Merge remote-tracking branch 'origin/priv-1.10' into HEAD

66 files changed, 763 insertions, 1063 deletions

diff --git a/benchmarks/Makefile b/benchmarks/Makefile
index 531d981..d1c2362 100644
--- a/benchmarks/Makefile
+++ b/benchmarks/Makefile
@@ -29,41 +29,29 @@ bmarks = \
 	spmv \
 	mt-vvadd \
 	mt-matmul \
-
-bmarks_host = \
-	median \
-	qsort \
-	towers \
-	vvadd \
-	multiply \
-	spmv \
-	vec-vvadd \
-	vec-cmplxmult \
-	vec-matmul \
+	pmp \
 
 #--------------------------------------------------------------------
 # Build rules
 #--------------------------------------------------------------------
 
-HOST_OPTS = -std=gnu99 -DPREALLOCATE=0 -DHOST_DEBUG=1
-HOST_COMP = gcc $(HOST_OPTS)
-
 RISCV_PREFIX ?= riscv$(XLEN)-unknown-elf-
 RISCV_GCC ?= $(RISCV_PREFIX)gcc
-RISCV_GCC_OPTS ?= -mcmodel=medany -static -std=gnu99 -O2 -ffast-math -fno-common -fno-builtin-printf
+RISCV_GCC_OPTS ?= -DPREALLOCATE=1 -mcmodel=medany -static -std=gnu99 -O2 -ffast-math -fno-common -fno-builtin-printf
 RISCV_LINK ?= $(RISCV_GCC) -T $(src_dir)/common/test.ld $(incs)
-RISCV_LINK_MT ?= $(RISCV_GCC) -T $(src_dir)/common/test-mt.ld
-RISCV_LINK_OPTS ?= -static -nostdlib -nostartfiles -lgcc
+RISCV_LINK_OPTS ?= -static -nostdlib -nostartfiles -lgcc -T $(src_dir)/common/test.ld
 RISCV_OBJDUMP ?= $(RISCV_PREFIX)objdump --disassemble-all --disassemble-zeroes --section=.text --section=.text.startup --section=.data
 RISCV_SIM ?= spike --isa=rv$(XLEN)gc
 
-VPATH += $(addprefix $(src_dir)/, $(bmarks))
-VPATH += $(src_dir)/common
-
 incs  += -I$(src_dir)/../env -I$(src_dir)/common $(addprefix -I$(src_dir)/, $(bmarks))
 objs  :=
 
-include $(patsubst %, $(src_dir)/%/bmark.mk, $(bmarks))
+define compile_template
+$(1).riscv: $(wildcard $(src_dir)/$(1)/*) $(wildcard $(src_dir)/common/*)
+	$$(RISCV_GCC) $$(incs) $$(RISCV_GCC_OPTS) $$(RISCV_LINK_OPTS) -o $$@ $(wildcard $(src_dir)/$(1)/*.c) $(wildcard $(src_dir)/common/*.c) $(wildcard $(src_dir)/common/*.S)
+endef
+
+$(foreach bmark,$(bmarks),$(eval $(call compile_template,$(bmark))))
 
 #------------------------------------------------------------
 # Build and run benchmarks on riscv simulator
@@ -72,47 +60,18 @@ bmarks_riscv_bin  = $(addsuffix .riscv,  $(bmarks))
 bmarks_riscv_dump = $(addsuffix .riscv.dump, $(bmarks))
 bmarks_riscv_out  = $(addsuffix .riscv.out,  $(bmarks))
 
-bmarks_defs   = -DPREALLOCATE=1 -DHOST_DEBUG=0
-bmarks_cycles = 80000
-
 $(bmarks_riscv_dump): %.riscv.dump: %.riscv
 	$(RISCV_OBJDUMP) $< > $@
 
 $(bmarks_riscv_out): %.riscv.out: %.riscv
 	$(RISCV_SIM) $< > $@
 
-%.o: %.c
-	$(RISCV_GCC) $(RISCV_GCC_OPTS) $(bmarks_defs) \
-	             -c $(incs) $< -o $@
-
-%.o: %.S
-	$(RISCV_GCC) $(RISCV_GCC_OPTS) $(bmarks_defs) -D__ASSEMBLY__=1 \
-	             -c $(incs) $< -o $@
-
 riscv: $(bmarks_riscv_dump)
-run-riscv: $(bmarks_riscv_out)
-	echo; perl -ne 'print "  [$$1] $$ARGV \t$$2\n" if /\*{3}(.{8})\*{3}(.*)/' \
-	       $(bmarks_riscv_out); echo;
+run: $(bmarks_riscv_out)
 
 junk += $(bmarks_riscv_bin) $(bmarks_riscv_dump) $(bmarks_riscv_hex) $(bmarks_riscv_out)
 
 #------------------------------------------------------------
-# Build and run benchmarks on host machine
-
-bmarks_host_bin = $(addsuffix .host, $(bmarks_host))
-bmarks_host_out = $(addsuffix .host.out, $(bmarks_host))
-
-$(bmarks_host_out): %.host.out: %.host
-	./$< > $@
-
-host: $(bmarks_host_bin)
-run-host: $(bmarks_host_out)
-	echo; perl -ne 'print "  [$$1] $$ARGV \t$$2\n" if /\*{3}(.{8})\*{3}(.*)/' \
-	       $(bmarks_host_out); echo;
-
-junk += $(bmarks_host_bin) $(bmarks_host_out)
-
-#------------------------------------------------------------
 # Default
 
 all: riscv
diff --git a/benchmarks/common/crt.S b/benchmarks/common/crt.S
index a821769..d75e81e 100644
--- a/benchmarks/common/crt.S
+++ b/benchmarks/common/crt.S
@@ -12,12 +12,9 @@
 # define REGBYTES 4
 #endif
 
-  .text
+  .section ".text.init"
   .globl _start
 _start:
-  la t0, trap_entry
-  csrw mtvec, t0
-
   li  x1, 0
   li  x2, 0
   li  x3, 0
@@ -55,20 +52,23 @@ _start:
   csrs mstatus, t0
 
   # make sure XLEN agrees with compilation choice
-  csrr t0, misa
+  li t0, 1
+  slli t0, t0, 31
 #if __riscv_xlen == 64
-  bltz t0, 1f
-#else
   bgez t0, 1f
+#else
+  bltz t0, 1f
 #endif
-  li a0, 1234
-  j tohost_exit
+2:
+  li a0, 1
+  sw a0, tohost, t0
+  j 2b
 1:
 
 #ifdef __riscv_flen
   # initialize FPU if we have one
-  andi t0, t0, 1 << ('f' - 'a')
-  beqz t0, 1f
+  la t0, 1f
+  csrw mtvec, t0
 
   fssr    x0
   fmv.s.x f0, x0
@@ -103,12 +103,18 @@ _start:
   fmv.s.x f29,x0
   fmv.s.x f30,x0
   fmv.s.x f31,x0
+1:
 #endif
 
-1:
+  # initialize trap vector
+  la t0, trap_entry
+  csrw mtvec, t0
 
   # initialize global pointer
-  la gp, _gp
+.option push
+.option norelax
+  la gp, __global_pointer$
+.option pop
 
   la  tp, _end + 63
   and tp, tp, -64
diff --git a/benchmarks/common/syscalls.c b/benchmarks/common/syscalls.c
index 3f15aac..4940aa2 100644
--- a/benchmarks/common/syscalls.c
+++ b/benchmarks/common/syscalls.c
@@ -8,15 +8,13 @@
 #include "util.h"
 
 #define SYS_write 64
-#define SYS_exit 93
-#define SYS_stats 1234
 
 #undef strcmp
 
 extern volatile uint64_t tohost;
 extern volatile uint64_t fromhost;
 
-static uintptr_t handle_frontend_syscall(uintptr_t which, uint64_t arg0, uint64_t arg1, uint64_t arg2)
+static uintptr_t syscall(uintptr_t which, uint64_t arg0, uint64_t arg1, uint64_t arg2)
 {
   volatile uint64_t magic_mem[8] __attribute__((aligned(64)));
   magic_mem[0] = which;
@@ -38,7 +36,7 @@ static uintptr_t handle_frontend_syscall(uintptr_t which, uint64_t arg0, uint64_
 static uintptr_t counters[NUM_COUNTERS];
 static char* counter_names[NUM_COUNTERS];
 
-static int handle_stats(int enable)
+void setStats(int enable)
 {
   int i = 0;
 #define READ_CTR(name) do { \
@@ -52,7 +50,6 @@ static int handle_stats(int enable)
   READ_CTR(minstret);
 
 #undef READ_CTR
-  return 0;
 }
 
 void __attribute__((noreturn)) tohost_exit(uintptr_t code)
@@ -61,39 +58,14 @@ void __attribute__((noreturn)) tohost_exit(uintptr_t code)
   while (1);
 }
 
-uintptr_t handle_trap(uintptr_t cause, uintptr_t epc, uintptr_t regs[32])
-{
-  if (cause != CAUSE_MACHINE_ECALL)
-    tohost_exit(1337);
-  else if (regs[17] == SYS_exit)
-    tohost_exit(regs[10]);
-  else if (regs[17] == SYS_stats)
-    regs[10] = handle_stats(regs[10]);
-  else
-    regs[10] = handle_frontend_syscall(regs[17], regs[10], regs[11], regs[12]);
-
-  return epc + ((*(unsigned short*)epc & 3) == 3 ? 4 : 2);
-}
-
-static uintptr_t syscall(uintptr_t num, uintptr_t arg0, uintptr_t arg1, uintptr_t arg2)
+uintptr_t __attribute__((weak)) handle_trap(uintptr_t cause, uintptr_t epc, uintptr_t regs[32])
 {
-  register uintptr_t a7 asm("a7") = num;
-  register uintptr_t a0 asm("a0") = arg0;
-  register uintptr_t a1 asm("a1") = arg1;
-  register uintptr_t a2 asm("a2") = arg2;
-  asm volatile ("scall" : "+r"(a0) : "r"(a1), "r"(a2), "r"(a7));
-  return a0;
+  tohost_exit(1337);
 }
 
 void exit(int code)
 {
-  syscall(SYS_exit, code, 0, 0);
-  while (1);
-}
-
-void setStats(int enable)
-{
-  syscall(SYS_stats, enable, 0, 0);
+  tohost_exit(code);
 }
 
 void printstr(const char* s)
diff --git a/benchmarks/common/test.ld b/benchmarks/common/test.ld
index dd32bb1..a08d71d 100644
--- a/benchmarks/common/test.ld
+++ b/benchmarks/common/test.ld
@@ -22,7 +22,7 @@ SECTIONS
 
   /* text: test code section */
   . = 0x80000000;
-  .text.init : { crt.o(.text) }
+  .text.init : { *(.text.init) }
 
   .tohost ALIGN(0x1000) : { *(.tohost) }
 
@@ -32,7 +32,7 @@ SECTIONS
   .data : { *(.data) }
 
   .sdata : {
-    _gp = . + 0x800;
+    __global_pointer$ = . + 0x800;
     *(.srodata.cst16) *(.srodata.cst8) *(.srodata.cst4) *(.srodata.cst2) *(.srodata*)
     *(.sdata .sdata.* .gnu.linkonce.s.*)
   }
@@ -48,14 +48,14 @@ SECTIONS
   .tdata :
   {
     _tls_data = .;
-    crt.o(.tdata.begin)
+    *(.tdata.begin)
     *(.tdata)
-    crt.o(.tdata.end)
+    *(.tdata.end)
   }
   .tbss :
   {
     *(.tbss)
-    crt.o(.tbss.end)
+    *(.tbss.end)
   }
 
   /* End of uninitalized data segement */
diff --git a/benchmarks/common/util.h b/benchmarks/common/util.h
index 22f81cf..081cfd6 100644
--- a/benchmarks/common/util.h
+++ b/benchmarks/common/util.h
@@ -3,62 +3,12 @@
 #ifndef __UTIL_H
 #define __UTIL_H
 
-//--------------------------------------------------------------------------
-// Macros
-
-// Set HOST_DEBUG to 1 if you are going to compile this for a host
-// machine (ie Athena/Linux) for debug purposes and set HOST_DEBUG
-// to 0 if you are compiling with the smips-gcc toolchain.
-
-#ifndef HOST_DEBUG
-#define HOST_DEBUG 0
-#endif
-
-// Set PREALLOCATE to 1 if you want to preallocate the benchmark
-// function before starting stats. If you have instruction/data
-// caches and you don't want to count the overhead of misses, then
-// you will need to use preallocation.
-
-#ifndef PREALLOCATE
-#define PREALLOCATE 0
-#endif
-
-// Set SET_STATS to 1 if you want to carve out the piece that actually
-// does the computation.
-
-#if HOST_DEBUG
-#include <stdio.h>
-static void setStats(int enable) {}
-#else
 extern void setStats(int enable);
-#endif
 
 #include <stdint.h>
 
 #define static_assert(cond) switch(0) { case 0: case !!(long)(cond): ; }
 
-static void printArray(const char name[], int n, const int arr[])
-{
-#if HOST_DEBUG
-  int i;
-  printf( " %10s :", name );
-  for ( i = 0; i < n; i++ )
-    printf( " %3d ", arr[i] );
-  printf( "\n" );
-#endif
-}
-
-static void printDoubleArray(const char name[], int n, const double arr[])
-{
-#if HOST_DEBUG
-  int i;
-  printf( " %10s :", name );
-  for ( i = 0; i < n; i++ )
-    printf( " %g ", arr[i] );
-  printf( "\n" );
-#endif
-}
-
 static int verify(int n, const volatile int* test, const int* verify)
 {
   int i;
@@ -117,6 +67,11 @@ static uint64_t lfsr(uint64_t x)
   return (x >> 1) | (bit << 62);
 }
 
+static uintptr_t insn_len(uintptr_t pc)
+{
+  return (*(unsigned short*)pc & 3) ? 4 : 2;
+}
+
 #ifdef __riscv
 #include "encoding.h"
 #endif
diff --git a/benchmarks/dhrystone/bmark.mk b/benchmarks/dhrystone/bmark.mk
deleted file mode 100644
index 06b4ab8..0000000
--- a/benchmarks/dhrystone/bmark.mk
+++ /dev/null
@@ -1,32 +0,0 @@
-#=======================================================================
-# UCB CS250 Makefile fragment for benchmarks
-#-----------------------------------------------------------------------
-#
-# Each benchmark directory should have its own fragment which
-# essentially lists what the source files are and how to link them
-# into an riscv and/or host executable. All variables should include
-# the benchmark name as a prefix so that they are unique.
-#
-
-dhrystone_c_src = \
-	dhrystone_main.c \
-	dhrystone.c \
-	syscalls.c \
-
-dhrystone_riscv_src = \
-	crt.S \
-
-dhrystone_c_objs     = $(patsubst %.c, %.o, $(dhrystone_c_src))
-dhrystone_riscv_objs = $(patsubst %.S, %.o, $(dhrystone_riscv_src))
-
-dhrystone_host_bin = dhrystone.host
-$(dhrystone_host_bin): $(dhrystone_c_src)
-	$(HOST_COMP) $^ -o $(dhrystone_host_bin)
-
-dhrystone_riscv_bin = dhrystone.riscv
-$(dhrystone_riscv_bin): $(dhrystone_c_objs) $(dhrystone_riscv_objs)
-	$(RISCV_LINK) $(dhrystone_c_objs) $(dhrystone_riscv_objs) \
-    -o $(dhrystone_riscv_bin) $(RISCV_LINK_OPTS)
-
-junk += $(dhrystone_c_objs) $(dhrystone_riscv_objs) \
-        $(dhrystone_host_bin) $(dhrystone_riscv_bin)
diff --git a/benchmarks/dhrystone/dhrystone_main.c b/benchmarks/dhrystone/dhrystone_main.c
index 492270a..9c7bcf5 100644
--- a/benchmarks/dhrystone/dhrystone_main.c
+++ b/benchmarks/dhrystone/dhrystone_main.c
@@ -5,45 +5,13 @@
 //--------------------------------------------------------------------------
 //
 // This is the classic Dhrystone synthetic integer benchmark.
-// You should not change anything except the HOST_DEBUG and
-// PREALLOCATE macros for your timing run.
+//
 
 #pragma GCC optimize ("no-inline")
 
 #include "dhrystone.h"
 
-//--------------------------------------------------------------------------
-// Macros
-
-// Set HOST_DEBUG to 1 if you are going to compile this for a host
-// machine (ie Athena/Linux) for debug purposes and set HOST_DEBUG
-// to 0 if you are compiling with the smips-gcc toolchain.
-
-#ifndef HOST_DEBUG
-#define HOST_DEBUG 0
-#endif
-
-// Set PREALLOCATE to 1 if you want to preallocate the benchmark
-// function before starting stats. If you have instruction/data
-// caches and you don't want to count the overhead of misses, then
-// you will need to use preallocation.
-
-#ifndef PREALLOCATE
-#define PREALLOCATE 0
-#endif
-
-// Set SET_STATS to 1 if you want to carve out the piece that actually
-// does the computation.
-
-#ifndef SET_STATS
-#define SET_STATS 0
-#endif
-
-#if HOST_DEBUG
-# define debug_printf printf
-#else
 void debug_printf(const char* str, ...);
-#endif
 
 #include "util.h"
 
@@ -101,21 +69,7 @@ int main (int argc, char** argv)
   REG   int             Number_Of_Runs;
 
   /* Arguments */
-#if HOST_DEBUG
-  if (argc > 2)
-  {
-     printf("Usage: %s [number of loops]\n", argv[0]);
-     exit (1);
-  }
-  if (argc == 2)
-  {
-     Number_Of_Runs = atoi (argv[1]);
-  }
-  else if (Number_Of_Runs <= 0)
-#endif
-  {
-     Number_Of_Runs = NUMBER_OF_RUNS;
-  }
+  Number_Of_Runs = NUMBER_OF_RUNS;
 
   /* Initializations */
 
diff --git a/benchmarks/median/bmark.mk b/benchmarks/median/bmark.mk
deleted file mode 100644
index b489a67..0000000
--- a/benchmarks/median/bmark.mk
+++ /dev/null
@@ -1,31 +0,0 @@
-#=======================================================================
-# UCB CS250 Makefile fragment for benchmarks
-#-----------------------------------------------------------------------
-#
-# Each benchmark directory should have its own fragment which
-# essentially lists what the source files are and how to link them
-# into an riscv and/or host executable. All variables should include
-# the benchmark name as a prefix so that they are unique.
-#
-
-median_c_src = \
-	median_main.c \
-	median.c \
-	syscalls.c \
-
-median_riscv_src = \
-	crt.S \
-
-median_c_objs     = $(patsubst %.c, %.o, $(median_c_src))
-median_riscv_objs = $(patsubst %.S, %.o, $(median_riscv_src))
-
-median_host_bin = median.host
-$(median_host_bin): $(median_c_src)
-	$(HOST_COMP) $^ -o $(median_host_bin)
-
-median_riscv_bin = median.riscv
-$(median_riscv_bin): $(median_c_objs) $(median_riscv_objs)
-	$(RISCV_LINK) $(median_c_objs) $(median_riscv_objs) -o $(median_riscv_bin) $(RISCV_LINK_OPTS)
-
-junk += $(median_c_objs) $(median_riscv_objs) \
-        $(median_host_bin) $(median_riscv_bin)
diff --git a/benchmarks/median/median_main.c b/benchmarks/median/median_main.c
index c7fd6a6..b62ecd1 100644
--- a/benchmarks/median/median_main.c
+++ b/benchmarks/median/median_main.c
@@ -7,8 +7,7 @@
 // This benchmark performs a 1D three element median filter. The
 // input data (and reference data) should be generated using the
 // median_gendata.pl perl script and dumped to a file named
-// dataset1.h You should not change anything except the
-// HOST_DEBUG and PREALLOCATE macros for your timing run.
+// dataset1.h.
 
 #include "util.h"
 
@@ -26,10 +25,6 @@ int main( int argc, char* argv[] )
 {
   int results_data[DATA_SIZE];
 
-  // Output the input array
-  printArray( "input",  DATA_SIZE, input_data  );
-  printArray( "verify", DATA_SIZE, verify_data );
-
 #if PREALLOCATE
   // If needed we preallocate everything in the caches
   median( DATA_SIZE, input_data, results_data );
@@ -40,9 +35,6 @@ int main( int argc, char* argv[] )
   median( DATA_SIZE, input_data, results_data );
   setStats(0);
 
-  // Print out the results
-  printArray( "results", DATA_SIZE, results_data );
-
   // Check the results
   return verify( DATA_SIZE, results_data, verify_data );
 }
diff --git a/benchmarks/mm/bmark.mk b/benchmarks/mm/bmark.mk
deleted file mode 100644
index ec8fd16..0000000
--- a/benchmarks/mm/bmark.mk
+++ /dev/null
@@ -1,31 +0,0 @@
-#=======================================================================
-# UCB CS250 Makefile fragment for benchmarks
-#-----------------------------------------------------------------------
-#
-# Each benchmark directory should have its own fragment which
-# essentially lists what the source files are and how to link them
-# into an riscv and/or host executable. All variables should include
-# the benchmark name as a prefix so that they are unique.
-#
-
-mm_c_src = \
-	mm_main.c \
-	mm.c \
-	syscalls.c \
-
-mm_riscv_src = \
-	crt.S \
-
-mm_c_objs     = $(patsubst %.c, %.o, $(mm_c_src))
-mm_riscv_objs = $(patsubst %.S, %.o, $(mm_riscv_src))
-
-mm_host_bin = mm.host
-$(mm_host_bin) : $(mm_c_src)
-	$(HOST_COMP) $^ -o $(mm_host_bin)
-
-mm_riscv_bin = mm.riscv
-$(mm_riscv_bin) : $(mm_c_objs) $(mm_riscv_objs)
-	$(RISCV_LINK) $(mm_c_objs) $(mm_riscv_objs) -o $(mm_riscv_bin) $(RISCV_LINK_OPTS)
-
-junk += $(mm_c_objs) $(mm_riscv_objs) \
-        $(mm_host_bin) $(mm_riscv_bin)
diff --git a/benchmarks/mt-matmul/bmark.mk b/benchmarks/mt-matmul/bmark.mk
deleted file mode 100644
index 6a7140f..0000000
--- a/benchmarks/mt-matmul/bmark.mk
+++ /dev/null
@@ -1,31 +0,0 @@
-#=======================================================================
-# UCB CS250 Makefile fragment for benchmarks
-#-----------------------------------------------------------------------
-#
-# Each benchmark directory should have its own fragment which
-# essentially lists what the source files are and how to link them
-# into an riscv and/or host executable. All variables should include
-# the benchmark name as a prefix so that they are unique.
-#
-
-mt_matmul_c_src = \
-	mt-matmul.c \
-	matmul.c \
-	syscalls.c \
-
-mt_matmul_riscv_src = \
-	crt.S \
-
-mt_matmul_c_objs     = $(patsubst %.c, %.o, $(mt_matmul_c_src))
-mt_matmul_riscv_objs = $(patsubst %.S, %.o, $(mt_matmul_riscv_src))
-
-mt_matmul_host_bin = mt-matmul.host
-$(mt_matmul_host_bin) : $(mt_matmul_c_src)
-	$(HOST_COMP) $^ -o $(mt_matmul_host_bin)
-
-mt_matmul_riscv_bin = mt-matmul.riscv
-$(mt_matmul_riscv_bin) : $(mt_matmul_c_objs) $(mt_matmul_riscv_objs)
-	$(RISCV_LINK) $(mt_matmul_c_objs) $(mt_matmul_riscv_objs) $(RISCV_LINK_OPTS) -o $(mt_matmul_riscv_bin)
-
-junk += $(mt_matmul_c_objs) $(mt_matmul_riscv_objs) \
-        $(mt_matmul_host_bin) $(mt_matmul_riscv_bin)
diff --git a/benchmarks/mt-matmul/mt-matmul.c b/benchmarks/mt-matmul/mt-matmul.c
index 9c68ada..dbb7562 100644
--- a/benchmarks/mt-matmul/mt-matmul.c
+++ b/benchmarks/mt-matmul/mt-matmul.c
@@ -12,10 +12,6 @@
 // using the matmul_gendata.pl perl script and dumped to a file named
 // dataset.h. 
 
-
-// print out arrays, etc.
-//#define DEBUG
-
 //--------------------------------------------------------------------------
 // Includes 
 
@@ -57,10 +53,5 @@ void thread_entry(int cid, int nc)
  
    int res = verify(ARRAY_SIZE, results_data, verify_data);
 
-#ifdef DEBUG
-   printArray("results:", ARRAY_SIZE, results_data);
-   printArray("verify :", ARRAY_SIZE, verify_data);
-#endif
-
    exit(res);
 }
diff --git a/benchmarks/mt-vvadd/bmark.mk b/benchmarks/mt-vvadd/bmark.mk
deleted file mode 100644
index ff969c1..0000000
--- a/benchmarks/mt-vvadd/bmark.mk
+++ /dev/null
@@ -1,31 +0,0 @@
-#=======================================================================
-# UCB CS250 Makefile fragment for benchmarks
-#-----------------------------------------------------------------------
-#
-# Each benchmark directory should have its own fragment which
-# essentially lists what the source files are and how to link them
-# into an riscv and/or host executable. All variables should include
-# the benchmark name as a prefix so that they are unique.
-#
-
-mt_vvadd_c_src = \
-	mt-vvadd.c \
-	vvadd.c \
-	syscalls.c \
-
-mt_vvadd_riscv_src = \
-	crt.S \
-
-mt_vvadd_c_objs     = $(patsubst %.c, %.o, $(mt_vvadd_c_src))
-mt_vvadd_riscv_objs = $(patsubst %.S, %.o, $(mt_vvadd_riscv_src))
-
-mt_vvadd_host_bin = mt-vvadd.host
-$(mt_vvadd_host_bin) : $(mt_vvadd_c_src)
-	$(HOST_COMP) $^ -o $(mt_vvadd_host_bin)
-
-mt_vvadd_riscv_bin = mt-vvadd.riscv
-$(mt_vvadd_riscv_bin) : $(mt_vvadd_c_objs) $(mt_vvadd_riscv_objs)
-	$(RISCV_LINK) $(mt_vvadd_c_objs) $(mt_vvadd_riscv_objs) $(RISCV_LINK_OPTS) -o $(mt_vvadd_riscv_bin)
-
-junk += $(mt_vvadd_c_objs) $(mt_vvadd_riscv_objs) \
-        $(mt_vvadd_host_bin) $(mt_vvadd_riscv_bin)
diff --git a/benchmarks/mt-vvadd/mt-vvadd.c b/benchmarks/mt-vvadd/mt-vvadd.c
index 8709166..54c9602 100644
--- a/benchmarks/mt-vvadd/mt-vvadd.c
+++ b/benchmarks/mt-vvadd/mt-vvadd.c
@@ -12,9 +12,6 @@
 // generated using the vvadd_gendata.pl perl script and dumped
 // to a file named dataset.h 
 
-// to print out arrays, etc.
-//#define DEBUG
-
 //--------------------------------------------------------------------------
 // Includes 
 
@@ -57,10 +54,6 @@ void thread_entry(int cid, int nc)
    stats(vvadd(cid, nc, DATA_SIZE, input1_data, input2_data, results_data); barrier(nc), DATA_SIZE);
  
    if(cid == 0) {
-#ifdef DEBUG
-     printDoubleArray("out-of-place results: ", DATA_SIZE, results_data);
-     printDoubleArray("out-of-place verify : ", DATA_SIZE, verify_data);
-#endif
      int res = verifyDouble(DATA_SIZE, results_data, verify_data);
      if(res) exit(res);
    }
@@ -76,10 +69,6 @@ void thread_entry(int cid, int nc)
    stats(vvadd(cid, nc, DATA_SIZE, results_data, input2_data, results_data); barrier(nc), DATA_SIZE);
  
    if(cid == 0) {
-#ifdef DEBUG
-     printDoubleArray("in-place results: ", DATA_SIZE, results_data);
-     printDoubleArray("in-place verify : ", DATA_SIZE, verify_data);
-#endif
      int res = verifyDouble(DATA_SIZE, results_data, verify_data);
      if(res) exit(res);
    }
diff --git a/benchmarks/multiply/bmark.mk b/benchmarks/multiply/bmark.mk
deleted file mode 100644
index 93ba67f..0000000
--- a/benchmarks/multiply/bmark.mk
+++ /dev/null
@@ -1,31 +0,0 @@
-#=======================================================================
-# UCB CS250 Makefile fragment for benchmarks
-#-----------------------------------------------------------------------
-#
-# Each benchmark directory should have its own fragment which 
-# essentially lists what the source files are and how to link them
-# into an riscv and/or host executable. All variables should include 
-# the benchmark name as a prefix so that they are unique.
-#
-
-multiply_c_src = \
-	multiply_main.c \
-	multiply.c \
-	syscalls.c \
-
-multiply_riscv_src = \
-	crt.S \
-
-multiply_c_objs     = $(patsubst %.c, %.o, $(multiply_c_src))
-multiply_riscv_objs = $(patsubst %.S, %.o, $(multiply_riscv_src))
-
-multiply_host_bin = multiply.host
-$(multiply_host_bin): $(multiply_c_src)
-	$(HOST_COMP) $^ -o $(multiply_host_bin)
-
-multiply_riscv_bin = multiply.riscv
-$(multiply_riscv_bin): $(multiply_c_objs) $(multiply_riscv_objs)
-	$(RISCV_LINK) $(multiply_c_objs) $(multiply_riscv_objs) -o $(multiply_riscv_bin) $(RISCV_LINK_OPTS)
-
-junk += $(multiply_c_objs) $(multiply_riscv_objs) \
-        $(multiply_host_bin) $(multiply_riscv_bin)
diff --git a/benchmarks/multiply/multiply_main.c b/benchmarks/multiply/multiply_main.c
index aed3e0d..ea225eb 100644
--- a/benchmarks/multiply/multiply_main.c
+++ b/benchmarks/multiply/multiply_main.c
@@ -7,8 +7,7 @@
 // This benchmark tests the software multiply implemenation. The
 // input data (and reference data) should be generated using the
 // multiply_gendata.pl perl script and dumped to a file named
-// dataset1.h You should not change anything except the
-// HOST_DEBUG and VERIFY macros for your timing run.
+// dataset1.h
 
 #include "util.h"
 
@@ -27,11 +26,6 @@ int main( int argc, char* argv[] )
   int i;
   int results_data[DATA_SIZE];
 
-  // Output the input arrays
-  printArray( "input1",  DATA_SIZE, input_data1  );
-  printArray( "input2",  DATA_SIZE, input_data2  );
-  printArray( "verify", DATA_SIZE, verify_data );
-
 #if PREALLOCATE
   for (i = 0; i < DATA_SIZE; i++)
   {
@@ -45,9 +39,6 @@ int main( int argc, char* argv[] )
     results_data[i] = multiply( input_data1[i], input_data2[i] );
   }
   setStats(0);
-  
-  // Print out the results
-  printArray( "results", DATA_SIZE, results_data );
 
   // Check the results
   return verify( DATA_SIZE, results_data, verify_data );
diff --git a/benchmarks/pmp/pmp.c b/benchmarks/pmp/pmp.c
new file mode 100644
index 0000000..9d07ed8
--- /dev/null
+++ b/benchmarks/pmp/pmp.c
@@ -0,0 +1,198 @@
+// See LICENSE for license details.
+
+// Test of PMP functionality.
+
+#include <stdint.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include "util.h"
+
+volatile int trap_expected;
+
+#define INLINE inline __attribute__((always_inline))
+
+uintptr_t handle_trap(uintptr_t cause, uintptr_t epc, uintptr_t regs[32])
+{
+  if (cause == CAUSE_ILLEGAL_INSTRUCTION)
+    exit(0); // no PMP support
+
+  if (!trap_expected || cause != CAUSE_LOAD_ACCESS)
+    exit(1);
+  trap_expected = 0;
+  return epc + insn_len(epc);
+}
+
+#define SCRATCH RISCV_PGSIZE
+uintptr_t scratch[RISCV_PGSIZE / sizeof(uintptr_t)] __attribute__((aligned(RISCV_PGSIZE)));
+uintptr_t l1pt[RISCV_PGSIZE / sizeof(uintptr_t)] __attribute__((aligned(RISCV_PGSIZE)));
+uintptr_t l2pt[RISCV_PGSIZE / sizeof(uintptr_t)] __attribute__((aligned(RISCV_PGSIZE)));
+#if __riscv_xlen == 64
+uintptr_t l3pt[RISCV_PGSIZE / sizeof(uintptr_t)] __attribute__((aligned(RISCV_PGSIZE)));
+#else
+#define l3pt l2pt
+#endif
+
+static void init_pt()
+{
+  l1pt[0] = ((uintptr_t)l2pt >> RISCV_PGSHIFT << PTE_PPN_SHIFT) | PTE_V;
+  l3pt[SCRATCH / RISCV_PGSIZE] = ((uintptr_t)scratch >> RISCV_PGSHIFT << PTE_PPN_SHIFT) | PTE_A | PTE_D | PTE_V | PTE_R | PTE_W;
+#if __riscv_xlen == 64
+  l2pt[0] = ((uintptr_t)l3pt >> RISCV_PGSHIFT << PTE_PPN_SHIFT) | PTE_V;
+  uintptr_t vm_choice = SPTBR_MODE_SV39;
+#else
+  uintptr_t vm_choice = SPTBR_MODE_SV32;
+#endif
+  write_csr(sptbr, ((uintptr_t)l1pt >> RISCV_PGSHIFT) |
+                   (vm_choice * (SPTBR_MODE & ~(SPTBR_MODE<<1))));
+  write_csr(pmpcfg0, (PMP_NAPOT | PMP_R) << 16);
+  write_csr(pmpaddr2, -1);
+}
+
+INLINE uintptr_t va2pa(uintptr_t va)
+{
+  if (va < SCRATCH || va >= SCRATCH + RISCV_PGSIZE)
+    exit(3);
+  return va - SCRATCH + (uintptr_t)scratch;
+}
+
+#define GRANULE (1UL << PMP_SHIFT)
+
+typedef struct {
+  uintptr_t cfg;
+  uintptr_t a0;
+  uintptr_t a1;
+} pmpcfg_t;
+
+INLINE int pmp_ok(pmpcfg_t p, uintptr_t addr, uintptr_t size)
+{
+  if ((p.cfg & PMP_A) == 0)
+    return 1;
+
+  if ((p.cfg & PMP_A) != PMP_TOR) {
+    uintptr_t range = 1;
+
+    if ((p.cfg & PMP_A) == PMP_NAPOT) {
+      range <<= 1;
+      for (uintptr_t i = 1; i; i <<= 1) {
+        if ((p.a1 & i) == 0)
+          break;
+        p.a1 &= ~i;
+        range <<= 1;
+      }
+    }
+
+    p.a0 = p.a1;
+    p.a1 = p.a0 + range;
+  }
+
+  p.a0 *= GRANULE;
+  p.a1 *= GRANULE;
+  addr = va2pa(addr);
+
+  uintptr_t hits = 0;
+  for (uintptr_t i = 0; i < size; i += GRANULE) {
+    if (p.a0 <= addr + i && addr + i < p.a1)
+      hits += GRANULE;
+  }
+
+  return hits == 0 || hits >= size;
+}
+
+INLINE void test_one(uintptr_t addr, uintptr_t size)
+{
+  uintptr_t new_mstatus = (read_csr(mstatus) & ~MSTATUS_MPP) | (MSTATUS_MPP & (MSTATUS_MPP >> 1)) | MSTATUS_MPRV;
+  switch (size) {
+    case 1: asm volatile ("csrrw %0, mstatus, %0; lb x0, (%1); csrw mstatus, %0" : "+&r" (new_mstatus) : "r" (addr)); break;
+    case 2: asm volatile ("csrrw %0, mstatus, %0; lh x0, (%1); csrw mstatus, %0" : "+&r" (new_mstatus) : "r" (addr)); break;
+    case 4: asm volatile ("csrrw %0, mstatus, %0; lw x0, (%1); csrw mstatus, %0" : "+&r" (new_mstatus) : "r" (addr)); break;
+#if __riscv_xlen >= 64
+    case 8: asm volatile ("csrrw %0, mstatus, %0; ld x0, (%1); csrw mstatus, %0" : "+&r" (new_mstatus) : "r" (addr)); break;
+#endif
+    default: __builtin_unreachable();
+  }
+}
+
+INLINE void test_all_sizes(pmpcfg_t p, uintptr_t addr)
+{
+  for (size_t size = 1; size <= sizeof(uintptr_t); size *= 2) {
+    if (addr & (size - 1))
+      continue;
+    trap_expected = !pmp_ok(p, addr, size);
+    test_one(addr, size);
+    if (trap_expected)
+      exit(2);
+  }
+}
+
+INLINE void test_range_once(pmpcfg_t p, uintptr_t base, uintptr_t range)
+{
+  for (uintptr_t addr = base; addr < base + range; addr += GRANULE)
+    test_all_sizes(p, addr);
+}
+
+INLINE pmpcfg_t set_pmp(pmpcfg_t p)
+{
+  uintptr_t cfg0 = read_csr(pmpcfg0);
+  write_csr(pmpcfg0, cfg0 & ~0xff00);
+  write_csr(pmpaddr0, p.a0);
+  write_csr(pmpaddr1, p.a1);
+  write_csr(pmpcfg0, ((p.cfg << 8) & 0xff00) | (cfg0 & ~0xff00));
+  asm volatile ("sfence.vma");
+  return p;
+}
+
+INLINE pmpcfg_t set_pmp_range(uintptr_t base, uintptr_t range)
+{
+  pmpcfg_t p;
+  p.cfg = PMP_TOR | PMP_R;
+  p.a0 = base >> PMP_SHIFT;
+  p.a1 = (base + range) >> PMP_SHIFT;
+  return set_pmp(p);
+}
+
+INLINE pmpcfg_t set_pmp_napot(uintptr_t base, uintptr_t range)
+{
+  pmpcfg_t p;
+  p.cfg = PMP_R | (range > GRANULE ? PMP_NAPOT : PMP_NA4);
+  p.a0 = 0;
+  p.a1 = (base + (range/2 - 1)) >> PMP_SHIFT;
+  return set_pmp(p);
+}
+
+static void test_range(uintptr_t addr, uintptr_t range)
+{
+  pmpcfg_t p = set_pmp_range(va2pa(addr), range);
+  test_range_once(p, addr, range);
+
+  if ((range & (range - 1)) == 0 && (addr & (range - 1)) == 0) {
+    p = set_pmp_napot(va2pa(addr), range);
+    test_range_once(p, addr, range);
+  }
+}
+
+static void test_ranges(uintptr_t addr, uintptr_t size)
+{
+  for (uintptr_t range = GRANULE; range <= size; range += GRANULE)
+    test_range(addr, range);
+}
+
+static void exhaustive_test(uintptr_t addr, uintptr_t size)
+{
+  for (uintptr_t base = addr; base < addr + size; base += GRANULE)
+    test_ranges(base, size - (base - addr));
+}
+
+int main()
+{
+  init_pt();
+
+  const int max_exhaustive = 32;
+  exhaustive_test(SCRATCH, max_exhaustive);
+  exhaustive_test(SCRATCH + RISCV_PGSIZE - max_exhaustive, max_exhaustive);
+
+  test_range(SCRATCH, RISCV_PGSIZE);
+  test_range(SCRATCH, RISCV_PGSIZE / 2);
+  test_range(SCRATCH + RISCV_PGSIZE / 2, RISCV_PGSIZE / 2);
+
+  return 0;
+}
diff --git a/benchmarks/qsort/bmark.mk b/benchmarks/qsort/bmark.mk
deleted file mode 100644
index 4b39d96..0000000
--- a/benchmarks/qsort/bmark.mk
+++ /dev/null
@@ -1,30 +0,0 @@
-#=======================================================================
-# UCB CS250 Makefile fragment for benchmarks
-#-----------------------------------------------------------------------
-#
-# Each benchmark directory should have its own fragment which
-# essentially lists what the source files are and how to link them
-# into an riscv and/or host executable. All variables should include
-# the benchmark name as a prefix so that they are unique.
-#
-
-qsort_c_src = \
-	qsort_main.c \
-	syscalls.c \
-
-qsort_riscv_src = \
-	crt.S \
-
-qsort_c_objs     = $(patsubst %.c, %.o, $(qsort_c_src))
-qsort_riscv_objs = $(patsubst %.S, %.o, $(qsort_riscv_src))
-
-qsort_host_bin = qsort.host
-$(qsort_host_bin) : $(qsort_c_src)
-	$(HOST_COMP) $^ -o $(qsort_host_bin)
-
-qsort_riscv_bin = qsort.riscv
-$(qsort_riscv_bin) : $(qsort_c_objs) $(qsort_riscv_objs)
-	$(RISCV_LINK) $(qsort_c_objs) $(qsort_riscv_objs) -o $(qsort_riscv_bin) $(RISCV_LINK_OPTS)
-
-junk += $(qsort_c_objs) $(qsort_riscv_objs) \
-        $(qsort_host_bin) $(qsort_riscv_bin)
diff --git a/benchmarks/qsort/qsort_main.c b/benchmarks/qsort/qsort_main.c
index e962e82..18adc25 100644
--- a/benchmarks/qsort/qsort_main.c
+++ b/benchmarks/qsort/qsort_main.c
@@ -8,10 +8,7 @@
 // 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 The smips-gcc toolchain does not support system calls
-// so printf's can only be used on a host system, not on the smips
-// processor simulator itself. You should not change anything except
-// the HOST_DEBUG and PREALLOCATE macros for your timing run.
+// dataset1.h.
 
 #include "util.h"
 #include <string.h>
@@ -76,10 +73,6 @@ void sort(size_t n, type arr[])
 
   for (;;)
   {
-#if HOST_DEBUG
-    printArray( "", n, arr );
-#endif
-
     // Insertion sort when subarray small enough.
     if ( ir-l < INSERTION_THRESHOLD )
     {
@@ -122,10 +115,6 @@ void sort(size_t n, type arr[])
       // Push pointers to larger subarray on stack,
       // process smaller subarray immediately.
 
-#if HOST_DEBUG
-      assert(stackp < stack+NSTACK);
-#endif
-
       if ( ir-i+1 >= j-l )
       {
         stackp[0] = ir;
@@ -147,10 +136,6 @@ void sort(size_t n, type arr[])
 
 int main( int argc, char* argv[] )
 {
-  // Output the input array
-  printArray( "input", DATA_SIZE, input_data );
-  printArray( "verify", DATA_SIZE, verify_data );
-
 #if PREALLOCATE
   // If needed we preallocate everything in the caches
   sort(DATA_SIZE, verify_data);
@@ -163,9 +148,6 @@ int main( int argc, char* argv[] )
   sort( DATA_SIZE, input_data );
   setStats(0);
 
-  // Print out the results
-  printArray( "test", DATA_SIZE, input_data );
-
   // Check the results
   return verify( DATA_SIZE, input_data, verify_data );
 }
diff --git a/benchmarks/readme.txt b/benchmarks/readme.txt
index a14780f..d90fe2d 100644
--- a/benchmarks/readme.txt
+++ b/benchmarks/readme.txt
@@ -37,13 +37,6 @@ points to make about the toolchain.
    about how to write assembly in C here:
    http://gcc.gnu.org/onlinedocs/gcc/Extended-Asm.html
 
- + If you look at the example benchmarks you will see that I have two
-   important macros HOST_DEBUG and VERIFY. Use HOST_DEBUG to compile the
-   benchmark on your host workstation (ie use standard gcc on Athena/Linux
-   box) and then debug the benchmark. Since you are using standard gcc you
-   can use printf's to make sure that your benchmark actually works before
-   trying it out on your RISCV processor.
-
  + Debugging C compiled code on the RISCV processor is a real pain. It is
    hard to associate the assembly with the C code and there is no
    debugger. So if you encounter a bug in your processor when running a C
diff --git a/benchmarks/rsort/bmark.mk b/benchmarks/rsort/bmark.mk
deleted file mode 100644
index 0fa713c..0000000
--- a/benchmarks/rsort/bmark.mk
+++ /dev/null
@@ -1,30 +0,0 @@
-#=======================================================================
-# UCB CS250 Makefile fragment for benchmarks
-#-----------------------------------------------------------------------
-#
-# Each benchmark directory should have its own fragment which
-# essentially lists what the source files are and how to link them
-# into an riscv and/or host executable. All variables should include
-# the benchmark name as a prefix so that they are unique.
-#
-
-rsort_c_src = \
-	rsort.c \
-	syscalls.c \
-
-rsort_riscv_src = \
-	crt.S \
-
-rsort_c_objs     = $(patsubst %.c, %.o, $(rsort_c_src))
-rsort_riscv_objs = $(patsubst %.S, %.o, $(rsort_riscv_src))
-
-rsort_host_bin = rsort.host
-$(rsort_host_bin) : $(rsort_c_src)
-	$(HOST_COMP) $^ -o $(rsort_host_bin)
-
-rsort_riscv_bin = rsort.riscv
-$(rsort_riscv_bin) : $(rsort_c_objs) $(rsort_riscv_objs)
-	$(RISCV_LINK) $(rsort_c_objs) $(rsort_riscv_objs) -o $(rsort_riscv_bin) $(RISCV_LINK_OPTS)
-
-junk += $(rsort_c_objs) $(rsort_riscv_objs) \
-        $(rsort_host_bin) $(rsort_riscv_bin)
diff --git a/benchmarks/rsort/rsort.c b/benchmarks/rsort/rsort.c
index dc53642..7ffc12a 100644
--- a/benchmarks/rsort/rsort.c
+++ b/benchmarks/rsort/rsort.c
@@ -8,10 +8,7 @@
 // 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 The smips-gcc toolchain does not support system calls
-// so printf's can only be used on a host system, not on the smips
-// processor simulator itself. You should not change anything except
-// the HOST_DEBUG and PREALLOCATE macros for your timing run.
+// dataset1.h
 
 #include "util.h"
 #include <string.h>
@@ -107,9 +104,6 @@ void sort(size_t n, type* arrIn, type* scratchIn)
 int main( int argc, char* argv[] )
 {
   static type scratch[DATA_SIZE];
-  // Output the input array
-  printArray( "input", DATA_SIZE, input_data );
-  printArray( "verify", DATA_SIZE, verify_data );
 
 #if PREALLOCATE
   // If needed we preallocate everything in the caches
@@ -123,9 +117,6 @@ int main( int argc, char* argv[] )
   sort(DATA_SIZE, input_data, scratch);
   setStats(0);
 
-  // Print out the results
-  printArray( "test", DATA_SIZE, input_data );
-
   // Check the results
   return verify( DATA_SIZE, input_data, verify_data );
 }
diff --git a/benchmarks/spmv/bmark.mk b/benchmarks/spmv/bmark.mk
deleted file mode 100644
index dcfdb19..0000000
--- a/benchmarks/spmv/bmark.mk
+++ /dev/null
@@ -1,30 +0,0 @@
-#=======================================================================
-# UCB CS250 Makefile fragment for benchmarks
-#-----------------------------------------------------------------------
-#
-# Each benchmark directory should have its own fragment which
-# essentially lists what the source files are and how to link them
-# into an riscv and/or host executable. All variables should include
-# the benchmark name as a prefix so that they are unique.
-#
-
-spmv_c_src = \
-	spmv_main.c \
-	syscalls.c \
-
-spmv_riscv_src = \
-	crt.S \
-
-spmv_c_objs     = $(patsubst %.c, %.o, $(spmv_c_src))
-spmv_riscv_objs = $(patsubst %.S, %.o, $(spmv_riscv_src))
-
-spmv_host_bin = spmv.host
-$(spmv_host_bin) : $(spmv_c_src)
-	$(HOST_COMP) $^ -o $(spmv_host_bin)
-
-spmv_riscv_bin = spmv.riscv
-$(spmv_riscv_bin) : $(spmv_c_objs) $(spmv_riscv_objs)
-	$(RISCV_LINK) $(spmv_c_objs) $(spmv_riscv_objs) -o $(spmv_riscv_bin) $(RISCV_LINK_OPTS)
-
-junk += $(spmv_c_objs) $(spmv_riscv_objs) \
-        $(spmv_host_bin) $(spmv_riscv_bin)
diff --git a/benchmarks/towers/bmark.mk b/benchmarks/towers/bmark.mk
deleted file mode 100644
index b16bf04..0000000
--- a/benchmarks/towers/bmark.mk
+++ /dev/null
@@ -1,30 +0,0 @@
-#=======================================================================
-# UCB CS250 Makefile fragment for benchmarks
-#-----------------------------------------------------------------------
-#
-# Each benchmark directory should have its own fragment which
-# essentially lists what the source files are and how to link them
-# into an riscv and/or host executable. All variables should include
-# the benchmark name as a prefix so that they are unique.
-#
-
-towers_c_src = \
-	towers_main.c \
-	syscalls.c \
-
-towers_riscv_src = \
-	crt.S \
-
-towers_c_objs     = $(patsubst %.c, %.o, $(towers_c_src))
-towers_riscv_objs = $(patsubst %.S, %.o, $(towers_riscv_src))
-
-towers_host_bin = towers.host
-$(towers_host_bin) : $(towers_c_src)
-	$(HOST_COMP) $^ -o $(towers_host_bin)
-
-towers_riscv_bin = towers.riscv
-$(towers_riscv_bin) : $(towers_c_objs) $(towers_riscv_objs)
-	$(RISCV_LINK) $(towers_c_objs) $(towers_riscv_objs) -o $(towers_riscv_bin) $(RISCV_LINK_OPTS)
-
-junk += $(towers_c_objs) $(towers_riscv_objs) \
-        $(towers_host_bin) $(towers_riscv_bin)
diff --git a/benchmarks/towers/towers_main.c b/benchmarks/towers/towers_main.c
index 03de47d..8c2355a 100644
--- a/benchmarks/towers/towers_main.c
+++ b/benchmarks/towers/towers_main.c
@@ -13,10 +13,7 @@
 // disc on top of a smaller disc.
 //
 // This implementation starts with NUM_DISC discs and uses a recursive
-// algorithm to sovel the puzzle.  The smips-gcc toolchain does not support
-// system calls so printf's can only be used on a host system, not on the
-// smips processor simulator itself. You should not change anything except
-// the HOST_DEBUG and PREALLOCATE macros for your timing run.
+// algorithm to sovel the puzzle.
 
 #include "util.h"
 
@@ -140,34 +137,6 @@ void towers_clear( struct Towers* this )
 
 }
 
-#if HOST_DEBUG
-void towers_print( struct Towers* this )
-{
-  struct Node* ptr;
-  int i, numElements;
-
-  printf( "  pegA: " );
-  for ( i = 0; i < ((this->numDiscs)-list_getSize(&(this->pegA))); i++ )
-    printf( ". " );
-  for ( ptr = this->pegA.head; ptr != 0; ptr = ptr->next )
-    printf( "%d ", ptr->val );
-
-  printf( "  pegB: " );
-  for ( i = 0; i < ((this->numDiscs)-list_getSize(&(this->pegB))); i++ )
-    printf( ". " );
-  for ( ptr = this->pegB.head; ptr != 0; ptr = ptr->next )
-    printf( "%d ", ptr->val );
-
-  printf( "  pegC: " );
-  for ( i = 0; i < ((this->numDiscs)-list_getSize(&(this->pegC))); i++ )
-    printf( ". " );
-  for ( ptr = this->pegC.head; ptr != 0; ptr = ptr->next )
-    printf( "%d ", ptr->val );
-
-  printf( "\n" );
-}
-#endif
-
 void towers_solve_h( struct Towers* this, int n,
                      struct List* startPeg,
                      struct List* tempPeg,
@@ -176,9 +145,6 @@ void towers_solve_h( struct Towers* this, int n,
   int val;
 
   if ( n == 1 ) {
-#if HOST_DEBUG
-    towers_print(this);
-#endif
     val = list_pop(startPeg);
     list_push(destPeg,val);
     this->numMoves++;
@@ -202,43 +168,25 @@ int towers_verify( struct Towers* this )
   int numDiscs = 0;
 
   if ( list_getSize(&this->pegA) != 0 ) {
-#if HOST_DEBUG
-    printf( "ERROR: Peg A is not empty!\n" );
-#endif
     return 2;
   }
 
   if ( list_getSize(&this->pegB) != 0 ) {
-#if HOST_DEBUG
-    printf( "ERROR: Peg B is not empty!\n" );
-#endif
     return 3;
   }
 
   if ( list_getSize(&this->pegC) != this->numDiscs ) {
-#if HOST_DEBUG
-    printf( " ERROR: Expected %d discs but found only %d discs!\n", \
-            this->numDiscs, list_getSize(&this->pegC) );
-#endif
     return 4;
   }
 
   for ( ptr = this->pegC.head; ptr != 0; ptr = ptr->next ) {
     numDiscs++;
     if ( ptr->val != numDiscs ) {
-#if HOST_DEBUG
-      printf( " ERROR: Expecting disc %d on peg C but found disc %d instead!\n", \
-              numDiscs, ptr->val );
-#endif
       return 5;
     }
   }
 
   if ( this->numMoves != ((1 << this->numDiscs) - 1) ) {
-#if HOST_DEBUG
-    printf( " ERROR: Expecting %d num moves but found %d instead!\n", \
-            ((1 << this->numDiscs) - 1), this->numMoves );
-#endif
     return 6;
   }
 
@@ -280,12 +228,6 @@ int main( int argc, char* argv[] )
   towers_solve( &towers );
   setStats(0);
 
-  // Print out the results
-
-#if HOST_DEBUG
-  towers_print( &towers );
-#endif
-
   // Check the results
   return towers_verify( &towers );
 }
diff --git a/benchmarks/vvadd/bmark.mk b/benchmarks/vvadd/bmark.mk
deleted file mode 100644
index 5ab99de..0000000
--- a/benchmarks/vvadd/bmark.mk
+++ /dev/null
@@ -1,30 +0,0 @@
-#=======================================================================
-# UCB CS250 Makefile fragment for benchmarks
-#-----------------------------------------------------------------------
-#
-# Each benchmark directory should have its own fragment which
-# essentially lists what the source files are and how to link them
-# into an riscv and/or host executable. All variables should include
-# the benchmark name as a prefix so that they are unique.
-#
-
-vvadd_c_src = \
-	vvadd_main.c \
-	syscalls.c \
-
-vvadd_riscv_src = \
-	crt.S \
-
-vvadd_c_objs     = $(patsubst %.c, %.o, $(vvadd_c_src))
-vvadd_riscv_objs = $(patsubst %.S, %.o, $(vvadd_riscv_src))
-
-vvadd_host_bin = vvadd.host
-$(vvadd_host_bin) : $(vvadd_c_src)
-	$(HOST_COMP) $^ -o $(vvadd_host_bin)
-
-vvadd_riscv_bin = vvadd.riscv
-$(vvadd_riscv_bin) : $(vvadd_c_objs) $(vvadd_riscv_objs)
-	$(RISCV_LINK) $(vvadd_c_objs) $(vvadd_riscv_objs) -o $(vvadd_riscv_bin) $(RISCV_LINK_OPTS)
-
-junk += $(vvadd_c_objs) $(vvadd_riscv_objs) \
-        $(vvadd_host_bin) $(vvadd_riscv_bin)
diff --git a/benchmarks/vvadd/vvadd_main.c b/benchmarks/vvadd/vvadd_main.c
index 00e1593..d8ff01c 100644
--- a/benchmarks/vvadd/vvadd_main.c
+++ b/benchmarks/vvadd/vvadd_main.c
@@ -7,11 +7,7 @@
 // This benchmark uses adds to vectors and writes the results to a
 // third vector. The input data (and reference data) should be
 // generated using the vvadd_gendata.pl perl script and dumped
-// to a file named dataset1.h The smips-gcc toolchain does not
-// support system calls so printf's can only be used on a host system,
-// not on the smips processor simulator itself. You should not change
-// anything except the HOST_DEBUG and PREALLOCATE macros for your timing
-// runs.
+// to a file named dataset1.h.
  
 #include "util.h"
 
@@ -37,11 +33,6 @@ int main( int argc, char* argv[] )
 {
   int results_data[DATA_SIZE];
 
-  // Output the input array
-  printArray( "input1", DATA_SIZE, input1_data );
-  printArray( "input2", DATA_SIZE, input2_data );
-  printArray( "verify", DATA_SIZE, verify_data );
-
 #if PREALLOCATE
   // If needed we preallocate everything in the caches
   vvadd( DATA_SIZE, input1_data, input2_data, results_data );
@@ -52,9 +43,6 @@ int main( int argc, char* argv[] )
   vvadd( DATA_SIZE, input1_data, input2_data, results_data );
   setStats(0);
 
-  // Print out the results
-  printArray( "results", DATA_SIZE, results_data );
-
   // Check the results
   return verify( DATA_SIZE, results_data, verify_data );
 }
diff --git a/debug/programs/entry.S b/debug/programs/entry.S
index a5e745d..e021860 100755
--- a/debug/programs/entry.S
+++ b/debug/programs/entry.S
@@ -35,7 +35,10 @@ handle_reset:
   csrwi mie, 0
 
   # initialize global pointer
-  la gp, _gp
+.option push
+.option norelax
+  la gp, __global_pointer$
+.option pop
 
   # initialize stack pointer
   la sp, stack_top
diff --git a/debug/targets/freedom-e300-sim/link.lds b/debug/targets/freedom-e300-sim/link.lds
index 1dbb99c..1e0645a 100755
--- a/debug/targets/freedom-e300-sim/link.lds
+++ b/debug/targets/freedom-e300-sim/link.lds
@@ -13,7 +13,7 @@ SECTIONS
   .data : { *(.data) }
 
   .sdata : {
-    _gp = . + 0x800;
+    __global_pointer$ = . + 0x800;
     *(.srodata.cst16) *(.srodata.cst8) *(.srodata.cst4) *(.srodata.cst2)
     *(.srodata*)
     *(.sdata .sdata.* .gnu.linkonce.s.*)
diff --git a/debug/targets/freedom-e300/link.lds b/debug/targets/freedom-e300/link.lds
index 1dbb99c..1e0645a 100755
--- a/debug/targets/freedom-e300/link.lds
+++ b/debug/targets/freedom-e300/link.lds
@@ -13,7 +13,7 @@ SECTIONS
   .data : { *(.data) }
 
   .sdata : {
-    _gp = . + 0x800;
+    __global_pointer$ = . + 0x800;
     *(.srodata.cst16) *(.srodata.cst8) *(.srodata.cst4) *(.srodata.cst2)
     *(.srodata*)
     *(.sdata .sdata.* .gnu.linkonce.s.*)
diff --git a/debug/targets/freedom-u500-sim/link.lds b/debug/targets/freedom-u500-sim/link.lds
index 1dbb99c..1e0645a 100755
--- a/debug/targets/freedom-u500-sim/link.lds
+++ b/debug/targets/freedom-u500-sim/link.lds
@@ -13,7 +13,7 @@ SECTIONS
   .data : { *(.data) }
 
   .sdata : {
-    _gp = . + 0x800;
+    __global_pointer$ = . + 0x800;
     *(.srodata.cst16) *(.srodata.cst8) *(.srodata.cst4) *(.srodata.cst2)
     *(.srodata*)
     *(.sdata .sdata.* .gnu.linkonce.s.*)
diff --git a/debug/targets/freedom-u500/link.lds b/debug/targets/freedom-u500/link.lds
index 1dbb99c..1e0645a 100755
--- a/debug/targets/freedom-u500/link.lds
+++ b/debug/targets/freedom-u500/link.lds
@@ -13,7 +13,7 @@ SECTIONS
   .data : { *(.data) }
 
   .sdata : {
-    _gp = . + 0x800;
+    __global_pointer$ = . + 0x800;
     *(.srodata.cst16) *(.srodata.cst8) *(.srodata.cst4) *(.srodata.cst2)
     *(.srodata*)
     *(.sdata .sdata.* .gnu.linkonce.s.*)
diff --git a/debug/targets/spike/link.lds b/debug/targets/spike/link.lds
index 52e4472..b86b2b7 100755
--- a/debug/targets/spike/link.lds
+++ b/debug/targets/spike/link.lds
@@ -15,7 +15,7 @@ SECTIONS
   .data : { *(.data) }
 
   .sdata : {
-    _gp = . + 0x800;
+    __global_pointer$ = . + 0x800;
     *(.srodata.cst16) *(.srodata.cst8) *(.srodata.cst4) *(.srodata.cst2)
     *(.srodata*)
     *(.sdata .sdata.* .gnu.linkonce.s.*)
diff --git a/env b/env
-Subproject 9e219c9ca70459bfda9067d637bb8bf52c5f032
+Subproject 3f1d04a2e8f849c306bdca7c7115462c3a60a6e
diff --git a/isa/Makefile b/isa/Makefile
index 3ed1b60..7f22c53 100644
--- a/isa/Makefile
+++ b/isa/Makefile
@@ -55,7 +55,7 @@ $$($(1)_p_tests): $(1)-p-%: $(1)/%.S
 $(1)_tests += $$($(1)_p_tests)
 
 $$($(1)_v_tests): $(1)-v-%: $(1)/%.S
-	$$(RISCV_GCC) $(2) $$(RISCV_GCC_OPTS) -DENTROPY=0x$$(shell echo \$$@ | md5sum | cut -c 1-7) -std=gnu99 -O2 -I$(src_dir)/../env/v -I$(src_dir)/macros/scalar -T$(src_dir)/../env/v/link.ld $(src_dir)/../env/v/entry.S $(src_dir)/../env/v/*.c $$< -lc -o $$@
+	$$(RISCV_GCC) $(2) $$(RISCV_GCC_OPTS) -DENTROPY=0x$$(shell echo \$$@ | md5sum | cut -c 1-7) -std=gnu99 -O2 -I$(src_dir)/../env/v -I$(src_dir)/macros/scalar -T$(src_dir)/../env/v/link.ld $(src_dir)/../env/v/entry.S $(src_dir)/../env/v/*.c $$< -o $$@
 $(1)_tests += $$($(1)_v_tests)
 
 $(1)_tests_dump = $$(addsuffix .dump, $$($(1)_tests))
@@ -92,8 +92,6 @@ tests_out = $(addsuffix .out, $(spike_tests))
 tests32_out = $(addsuffix .out32, $(spike32_tests))
 
 run: $(tests_out) $(tests32_out)
-	echo; perl -ne 'print "  [$$1] $$ARGV \t$$2\n" if /\*{3}(.{8})\*{3}(.*)/' \
-	       $(tests_out) $(tests32_out); echo;
 
 junk += $(tests) $(tests_dump) $(tests_hex) $(tests_out) $(tests32_out)
 
diff --git a/isa/macros/scalar/test_macros.h b/isa/macros/scalar/test_macros.h
index 006f419..6650fa8 100644
--- a/isa/macros/scalar/test_macros.h
+++ b/isa/macros/scalar/test_macros.h
@@ -44,9 +44,9 @@ test_ ## testnum: \
 #define SEXT_IMM(x) ((x) | (-(((x) >> 11) & 1) << 11))
 
 #define TEST_IMM_OP( testnum, inst, result, val1, imm ) \
-    TEST_CASE( testnum, x3, result, \
+    TEST_CASE( testnum, x30, result, \
       li  x1, MASK_XLEN(val1); \
-      inst x3, x1, SEXT_IMM(imm); \
+      inst x30, x1, SEXT_IMM(imm); \
     )
 
 #define TEST_IMM_SRC1_EQ_DEST( testnum, inst, result, val1, imm ) \
@@ -59,20 +59,20 @@ test_ ## testnum: \
     TEST_CASE( testnum, x6, result, \
       li  x4, 0; \
 1:    li  x1, MASK_XLEN(val1); \
-      inst x3, x1, SEXT_IMM(imm); \
+      inst x30, x1, SEXT_IMM(imm); \
       TEST_INSERT_NOPS_ ## nop_cycles \
-      addi  x6, x3, 0; \
+      addi  x6, x30, 0; \
       addi  x4, x4, 1; \
       li  x5, 2; \
       bne x4, x5, 1b \
     )
 
 #define TEST_IMM_SRC1_BYPASS( testnum, nop_cycles, inst, result, val1, imm ) \
-    TEST_CASE( testnum, x3, result, \
+    TEST_CASE( testnum, x30, result, \
       li  x4, 0; \
 1:    li  x1, MASK_XLEN(val1); \
       TEST_INSERT_NOPS_ ## nop_cycles \
-      inst x3, x1, SEXT_IMM(imm); \
+      inst x30, x1, SEXT_IMM(imm); \
       addi  x4, x4, 1; \
       li  x5, 2; \
       bne x4, x5, 1b \
@@ -94,9 +94,9 @@ test_ ## testnum: \
 #-----------------------------------------------------------------------
 
 #define TEST_R_OP( testnum, inst, result, val1 ) \
-    TEST_CASE( testnum, x3, result, \
+    TEST_CASE( testnum, x30, result, \
       li  x1, val1; \
-      inst x3, x1; \
+      inst x30, x1; \
     )
 
 #define TEST_R_SRC1_EQ_DEST( testnum, inst, result, val1 ) \
@@ -109,9 +109,9 @@ test_ ## testnum: \
     TEST_CASE( testnum, x6, result, \
       li  x4, 0; \
 1:    li  x1, val1; \
-      inst x3, x1; \
+      inst x30, x1; \
       TEST_INSERT_NOPS_ ## nop_cycles \
-      addi  x6, x3, 0; \
+      addi  x6, x30, 0; \
       addi  x4, x4, 1; \
       li  x5, 2; \
       bne x4, x5, 1b \
@@ -122,10 +122,10 @@ test_ ## testnum: \
 #-----------------------------------------------------------------------
 
 #define TEST_RR_OP( testnum, inst, result, val1, val2 ) \
-    TEST_CASE( testnum, x3, result, \
+    TEST_CASE( testnum, x30, result, \
       li  x1, MASK_XLEN(val1); \
       li  x2, MASK_XLEN(val2); \
-      inst x3, x1, x2; \
+      inst x30, x1, x2; \
     )
 
 #define TEST_RR_SRC1_EQ_DEST( testnum, inst, result, val1, val2 ) \
@@ -153,35 +153,35 @@ test_ ## testnum: \
       li  x4, 0; \
 1:    li  x1, MASK_XLEN(val1); \
       li  x2, MASK_XLEN(val2); \
-      inst x3, x1, x2; \
+      inst x30, x1, x2; \
       TEST_INSERT_NOPS_ ## nop_cycles \
-      addi  x6, x3, 0; \
+      addi  x6, x30, 0; \
       addi  x4, x4, 1; \
       li  x5, 2; \
       bne x4, x5, 1b \
     )
 
 #define TEST_RR_SRC12_BYPASS( testnum, src1_nops, src2_nops, inst, result, val1, val2 ) \
-    TEST_CASE( testnum, x3, result, \
+    TEST_CASE( testnum, x30, result, \
       li  x4, 0; \
 1:    li  x1, MASK_XLEN(val1); \
       TEST_INSERT_NOPS_ ## src1_nops \
       li  x2, MASK_XLEN(val2); \
       TEST_INSERT_NOPS_ ## src2_nops \
-      inst x3, x1, x2; \
+      inst x30, x1, x2; \
       addi  x4, x4, 1; \
       li  x5, 2; \
       bne x4, x5, 1b \
     )
 
 #define TEST_RR_SRC21_BYPASS( testnum, src1_nops, src2_nops, inst, result, val1, val2 ) \
-    TEST_CASE( testnum, x3, result, \
+    TEST_CASE( testnum, x30, result, \
       li  x4, 0; \
 1:    li  x2, MASK_XLEN(val2); \
       TEST_INSERT_NOPS_ ## src1_nops \
       li  x1, MASK_XLEN(val1); \
       TEST_INSERT_NOPS_ ## src2_nops \
-      inst x3, x1, x2; \
+      inst x30, x1, x2; \
       addi  x4, x4, 1; \
       li  x5, 2; \
       bne x4, x5, 1b \
@@ -216,17 +216,17 @@ test_ ## testnum: \
 #-----------------------------------------------------------------------
 
 #define TEST_LD_OP( testnum, inst, result, offset, base ) \
-    TEST_CASE( testnum, x3, result, \
+    TEST_CASE( testnum, x30, result, \
       la  x1, base; \
-      inst x3, offset(x1); \
+      inst x30, offset(x1); \
     )
 
 #define TEST_ST_OP( testnum, load_inst, store_inst, result, offset, base ) \
-    TEST_CASE( testnum, x3, result, \
+    TEST_CASE( testnum, x30, result, \
       la  x1, base; \
       li  x2, result; \
       store_inst x2, offset(x1); \
-      load_inst x3, offset(x1); \
+      load_inst x30, offset(x1); \
     )
 
 #define TEST_LD_DEST_BYPASS( testnum, nop_cycles, inst, result, offset, base ) \
@@ -234,9 +234,9 @@ test_ ## testnum: \
     li  TESTNUM, testnum; \
     li  x4, 0; \
 1:  la  x1, base; \
-    inst x3, offset(x1); \
+    inst x30, offset(x1); \
     TEST_INSERT_NOPS_ ## nop_cycles \
-    addi  x6, x3, 0; \
+    addi  x6, x30, 0; \
     li  x29, result; \
     bne x6, x29, fail; \
     addi  x4, x4, 1; \
@@ -249,9 +249,9 @@ test_ ## testnum: \
     li  x4, 0; \
 1:  la  x1, base; \
     TEST_INSERT_NOPS_ ## nop_cycles \
-    inst x3, offset(x1); \
+    inst x30, offset(x1); \
     li  x29, result; \
-    bne x3, x29, fail; \
+    bne x30, x29, fail; \
     addi  x4, x4, 1; \
     li  x5, 2; \
     bne x4, x5, 1b \
@@ -265,9 +265,9 @@ test_ ## testnum: \
     la  x2, base; \
     TEST_INSERT_NOPS_ ## src2_nops \
     store_inst x1, offset(x2); \
-    load_inst x3, offset(x2); \
+    load_inst x30, offset(x2); \
     li  x29, result; \
-    bne x3, x29, fail; \
+    bne x30, x29, fail; \
     addi  x4, x4, 1; \
     li  x5, 2; \
     bne x4, x5, 1b \
@@ -281,9 +281,9 @@ test_ ## testnum: \
     li  x1, result; \
     TEST_INSERT_NOPS_ ## src2_nops \
     store_inst x1, offset(x2); \
-    load_inst x3, offset(x2); \
+    load_inst x30, offset(x2); \
     li  x29, result; \
-    bne x3, x29, fail; \
+    bne x30, x29, fail; \
     addi  x4, x4, 1; \
     li  x5, 2; \
     bne x4, x5, 1b \
diff --git a/isa/rv32mi/Makefrag b/isa/rv32mi/Makefrag
index a726244..5ee7edc 100644
--- a/isa/rv32mi/Makefrag
+++ b/isa/rv32mi/Makefrag
@@ -13,9 +13,6 @@ rv32mi_sc_tests = \
 	sbreak \
 	shamt \
 
-rv32mi_mc_tests = \
-	ipi \
-
 rv32mi_p_tests = $(addprefix rv32mi-p-, $(rv32mi_sc_tests))
 
 spike32_tests += $(rv32mi_p_tests)
diff --git a/isa/rv32mi/ipi.S b/isa/rv32mi/ipi.S
deleted file mode 100644
index c39fc29..0000000
--- a/isa/rv32mi/ipi.S
+++ /dev/null
@@ -1,7 +0,0 @@
-# See LICENSE for license details.
-
-#include "riscv_test.h"
-#undef RVTEST_RV64M
-#define RVTEST_RV64M RVTEST_RV32M
-
-#include "../rv64mi/ipi.S"
diff --git a/isa/rv32si/dirty.S b/isa/rv32si/dirty.S
index 7692b28..9340a85 100644
--- a/isa/rv32si/dirty.S
+++ b/isa/rv32si/dirty.S
@@ -4,7 +4,7 @@
 #undef RVTEST_RV64M
 #define RVTEST_RV64M RVTEST_RV32M
 
-#undef VM_SV39
-#define VM_SV39 VM_SV32
+#undef SPTBR_MODE_SV39
+#define SPTBR_MODE_SV39 SPTBR_MODE_SV32
 
 #include "../rv64si/dirty.S"
diff --git a/isa/rv32uf/Makefrag b/isa/rv32uf/Makefrag
index bc958a7..7dde664 100644
--- a/isa/rv32uf/Makefrag
+++ b/isa/rv32uf/Makefrag
@@ -3,7 +3,7 @@
 #-----------------------------------------------------------------------
 
 rv32uf_sc_tests = \
-	fadd fdiv fclass fcmp fcvt fcvt_w fmadd fmin fsgnj \
+	fadd fdiv fclass fcmp fcvt fcvt_w fmadd fmin \
 	ldst move recoding \
 
 rv32uf_p_tests = $(addprefix rv32uf-p-, $(rv32uf_sc_tests))
diff --git a/isa/rv32uf/fsgnj.S b/isa/rv32uf/fsgnj.S
deleted file mode 100644
index 6d05a23..0000000
--- a/isa/rv32uf/fsgnj.S
+++ /dev/null
@@ -1,7 +0,0 @@
-# See LICENSE for license details.
-
-#include "riscv_test.h"
-#undef RVTEST_RV64UF
-#define RVTEST_RV64UF RVTEST_RV32UF
-
-#include "../rv64uf/fsgnj.S"
diff --git a/isa/rv64mi/Makefrag b/isa/rv64mi/Makefrag
index e4fa426..fb38e5f 100644
--- a/isa/rv64mi/Makefrag
+++ b/isa/rv64mi/Makefrag
@@ -12,9 +12,6 @@ rv64mi_sc_tests = \
 	scall \
 	sbreak \
 
-rv64mi_mc_tests = \
-	ipi \
-
 rv64mi_p_tests = $(addprefix rv64mi-p-, $(rv64mi_sc_tests))
 
 spike_tests += $(rv64mi_p_tests)
diff --git a/isa/rv64mi/illegal.S b/isa/rv64mi/illegal.S
index 30c22b2..a1b445f 100644
--- a/isa/rv64mi/illegal.S
+++ b/isa/rv64mi/illegal.S
@@ -13,23 +13,165 @@
 RVTEST_RV64M
 RVTEST_CODE_BEGIN
 
+  .align 2
+  .option norvc
+
   li TESTNUM, 2
+bad2:
   .word 0
   j fail
 
-  j pass
+  # Skip the rest of the test if S-mode is not present.
+  li t0, MSTATUS_MPP
+  csrc mstatus, t0
+  li t1, (MSTATUS_MPP & -MSTATUS_MPP) * PRV_S
+  csrs mstatus, t1
+  csrr t2, mstatus
+  and t2, t2, t0
+  bne t1, t2, pass
+
+  # Test vectored interrupts if they are supported.
+test_vectored_interrupts:
+  csrwi mip, MIP_SSIP
+  csrwi mie, MIP_SSIP
+  la t0, mtvec_handler + 1
+  csrrw s0, mtvec, t0
+  csrr t0, mtvec
+  andi t0, t0, 1
+  beqz t0, msip
+  csrsi mstatus, MSTATUS_MIE
+1:
+  j 1b
+msip:
+  csrw mtvec, s0
+
+  # Delegate supervisor software interrupts so WFI won't stall.
+  csrwi mideleg, MIP_SSIP
+  # Enter supervisor mode.
+  la t0, 1f
+  csrw mepc, t0
+  li t0, MSTATUS_MPP
+  csrc mstatus, t0
+  li t1, (MSTATUS_MPP & -MSTATUS_MPP) * PRV_S
+  csrs mstatus, t1
+  mret
+
+1:
+  # Make sure WFI doesn't trap when TW=0.
+  wfi
+bad3:
+  .word 0
+  j fail
+
+bad4:
+  # Make sure WFI does trap when TW=1.
+  wfi
+  j fail
+
+  # Make sure SFENCE.VMA and sptbr don't trap when TVM=0.
+  sfence.vma
+  csrr t0, sptbr
+bad5:
+  .word 0
+  j fail
+
+bad6:
+  # Make sure SFENCE.VMA and sptbr do trap when TVM=1.
+  sfence.vma
+  j fail
+bad7:
+  csrr t0, sptbr
+  j fail
+
+  # Make sure SRET doesn't trap when TSR=0.
+  la t0, bad8
+  csrw sepc, t0
+  li t0, SSTATUS_SPP
+  csrs sstatus, t0
+  li t0, SSTATUS_SPIE
+  csrc sstatus, t0
+  sret
+bad8:
+  .word 0
+  j fail
+
+  # Make sure SRET does trap when TSR=1.
+  la t0, 1f
+  csrw sepc, t0
+bad9:
+  sret
+1:
+  j fail
 
   TEST_PASSFAIL
 
+  .align 8
 mtvec_handler:
+  j synchronous_exception
+  j msip
+  j fail
+  j fail
+  j fail
+  j fail
+  j fail
+  j fail
+  j fail
+  j fail
+  j fail
+  j fail
+  j fail
+  j fail
+  j fail
+  j fail
+
+synchronous_exception:
   li t1, CAUSE_ILLEGAL_INSTRUCTION
   csrr t0, mcause
   bne t0, t1, fail
   csrr t0, mepc
+  la t1, bad2
+  beq t0, t1, 2f
+  la t1, bad3
+  beq t0, t1, 3f
+  la t1, bad4
+  beq t0, t1, 4f
+  la t1, bad5
+  beq t0, t1, 5f
+  la t1, bad6
+  beq t0, t1, 6f
+  la t1, bad7
+  beq t0, t1, 7f
+  la t1, bad8
+  beq t0, t1, 8f
+  la t1, bad9
+  beq t0, t1, 9f
+  j fail
+2:
+4:
+6:
+7:
   addi t0, t0, 8
   csrw mepc, t0
   mret
 
+3:
+  li t1, MSTATUS_TW
+  csrs mstatus, t1
+  j 2b
+
+5:
+  li t1, MSTATUS_TVM
+  csrs mstatus, t1
+  j 2b
+
+8:
+  li t1, MSTATUS_TSR
+  csrs mstatus, t1
+  j 2b
+
+9:
+  j 2b
+
 RVTEST_CODE_END
 
   .data
diff --git a/isa/rv64mi/ipi.S b/isa/rv64mi/ipi.S
deleted file mode 100644
index 7178310..0000000
--- a/isa/rv64mi/ipi.S
+++ /dev/null
@@ -1,49 +0,0 @@
-# See LICENSE for license details.
-
-#*****************************************************************************
-# ipi.S
-#-----------------------------------------------------------------------------
-#
-# Test interprocessor interrupts.
-#
-
-#include "riscv_test.h"
-#include "test_macros.h"
-
-RVTEST_RV64M
-RVTEST_CODE_BEGIN
-
-  # enable interrupts
-  csrs mstatus, MSTATUS_MIE
-  csrs mie, MIP_MSIP
-
-  # get a unique core id
-  la a0, coreid
-  li a1, 1
-  amoadd.w a2, a1, (a0)
-  
-  # for now, only run this on core 0
-  1:li a3, 1
-  bgeu a2, a3, 1b
-  
-  # send a self-IPI
-  csrwi mipi, 1
-  1: j 1b
-
-mtvec_handler:
-  bnez a2, fail
-  RVTEST_PASS
-
-  TEST_PASSFAIL
-
-RVTEST_CODE_END
-
-  .data
-RVTEST_DATA_BEGIN
-
-  TEST_DATA
-
-coreid: .word 0
-foo: .word 0
-
-RVTEST_DATA_END
diff --git a/isa/rv64mi/ma_addr.S b/isa/rv64mi/ma_addr.S
index be3572f..1e5ab68 100644
--- a/isa/rv64mi/ma_addr.S
+++ b/isa/rv64mi/ma_addr.S
@@ -20,47 +20,73 @@ RVTEST_CODE_BEGIN
   # indicate it's a load test
   li s1, CAUSE_MISALIGNED_LOAD
 
-#define MISALIGNED_LDST_TEST(testnum, insn, base, offset) \
-  li TESTNUM, testnum; \
-  insn x0, offset(base); \
-  j fail \
+#define SEXT(x, n) ((-((x) >> ((n)-1)) << (n)) | ((x) & ((1 << (n))-1)))
 
-  MISALIGNED_LDST_TEST(2,  lh,  s0, 1)
-  MISALIGNED_LDST_TEST(3,  lhu, s0, 1)
-  MISALIGNED_LDST_TEST(4,  lw,  s0, 1)
-  MISALIGNED_LDST_TEST(5,  lw,  s0, 2)
-  MISALIGNED_LDST_TEST(6,  lw,  s0, 3)
+/* Check that a misaligned load either writes the correct value, or
+   takes an exception and performs no writeback.  */
+#define MISALIGNED_LOAD_TEST(testnum, insn, base, offset, res) \
+  li TESTNUM, testnum; \
+  la t2, 1f; \
+  addi t1, base, offset; \
+  insn t1, offset(base); \
+  li t2, res; \
+  bne t1, t2, fail; \
+1:
+
+  MISALIGNED_LOAD_TEST(2,  lh,  s0, 1, SEXT(0xbbcc, 16))
+  MISALIGNED_LOAD_TEST(3,  lhu, s0, 1, 0xbbcc)
+  MISALIGNED_LOAD_TEST(4,  lw,  s0, 1, SEXT(0x99aabbcc, 32))
+  MISALIGNED_LOAD_TEST(5,  lw,  s0, 2, SEXT(0x8899aabb, 32))
+  MISALIGNED_LOAD_TEST(6,  lw,  s0, 3, SEXT(0x778899aa, 32))
 
 #if __riscv_xlen == 64
-  MISALIGNED_LDST_TEST(7,  lwu, s0, 1)
-  MISALIGNED_LDST_TEST(8,  lwu, s0, 2)
-  MISALIGNED_LDST_TEST(9,  lwu, s0, 3)
-
-  MISALIGNED_LDST_TEST(10, ld, s0, 1)
-  MISALIGNED_LDST_TEST(11, ld, s0, 2)
-  MISALIGNED_LDST_TEST(12, ld, s0, 3)
-  MISALIGNED_LDST_TEST(13, ld, s0, 4)
-  MISALIGNED_LDST_TEST(14, ld, s0, 5)
-  MISALIGNED_LDST_TEST(15, ld, s0, 6)
-  MISALIGNED_LDST_TEST(16, ld, s0, 7)
+  MISALIGNED_LOAD_TEST(7,  lwu, s0, 1, 0x99aabbcc)
+  MISALIGNED_LOAD_TEST(8,  lwu, s0, 2, 0x8899aabb)
+  MISALIGNED_LOAD_TEST(9,  lwu, s0, 3, 0x778899aa)
+
+  MISALIGNED_LOAD_TEST(10, ld, s0, 1, 0x5566778899aabbcc)
+  MISALIGNED_LOAD_TEST(11, ld, s0, 2, 0x445566778899aabb)
+  MISALIGNED_LOAD_TEST(12, ld, s0, 3, 0x33445566778899aa)
+  MISALIGNED_LOAD_TEST(13, ld, s0, 4, 0x2233445566778899)
+  MISALIGNED_LOAD_TEST(14, ld, s0, 5, 0x1122334455667788)
+  MISALIGNED_LOAD_TEST(15, ld, s0, 6, 0xee11223344556677)
+  MISALIGNED_LOAD_TEST(16, ld, s0, 7, 0xffee112233445566)
 #endif
 
   # indicate it's a store test
   li s1, CAUSE_MISALIGNED_STORE
 
-  MISALIGNED_LDST_TEST(22,  sh,  s0, 1)
-  MISALIGNED_LDST_TEST(23,  sw,  s0, 1)
-  MISALIGNED_LDST_TEST(24,  sw,  s0, 2)
-  MISALIGNED_LDST_TEST(25,  sw,  s0, 3)
+/* Check that a misaligned store has some effect and takes no exception,
+   or takes no effect and generates an exception.  This is not very
+   thorough.  */
+#define MISALIGNED_STORE_TEST(testnum, insn, base, offset, size) \
+  li TESTNUM, testnum; \
+  la t2, 1f; \
+  addi t1, base, offset; \
+  insn x0, offset(base); \
+  lb t1, (offset - 1)(base); \
+  beqz t1, fail; \
+  lb t1, (offset + size)(base); \
+  beqz t1, fail; \
+  lb t1, (offset + 0)(base); \
+  bnez t1, fail; \
+  lb t1, (offset + size - 1)(base); \
+  bnez t1, fail; \
+1:
+
+  MISALIGNED_STORE_TEST(22,  sh,  s0, 1, 2)
+  MISALIGNED_STORE_TEST(23,  sw,  s0, 5, 4)
+  MISALIGNED_STORE_TEST(24,  sw,  s0, 10, 4)
+  MISALIGNED_STORE_TEST(25,  sw,  s0, 15, 4)
 
 #if __riscv_xlen == 64
-  MISALIGNED_LDST_TEST(26, sd, s0, 1)
-  MISALIGNED_LDST_TEST(27, sd, s0, 2)
-  MISALIGNED_LDST_TEST(28, sd, s0, 3)
-  MISALIGNED_LDST_TEST(29, sd, s0, 4)
-  MISALIGNED_LDST_TEST(30, sd, s0, 5)
-  MISALIGNED_LDST_TEST(31, sd, s0, 6)
-  MISALIGNED_LDST_TEST(32, sd, s0, 7)
+  MISALIGNED_STORE_TEST(26, sd, s0, 25, 8)
+  MISALIGNED_STORE_TEST(27, sd, s0, 34, 8)
+  MISALIGNED_STORE_TEST(28, sd, s0, 43, 8)
+  MISALIGNED_STORE_TEST(29, sd, s0, 52, 8)
+  MISALIGNED_STORE_TEST(30, sd, s0, 61, 8)
+  MISALIGNED_STORE_TEST(31, sd, s0, 70, 8)
+  MISALIGNED_STORE_TEST(32, sd, s0, 79, 8)
 #endif
 
   TEST_PASSFAIL
@@ -70,9 +96,13 @@ mtvec_handler:
   csrr t0, mcause
   bne t0, s1, fail
 
-  csrr t0, mepc
-  addi t0, t0, 8
-  csrw mepc, t0
+  csrr t0, mbadaddr
+  bne t0, t1, fail
+
+  lb t0, (t0)
+  beqz t0, fail
+
+  csrw mepc, t2
   mret
 
 RVTEST_CODE_END
@@ -81,7 +111,13 @@ RVTEST_CODE_END
 RVTEST_DATA_BEGIN
 
 data:
-  .dword 0
+  .align 3
+.word 0xaabbccdd
+.word 0x66778899
+.word 0x22334455
+.word 0xeeffee11
+.fill 0xff, 1, 80
+
 
   TEST_DATA
 
diff --git a/isa/rv64si/dirty.S b/isa/rv64si/dirty.S
index 0314cf5..17aa57f 100644
--- a/isa/rv64si/dirty.S
+++ b/isa/rv64si/dirty.S
@@ -13,37 +13,45 @@
 RVTEST_RV64M
 RVTEST_CODE_BEGIN
 
-  # Turn on VM with superpage identity mapping
+  # Turn on VM
+  li a0, (SPTBR_MODE & ~(SPTBR_MODE<<1)) * SPTBR_MODE_SV39
   la a1, page_table_1
   srl a1, a1, RISCV_PGSHIFT
-  la a2, page_table_2
-  srl a2, a2, RISCV_PGSHIFT
+  or a1, a1, a0
   csrw sptbr, a1
-  sfence.vm
-  li a1, ((MSTATUS_VM & ~(MSTATUS_VM<<1)) * VM_SV39) | ((MSTATUS_MPP & ~(MSTATUS_MPP<<1)) * PRV_S)
+  sfence.vma
+
+  # Set up MPRV with MPP=S, so loads and stores use S-mode
+  li a1, ((MSTATUS_MPP & ~(MSTATUS_MPP<<1)) * PRV_S) | MSTATUS_MPRV
   csrs mstatus, a1
-  la a1, 1f - DRAM_BASE
-  csrw mepc, a1
-  la a1, stvec_handler - DRAM_BASE
-  csrw stvec, a1
-  mret
-1:
 
   # Try a faulting store to make sure dirty bit is not set
   li TESTNUM, 2
-  li t0, 1
-  sw t0, dummy, t1
+  li t2, 1
+  sw t2, dummy - DRAM_BASE, t1
 
-  # Load new page table
+  # Set SUM=1 so user memory access is permitted
   li TESTNUM, 3
-  csrw sptbr, a2
-  sfence.vm
+  li a1, ((MSTATUS_MPP & ~(MSTATUS_MPP<<1)) * PRV_S) | MSTATUS_SUM
+  csrs mstatus, a1
+
+  # Make sure SUM=1 works
+  lw t0, dummy - DRAM_BASE
+  bnez t0, die
 
   # Try a non-faulting store to make sure dirty bit is set
-  sw t0, dummy, t1
+  sw t2, dummy - DRAM_BASE, t1
+
+  # Make sure it succeeded
+  lw t0, dummy - DRAM_BASE
+  bne t0, t2, die
 
-  # Make sure R and D bits are set
-  lw t0, page_table_2
+  # Leave MPRV
+  li t0, MSTATUS_MPRV
+  csrc mstatus, t0
+
+  # Make sure D bit is set
+  lw t0, page_table_1
   li t1, PTE_A | PTE_D
   and t0, t0, t1
   bne t0, t1, die
@@ -53,27 +61,34 @@ RVTEST_CODE_BEGIN
   TEST_PASSFAIL
 
   .align 2
-stvec_handler:
-  csrr t0, scause
+mtvec_handler:
+  csrr t0, mcause
+  add t0, t0, -CAUSE_STORE_PAGE_FAULT
+  bnez t0, die
+
   li t1, 2
   bne TESTNUM, t1, 1f
-  # Make sure R bit is set
-  lw t0, page_table_1
-  li t1, PTE_A
-  and t0, t0, t1
-  bne t0, t1, die
-
   # Make sure D bit is clear
   lw t0, page_table_1
-  li t1, PTE_D
-  and t0, t0, t1
-  beq t0, t1, die
-
-  csrr t0, sepc
+  and t1, t0, PTE_D
+  bnez t1, die
+skip:
+  csrr t0, mepc
   add t0, t0, 4
-  csrw sepc, t0
-  sret
+  csrw mepc, t0
+  mret
 
+1:
+  li t1, 3
+  bne TESTNUM, t1, 1f
+  # The implementation doesn't appear to set D bits in HW.  Skip the test,
+  # after making sure the D bit is clear.
+  lw t0, page_table_1
+  and t1, t0, PTE_D
+  bnez t1, die
+  j pass
+
+1:
 die:
   RVTEST_FAIL
 
@@ -85,9 +100,7 @@ RVTEST_DATA_BEGIN
   TEST_DATA
 
 .align 12
-page_table_1: .dword (DRAM_BASE/RISCV_PGSIZE << PTE_PPN_SHIFT) | PTE_V | PTE_U | PTE_R | PTE_X
+page_table_1: .dword (DRAM_BASE/RISCV_PGSIZE << PTE_PPN_SHIFT) | PTE_V | PTE_U | PTE_R | PTE_W | PTE_X | PTE_A
 dummy: .dword 0
-.align 12
-page_table_2: .dword (DRAM_BASE/RISCV_PGSIZE << PTE_PPN_SHIFT) | PTE_V | PTE_U | PTE_R | PTE_X | PTE_W
 
 RVTEST_DATA_END
diff --git a/isa/rv64ud/Makefrag b/isa/rv64ud/Makefrag
index 6e8be9c..9cffb5d 100644
--- a/isa/rv64ud/Makefrag
+++ b/isa/rv64ud/Makefrag
@@ -3,7 +3,7 @@
 #-----------------------------------------------------------------------
 
 rv64ud_sc_tests = \
-	fadd fdiv fclass fcmp fcvt fcvt_w fmadd fmin fsgnj \
+	fadd fdiv fclass fcmp fcvt fcvt_w fmadd fmin \
 	ldst move structural recoding \
 
 rv64ud_p_tests = $(addprefix rv64ud-p-, $(rv64ud_sc_tests))
diff --git a/isa/rv64ud/fsgnj.S b/isa/rv64ud/fsgnj.S
deleted file mode 100644
index e914777..0000000
--- a/isa/rv64ud/fsgnj.S
+++ /dev/null
@@ -1,44 +0,0 @@
-# See LICENSE for license details.
-
-#*****************************************************************************
-# fsgnj.S
-#-----------------------------------------------------------------------------
-#
-# Test fsgn{j|jn|x}.d instructions.
-#
-
-#include "riscv_test.h"
-#include "test_macros.h"
-
-RVTEST_RV64UF
-RVTEST_CODE_BEGIN
-
-  #-------------------------------------------------------------
-  # Arithmetic tests
-  #-------------------------------------------------------------
-
-  TEST_FP_OP2_D( 2,  fsgnj.d, 0, -6.3,  6.3, -1.0 );
-  TEST_FP_OP2_D( 3,  fsgnj.d, 0,  7.3,  7.3,  2.0 );
-  TEST_FP_OP2_D( 4,  fsgnj.d, 0, -8.3, -8.3, -3.0 );
-  TEST_FP_OP2_D( 5,  fsgnj.d, 0,  9.3, -9.3,  4.0 );
-
-  TEST_FP_OP2_D(12, fsgnjn.d, 0,  6.3,  6.3, -1.0 );
-  TEST_FP_OP2_D(13, fsgnjn.d, 0, -7.3,  7.3,  2.0 );
-  TEST_FP_OP2_D(14, fsgnjn.d, 0,  8.3, -8.3, -3.0 );
-  TEST_FP_OP2_D(15, fsgnjn.d, 0, -9.3, -9.3,  4.0 );
-
-  TEST_FP_OP2_D(22, fsgnjx.d, 0, -6.3,  6.3, -1.0 );
-  TEST_FP_OP2_D(23, fsgnjx.d, 0,  7.3,  7.3,  2.0 );
-  TEST_FP_OP2_D(24, fsgnjx.d, 0,  8.3, -8.3, -3.0 );
-  TEST_FP_OP2_D(25, fsgnjx.d, 0, -9.3, -9.3,  4.0 );
-
-  TEST_PASSFAIL
-
-RVTEST_CODE_END
-
-  .data
-RVTEST_DATA_BEGIN
-
-  TEST_DATA
-
-RVTEST_DATA_END
diff --git a/isa/rv64ud/ldst.S b/isa/rv64ud/ldst.S
index 59084e3..9629341 100644
--- a/isa/rv64ud/ldst.S
+++ b/isa/rv64ud/ldst.S
@@ -13,8 +13,12 @@
 RVTEST_RV64UF
 RVTEST_CODE_BEGIN
 
-  TEST_CASE(2, a0, 0x40000000bf800000, la a1, tdat; fld f2, 0(a1); fsd f2, 16(a1); ld a0, 16(a1))
-  TEST_CASE(3, a0, 0xc080000040400000, la a1, tdat; fld f2, 8(a1); fsd f2, 16(a1); ld a0, 16(a1))
+  la s0, tdat
+  TEST_CASE(2, a0, 0x40000000bf800000, fld f2, 0(s0); fsd f2, 16(s0); ld a0, 16(s0))
+  TEST_CASE(3, a0, 0x40000000bf800000, fld f2, 0(s0); fsw f2, 16(s0); ld a0, 16(s0))
+  TEST_CASE(4, a0, 0x40000000bf800000, flw f2, 0(s0); fsw f2, 16(s0); ld a0, 16(s0))
+  TEST_CASE(5, a0, 0xc080000040400000, fld f2, 8(s0); fsd f2, 16(s0); ld a0, 16(s0))
+  TEST_CASE(6, a0, 0xffffffff40400000, flw f2, 8(s0); fsd f2, 16(s0); ld a0, 16(s0))
 
   TEST_PASSFAIL
 
diff --git a/isa/rv64ud/move.S b/isa/rv64ud/move.S
index 806d4de..ccc41b5 100644
--- a/isa/rv64ud/move.S
+++ b/isa/rv64ud/move.S
@@ -4,8 +4,7 @@
 # move.S
 #-----------------------------------------------------------------------------
 #
-# This test verifies that mxtf.[s,d], mftx.[s,d], fssr, frsr,
-# and fsgnj[x|n].d work properly.
+# This test verifies that fmv.d.x, fmv.x.d, and fsgnj[x|n].d work properly.
 #
 
 #include "riscv_test.h"
@@ -14,15 +13,92 @@
 RVTEST_RV64UF
 RVTEST_CODE_BEGIN
 
-li a0, 1
-fssr a0
+#define TEST_FSGNJD(n, insn, new_sign, rs1_sign, rs2_sign) \
+  TEST_CASE(n, a0, 0x123456789abcdef0 | (-(new_sign) << 63), \
+    li a1, ((rs1_sign) << 63) | 0x123456789abcdef0; \
+    li a2, -(rs2_sign); \
+    fmv.d.x f1, a1; \
+    fmv.d.x f2, a2; \
+    insn f0, f1, f2; \
+    fmv.x.d a0, f0)
 
-  TEST_CASE(2, a1, 1, li a0, 0x1234; fssr a1, a0)
-  TEST_CASE(3, a0, 0x34, frsr a0)
-  TEST_CASE(4, a0, 0x34, frsr a0)
+  TEST_FSGNJD(10, fsgnj.d, 0, 0, 0)
+  TEST_FSGNJD(11, fsgnj.d, 1, 0, 1)
+  TEST_FSGNJD(12, fsgnj.d, 0, 1, 0)
+  TEST_FSGNJD(13, fsgnj.d, 1, 1, 1)
 
-  TEST_CASE(5, a0, 0x3FF02468A0000000, li a1, 0x3FF02468A0000000; fmv.d.x f1, a1; fmv.x.d a0, f1)
-  TEST_CASE(6, a0, 0xBFF02468A0001000, li a1, 0x3FF02468A0001000; li a2, -1; fmv.d.x f1, a1; fmv.d.x f2, a2; fsgnj.d f0, f1, f2; fmv.x.d a0, f0)
+  TEST_FSGNJD(20, fsgnjn.d, 1, 0, 0)
+  TEST_FSGNJD(21, fsgnjn.d, 0, 0, 1)
+  TEST_FSGNJD(22, fsgnjn.d, 1, 1, 0)
+  TEST_FSGNJD(23, fsgnjn.d, 0, 1, 1)
+
+  TEST_FSGNJD(30, fsgnjx.d, 0, 0, 0)
+  TEST_FSGNJD(31, fsgnjx.d, 1, 0, 1)
+  TEST_FSGNJD(32, fsgnjx.d, 1, 1, 0)
+  TEST_FSGNJD(33, fsgnjx.d, 0, 1, 1)
+
+// Test fsgnj.s in conjunction with double-precision moves
+#define TEST_FSGNJS(n, rd, rs1, rs2) \
+  TEST_CASE(n, a0, (rd) | (-((rd) >> 31) << 32), \
+    li a1, rs1; \
+    li a2, rs2; \
+    fmv.d.x f1, a1; \
+    fmv.d.x f2, a2; \
+    fsgnj.s f0, f1, f2; \
+    fmv.x.s a0, f0); \
+  TEST_CASE(1##n, a0, (rd) | 0xffffffff00000000, \
+    li a1, rs1; \
+    li a2, rs2; \
+    fmv.d.x f1, a1; \
+    fmv.d.x f2, a2; \
+    fsgnj.s f0, f1, f2; \
+    fmv.x.d a0, f0)
+
+  TEST_FSGNJS(40, 0x7fc00000, 0x7ffffffe12345678, 0)
+  TEST_FSGNJS(41, 0x7fc00000, 0xfffffffe12345678, 0)
+  TEST_FSGNJS(42, 0x7fc00000, 0x7fffffff12345678, 0)
+  TEST_FSGNJS(43, 0x12345678, 0xffffffff12345678, 0)
+
+  TEST_FSGNJS(50, 0x7fc00000, 0x7ffffffe12345678, 0x80000000)
+  TEST_FSGNJS(51, 0x7fc00000, 0xfffffffe12345678, 0x80000000)
+  TEST_FSGNJS(52, 0x7fc00000, 0x7fffffff12345678, 0x80000000)
+  TEST_FSGNJS(53, 0x12345678, 0xffffffff12345678, 0x80000000)
+
+  TEST_FSGNJS(60, 0xffc00000, 0x7ffffffe12345678, 0xffffffff80000000)
+  TEST_FSGNJS(61, 0xffc00000, 0xfffffffe12345678, 0xffffffff80000000)
+  TEST_FSGNJS(62, 0x92345678, 0xffffffff12345678, 0xffffffff80000000)
+  TEST_FSGNJS(63, 0x12345678, 0xffffffff12345678, 0x7fffffff80000000)
+
+// Test fsgnj.d in conjunction with single-precision moves
+#define TEST_FSGNJD_SP(n, isnan, rd, rs1, rs2) \
+  TEST_CASE(n, a0, ((rd) & 0xffffffff) | (-(((rd) >> 31) & 1) << 32), \
+    li a1, rs1; \
+    li a2, rs2; \
+    fmv.d.x f1, a1; \
+    fmv.d.x f2, a2; \
+    fsgnj.d f0, f1, f2; \
+    feq.s a0, f0, f0; \
+    addi a0, a0, -!(isnan); \
+    bnez a0, 1f; \
+    fmv.x.s a0, f0; \
+    1:); \
+  TEST_CASE(1##n, a0, rd, \
+    li a1, rs1; \
+    li a2, rs2; \
+    fmv.d.x f1, a1; \
+    fmv.d.x f2, a2; \
+    fsgnj.d f0, f1, f2; \
+    fmv.x.d a0, f0; \
+    1:)
+
+  TEST_FSGNJD_SP(70, 0, 0xffffffff11111111, 0xffffffff11111111, 0xffffffff11111111)
+  TEST_FSGNJD_SP(71, 1, 0x7fffffff11111111, 0xffffffff11111111, 0x7fffffff11111111)
+  TEST_FSGNJD_SP(72, 0, 0xffffffff11111111, 0xffffffff11111111, 0xffffffff91111111)
+  TEST_FSGNJD_SP(73, 0, 0xffffffff11111111, 0xffffffff11111111, 0x8000000000000000)
+  TEST_FSGNJD_SP(74, 0, 0xffffffff11111111, 0x7fffffff11111111, 0xffffffff11111111)
+  TEST_FSGNJD_SP(75, 1, 0x7fffffff11111111, 0x7fffffff11111111, 0x7fffffff11111111)
+  TEST_FSGNJD_SP(76, 0, 0xffffffff11111111, 0x7fffffff11111111, 0xffffffff91111111)
+  TEST_FSGNJD_SP(77, 0, 0xffffffff11111111, 0x7fffffff11111111, 0x8000000000000000)
 
   TEST_PASSFAIL
 
diff --git a/isa/rv64ud/structural.S b/isa/rv64ud/structural.S
index 76c6691..5ecbb96 100644
--- a/isa/rv64ud/structural.S
+++ b/isa/rv64ud/structural.S
@@ -30,8 +30,8 @@ li x1, 0x3F800000
   nops          ;\
   fsgnj.s f3, f1, f1 ;\
   fmv.x.d  x4, f4    ;\
-  fmv.x.s  x3, f3    ;\
-  beq     x1, x3, 2f  ;\
+  fmv.x.s  x5, f3    ;\
+  beq     x1, x5, 2f  ;\
   RVTEST_FAIL ;\
 2:beq     x2, x4, 2f  ;\
   RVTEST_FAIL; \
diff --git a/isa/rv64uf/Makefrag b/isa/rv64uf/Makefrag
index d3c3f23..33c11db 100644
--- a/isa/rv64uf/Makefrag
+++ b/isa/rv64uf/Makefrag
@@ -3,7 +3,7 @@
 #-----------------------------------------------------------------------
 
 rv64uf_sc_tests = \
-	fadd fdiv fclass fcmp fcvt fcvt_w fmadd fmin fsgnj \
+	fadd fdiv fclass fcmp fcvt fcvt_w fmadd fmin \
 	ldst move recoding \
 
 rv64uf_p_tests = $(addprefix rv64uf-p-, $(rv64uf_sc_tests))
diff --git a/isa/rv64uf/fsgnj.S b/isa/rv64uf/fsgnj.S
deleted file mode 100644
index 6d4bdb4..0000000
--- a/isa/rv64uf/fsgnj.S
+++ /dev/null
@@ -1,44 +0,0 @@
-# See LICENSE for license details.
-
-#*****************************************************************************
-# fsgnj.S
-#-----------------------------------------------------------------------------
-#
-# Test fsgn{j|jn|x}.s instructions.
-#
-
-#include "riscv_test.h"
-#include "test_macros.h"
-
-RVTEST_RV64UF
-RVTEST_CODE_BEGIN
-
-  #-------------------------------------------------------------
-  # Arithmetic tests
-  #-------------------------------------------------------------
-
-  TEST_FP_OP2_S( 2,  fsgnj.s, 0, -6.3,  6.3, -1.0 );
-  TEST_FP_OP2_S( 3,  fsgnj.s, 0,  7.3,  7.3,  2.0 );
-  TEST_FP_OP2_S( 4,  fsgnj.s, 0, -8.3, -8.3, -3.0 );
-  TEST_FP_OP2_S( 5,  fsgnj.s, 0,  9.3, -9.3,  4.0 );
-
-  TEST_FP_OP2_S(12, fsgnjn.s, 0,  6.3,  6.3, -1.0 );
-  TEST_FP_OP2_S(13, fsgnjn.s, 0, -7.3,  7.3,  2.0 );
-  TEST_FP_OP2_S(14, fsgnjn.s, 0,  8.3, -8.3, -3.0 );
-  TEST_FP_OP2_S(15, fsgnjn.s, 0, -9.3, -9.3,  4.0 );
-
-  TEST_FP_OP2_S(22, fsgnjx.s, 0, -6.3,  6.3, -1.0 );
-  TEST_FP_OP2_S(23, fsgnjx.s, 0,  7.3,  7.3,  2.0 );
-  TEST_FP_OP2_S(24, fsgnjx.s, 0,  8.3, -8.3, -3.0 );
-  TEST_FP_OP2_S(25, fsgnjx.s, 0, -9.3, -9.3,  4.0 );
-
-  TEST_PASSFAIL
-
-RVTEST_CODE_END
-
-  .data
-RVTEST_DATA_BEGIN
-
-  TEST_DATA
-
-RVTEST_DATA_END
diff --git a/isa/rv64uf/move.S b/isa/rv64uf/move.S
index a94af55..60f7cf3 100644
--- a/isa/rv64uf/move.S
+++ b/isa/rv64uf/move.S
@@ -4,8 +4,8 @@
 # move.S
 #-----------------------------------------------------------------------------
 #
-# This test verifies that mxtf.[s,d], mftx.[s,d], fssr, frsr,
-# and fsgnj[x|n].s work properly.
+# This test verifies that the fmv.s.x, fmv.x.s, and fsgnj[x|n].d instructions
+# and the fcsr work properly.
 #
 
 #include "riscv_test.h"
@@ -14,18 +14,37 @@
 RVTEST_RV64UF
 RVTEST_CODE_BEGIN
 
-li a0, 1
-fssr a0
-
-  TEST_CASE(2, a1, 1, li a0, 0x1234; fssr a1, a0)
+  TEST_CASE(2, a1, 1, csrwi fcsr, 1; li a0, 0x1234; fssr a1, a0)
   TEST_CASE(3, a0, 0x34, frsr a0)
-  TEST_CASE(4, a0, 0x34, frsr a0)
-
-  TEST_CASE(5, a0, 0xFFFFFFFFBF812345, li a1, 0xFFFFFFFFBF812345; fmv.s.x f0, a1; fmv.x.s a0, f0)
-
-  TEST_CASE(6, a0, 0xFFFFFFFFBF812345, li a1, 0xFFFFFFFFBF812345; fmv.s.x f0, a1; fsgnj.s f1, f0, f0; fmv.x.s a0, f1)
-  TEST_CASE(7, a0, 0x000000004BA98765, li a1, 0xFFFFFFFFCBA98765; fmv.s.x f0, a1; fsgnjx.s f1, f0, f0; fmv.x.s a0, f1)
-  TEST_CASE(8, a0, 0x000000005EADBEEF, li a1, 0xFFFFFFFFDEADBEEF; fmv.s.x f0, a1; fsgnjn.s f1, f0, f0; fmv.x.s a0, f1)
+  TEST_CASE(4, a0, 0x14, frflags a0)
+  TEST_CASE(5, a0, 0x01, csrrwi a0, frm, 2)
+  TEST_CASE(6, a0, 0x54, frsr a0)
+  TEST_CASE(7, a0, 0x14, csrrci a0, fflags, 4)
+  TEST_CASE(8, a0, 0x50, frsr a0)
+
+#define TEST_FSGNJS(n, insn, new_sign, rs1_sign, rs2_sign) \
+  TEST_CASE(n, a0, 0x12345678 | (-(new_sign) << 31), \
+    li a1, ((rs1_sign) << 31) | 0x12345678; \
+    li a2, -(rs2_sign); \
+    fmv.s.x f1, a1; \
+    fmv.s.x f2, a2; \
+    insn f0, f1, f2; \
+    fmv.x.s a0, f0)
+
+  TEST_FSGNJS(10, fsgnj.s, 0, 0, 0)
+  TEST_FSGNJS(11, fsgnj.s, 1, 0, 1)
+  TEST_FSGNJS(12, fsgnj.s, 0, 1, 0)
+  TEST_FSGNJS(13, fsgnj.s, 1, 1, 1)
+
+  TEST_FSGNJS(20, fsgnjn.s, 1, 0, 0)
+  TEST_FSGNJS(21, fsgnjn.s, 0, 0, 1)
+  TEST_FSGNJS(22, fsgnjn.s, 1, 1, 0)
+  TEST_FSGNJS(23, fsgnjn.s, 0, 1, 1)
+
+  TEST_FSGNJS(30, fsgnjx.s, 0, 0, 0)
+  TEST_FSGNJS(31, fsgnjx.s, 1, 0, 1)
+  TEST_FSGNJS(32, fsgnjx.s, 1, 1, 0)
+  TEST_FSGNJS(33, fsgnjx.s, 0, 1, 1)
 
   TEST_PASSFAIL
 
diff --git a/isa/rv64ui/jal.S b/isa/rv64ui/jal.S
index f7f299d..00c65d8 100644
--- a/isa/rv64ui/jal.S
+++ b/isa/rv64ui/jal.S
@@ -21,7 +21,7 @@ test_2:
   li  TESTNUM, 2
   li  ra, 0
 
-  jal x3, target_2
+  jal x4, target_2
 linkaddr_2:
   nop
   nop
@@ -30,7 +30,7 @@ linkaddr_2:
 
 target_2:
   la  x2, linkaddr_2
-  bne x2, x3, fail
+  bne x2, x4, fail
 
   #-------------------------------------------------------------
   # Test delay slot instructions not executed nor bypassed
diff --git a/isa/rv64ui/lb.S b/isa/rv64ui/lb.S
index 277b03e..856dfe9 100644
--- a/isa/rv64ui/lb.S
+++ b/isa/rv64ui/lb.S
@@ -31,18 +31,18 @@ RVTEST_CODE_BEGIN
 
   # Test with a negative base
 
-  TEST_CASE( 10, x3, 0xffffffffffffffff, \
+  TEST_CASE( 10, x5, 0xffffffffffffffff, \
     la  x1, tdat; \
     addi x1, x1, -32; \
-    lb x3, 32(x1); \
+    lb x5, 32(x1); \
   )
 
   # Test with unaligned base
 
-  TEST_CASE( 11, x3, 0x0000000000000000, \
+  TEST_CASE( 11, x5, 0x0000000000000000, \
     la  x1, tdat; \
     addi x1, x1, -6; \
-    lb x3, 7(x1); \
+    lb x5, 7(x1); \
   )
 
   #-------------------------------------------------------------
@@ -62,14 +62,14 @@ RVTEST_CODE_BEGIN
   #-------------------------------------------------------------
 
   TEST_CASE( 18, x2, 2, \
-    la  x3, tdat; \
-    lb  x2, 0(x3); \
+    la  x5, tdat; \
+    lb  x2, 0(x5); \
     li  x2, 2; \
   )
 
   TEST_CASE( 19, x2, 2, \
-    la  x3, tdat; \
-    lb  x2, 0(x3); \
+    la  x5, tdat; \
+    lb  x2, 0(x5); \
     nop; \
     li  x2, 2; \
   )
diff --git a/isa/rv64ui/lbu.S b/isa/rv64ui/lbu.S
index 5f4c2fe..adc3a05 100644
--- a/isa/rv64ui/lbu.S
+++ b/isa/rv64ui/lbu.S
@@ -31,18 +31,18 @@ RVTEST_CODE_BEGIN
 
   # Test with a negative base
 
-  TEST_CASE( 10, x3, 0x00000000000000ff, \
+  TEST_CASE( 10, x5, 0x00000000000000ff, \
     la  x1, tdat; \
     addi x1, x1, -32; \
-    lbu x3, 32(x1); \
+    lbu x5, 32(x1); \
   )
 
   # Test with unaligned base
 
-  TEST_CASE( 11, x3, 0x0000000000000000, \
+  TEST_CASE( 11, x5, 0x0000000000000000, \
     la  x1, tdat; \
     addi x1, x1, -6; \
-    lbu x3, 7(x1); \
+    lbu x5, 7(x1); \
   )
 
   #-------------------------------------------------------------
@@ -62,14 +62,14 @@ RVTEST_CODE_BEGIN
   #-------------------------------------------------------------
 
   TEST_CASE( 18, x2, 2, \
-    la  x3, tdat; \
-    lbu  x2, 0(x3); \
+    la  x5, tdat; \
+    lbu  x2, 0(x5); \
     li  x2, 2; \
   )
 
   TEST_CASE( 19, x2, 2, \
-    la  x3, tdat; \
-    lbu  x2, 0(x3); \
+    la  x5, tdat; \
+    lbu  x2, 0(x5); \
     nop; \
     li  x2, 2; \
   )
diff --git a/isa/rv64ui/ld.S b/isa/rv64ui/ld.S
index 62fe4e5..948c34b 100644
--- a/isa/rv64ui/ld.S
+++ b/isa/rv64ui/ld.S
@@ -31,18 +31,18 @@ RVTEST_CODE_BEGIN
 
   # Test with a negative base
 
-  TEST_CASE( 10, x3, 0x00ff00ff00ff00ff, \
+  TEST_CASE( 10, x5, 0x00ff00ff00ff00ff, \
     la  x1, tdat; \
     addi x1, x1, -32; \
-    ld x3, 32(x1); \
+    ld x5, 32(x1); \
   )
 
   # Test with unaligned base
 
-  TEST_CASE( 11, x3, 0xff00ff00ff00ff00, \
+  TEST_CASE( 11, x5, 0xff00ff00ff00ff00, \
     la  x1, tdat; \
     addi x1, x1, -3; \
-    ld x3, 11(x1); \
+    ld x5, 11(x1); \
   )
 
   #-------------------------------------------------------------
@@ -62,14 +62,14 @@ RVTEST_CODE_BEGIN
   #-------------------------------------------------------------
 
   TEST_CASE( 18, x2, 2, \
-    la  x3, tdat; \
-    ld  x2, 0(x3); \
+    la  x5, tdat; \
+    ld  x2, 0(x5); \
     li  x2, 2; \
   )
 
   TEST_CASE( 19, x2, 2, \
-    la  x3, tdat; \
-    ld  x2, 0(x3); \
+    la  x5, tdat; \
+    ld  x2, 0(x5); \
     nop; \
     li  x2, 2; \
   )
diff --git a/isa/rv64ui/lh.S b/isa/rv64ui/lh.S
index decacda..338ed69 100644
--- a/isa/rv64ui/lh.S
+++ b/isa/rv64ui/lh.S
@@ -31,18 +31,18 @@ RVTEST_CODE_BEGIN
 
   # Test with a negative base
 
-  TEST_CASE( 10, x3, 0x00000000000000ff, \
+  TEST_CASE( 10, x5, 0x00000000000000ff, \
     la  x1, tdat; \
     addi x1, x1, -32; \
-    lh x3, 32(x1); \
+    lh x5, 32(x1); \
   )
 
   # Test with unaligned base
 
-  TEST_CASE( 11, x3, 0xffffffffffffff00, \
+  TEST_CASE( 11, x5, 0xffffffffffffff00, \
     la  x1, tdat; \
     addi x1, x1, -5; \
-    lh x3, 7(x1); \
+    lh x5, 7(x1); \
   )
 
   #-------------------------------------------------------------
@@ -62,14 +62,14 @@ RVTEST_CODE_BEGIN
   #-------------------------------------------------------------
 
   TEST_CASE( 18, x2, 2, \
-    la  x3, tdat; \
-    lh  x2, 0(x3); \
+    la  x5, tdat; \
+    lh  x2, 0(x5); \
     li  x2, 2; \
   )
 
   TEST_CASE( 19, x2, 2, \
-    la  x3, tdat; \
-    lh  x2, 0(x3); \
+    la  x5, tdat; \
+    lh  x2, 0(x5); \
     nop; \
     li  x2, 2; \
   )
diff --git a/isa/rv64ui/lhu.S b/isa/rv64ui/lhu.S
index 5a55724..a4cc49b 100644
--- a/isa/rv64ui/lhu.S
+++ b/isa/rv64ui/lhu.S
@@ -31,18 +31,18 @@ RVTEST_CODE_BEGIN
 
   # Test with a negative base
 
-  TEST_CASE( 10, x3, 0x00000000000000ff, \
+  TEST_CASE( 10, x5, 0x00000000000000ff, \
     la  x1, tdat; \
     addi x1, x1, -32; \
-    lhu x3, 32(x1); \
+    lhu x5, 32(x1); \
   )
 
   # Test with unaligned base
 
-  TEST_CASE( 11, x3, 0x000000000000ff00, \
+  TEST_CASE( 11, x5, 0x000000000000ff00, \
     la  x1, tdat; \
     addi x1, x1, -5; \
-    lhu x3, 7(x1); \
+    lhu x5, 7(x1); \
   )
 
   #-------------------------------------------------------------
@@ -62,14 +62,14 @@ RVTEST_CODE_BEGIN
   #-------------------------------------------------------------
 
   TEST_CASE( 18, x2, 2, \
-    la  x3, tdat; \
-    lhu  x2, 0(x3); \
+    la  x5, tdat; \
+    lhu  x2, 0(x5); \
     li  x2, 2; \
   )
 
   TEST_CASE( 19, x2, 2, \
-    la  x3, tdat; \
-    lhu  x2, 0(x3); \
+    la  x5, tdat; \
+    lhu  x2, 0(x5); \
     nop; \
     li  x2, 2; \
   )
diff --git a/isa/rv64ui/lw.S b/isa/rv64ui/lw.S
index 02a12c0..40a73f1 100644
--- a/isa/rv64ui/lw.S
+++ b/isa/rv64ui/lw.S
@@ -31,18 +31,18 @@ RVTEST_CODE_BEGIN
 
   # Test with a negative base
 
-  TEST_CASE( 10, x3, 0x0000000000ff00ff, \
+  TEST_CASE( 10, x5, 0x0000000000ff00ff, \
     la  x1, tdat; \
     addi x1, x1, -32; \
-    lw x3, 32(x1); \
+    lw x5, 32(x1); \
   )
 
   # Test with unaligned base
 
-  TEST_CASE( 11, x3, 0xffffffffff00ff00, \
+  TEST_CASE( 11, x5, 0xffffffffff00ff00, \
     la  x1, tdat; \
     addi x1, x1, -3; \
-    lw x3, 7(x1); \
+    lw x5, 7(x1); \
   )
 
   #-------------------------------------------------------------
@@ -62,14 +62,14 @@ RVTEST_CODE_BEGIN
   #-------------------------------------------------------------
 
   TEST_CASE( 18, x2, 2, \
-    la  x3, tdat; \
-    lw  x2, 0(x3); \
+    la  x5, tdat; \
+    lw  x2, 0(x5); \
     li  x2, 2; \
   )
 
   TEST_CASE( 19, x2, 2, \
-    la  x3, tdat; \
-    lw  x2, 0(x3); \
+    la  x5, tdat; \
+    lw  x2, 0(x5); \
     nop; \
     li  x2, 2; \
   )
diff --git a/isa/rv64ui/lwu.S b/isa/rv64ui/lwu.S
index 1ca17b1..9f7cf67 100644
--- a/isa/rv64ui/lwu.S
+++ b/isa/rv64ui/lwu.S
@@ -31,18 +31,18 @@ RVTEST_CODE_BEGIN
 
   # Test with a negative base
 
-  TEST_CASE( 10, x3, 0x0000000000ff00ff, \
+  TEST_CASE( 10, x5, 0x0000000000ff00ff, \
     la  x1, tdat; \
     addi x1, x1, -32; \
-    lwu x3, 32(x1); \
+    lwu x5, 32(x1); \
   )
 
   # Test with unaligned base
 
-  TEST_CASE( 11, x3, 0x00000000ff00ff00, \
+  TEST_CASE( 11, x5, 0x00000000ff00ff00, \
     la  x1, tdat; \
     addi x1, x1, -3; \
-    lwu x3, 7(x1); \
+    lwu x5, 7(x1); \
   )
 
   #-------------------------------------------------------------
@@ -62,14 +62,14 @@ RVTEST_CODE_BEGIN
   #-------------------------------------------------------------
 
   TEST_CASE( 18, x2, 2, \
-    la  x3, tdat; \
-    lwu x2, 0(x3); \
+    la  x5, tdat; \
+    lwu x2, 0(x5); \
     li  x2, 2; \
   )
 
   TEST_CASE( 19, x2, 2, \
-    la  x3, tdat; \
-    lwu x2, 0(x3); \
+    la  x5, tdat; \
+    lwu x2, 0(x5); \
     nop; \
     li  x2, 2; \
   )
diff --git a/isa/rv64ui/sb.S b/isa/rv64ui/sb.S
index 17ab2e2..19e32d6 100644
--- a/isa/rv64ui/sb.S
+++ b/isa/rv64ui/sb.S
@@ -31,23 +31,23 @@ RVTEST_CODE_BEGIN
 
   # Test with a negative base
 
-  TEST_CASE( 10, x3, 0x78, \
+  TEST_CASE( 10, x5, 0x78, \
     la  x1, tdat9; \
     li  x2, 0x12345678; \
     addi x4, x1, -32; \
     sb x2, 32(x4); \
-    lb x3, 0(x1); \
+    lb x5, 0(x1); \
   )
 
   # Test with unaligned base
 
-  TEST_CASE( 11, x3, 0xffffffffffffff98, \
+  TEST_CASE( 11, x5, 0xffffffffffffff98, \
     la  x1, tdat9; \
     li  x2, 0x00003098; \
     addi x1, x1, -6; \
     sb x2, 7(x1); \
     la  x4, tdat10; \
-    lb x3, 0(x4); \
+    lb x5, 0(x4); \
   )
 
   #-------------------------------------------------------------
diff --git a/isa/rv64ui/sd.S b/isa/rv64ui/sd.S
index f9d83b7..b6fd66d 100644
--- a/isa/rv64ui/sd.S
+++ b/isa/rv64ui/sd.S
@@ -31,23 +31,23 @@ RVTEST_CODE_BEGIN
 
   # Test with a negative base
 
-  TEST_CASE( 10, x3, 0x1234567812345678, \
+  TEST_CASE( 10, x5, 0x1234567812345678, \
     la  x1, tdat9; \
     li  x2, 0x1234567812345678; \
     addi x4, x1, -32; \
     sd x2, 32(x4); \
-    ld x3, 0(x1); \
+    ld x5, 0(x1); \
   )
 
   # Test with unaligned base
 
-  TEST_CASE( 11, x3, 0x5821309858213098, \
+  TEST_CASE( 11, x5, 0x5821309858213098, \
     la  x1, tdat9; \
     li  x2, 0x5821309858213098; \
     addi x1, x1, -3; \
     sd x2, 11(x1); \
     la  x4, tdat10; \
-    ld x3, 0(x4); \
+    ld x5, 0(x4); \
   )
 
   #-------------------------------------------------------------
diff --git a/isa/rv64ui/sh.S b/isa/rv64ui/sh.S
index 10897d4..ea9eb23 100644
--- a/isa/rv64ui/sh.S
+++ b/isa/rv64ui/sh.S
@@ -31,23 +31,23 @@ RVTEST_CODE_BEGIN
 
   # Test with a negative base
 
-  TEST_CASE( 10, x3, 0x5678, \
+  TEST_CASE( 10, x5, 0x5678, \
     la  x1, tdat9; \
     li  x2, 0x12345678; \
     addi x4, x1, -32; \
     sh x2, 32(x4); \
-    lh x3, 0(x1); \
+    lh x5, 0(x1); \
   )
 
   # Test with unaligned base
 
-  TEST_CASE( 11, x3, 0x3098, \
+  TEST_CASE( 11, x5, 0x3098, \
     la  x1, tdat9; \
     li  x2, 0x00003098; \
     addi x1, x1, -5; \
     sh x2, 7(x1); \
     la  x4, tdat10; \
-    lh x3, 0(x4); \
+    lh x5, 0(x4); \
   )
 
   #-------------------------------------------------------------
diff --git a/isa/rv64ui/sw.S b/isa/rv64ui/sw.S
index 86b62fc..ab094b3 100644
--- a/isa/rv64ui/sw.S
+++ b/isa/rv64ui/sw.S
@@ -31,23 +31,23 @@ RVTEST_CODE_BEGIN
 
   # Test with a negative base
 
-  TEST_CASE( 10, x3, 0x12345678, \
+  TEST_CASE( 10, x5, 0x12345678, \
     la  x1, tdat9; \
     li  x2, 0x12345678; \
     addi x4, x1, -32; \
     sw x2, 32(x4); \
-    lw x3, 0(x1); \
+    lw x5, 0(x1); \
   )
 
   # Test with unaligned base
 
-  TEST_CASE( 11, x3, 0x58213098, \
+  TEST_CASE( 11, x5, 0x58213098, \
     la  x1, tdat9; \
     li  x2, 0x58213098; \
     addi x1, x1, -3; \
     sw x2, 7(x1); \
     la  x4, tdat10; \
-    lw x3, 0(x4); \
+    lw x5, 0(x4); \
   )
 
   #-------------------------------------------------------------