[libc-commits] [PATCH] D130498: [libc] Use nearest_integer instructions to improve expf performance.

[Tue Ly via Phabricator via libc-commits]

lntue updated this revision to Diff 447412.
lntue added a comment.

Add multiply_add to dependency.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D130498/new/

https://reviews.llvm.org/D130498

Files:
  libc/docs/math.rst
  libc/src/math/generic/CMakeLists.txt
  libc/src/math/generic/expf.cpp


Index: libc/src/math/generic/expf.cpp
===================================================================
--- libc/src/math/generic/expf.cpp
+++ libc/src/math/generic/expf.cpp
@@ -10,9 +10,10 @@
 #include "common_constants.h" // Lookup tables EXP_M1 and EXP_M2.
 #include "src/__support/FPUtil/BasicOperations.h"
 #include "src/__support/FPUtil/FEnvImpl.h"
-#include "src/__support/FPUtil/FMA.h"
 #include "src/__support/FPUtil/FPBits.h"
 #include "src/__support/FPUtil/PolyEval.h"
+#include "src/__support/FPUtil/multiply_add.h"
+#include "src/__support/FPUtil/nearest_integer.h"
 #include "src/__support/common.h"
 
 #include <errno.h>
@@ -80,13 +81,10 @@
   // generated by Sollya.
 
   // x_hi = (hi + mid) * 2^7 = round(x * 2^7).
-  // The default rounding mode for float-to-int conversion in C++ is
-  // round-toward-zero. To make it round-to-nearest, we add (-1)^sign(x) * 0.5
-  // before conversion.
-  int x_hi = static_cast<int>(x * 0x1.0p7f + (xbits.get_sign() ? -0.5f : 0.5f));
+  float kf = fputil::nearest_integer(x * 0x1.0p7f);
   // Subtract (hi + mid) from x to get lo.
-  x -= static_cast<float>(x_hi) * 0x1.0p-7f;
-  double xd = static_cast<double>(x);
+  double xd = static_cast<double>(fputil::multiply_add(kf, -0x1.0p-7f, x));
+  int x_hi = static_cast<int>(kf);
   x_hi += 104 << 7;
   // hi = x_hi >> 7
   double exp_hi = EXP_M1[x_hi >> 7];
Index: libc/src/math/generic/CMakeLists.txt
===================================================================
--- libc/src/math/generic/CMakeLists.txt
+++ libc/src/math/generic/CMakeLists.txt
@@ -483,6 +483,8 @@
   DEPENDS
     .common_constants
     libc.src.__support.FPUtil.fputil
+    libc.src.__support.FPUtil.multiply_add
+    libc.src.__support.FPUtil.nearest_integer
     libc.src.__support.FPUtil.polyeval
     libc.include.math
   COMPILE_OPTIONS
Index: libc/docs/math.rst
===================================================================
--- libc/docs/math.rst
+++ libc/docs/math.rst
@@ -199,7 +199,7 @@
 +==============+===========+===================+===========+===================+=====================================+============+=========================+==============+===============+
 | cosf         |        37 |                32 |        73 |                72 | :math:`[0, 2\pi]`                   | Ryzen 1700 | Ubuntu 20.04 LTS x86_64 | Clang 12.0.0 |               |
 +--------------+-----------+-------------------+-----------+-------------------+-------------------------------------+------------+-------------------------+--------------+---------------+
-| expf         |        14 |                 9 |        58 |                42 | :math:`[-10, 10]`                   | Ryzen 1700 | Ubuntu 20.04 LTS x86_64 | Clang 12.0.0 | FMA           |
+| expf         |         9 |                 7 |        44 |                38 | :math:`[-10, 10]`                   | Ryzen 1700 | Ubuntu 20.04 LTS x86_64 | Clang 12.0.0 | FMA           |
 +--------------+-----------+-------------------+-----------+-------------------+-------------------------------------+------------+-------------------------+--------------+---------------+
 | exp2f        |        25 |                 8 |        81 |                37 | :math:`[-10, 10]`                   | Ryzen 1700 | Ubuntu 20.04 LTS x86_64 | Clang 12.0.0 | FMA           |
 +--------------+-----------+-------------------+-----------+-------------------+-------------------------------------+------------+-------------------------+--------------+---------------+


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D130498.447412.patch
Type: text/x-patch
Size: 3506 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/libc-commits/attachments/20220725/97f3c69e/attachment.bin>