From 72286b36e8ce478af9f62507b5aea762db4ea46b Mon Sep 17 00:00:00 2001
From: Adrian Kummerlaender
Date: Sun, 6 Oct 2019 14:03:02 +0200
Subject: Use OpenCL access qualifiers only for image objects

It seems I was overeager in adding those qualifiers to non-image buffers
as they are only defined by the standard in relation to image objects.
Adding the qualifiers to normal buffers causes no observable performance
difference on Nvidia targets and fails compilation when targeting AMD or
Intel.
---
 template/kernel.mako | 24 ++++++++++++------------
 1 file changed, 12 insertions(+), 12 deletions(-)

(limited to 'template/kernel.mako')

diff --git a/template/kernel.mako b/template/kernel.mako
index 025baf5..bf7b29b 100644
--- a/template/kernel.mako
+++ b/template/kernel.mako
@@ -9,13 +9,13 @@ def pop_offset(i):
     return i * memory.volume
 %>
 
-__kernel void equilibrilize(__global __write_only ${float_type}* f_next,
-                            __global __write_only ${float_type}* f_prev)
+__kernel void equilibrilize(__global ${float_type}* f_next,
+                            __global ${float_type}* f_prev)
 {
     const unsigned int gid = ${gid()};
 
-    __global __write_only ${float_type}* preshifted_f_next = f_next + gid;
-    __global __write_only ${float_type}* preshifted_f_prev = f_prev + gid;
+    __global ${float_type}* preshifted_f_next = f_next + gid;
+    __global ${float_type}* preshifted_f_prev = f_prev + gid;
 
 % if pop_eq_src == '':
 %     for i, w_i in enumerate(descriptor.w):
@@ -36,9 +36,9 @@ def neighbor_offset(c_i):
 
 %>
 
-__kernel void collide_and_stream(__global __write_only ${float_type}* f_next,
-                                 __global __read_only  ${float_type}* f_prev,
-                                 __global __read_only  int* material,
+__kernel void collide_and_stream(__global ${float_type}* f_next,
+                                 __global ${float_type}* f_prev,
+                                 __global int* material,
                                  unsigned int time)
 {
     const unsigned int gid = ${gid()};
@@ -49,8 +49,8 @@ __kernel void collide_and_stream(__global __write_only ${float_type}* f_next,
         return;
     }
 
-    __global __write_only ${float_type}* preshifted_f_next = f_next + gid;
-    __global __read_only  ${float_type}* preshifted_f_prev = f_prev + gid;
+    __global ${float_type}* preshifted_f_next = f_next + gid;
+    __global ${float_type}* preshifted_f_prev = f_prev + gid;
 
 % for i, c_i in enumerate(descriptor.c):
     const ${float_type} f_curr_${i} = preshifted_f_prev[${pop_offset(i) + neighbor_offset(-c_i)}];
@@ -79,12 +79,12 @@ __kernel void collide_and_stream(__global __write_only ${float_type}* f_next,
 % endfor
 }
 
-__kernel void collect_moments(__global __read_only  ${float_type}* f,
-                              __global __write_only ${float_type}* moments)
+__kernel void collect_moments(__global ${float_type}* f,
+                              __global ${float_type}* moments)
 {
     const unsigned int gid = ${gid()};
 
-    __global __read_only ${float_type}* preshifted_f = f + gid;
+    __global ${float_type}* preshifted_f = f + gid;
 
 % for i in range(0,descriptor.q):
     const ${float_type} f_curr_${i} = preshifted_f[${pop_offset(i)}];
-- 
cgit v1.2.3