add rsort

benchmarks/Makefile

@@ -25,12 +25,14 @@ bmarks = \
 	cmplxmult \
 	imax \
 	spmv \
+	rsort \
 	vec-vvadd \
 	vec-csaxpy \
 	vec-dgemv \
 	vec-dgemm \
 	vec-imax \
 	vec-spmv \
+	vec-rsort \
 
 #--------------------------------------------------------------------
 # Build rules

benchmarks/vec-rsort/dataset1-test.h

@@ -0,0 +1 @@
+../rsort/dataset1-test.h

benchmarks/vec-rsort/dataset1.h

@@ -0,0 +1 @@
+../rsort/dataset1.h

benchmarks/vec-rsort/vec-rsort.S

@@ -0,0 +1,135 @@
+#include "rvv.h"
+
+.text
+.align 2
+
+.globl vec_rsort
+.type  vec_rsort,@function
+
+# assumes calling convention:
+# a0 has int n
+# a1 has type* arrIn  <---
+# a2 has type* scratchIn
+# a3 has int* buckets
+# a4 has int log_base
+vec_rsort:
+    # TODO: prologue
+	addi sp, sp, -64
+	sd s0, 56(sp)
+	sd s1, 48(sp)
+	sd s2, 40(sp)
+   sd s3, 32(sp)
+	sd s4, 24(sp)
+	sd s5, 16(sp)
+	sd s6, 8(sp)
+	sd s7, 0(sp)
+
+    # Step 1: Configure machine
+    # TODO: change configurations if necessary
+    # configuration:
+    # v0-v7:   32 bit int vector
+    # v8-v15:  32 bit int scalar
+    # v16-v31: 32 bit int vector
+    setvcfg(vcfg0, \
+      VECTOR | INT | W32, \
+		VECTOR | INT | W32, \
+		VECTOR | INT | W32, \
+		VECTOR | INT | W32)
+    setvcfg(vcfg2, \
+      VECTOR | INT | W32, \
+		VECTOR | INT | W32, \
+		VECTOR | INT | W32, \
+		VECTOR | INT | W32)
+    setvcfg(vcfg4, \
+      SCALAR | INT | W32, \
+		SCALAR | INT | W32, \
+		SCALAR | INT | W32, \
+		SCALAR | INT | W32)
+    setvcfg(vcfg6, \
+      SCALAR | INT | W32, \
+		SCALAR | INT | W32, \
+		SCALAR | INT | W32, \
+		SCALAR | INT | W32)
+    setvcfg(vcfg8, \
+      VECTOR | INT | W32, \
+      VECTOR | INT | W32, \
+      VECTOR | INT | W32, \
+      VECTOR | INT | W32)
+    setvcfg(vcfg10, \
+      VECTOR | INT | W32, \
+      VECTOR | INT | W32, \
+      VECTOR | INT | W32, \
+      VECTOR | INT | W32)
+    setvcfg(vcfg12, \
+      VECTOR | INT | W32, \
+      VECTOR | INT | W32, \
+      VECTOR | INT | W32, \
+      VECTOR | INT | W32)
+    setvcfg(vcfg14, \
+      VECTOR | INT | W32, \
+      VECTOR | INT | W32, \
+      VECTOR | INT | W32, \
+      VECTOR | INT | W32)
+
+	# s0 = base
+   li s0, 1
+	sll s0, s0, a4
+	mv s1, x0           # s1 = log_exp
+   mv s2, a1           # s2 = arr
+	mv s3, a2           # s3 = scratch
+	li s4, 1            # s4 = 1
+	li s5, -1           # s5 = 1
+	li s6, 32           # s6 = CHAR_BIT * sizeof(type)
+
+while_loop:
+	# Step 2: Reset buckets
+bucket_init:
+
+	# Step 3: Assign buckets
+to_buckets:
+   setvl(t0, t1)   # set VL
+	# TODO: find digits
+	# TODO: count and prune duplicate digits
+	# TODO: increment buckets
+
+	# Step 4: Add offsets to buckets
+
+   # Step 5: Merge buckets
+from_buckets:
+	setvl(t0, t1)   # set VL  
+
+	# TODO: find digits 
+	# TODO: count and prune duplicate digits
+	# TODO: decrement buckets
+	# TODO: store scratch
+
+   # Step 6: Swap arr & scratch
+
+   # Step 7: Loop arithmatic
+
+   # Call memcpy before return
+   beq a1, s2, rsort_done
+	mv t1, a0
+	sll t1, t1, 2
+	mv a0, a1  # arrIn
+	mv a1, s2  # arr
+	mv a2, t1  # n * sizeof(type)
+
+	addi sp, sp, -8
+	sd ra, 0(sp)
+   jal memcpy
+	ld ra, 0(sp)
+	addi sp, sp, 8
+
+rsort_done:
+	# TODO: epilogue
+	ld s0, 56(sp)
+	ld s1, 48(sp)
+	ld s2, 40(sp)
+   ld s3, 32(sp)
+	ld s4, 24(sp)
+	ld s5, 16(sp)
+	ld s6, 8(sp)
+	ld s7, 0(sp)
+	addi sp, sp, 64
+	ret	

benchmarks/vec-rsort/vec-rsort.c

@@ -0,0 +1,44 @@
+// See LICENSE for license details.
+
+//**************************************************************************
+// Quicksort benchmark
+//--------------------------------------------------------------------------
+//
+// This benchmark uses quicksort to sort an array of integers. The
+// implementation is largely adapted from Numerical Recipes for C. The
+// input data (and reference data) should be generated using the
+// qsort_gendata.pl perl script and dumped to a file named
+// dataset1.h
+
+#include "util.h"
+#include <string.h>
+#include <limits.h>
+
+//--------------------------------------------------------------------------
+// Input/Reference Data
+
+#define type unsigned int
+#include "dataset1.h"
+// #include "dataset1-test.h"
+
+#define LOG_BASE 3
+#define BASE (1 << LOG_BASE)
+
+void vec_rsort(int n, type* arrIn, type* scratchIn, int* buckets, int log_base);
+
+//--------------------------------------------------------------------------
+// Main
+
+int main( int argc, char* argv[] )
+{
+  static type scratch[DATA_SIZE];
+  static int buckets[BASE];
+
+  // Do the sort
+  setStats(1);
+  vec_rsort(DATA_SIZE, input_data, scratch, buckets, LOG_BASE);
+  setStats(0);
+
+  // Check the results
+  return verify( DATA_SIZE, input_data, verify_data );
+}