From 15a3ef370a433eedcf6e6650f07ec81775d0322d Mon Sep 17 00:00:00 2001
From: Matt Pharr <mpharr@google.com>
Date: Tue, 23 Jul 2013 17:11:01 -0700
Subject: [PATCH] Use @llvm.readcyclecounter to implement stdlib clock()
 function.

Also added a test for the clock builtin.
---
 builtins/util.m4 | 14 ++++----------
 tests/clock.ispc | 14 ++++++++++++++
 2 files changed, 18 insertions(+), 10 deletions(-)
 create mode 100644 tests/clock.ispc

diff --git a/builtins/util.m4 b/builtins/util.m4
index d6f3e5c3..8c379781 100644
--- a/builtins/util.m4
+++ b/builtins/util.m4
@@ -2891,17 +2891,11 @@ m4exit(`1')
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ;; read hw clock
 
+declare i64 @llvm.readcyclecounter()
+
 define i64 @__clock() nounwind {
-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
+  %r = call i64 @llvm.readcyclecounter()
+  ret i64 %r
 }
 
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
diff --git a/tests/clock.ispc b/tests/clock.ispc
new file mode 100644
index 00000000..0e95379b
--- /dev/null
+++ b/tests/clock.ispc
@@ -0,0 +1,14 @@
+
+export uniform int width() { return programCount; }
+
+
+export void f_f(uniform float RET[], uniform float aFOO[]) {
+  unsigned uniform int64 a = clock();
+  float x = pow(sqrt(aFOO[programIndex]), 5.5);
+  unsigned uniform int64 b = clock();
+  RET[programIndex] = (b - a) > 0 ? 1 : 0;
+}
+
+export void result(uniform float RET[]) {
+  RET[programIndex] = 1;
+}