py/misc: Add a popcount(uint32_t) implementation.

This makes the existing popcount(uint32_t) implementation found in the
RV32 emitter available to the rest of the codebase.  This version of
popcount will use intrinsic or builtin implementations if they are
available, falling back to a generic implementation if that is not the
case.

Signed-off-by: Alessandro Gatti <a.gatti@frob.it>

Author: agatti

py/asmrv32.c

@@ -29,6 +29,7 @@
 #include <string.h>
 
 #include "py/emit.h"
+#include "py/misc.h"
 #include "py/mpconfig.h"
 
 // wrapper around everything in this file
@@ -43,27 +44,6 @@
 #define DEBUG_printf(...) (void)0
 #endif
 
-#ifndef MP_POPCOUNT
-#ifdef _MSC_VER
-#include <intrin.h>
-#define MP_POPCOUNT __popcnt
-#else
-#if defined __has_builtin
-#if __has_builtin(__builtin_popcount)
-#define MP_POPCOUNT __builtin_popcount
-#endif
-#else
-static uint32_t fallback_popcount(uint32_t value) {
-    value = value - ((value >> 1) & 0x55555555);
-    value = (value & 0x33333333) + ((value >> 2) & 0x33333333);
-    value = (value + (value >> 4)) & 0x0F0F0F0F;
-    return value * 0x01010101;
-}
-#define MP_POPCOUNT fallback_popcount
-#endif
-#endif
-#endif
-
 #define INTERNAL_TEMPORARY ASM_RV32_REG_S0
 #define AVAILABLE_REGISTERS_COUNT 32
 
@@ -249,7 +229,7 @@ static void adjust_stack(asm_rv32_t *state, mp_int_t stack_size) {
 // stack to hold all the tainted registers and an arbitrary amount of space
 // for locals.
 static void emit_function_prologue(asm_rv32_t *state, mp_uint_t registers) {
-    mp_uint_t registers_count = MP_POPCOUNT(registers);
+    mp_uint_t registers_count = mp_popcount(registers);
     state->stack_size = (registers_count + state->locals_count) * sizeof(uint32_t);
     mp_uint_t old_saved_registers_mask = state->saved_registers_mask;
     // Move stack pointer up.

py/misc.h

@@ -370,12 +370,29 @@ static inline uint32_t mp_ctz(uint32_t x) {
 static inline bool mp_check(bool value) {
     return value;
 }
+
+static inline uint32_t mp_popcount(uint32_t x) {
+    return __popcnt(x);
+}
 #else
 #define mp_clz(x) __builtin_clz(x)
 #define mp_clzl(x) __builtin_clzl(x)
 #define mp_clzll(x) __builtin_clzll(x)
 #define mp_ctz(x) __builtin_ctz(x)
 #define mp_check(x) (x)
+#if defined __has_builtin
+#if __has_builtin(__builtin_popcount)
+#define mp_popcount(x) __builtin_popcount(x)
+#endif
+#endif
+#if !defined(mp_popcount)
+static inline uint32_t mp_popcount(uint32_t x) {
+    x = x - ((x >> 1) & 0x55555555);
+    x = (x & 0x33333333) + ((x >> 2) & 0x33333333);
+    x = (x + (x >> 4)) & 0x0F0F0F0F;
+    return x * 0x01010101;
+}
+#endif
 #endif
 
 // mp_int_t can be larger than long, i.e. Windows 64-bit, nan-box variants