gather_dependent.br

用于GPU通用计算的编程语言BrookGPU 0.4

// gather_dependent.br
// Tests that gather operations produce correct
// results when the index is non-constant.
// This utilizes the dependent texture read
// capability of the GPU.

#include <stdio.h>

kernel void gather1( float index<>, float a[100], out float b<> ) {
	b = a[ index.x ];
}

kernel void gather2( float2 index<>, float a[10][10], out float b<> ) {
	b = a[ index.xy ];
}

int main() {
	int i, j;
	float index1< 100 >;
	float a1< 100 >;
	float b1< 100 >;
	float2 index2< 10, 10 >;
	float a2< 10, 10 >;
	float b2< 10, 10 >;
	float input_index1[100];
	float input_a1[100];
	float output_b1[100];
	float2 input_index2[10][10];
	float input_a2[10][10];
	float output_b2[10][10];
	
	// initialize data
	for( i = 0; i < 10; i++ )
	{
		for( j = 0; j < 10; j++ )
		{
			input_index1[10*i+j] = (float)(10*i+j);
			input_a1[10*i+j] = (float)(99 - (10*i+j));
			input_index2[i][j].x = (float)j;
			input_index2[i][j].y = (float)i;
			input_a2[i][j] = (float)(99 - (10*i+j));
		}
	}
	streamRead( index1, input_index1 );
	streamRead( a1, input_a1 );
	streamRead( index2, input_index2 );
	streamRead( a2, input_a2 );
	
	gather1( index1, a1, b1 );
	gather2( index2, a2, b2 );
	
	streamWrite( b1, output_b1 );
	streamWrite( b2, output_b2 );
	
	for( i = 0; i < 10; i++ )
	{
		for( j = 0; j < 10; j++ )
			printf( "%3.2f ", output_b1[10*i+j] );
		printf("\n");
	}
	
	for( i = 0; i < 10; i++ )
	{
		for( j = 0; j < 10; j++ )
			printf( "%3.2f ", output_b2[i][j] );
		printf("\n");
	}
	
	return 0;
}