From d805e8b18386e5635e617a2d5188a13ce8c40c31 Mon Sep 17 00:00:00 2001
From: Matt Pharr <matt.pharr@intel.com>
Date: Sun, 22 Jan 2012 13:05:27 -0800
Subject: [PATCH] Add clock() function to standard library.

Also corrected the declaration of num_cores() to return a
uniform value.
---
 builtins/util.m4 | 16 ++++++++++++++++
 docs/ispc.rst    | 21 ++++++++++++++++++---
 stdlib.ispc      |  6 +++++-
 3 files changed, 39 insertions(+), 4 deletions(-)

diff --git a/builtins/util.m4 b/builtins/util.m4
index 883cfb4c..64e3a130 100644
--- a/builtins/util.m4
+++ b/builtins/util.m4
@@ -1811,6 +1811,22 @@ ok:
   ret void
 }
 
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;; read hw clock
+
+define i64 @__clock() nounwind uwtable ssp {
+entry:
+  tail call void asm sideeffect "xorl %eax,%eax \0A    cpuid", "~{rax},~{rbx},~{rcx},~{rdx},~{dirflag},~{fpsr},~{flags}"() nounwind
+  %0 = tail call { i32, i32 } asm sideeffect "rdtsc", "={ax},={dx},~{dirflag},~{fpsr},~{flags}"() nounwind
+  %asmresult = extractvalue { i32, i32 } %0, 0
+  %asmresult1 = extractvalue { i32, i32 } %0, 1
+  %conv = zext i32 %asmresult1 to i64
+  %shl = shl nuw i64 %conv, 32
+  %conv2 = zext i32 %asmresult to i64
+  %or = or i64 %shl, %conv2
+  ret i64 %or
+}
+
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ;; stdlib transcendentals
 ;;
diff --git a/docs/ispc.rst b/docs/ispc.rst
index ddac4bcd..c0dcd6df 100644
--- a/docs/ispc.rst
+++ b/docs/ispc.rst
@@ -3444,12 +3444,27 @@ pointer types.
 System Information
 ------------------
 
-A routine is available to find the number of CPU cores available in the
-system:
+The value of a  high-precision hardware clock counter is returned by the
+``clock()`` routine; its value increments by one each processor cycle.
+Thus, taking the difference between the values returned by ``clock()`` and
+different points in program execution gives the number of cycles between
+those points in the program.
 
 ::
 
-    int num_cores()
+    uniform int64 clock()
+
+Note that ``clock()`` flushes the processor pipeline.  It has an overhead
+of a hundred or so cycles, so for very fine-grained measurements, it may be
+worthwhile to measure the cost of calling ``clock()`` and subtracting that
+value from reported results.
+    
+A routine is also available to find the number of CPU cores available in
+the system:
+
+::
+
+    uniform int num_cores()
 
 This value can be useful for adapting the granularity of parallel task
 decomposition depending on the number of processors in the system.
diff --git a/stdlib.ispc b/stdlib.ispc
index cae63abe..8a7daf49 100644
--- a/stdlib.ispc
+++ b/stdlib.ispc
@@ -787,10 +787,14 @@ packed_store_active(uniform int * uniform a, int vals) {
 ///////////////////////////////////////////////////////////////////////////
 // System information
 
-static inline int num_cores() {
+static inline uniform int num_cores() {
     return __num_cores();
 }
 
+static inline uniform int64 clock() {
+    return __clock();
+}
+
 ///////////////////////////////////////////////////////////////////////////
 // Atomics and memory barriers