bones-compiler 1.1.0

data/examples/benchmarks/jacobi-2d-imper.c

@@ -0,0 +1,61 @@
+//
+// This file is part of the Bones source-to-source compiler examples. The C-code
+// is largely identical in terms of functionality and variable naming to the code
+// found in PolyBench/C version 3.2. For more information on PolyBench/C or Bones
+// please use the contact information below.
+//
+// == More information on PolyBench/C
+// Contact............Louis-Noel Pouchet <pouchet@cse.ohio-state.edu>
+// Web address........http://polybench.sourceforge.net/
+// 
+// == More information on Bones
+// Contact............Cedric Nugteren <c.nugteren@tue.nl>
+// Web address........http://parse.ele.tue.nl/bones/
+//
+// == File information
+// Filename...........benchmark/jacobi-2d-imper.c
+// Author.............Cedric Nugteren
+// Last modified on...03-April-2012
+//
+
+#include "common.h"
+
+// This is 'jacobi-2d-imper', a 2D Jacobi stencil computation
+int main(void) {
+	int i,j,t;
+	
+	// Declare arrays on the stack
+	float A[N][N];
+	float B[N][N];
+	
+	// Set the input data
+	for (i=0; i<N; i++) {
+		for (j=0; j<N; j++) {
+			A[i][j] = ((float) i*(j+2) + 2) / N;
+			B[i][j] = ((float) i*(j+3) + 3) / N;
+		}
+	}
+	
+	// Perform the computation
+	for (t=0; t<TSTEPS; t++) {
+		#pragma species kernel 1:N-2,1:N-2|neighbourhood(-1:1,-1:1) -> 1:N-2,1:N-2|element
+		for (i=1; i<N-1; i++) {
+			for (j=1; j<N-1; j++) {
+				B[i][j] = 0.2 * (A[i][j] + A[i][j-1] + A[i][1+j] + A[1+i][j] + A[i-1][j]);
+			}
+		}
+		#pragma species endkernel jacobi-2d-imper-part1
+		#pragma species kernel 1:N-2,1:N-2|element -> 1:N-2,1:N-2|element
+		for (i=1; i<N-1; i++) {
+			for (j=1; j<N-1; j++) {
+				A[i][j] = B[i][j];
+			}
+		}
+		#pragma species endkernel jacobi-2d-imper-part2
+	}
+	
+	// Clean-up and exit the function
+	fflush(stdout);
+	return 0;
+}
+

data/examples/benchmarks/lu.c

@@ -0,0 +1,57 @@
+//
+// This file is part of the Bones source-to-source compiler examples. The C-code
+// is largely identical in terms of functionality and variable naming to the code
+// found in PolyBench/C version 3.2. For more information on PolyBench/C or Bones
+// please use the contact information below.
+//
+// == More information on PolyBench/C
+// Contact............Louis-Noel Pouchet <pouchet@cse.ohio-state.edu>
+// Web address........http://polybench.sourceforge.net/
+// 
+// == More information on Bones
+// Contact............Cedric Nugteren <c.nugteren@tue.nl>
+// Web address........http://parse.ele.tue.nl/bones/
+//
+// == File information
+// Filename...........benchmark/lu.c
+// Author.............Cedric Nugteren
+// Last modified on...26-Jun-2012
+//
+
+#include "common.h"
+
+// This is 'lu', an LU decomposition kernel
+int main(void) {
+	int i,j,k;
+	
+	// Declare arrays on the stack
+	float A[N][N];
+	
+	// Set the input data
+	for (i=0; i<N; i++) {
+		for (j=0; j<N; j++) {
+			A[i][j] = ((float) (i+1)*(j+1)) / N;
+		}
+	}
+	
+	// Perform the computation
+	for (k=0; k<N; k++) {
+		#pragma species kernel k:k,k+1:N-1|element -> k:k,k+1:N-1|element
+		for (j=k+1; j<N; j++) {
+			A[k][j] = A[k][j] / A[k][k];
+		}
+		#pragma species endkernel lu-part1
+		#pragma species kernel k+1:N-1,k:k|element ^ k:k,k+1:N-1|element ^ k+1:N-1,k+1:N-1|element -> k+1:N-1,k+1:N-1|element
+		for(i=k+1; i<N; i++) {
+			for (j=k+1; j<N; j++) {
+				A[i][j] = A[i][j] - A[i][k] * A[k][j];
+			}
+		}
+		#pragma species endkernel lu-part2
+	}
+	
+	// Clean-up and exit the function
+	fflush(stdout);
+	return 0;
+}
+

data/examples/benchmarks/ludcmp.c

@@ -0,0 +1,91 @@
+//
+// This file is part of the Bones source-to-source compiler examples. The C-code
+// is largely identical in terms of functionality and variable naming to the code
+// found in PolyBench/C version 3.2. For more information on PolyBench/C or Bones
+// please use the contact information below.
+//
+// == More information on PolyBench/C
+// Contact............Louis-Noel Pouchet <pouchet@cse.ohio-state.edu>
+// Web address........http://polybench.sourceforge.net/
+// 
+// == More information on Bones
+// Contact............Cedric Nugteren <c.nugteren@tue.nl>
+// Web address........http://parse.ele.tue.nl/bones/
+//
+// == File information
+// Filename...........benchmark/ludcmp.c
+// Author.............Cedric Nugteren
+// Last modified on...23-May-2012
+//
+
+#include "common.h"
+
+// This is 'ludcmp', an LU decomposition kernel
+int main(void) {
+	int i,j,k;
+	float w[1];
+	
+	// Declare arrays on the stack
+	float A[N+1][N+1];
+	float b[N+1];
+	float x[N+1];
+	float y[N+1];
+	
+	// Set the input data
+	for (i=0; i<=N; i++) {
+		x[i] = i+1;
+		y[i] = (i+1)/(float)(N*40) + 1;
+		b[i] = (i+1)/(float)(N*20) + 42;
+		for (j=0; j<=N; j++) {
+			A[i][j] = (i+1)/(float)(10*N) + (j+1)/(float)(5*N);
+		}
+	}
+	
+	// Perform the computation
+	b[0] = 1.0;
+	for (i=0; i<N; i++) {
+		for (j=i+1; j<=N; j++) {
+			w[0] = A[j][i];
+			#pragma species kernel j:j,0:i-1|element ^ 0:i-1,i:i|element -> 0:0|shared
+			for (k=0; k<i; k++) {
+				w[0] = w[0] - A[j][k] * A[k][i];
+			}
+			#pragma species endkernel ludcmp-part1
+			A[j][i] = w[0] / A[i][i];
+		}
+		for (j=i+1; j<=N; j++) {
+			w[0] = A[i+1][j];
+			#pragma species kernel i+1:i+1,0:i|element ^ 0:i,j:j|element -> 0:0|shared
+			for (k=0; k<=i; k++) {
+				w[0] = w[0] - A[i+1][k] * A[k][j];
+			}
+			#pragma species endkernel ludcmp-part2
+			A[i+1][j] = w[0];
+		}
+	}
+	y[0] = b[0];
+	for (i=1; i<=N; i++) {
+		w[0] = b[i];
+		#pragma species kernel i:i,0:i-1|element ^ 0:i-1|element -> 0:0|shared
+		for (j=0; j<i; j++) {
+			w[0] = w[0] - A[i][j] * y[j];
+		}
+		#pragma species endkernel ludcmp-part3
+		y[i] = w[0];
+	}
+	x[N] = y[N] / A[N][N];
+	for (i=0; i<=N-1; i++) {
+		w[0] = y[N-1-i];
+		//#pragma species kernel N-1-i:N-1-i,N-i:N|element ^ N-i:N|element -> 0:0|shared
+		for (j=N-i; j<=N; j++) {
+			w[0] = w[0] - A[N-1-i][j] * x[j];
+		}
+		//#pragma species endkernel ludcmp-part4
+		x[N-1-i] = w[0] / A[N-1-i][N-1-i];
+	}
+	
+	// Clean-up and exit the function
+	fflush(stdout);
+	return 0;
+}
+

data/examples/benchmarks/mvt.c

@@ -0,0 +1,65 @@
+//
+// This file is part of the Bones source-to-source compiler examples. The C-code
+// is largely identical in terms of functionality and variable naming to the code
+// found in PolyBench/C version 3.2. For more information on PolyBench/C or Bones
+// please use the contact information below.
+//
+// == More information on PolyBench/C
+// Contact............Louis-Noel Pouchet <pouchet@cse.ohio-state.edu>
+// Web address........http://polybench.sourceforge.net/
+// 
+// == More information on Bones
+// Contact............Cedric Nugteren <c.nugteren@tue.nl>
+// Web address........http://parse.ele.tue.nl/bones/
+//
+// == File information
+// Filename...........benchmark/mvt.c
+// Author.............Cedric Nugteren
+// Last modified on...23-May-2012
+//
+
+#include "common.h"
+
+// This is 'mvt', a matrix vector product and transpose kernel
+int main(void) {
+	int i,j;
+	
+	// Declare arrays on the stack
+	float A[NX][NX];
+	float x1[NX];
+	float x2[NX];
+	float y_1[NX];
+	float y_2[NX];
+	
+	// Set the input data
+	for (i=0; i<NX; i++) {
+		x1[i] = ((float) i) / NX;
+		x2[i] = ((float) i + 1) / NX;
+		y_1[i] = ((float) i + 3) / NX;
+		y_2[i] = ((float) i + 4) / NX;
+		for (j=0; j<NX; j++) {
+			A[i][j] = ((float) i*j) / NX;
+		}
+	}
+	
+	// Perform the computation
+	#pragma species kernel 0:NX-1|element ^ 0:NX-1,0:NX-1|chunk(0:0,0:NX-1) ^ 0:NX-1|full -> 0:NX-1|element
+	for (i=0; i<NX; i++) {
+		for (j=0; j<NX; j++) {
+			x1[i] = x1[i] + A[i][j] * y_1[j];
+		}
+	}
+	#pragma species endkernel mvt-part1
+	#pragma species kernel 0:NX-1|element ^ 0:NX-1,0:NX-1|chunk(0:NX-1,0:0) ^ 0:NX-1|full -> 0:NX-1|element
+	for (i=0; i<NX; i++) {
+		for (j=0; j<NX; j++) {
+			x2[i] = x2[i] + A[j][i] * y_2[j];
+		}
+	}
+	#pragma species endkernel mvt-part2
+	
+	// Clean-up and exit the function
+	fflush(stdout);
+	return 0;
+}
+

data/examples/benchmarks/overview.txt

@@ -0,0 +1,38 @@
+
+== Benchmarks with full parallelism
+linear-algebra/kernels/2mm...........2 species...[fully classified]
+linear-algebra/kernels/3mm...........3 species...[fully classified]
+linear-algebra/kernels/atax..........2 species...[fully classified]
+linear-algebra/kernels/bicg..........2 species...[fully classified]
+linear-algebra/kernels/doitgen.......2 species...[fully classified]
+linear-algebra/kernels/gemm..........1 species...[fully classified]
+linear-algebra/kernels/gemver........4 species...[fully classified]
+linear-algebra/kernels/gesummv.......1 species...[fully classified]
+linear-algebra/kernels/mvt...........2 species...[fully classified]
+linear-algebra/kernels/syr2k.........1 species...[fully classified]
+linear-algebra/kernels/syrk..........1 species...[fully classified]
+stencils/fdtd-2d.....................4 species...[fully classified]
+stencils/jacobi-1d-imper.............2 species...[fully classified]
+stencils/jacobi-2d-imper.............2 species...[fully classified]
+
+== Benchmarks with significant parallelism
+linear-algebra/kernels/cholesky......3 species...[no outer-loop parallelism, inner-loops only classified]
+linear-algebra/kernels/symm..........1 species...[no outer-loop parallelism, inner-loops only classified]
+linear-algebra/kernels/trisolv.......1 species...[no outer-loop parallelism, inner-loops only classified]
+linear-algebra/kernels/trmm..........1 species...[no outer-loop parallelism, inner-loops only classified]
+linear-algebra/solvers/gramschmidt...4 species...[no outer-loop parallelism, inner-loops only classified]
+linear-algebra/solvers/lu............3 species...[no outer-loop parallelism, inner-loops only classified]
+linear-algebra/solvers/ludcmp........4 species...[no outer-loop parallelism, inner-loops only classified]
+datamining/correlation...............5 species...[most parts classified, final part inner-loop only]
+datamining/covariance................4 species...[most parts classified, final part inner-loop only]
+medley/reg_detect....................2 species...[partly classified, several parts have no parallelism]
+
+== Benchmarks with very little parallelism
+linear-algebra/solvers/durbin........2 species...[partly classified, most parts have no parallelism]
+linear-algebra/solvers/dynprog.......1 species...[partly classified, most parts have no parallelism]
+stencils/adi.........................2 species...[partly classified, most parts have no parallelism]
+stencils/fdtd-ampl...................1 species...[partly classified, most parts have no parallelism]
+
+== Benchmarks without parallelism
+medley/floyd-warshall................0 species...[no parallelism]
+stencils/seidel-2d...................0 species...[no parallelism]

data/examples/benchmarks/reg_detect.c

@@ -0,0 +1,82 @@
+//
+// This file is part of the Bones source-to-source compiler examples. The C-code
+// is largely identical in terms of functionality and variable naming to the code
+// found in PolyBench/C version 3.2. For more information on PolyBench/C or Bones
+// please use the contact information below.
+//
+// == More information on PolyBench/C
+// Contact............Louis-Noel Pouchet <pouchet@cse.ohio-state.edu>
+// Web address........http://polybench.sourceforge.net/
+// 
+// == More information on Bones
+// Contact............Cedric Nugteren <c.nugteren@tue.nl>
+// Web address........http://parse.ele.tue.nl/bones/
+//
+// == File information
+// Filename...........benchmark/reg_detect.c
+// Author.............Cedric Nugteren
+// Last modified on...26-Jun-2012
+//
+
+#include "common.h"
+
+// This is 'reg_detect', a regularity detection algorithm
+int main(void) {
+	int i,j,t,cnt;
+	float sum;
+	
+	// Declare arrays on the stack
+	float sum_tang[MAXGRID][MAXGRID];
+	float mean[MAXGRID][MAXGRID];
+	float path[MAXGRID][MAXGRID];
+	float diff[MAXGRID][MAXGRID][LENGTH];
+	float sum_diff[MAXGRID][MAXGRID][LENGTH];
+	
+	// Set the input data
+	for (i=0; i<MAXGRID; i++) {
+		for (j=0; j<MAXGRID; j++) {
+			sum_tang[i][j] = (float)((i+1)*(j+1));
+			mean[i][j] = ((float) i-j) / MAXGRID;
+			path[i][j] = ((float) i*(j-1)) / MAXGRID;
+		}
+	}
+	
+	// Perform the computation
+	for (t=0; t<ITER; t++) {
+		#pragma species kernel 0:MAXGRID-1,0:MAXGRID-1|element -> 0:MAXGRID-1,0:MAXGRID-1,0:LENGTH-1|chunk(0:0,0:0,0:LENGTH-1)
+		for (j=0; j<=MAXGRID-1; j++) {
+			for (i=0; i<=MAXGRID-1; i++) {
+				sum = sum_tang[j][i];
+				for (cnt=0; cnt<=LENGTH-1; cnt++) {
+					diff[j][i][cnt] = sum;
+				}
+			}
+		}
+		#pragma species endkernel reg-detect-part1
+		for (j=0; j<=MAXGRID-1; j++) {
+			for (i=j; i<=MAXGRID-1; i++) {
+				sum_diff[j][i][0] = diff[j][i][0];
+				for (cnt=1; cnt<=LENGTH-1; cnt++) {
+					sum_diff[j][i][cnt] = sum_diff[j][i][cnt-1] + diff[j][i][cnt];
+				}
+				mean[j][i] = sum_diff[j][i][LENGTH-1];
+			}
+		}
+		#pragma species kernel 0:0,0:MAXGRID-1|element -> 0:0,0:MAXGRID-1|element
+		for (i=0; i<=MAXGRID-1; i++) {
+			path[0][i] = mean[0][i];
+		}
+		#pragma species endkernel reg-detect-part2
+		for (j=1; j<=MAXGRID-1; j++) {
+			#pragma species kernel j-1:j-1,j-1:MAXGRID-2|element ^ j:j,j:MAXGRID-1|element -> j:j,j:MAXGRID-1|element
+			for (i=j; i<=MAXGRID-1; i++) {
+				path[j][i] = path[j-1][i-1] + mean[j][i];
+			}
+			#pragma species endkernel reg-detect-part3
+		}
+	}
+	
+	// Clean-up and exit the function
+	fflush(stdout);
+	return 0;
+}

data/examples/benchmarks/saxpy.c

@@ -0,0 +1,45 @@
+//
+// This file is part of the Bones source-to-source compiler examples. For more
+// information on Bones please use the contact information below.
+//
+// == More information on Bones
+// Contact............Cedric Nugteren <c.nugteren@tue.nl>
+// Web address........http://parse.ele.tue.nl/bones/
+//
+// == File information
+// Filename...........benchmark/saxpy.c
+// Author.............Cedric Nugteren
+// Last modified on...04-Jul-2012
+//
+
+#include "common.h"
+
+// This is 'saxpy', a scalar multiplication and vector addition kernel
+int main(void) {
+	int i;
+	
+	// Declare arrays on the stack
+	float x[LARGE_N];
+	float y[LARGE_N];
+	
+	// Set the input data
+	for (i=0; i<LARGE_N; i++) {
+		x[i] = i*1.4;
+		y[i] = i/0.9;
+	}
+	
+	// Set the constants
+	float a = 411.3;
+	
+	// Perform the computation (y := ax+y)
+	#pragma species kernel 0:LARGE_N-1|element ^ 0:LARGE_N-1|element -> 0:LARGE_N-1|element
+	for (i=0; i<LARGE_N; i++) {
+		y[i] = a*x[i] + y[i];
+	}
+	#pragma species endkernel saxpy
+	
+	// Clean-up and exit the function
+	fflush(stdout);
+	return 0;
+}
+