Merge pull request #14556 from benpicco/bitarithm_test_and_clear

core/bitarithm: add bitarithm_test_and_clear()

core/include/bitarithm.h

@@ -164,6 +164,45 @@ static inline unsigned bitarithm_lsb(unsigned v)
 }
 #endif
 
+/**
+ * @brief   Used for iterating over the bits in @p state.
+ *          Returns the index of a set bit in @p state, returns @p state with that bit cleared.
+ *
+ * @warning This is a low-level helper function, arguments are not checked.
+ *          It is intended to iterate over a bit map until all bits are cleared.
+ *          Whether it starts with the highest or lowest bit will depend on what is fastest
+ *          on the given hardware.
+ *
+ * @warning @p state must not be zero.
+ *
+ * @param[in]   state   Bit Map with at least one bit set
+ * @param[out]  index   Index of the first set bit. Must be initialized with 0 before the
+ *                      first call to this function, must not be modified between subsequent
+ *                      calls.
+ *
+ * @return      new state value - must be treated as opaque value
+ *
+ */
+static inline unsigned bitarithm_test_and_clear(unsigned state, uint8_t *index)
+{
+#if defined(BITARITHM_HAS_CLZ)
+    *index = 8 * sizeof(state) - __builtin_clz(state) - 1;
+    return state & ~(1 << *index);
+#elif defined(BITARITHM_LSB_LOOKUP)
+    /* Source: http://graphics.stanford.edu/~seander/bithacks.html#ZerosOnRightMultLookup */
+    extern const uint8_t MultiplyDeBruijnBitPosition[32];
+    uint32_t least_bit = state & -state;
+    *index = MultiplyDeBruijnBitPosition[(least_bit * 0x077CB531U) >> 27];
+    return state & ~least_bit;
+#else
+    while ((state & 1) == 0) {
+        *index += 1;
+        state >>= 1;
+    }
+    return state & ~1;
+#endif
+}
+
 #ifdef __cplusplus
 }
 #endif

cpu/arm7_common/include/arm_cpu.h

@@ -34,6 +34,17 @@ void arm_reset(void);
  */
 #define STACK_CANARY_WORD   (0xEAFFFFFEu)
 
+/**
+ * @brief   Select fastest bitarithm_lsb implementation
+ * @{
+ */
+#ifdef __ARM_FEATURE_CLZ
+#define BITARITHM_LSB_BUILTIN
+#define BITARITHM_HAS_CLZ
+#else
+#define BITARITHM_LSB_LOOKUP
+#endif
+/** @} */
 
 #ifdef __cplusplus
 }

cpu/cortexm_common/include/cpu_conf_common.h

@@ -63,6 +63,7 @@ extern "C" {
  */
 #ifdef __ARM_FEATURE_CLZ
 #define BITARITHM_LSB_BUILTIN
+#define BITARITHM_HAS_CLZ
 #else
 #define BITARITHM_LSB_LOOKUP
 #endif

tests/bitarithm_timings/main.c

@@ -40,6 +40,15 @@
 #define TIMEOUT (TIMEOUT_S * US_PER_SEC)
 #define PER_ITERATION (4)
 
+#if ARCH_32_BIT
+#define TEST_AND_CLEAR_TEST_MASK_0  (0xA2101045UL)
+#define TEST_AND_CLEAR_TEST_MASK_1  (0x22101044UL)
+#else
+#define TEST_AND_CLEAR_TEST_MASK_0  (0x9045UL)
+#define TEST_AND_CLEAR_TEST_MASK_1  (0x1244UL)
+
+#endif
+
 static atomic_bool done;
 
 static void callback(void *unused)
@@ -84,6 +93,36 @@ static void run_test(const char *name, unsigned (*test)(unsigned))
     printf("+ %s: %lu iterations per second\r\n", name, (4*PER_ITERATION) * count / TIMEOUT_S);
 }
 
+static unsigned do_test_and_clear(unsigned state)
+{
+    uint8_t index  = 0;
+    unsigned found = 0;
+
+    while (state) {
+        state = bitarithm_test_and_clear(state, &index);
+        found |= 1 << index;
+    }
+
+    return found;
+}
+
+static void run_test_test_and_clear(void)
+{
+    unsigned long count = 0;
+    atomic_store(&done, false);
+
+    xtimer_t xtimer = { .callback = callback };
+    xtimer_set(&xtimer, TIMEOUT);
+
+    do {
+        assert(do_test_and_clear(TEST_AND_CLEAR_TEST_MASK_0) == TEST_AND_CLEAR_TEST_MASK_0);
+        assert(do_test_and_clear(TEST_AND_CLEAR_TEST_MASK_1) == TEST_AND_CLEAR_TEST_MASK_1);
+        ++count;
+    } while (atomic_load(&done) == false);
+
+    printf("+ %s: %lu iterations per second\r\n", "bitarithm_test_and_clear", 2 * count / TIMEOUT_S);
+}
+
 #define run_test(test) run_test(#test, test)
 
 int main(void)
@@ -93,6 +132,7 @@ int main(void)
     run_test(bitarithm_msb);
     run_test(bitarithm_lsb);
     run_test(bitarithm_bits_set);
+    run_test_test_and_clear();
 
     printf("Done.\r\n");
     return 0;

tests/bitarithm_timings/tests/01-run.py

@@ -15,6 +15,7 @@ def testfunc(child):
     child.expect(r'\+ bitarithm_msb: \d+ iterations per second')
     child.expect(r'\+ bitarithm_lsb: \d+ iterations per second')
     child.expect(r'\+ bitarithm_bits_set: \d+ iterations per second')
+    child.expect(r'\+ bitarithm_test_and_clear: \d+ iterations per second')
     child.expect_exact("Done.")