir/ir_array.cc | 2 +-
+ 1 file changed, 1 insertion(+), 1 deletion(-)
+
+diff --git a/tensorflow/compiler/xla/service/llvm_ir/ir_array.cc b/tensorflow/compiler/xla/service/llvm_ir/ir_array.cc
+index b14104be6ad..e401305ae73 100644
+--- a/tensorflow/compiler/xla/service/llvm_ir/ir_array.cc
++++ b/tensorflow/compiler/xla/service/llvm_ir/ir_array.cc
+@@ -158,7 +158,7 @@ IrArray::Index IrArray::Index::SourceIndexOfTranspose(
+ tensorflow::gtl::ArraySlice<int64> dimension_mapping,
+ llvm::IRBuilder<>* builder) const {
+ std::vector<llvm::Value*> operand_multidim_index =
+- Permute<std::vector, llvm::Value*>(dimension_mapping, multidim());
++ Permute(dimension_mapping, multidim());
+ if (linear() != nullptr &&
+ ShapeUtil::TransposeIsBitcast(operand_shape, shape, dimension_mapping)) {
+ return Index(operand_multidim_index, linear(), operand_shape);
Modified: PKGBUILD
===================================================================
--- PKGBUILD 2017-06-22 06:24:07 UTC (rev 239451)
+++ PKGBUILD 2017-06-22 06:55:34 UTC (rev 239452)
@@ -4,19 +4,24 @@
pkgbase=tensorflow
pkgname=(tensorflow tensorflow-cuda python-tensorflow python-tensorflow-cuda)
-pkgver=1.1.0
-pkgrel=4
+pkgver=1.2.0
+pkgrel=1
pkgdesc="Library for computation using data flow graphs for scalable machine learning"
url="https://www.tensorflow.org/"
license=('Apache2')
arch=('x86_64')
makedepends=('git' 'bazel' 'python-numpy' 'gcc5' 'cuda' 'cudnn' 'python-pip' 'python-wheel' 'python-setuptools')
-source=("https://github.com/tensorflow/tensorflow/archive/v${pkgver}.tar.gz")
-sha512sums=('5dc0e13742c06cbc6407cc54de02d825928e9c16eeee169b198bbec95acd2a9af49bf3df092c99be1e3adadd40c5f4edf8735c6ad67a370c90c1e74146c32eac')
+source=("https://github.com/tensorflow/tensorflow/archive/v${pkgver}.tar.gz"
+ 10868.patch)
+sha512sums=('7f51060e35bdf313f043fdf99c174ec7a734463c5ea7436380428fd3900421c656b61d00fb9c7492b97172a0bd963502dc0ff4cabd31d91297ed99065842647d'
+ '90940827ea6cbc6694b14ea276a37c5f8af027317873d67a8b40802d8d75a81984c283335bf62cc5b5fa63023c9cc9b414a145d3019124127c5f15472bfdfcec')
prepare() {
+ patch -Np1 -d tensorflow-${pkgver} < ${srcdir}/10868.patch
+
cp -r tensorflow-${pkgver} tensorflow-${pkgver}-cuda
# These environment variables influence the behavior of the configure call below.
+ export TF_NEED_MKL=0
export PYTHON_BIN_PATH=/usr/bin/python
export USE_DEFAULT_PYTHON_LIB_PATH=1
export CC_OPT_FLAGS="-march=x86-64"
@@ -42,7 +47,7 @@
export TF_NEED_CUDA=1
export GCC_HOST_COMPILER_PATH=/usr/bin/gcc-5
# For next version instead of the gcc-5 stuff:
- # export TF_CUDA_CLANG=1
+ export TF_CUDA_CLANG=0
# export CLANG_CUDA_COMPILER_PATH=/usr/bin/clang
export CUDA_TOOLKIT_PATH=/opt/cuda
export TF_CUDA_VERSION=$($CUDA_TOOLKIT_PATH/bin/nvcc --version | sed -n 's/^.*release \(.*\),.*/\1/p')
More information about the arch-commits
mailing list