Try to add support code for riscv64

Makefile

@@ -26,6 +26,13 @@ ifeq ($(ARCH),aarch64)
  endif
 endif
 
+ifeq ($(ARCH),riscv64)
+ CAP ?= $(shell cat /proc/cpuinfo | grep -E 'zihintpause' | head -1)
+ ifneq (,$(findstring zihintpause,$(CAP)))
+  CFLAGS+=-march=rv64gc_zihintpause
+ endif
+endif
+
 EXE=multichase multiload fairness pingpong
 
 all: $(EXE)

br_asm.c

@@ -83,6 +83,65 @@ int convert_pointers_to_branches(void *head, int chunk_size) {
   return base_chunk_size;
 }
 
+#elif defined(__riscv) && __riscv_xlen == 64
+static char *riscv64_emit_lui_a0_imm64(char *p, uint64_t imm64) {
+  *p++ = 0x37; // opcode for LUI (Load Upper Immediate)
+  *p++ = 0x05; // rd = a0
+  *p++ = (imm64 >> 32) & 0xff;
+  *p++ = (imm64 >> 40) & 0xff;
+  *p++ = (imm64 >> 48) & 0xff;
+  *p++ = (imm64 >> 56) & 0xff;
+  return p;
+}
+
+static char *riscv64_emit_jalr_a0(char *p) {
+  *p++ = 0x67; // opcode for JALR (Jump and Link Register)
+  *p++ = 0x80; // rd = x1 (return address), rs1 = a0
+  *p++ = 0x00;
+  *p++ = 0x00;
+  return p;
+}
+
+static char *riscv64_emit_ret(char *p) {
+  *p++ = 0x80; // opcode for RETL (Return)
+  *p++ = 0x02;
+  *p++ = 0x10;
+  *p++ = 0x00;
+  return p;
+}
+
+int convert_pointers_to_branches(void *head, int chunk_size) {
+  int remain = cycle_len(head);
+  chunk_size = (remain < chunk_size)
+                   ? remain
+                   : remain / (1 << lround(log2(1.0 * remain / chunk_size)));
+  int base_chunk_size = chunk_size;
+  int chunks_remaining = remain / chunk_size;
+  int chunk_count = 0;
+  const int br_code_len = 20; // len(lui) + len(jalr)
+  char *p = (char *)head;
+  do {
+    if (!chunk_count) chunk_count = remain / chunks_remaining;
+    char *next = *((char **)p);
+    for (int i = 8; i < br_code_len; i++) {
+      if (p[i]) {
+        fprintf(stderr, "not enough space to convert a pointer to branches\n");
+        exit(1);
+      }
+    }
+    p = riscv64_emit_lui_a0_imm64(p, (intptr_t)next);
+    --remain;
+    if (--chunk_count == 0) {
+      p = riscv64_emit_ret(p);
+      --chunks_remaining;
+    } else {
+      p = riscv64_emit_jalr_a0(p);
+    }
+    p = next;
+  } while (p != head);
+  return base_chunk_size;
+}
+
 #elif defined(__x86_64__)
 
 static char *x64_emit_mov_imm64_rax(char *p, uint64_t imm64) {

cpu_util.h

@@ -40,6 +40,8 @@ static inline void cpu_relax(void) {
 }
 #elif defined(__aarch64__)
 #define cpu_relax() asm volatile("yield" ::: "memory")
+#elif defined(__riscv) && __riscv_xlen == 64
+#define cpu_relax() asm volatile("pause" ::: "memory")
 #else
 #warning "no cpu_relax for your cpu"
 #define cpu_relax() \

multichase.c

@@ -487,11 +487,12 @@ static void *thread_start(void *data) {
   if (!strcmp(args->x.chase->name, "branch")) {
     void *p = args->x.cycle[0];
     args->x.branch_chunk_size = convert_pointers_to_branches(p, 200);
-#if defined(__aarch64__)
+#if defined(__aarch64__) || (defined(__riscv) && __riscv_xlen == 64)
     __builtin___clear_cache(
         args->x.genchase_args->arena,
         args->x.genchase_args->arena + args->x.genchase_args->total_memory);
 #endif
+
   }
 
   // now flush our caches

multiload.c

@@ -601,6 +601,17 @@ static void load_stream_triad_nontemporal_injection_delay(per_thread_t *t) {
       if (i % num_elem_twocachelines == 0) delay_until_iteration(t->x.delay);
 #if defined(__aarch64__)
       asm volatile ("stnp %0, %1, [%2]" :: "r"(b[i]+c[i]), "r"(b[i+1]+c[i+1]), "r" (a+i));
+#elif defined(__riscv) && __riscv_xlen == 64
+      // sd (Store Doubleword) is a risc-v memory store instruction that stores a 64-bit (8-byte) value from a register to memory. 
+      // "sd %0, (%2)\n\t" stores 64-bit value from register %0 to memory address contained in register %2,
+      // "sd %1, 8(%2)" stores 64-bit value from register %1 to memory address [%2] + 8.
+      asm volatile ("sd %0, (%2)\n\t"
+              "sd %1, 8(%2)"
+              : // no outputs
+              : "r"(b[i]+c[i]),
+                "r"(b[i+1]+c[i+1]),
+                "r" (a+i)
+              : "memory");
 #elif defined(__x86_64__)
       _mm_stream_si64(&((long long*)a)[i], b[i]+c[i]);
       _mm_stream_si64(&((long long*)a)[i+1], b[i+1]+c[i+1]);