barracuda 1.1 → 1.2

data/README.md

@@ -58,16 +58,13 @@ EXAMPLE
 Consider the following example to sum a bunch of integers:
 
     program = Program.new <<-'eof'
-      __kernel sum(__global int *out, __global int *in, int total) {
-        int id = get_global_id(0);
-        if (id < total) atom_add(&out[0], in[id]); 
+      __kernel sum(__global int *in, __global int *out) {
+        atom_add(out, in[get_global_id(0)]); 
       }
     eof
     
-    arr    = (1..65536).to_a
-    input  = Buffer.new(arr)
     output = OutputBuffer.new(:int, 1)
-    program.sum(output, input, arr.size)
+    program.sum((1..65536).to_a, output)
     
     puts "The sum is: " + output.data[0].to_s
     
@@ -86,12 +83,6 @@ manually specify the work group size, call the kernel with an options hash:
 
     program.my_kernel_method(..., :times => 512)
     
-Note that the work group size must be a power of 2. Barracuda will increase
-the work group size to the next power of 2 if it needs to. This means your
-OpenCL program might run more iterations of your kernel method than you 
-request. Because we can't rely on the work group size, we pass in the total 
-data size to ensure we do not exceed the bounds of our data.
-
 CONVERTING TYPES
 ----------------
 

data/benchmarks/normalize.rb

@@ -31,12 +31,11 @@ prog = Program.new <<-'eof'
   }
 eof
 
-num_vecs = 100000
+num_vecs = 1000000
 arr = []
 num_vecs.times { arr.push(rand, rand, rand, 0.0) }
 output = OutputBuffer.new(:float, arr.size)
 
-
 Benchmark.bmbm do |x|
   x.report("cpu") { norm_all(arr) }
   x.report("gpu") { prog.norm(output, arr, num_vecs) }

data/benchmarks/sort.rb

@@ -9,7 +9,6 @@ prog = Program.new <<-'eof'
   __kernel sort(__global int *out, __global int *in, int total) {
     int i, final_index = 0, extra = 0;
     int id = get_global_id(0);
-    if (id >= total) return;
     int my_value = in[id];
     for (i = 0; i < total; i++) {
       if (in[i] < my_value) final_index++;
@@ -19,7 +18,7 @@ prog = Program.new <<-'eof'
   }
 eof
 
-max = 1000
+max = 10000
 arr = (1..max).map { (rand * max).to_i }
 output = OutputBuffer.new(:int, arr.size)
  

data/benchmarks/to_float.rb

@@ -6,9 +6,9 @@ require 'benchmark'
 include Barracuda
 
 prog = Program.new <<-'eof'
-  __kernel sum(__global float *out, __global int *in, int total) {
+  __kernel sum(__global float *out, __global int *in) {
     int i = get_global_id(0);
-    if (i < total) out[i] = ((float)in[i] + 0.5) / 3.8 + 2.0;
+    out[i] = ((float)in[i] + 0.5) / 3.8 + 2.0;
   }
 eof
 
@@ -16,9 +16,8 @@ arr = (1..3333333).to_a
 input = Buffer.new(arr)
 output = OutputBuffer.new(:float, arr.size)
  
-TIMES = 1
 Benchmark.bmbm do |x|
-  x.report("regular") { TIMES.times { arr.map {|x| (x.to_f + 0.5) / 3.8 + 2.0 } } }
-  x.report("opencl") { TIMES.times { prog.sum(output, input, arr.size); output.clear } }
+  x.report("regular") { arr.map {|x| (x.to_f + 0.5) / 3.8 + 2.0 } }
+  x.report("opencl") { prog.sum(output, input) }
 end
 

data/ext/barracuda.c

@@ -1,4 +1,5 @@
 #include <ruby.h>
+#include <math.h>
 #include <OpenCL/OpenCL.h>
 
 static VALUE rb_mBarracuda;
@@ -38,18 +39,15 @@ static VALUE buffer_data_set(VALUE self, VALUE new_value);
 
 static cl_device_id device_id = NULL;
 static cl_context context = NULL;
+static size_t max_work_group_size = 65535;
 static int err;
 
-#define VERSION_STRING "1.1"
+#define VERSION_STRING "1.2"
 
 struct program {
     cl_program program;
 };
 
-struct kernel {
-    cl_kernel kernel;
-};
-
 struct buffer {
     VALUE arr;
     ID type;
@@ -376,8 +374,6 @@ buffer_data_set(VALUE self, VALUE new_value)
 static VALUE
 buffer_initialize(int argc, VALUE *argv, VALUE self)
 {
-    GET_BUFFER();
-    
     if (argc == 0) {
         rb_raise(rb_eArgError, "no buffer data given");
     }
@@ -495,21 +491,13 @@ program_compile(VALUE self, VALUE source)
     return Qtrue;
 }
 
-#define CLEAN() program_clean(kernel, commands);
-#define ERROR(msg) if (err != CL_SUCCESS) { CLEAN(); rb_raise(rb_eOpenCLError, msg); }
-
-static void
-program_clean(cl_kernel kernel, cl_command_queue commands)
-{
-    clReleaseKernel(kernel);
-    clReleaseCommandQueue(commands);
-}
+#define CLEAN() { clReleaseKernel(kernel); clReleaseCommandQueue(commands); }
 
 static VALUE
 program_method_missing(int argc, VALUE *argv, VALUE self)
 {
     int i;
-    size_t local = 0, global = 0;
+    size_t global[3] = {1, 1, 1}, local;
     cl_kernel kernel;
     cl_command_queue commands;
     GET_PROGRAM();
@@ -533,7 +521,7 @@ program_method_missing(int argc, VALUE *argv, VALUE self)
         if (i == argc - 1 && TYPE(item) == T_HASH) {
             VALUE worker_size = rb_hash_aref(item, ID2SYM(id_times));
             if (RTEST(worker_size) && TYPE(worker_size) == T_FIXNUM) {
-                global = FIX2UINT(worker_size);
+                global[0] = FIX2UINT(worker_size);
             }
             else {
                 CLEAN();
@@ -553,8 +541,8 @@ program_method_missing(int argc, VALUE *argv, VALUE self)
             struct buffer *buffer;
             Data_Get_Struct(item, struct buffer, buffer);
             err = clSetKernelArg(kernel, i - 1, sizeof(cl_mem), &buffer->data);
-            if (buffer->num_items > global) {
-                global = buffer->num_items;
+            if (buffer->num_items > global[0]) {
+                global[0] = buffer->num_items;
             }
         }
         else if (CLASS_OF(item) == rb_cBuffer) {
@@ -565,8 +553,8 @@ program_method_missing(int argc, VALUE *argv, VALUE self)
             clEnqueueWriteBuffer(commands, buffer->data, CL_TRUE, 0, 
                 buffer->num_items * buffer->member_size, buffer->cachebuf, 0, NULL, NULL);
             err = clSetKernelArg(kernel, i - 1, sizeof(cl_mem), &buffer->data);
-            if (buffer->num_items > global) {
-                global = buffer->num_items;
+            if (buffer->num_items > global[0]) {
+                global[0] = buffer->num_items;
             }
         }
         else {
@@ -599,17 +587,16 @@ program_method_missing(int argc, VALUE *argv, VALUE self)
     }
     
     err = clGetKernelWorkGroupInfo(kernel, device_id, CL_KERNEL_WORK_GROUP_SIZE, sizeof(size_t), &local, NULL);
-    ERROR("failed to retrieve kernel work group info");
-    
-    { /* global work size must be power of 2, greater than 3 and not smaller than local */
-        size_t size = 4;
-        while (size < global) size *= 2;
-        global = size;
-        if (global < local) global = local;
+    err = clEnqueueNDRangeKernel(commands, kernel, 3, NULL, global, NULL, 0, NULL, NULL);
+    if (err != CL_SUCCESS) { 
+        CLEAN(); 
+        if (err == CL_INVALID_KERNEL_ARGS) {
+            rb_raise(rb_eArgError, "invalid arguments"); 
+        }
+        else {
+            rb_raise(rb_eOpenCLError, "failed to execute kernel method %d", err); 
+        }
     }
-
-    clEnqueueNDRangeKernel(commands, kernel, 1, NULL, &global, &local, 0, NULL, NULL);
-    if (err) { CLEAN(); rb_raise(rb_eOpenCLError, "failed to execute kernel method"); }
     
     clFinish(commands);
     
@@ -620,7 +607,10 @@ program_method_missing(int argc, VALUE *argv, VALUE self)
             Data_Get_Struct(item, struct buffer, buffer);
             err = clEnqueueReadBuffer(commands, buffer->data, CL_TRUE, 0, 
                 buffer->num_items * buffer->member_size, buffer->cachebuf, 0, NULL, NULL);
-            ERROR("failed to read output buffer");
+            if (err != CL_SUCCESS) {
+                CLEAN();
+                rb_raise(rb_eOpenCLError, "failed to read output buffer");
+            }
             buffer_read(item);
         }
     }
@@ -645,6 +635,10 @@ init_opencl()
             rb_raise(rb_eOpenCLError, "failed to create a program context");
         }
     }
+    
+    clGetDeviceInfo(device_id, CL_DEVICE_MAX_WORK_GROUP_SIZE, 
+        sizeof(size_t), &max_work_group_size, NULL);
+    max_work_group_size = 4096;
 }
 
 void

data/test/test_barracuda.rb

@@ -135,7 +135,7 @@ class TestProgram < Test::Unit::TestCase
   
   def test_kernel_run
     p = Program.new("__kernel x_y_z(int x) { }")
-    assert_nothing_raised { p.x_y_z }
+    assert_raise(ArgumentError) { p.x_y_z }
   end
   
   def test_kernel_missing
@@ -217,7 +217,7 @@ class TestProgram < Test::Unit::TestCase
     p = Program.new <<-'eof'
       __kernel sum(__global int* out, __global int* in, int total) {
         int id = get_global_id(0);
-        if (id < total) atom_add(&out[0], in[id]); 
+        if (id < total) atom_add(out, in[id]); 
       }
     eof
     
@@ -233,7 +233,7 @@ class TestProgram < Test::Unit::TestCase
     p = Program.new <<-'eof'
       __kernel sum(__global int* out, __global int* in, int total) {
         int id = get_global_id(0);
-        if (id < total) atom_add(&out[0], in[id]); 
+        if (id < total) atom_add(out, in[id]); 
       }
     eof
     
@@ -271,4 +271,21 @@ class TestProgram < Test::Unit::TestCase
     p.copy_to_out(out, [2.5, 2.5, 2.5, 2.5])
     assert_equal [3, 3, 3, 3], out.data
   end
+  
+  def test_program_no_total
+    p = Program.new <<-'eof'
+      __kernel copy(__global int *out, __global int *in) {
+        int i = get_global_id(0);
+        out[i] = in[i] + 1;
+      }
+    eof
+    
+    out = OutputBuffer.new(:int, 3)
+    p.copy(out, (1..3).to_a)
+    assert_equal (2..4).to_a, out.data
+
+    out = OutputBuffer.new(:int, 50446)
+    p.copy(out, (1..50446).to_a)
+    assert_equal (2..50447).to_a, out.data
+  end
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification 
 name: barracuda
 version: !ruby/object:Gem::Version 
-  version: "1.1"
+  version: "1.2"
 platform: ruby
 authors: 
 - Loren Segal
@@ -9,7 +9,7 @@ autorequire:
 bindir: bin
 cert_chain: []
 
-date: 2009-09-02 00:00:00 -04:00
+date: 2009-09-03 00:00:00 -04:00
 default_executable: 
 dependencies: []