diff --git a/src/bits.c b/src/bits.c
index 49a3380..09b6da7 100644
--- a/src/bits.c
+++ b/src/bits.c
@@ -167,7 +167,7 @@ static inline void ac_put(struct lc3_bits_buffer *buffer, int byte)
  * ac              Arithmetic coder
  * buffer          Bitstream buffer
  */
-static inline void ac_shift(
+LC3_HOT static inline void ac_shift(
     struct lc3_bits_ac *ac, struct lc3_bits_buffer *buffer)
 {
     if (ac->low < 0xff0000 || ac->carry)
@@ -262,7 +262,7 @@ void lc3_flush_bits(struct lc3_bits *bits)
  * Write from 1 to 32 bits,
  * exceeding the capacity of the accumulator
  */
-void lc3_put_bits_generic(struct lc3_bits *bits, unsigned v, int n)
+LC3_HOT void lc3_put_bits_generic(struct lc3_bits *bits, unsigned v, int n)
 {
     struct lc3_bits_accu *accu = &bits->accu;
 
@@ -285,7 +285,7 @@ void lc3_put_bits_generic(struct lc3_bits *bits, unsigned v, int n)
 /**
  * Arithmetic coder renormalization
  */
-void lc3_ac_write_renorm(struct lc3_bits *bits)
+LC3_HOT void lc3_ac_write_renorm(struct lc3_bits *bits)
 {
     struct lc3_bits_ac *ac = &bits->ac;
 
@@ -336,7 +336,7 @@ static inline void accu_load(struct lc3_bits_accu *accu,
  * Read from 1 to 32 bits,
  * exceeding the capacity of the accumulator
  */
-unsigned lc3_get_bits_generic(struct lc3_bits *bits, int n)
+LC3_HOT unsigned lc3_get_bits_generic(struct lc3_bits *bits, int n)
 {
     struct lc3_bits_accu *accu = &bits->accu;
     struct lc3_bits_buffer *buffer = &bits->buffer;
@@ -366,7 +366,7 @@ unsigned lc3_get_bits_generic(struct lc3_bits *bits, int n)
 /**
  * Arithmetic coder renormalization
  */
-void lc3_ac_read_renorm(struct lc3_bits *bits)
+LC3_HOT void lc3_ac_read_renorm(struct lc3_bits *bits)
 {
     struct lc3_bits_ac *ac = &bits->ac;
 
diff --git a/src/bits.h b/src/bits.h
index 8505cb3..5dd56cd 100644
--- a/src/bits.h
+++ b/src/bits.h
@@ -213,7 +213,7 @@ void lc3_ac_write_renorm(lc3_bits_t *bits);
 /**
  * Put a bit
  */
-static inline void lc3_put_bit(lc3_bits_t *bits, int v)
+LC3_HOT static inline void lc3_put_bit(lc3_bits_t *bits, int v)
 {
     lc3_put_bits(bits, v, 1);
 }
@@ -221,7 +221,8 @@ static inline void lc3_put_bit(lc3_bits_t *bits, int v)
 /**
  * Put from 1 to 32 bits
  */
-static inline void lc3_put_bits(struct lc3_bits *bits, unsigned v, int n)
+LC3_HOT static inline void lc3_put_bits(
+    struct lc3_bits *bits, unsigned v, int n)
 {
     struct lc3_bits_accu *accu = &bits->accu;
 
@@ -236,7 +237,7 @@ static inline void lc3_put_bits(struct lc3_bits *bits, unsigned v, int n)
 /**
  * Get a bit
  */
-static inline int lc3_get_bit(lc3_bits_t *bits)
+LC3_HOT static inline int lc3_get_bit(lc3_bits_t *bits)
 {
     return lc3_get_bits(bits, 1);
 }
@@ -244,7 +245,7 @@ static inline int lc3_get_bit(lc3_bits_t *bits)
 /**
  * Get from 1 to 32 bits
  */
-static inline unsigned lc3_get_bits(struct lc3_bits *bits, int n)
+LC3_HOT static inline unsigned lc3_get_bits(struct lc3_bits *bits, int n)
 {
     struct lc3_bits_accu *accu = &bits->accu;
 
@@ -260,7 +261,7 @@ static inline unsigned lc3_get_bits(struct lc3_bits *bits, int n)
 /**
  * Put arithmetic coder symbol
  */
-static inline void lc3_put_symbol(
+LC3_HOT static inline void lc3_put_symbol(
     struct lc3_bits *bits, const struct lc3_ac_model *model, unsigned s)
 {
     const struct lc3_ac_symbol *symbols = model->s;
@@ -280,7 +281,7 @@ static inline void lc3_put_symbol(
 /**
  * Get arithmetic coder symbol
  */
-static inline unsigned lc3_get_symbol(
+LC3_HOT static inline unsigned lc3_get_symbol(
     lc3_bits_t *bits, const struct lc3_ac_model *model)
 {
     const struct lc3_ac_symbol *symbols = model->s;
diff --git a/src/common.h b/src/common.h
index 2a4e0cb..d1f8274 100644
--- a/src/common.h
+++ b/src/common.h
@@ -35,6 +35,24 @@
 #endif
 
 
+/**
+ * Hot Function attribute
+ * Selectively disable sanitizer
+ */
+
+#ifdef __clang__
+
+#define LC3_HOT \
+    __attribute__((no_sanitize("bounds"))) \
+    __attribute__((no_sanitize("integer")))
+
+#else /* __clang__ */
+
+#define LC3_HOT
+
+#endif /* __clang__ */
+
+
 /**
  * Macros
  * MIN/MAX  Minimum and maximum between 2 values
diff --git a/src/ltpf.c b/src/ltpf.c
index 7c35890..d2a93c2 100644
--- a/src/ltpf.c
+++ b/src/ltpf.c
@@ -151,7 +151,7 @@ static const int16_t h_48k_12k8_q15[4*60] = {
  * xn              Input sample, in fixed Q30
  * return          Filtered sample, in fixed Q30
  */
-static inline int32_t filter_hp50(
+LC3_HOT static inline int32_t filter_hp50(
     struct lc3_ltpf_hp50_state *hp50, int32_t xn)
 {
     int32_t yn;
@@ -178,7 +178,7 @@ static inline int32_t filter_hp50(
  * The number of previous samples `d` accessed on `x` is :
  *   d: { 10, 20, 40 } - 1 for resampling factors 8, 4 and 2.
  */
-static inline void resample_x64k_12k8(const int p, const int16_t *h,
+LC3_HOT static inline void resample_x64k_12k8(const int p, const int16_t *h,
     struct lc3_ltpf_hp50_state *hp50, const int16_t *x, int16_t *y, int n)
 {
     const int w = 2*(40 / p);
@@ -220,7 +220,7 @@ static inline void resample_x64k_12k8(const int p, const int16_t *h,
  * The number of previous samples `d` accessed on `x` is :
  *   d: { 30, 60 } - 1 for resampling factors 8 and 4.
  */
-static inline void resample_x192k_12k8(const int p, const int16_t *h,
+LC3_HOT static inline void resample_x192k_12k8(const int p, const int16_t *h,
     struct lc3_ltpf_hp50_state *hp50, const int16_t *x, int16_t *y, int n)
 {
     const int w = 2*(120 / p);
@@ -264,7 +264,7 @@ static inline void resample_x192k_12k8(const int p, const int16_t *h,
  * The `x` vector is aligned on 32 bits
  */
 #ifndef resample_8k_12k8
-static void resample_8k_12k8(
+LC3_HOT static void resample_8k_12k8(
     struct lc3_ltpf_hp50_state *hp50, const int16_t *x, int16_t *y, int n)
 {
     resample_x64k_12k8(8, h_8k_12k8_q15, hp50, x, y, n);
@@ -280,7 +280,7 @@ static void resample_8k_12k8(
  * The `x` vector is aligned on 32 bits
  */
 #ifndef resample_16k_12k8
-static void resample_16k_12k8(
+LC3_HOT static void resample_16k_12k8(
     struct lc3_ltpf_hp50_state *hp50, const int16_t *x, int16_t *y, int n)
 {
     resample_x64k_12k8(4, h_16k_12k8_q15, hp50, x, y, n);
@@ -296,7 +296,7 @@ static void resample_16k_12k8(
  * The `x` vector is aligned on 32 bits
  */
 #ifndef resample_32k_12k8
-static void resample_32k_12k8(
+LC3_HOT static void resample_32k_12k8(
     struct lc3_ltpf_hp50_state *hp50, const int16_t *x, int16_t *y, int n)
 {
     resample_x64k_12k8(2, h_32k_12k8_q15, hp50, x, y, n);
@@ -312,7 +312,7 @@ static void resample_32k_12k8(
  * The `x` vector is aligned on 32 bits
  */
 #ifndef resample_24k_12k8
-static void resample_24k_12k8(
+LC3_HOT static void resample_24k_12k8(
     struct lc3_ltpf_hp50_state *hp50, const int16_t *x, int16_t *y, int n)
 {
     resample_x192k_12k8(8, h_24k_12k8_q15, hp50, x, y, n);
@@ -328,8 +328,8 @@ static void resample_24k_12k8(
 * The `x` vector is aligned on 32 bits
 */
 #ifndef resample_48k_12k8
-static void resample_48k_12k8(
-struct lc3_ltpf_hp50_state *hp50, const int16_t *x, int16_t *y, int n)
+LC3_HOT static void resample_48k_12k8(
+    struct lc3_ltpf_hp50_state *hp50, const int16_t *x, int16_t *y, int n)
 {
     resample_x192k_12k8(4, h_48k_12k8_q15, hp50, x, y, n);
 }
@@ -343,7 +343,7 @@ struct lc3_ltpf_hp50_state *hp50, const int16_t *x, int16_t *y, int n)
 * The `x` vector is aligned on 32 bits
  */
 #ifndef resample_6k4
-static void resample_6k4(const int16_t *x, int16_t *y, int n)
+LC3_HOT static void resample_6k4(const int16_t *x, int16_t *y, int n)
 {
     static const int16_t h[] = { 18477, 15424, 8105 };
     const int16_t *ye = y + n;
@@ -381,7 +381,7 @@ static void (* const resample_12k8[])
  * The size `n` of vectors must be multiple of 16, and less or equal to 128
 */
 #ifndef dot
-static inline float dot(const int16_t *a, const int16_t *b, int n)
+LC3_HOT static inline float dot(const int16_t *a, const int16_t *b, int n)
 {
     int64_t v = 0;
 
@@ -403,7 +403,7 @@ static inline float dot(const int16_t *a, const int16_t *b, int n)
  * The size `n` of vectors is multiple of 16, and less or equal to 128
  */
 #ifndef correlate
-static void correlate(
+LC3_HOT static void correlate(
     const int16_t *a, const int16_t *b, int n, float *y, int nc)
 {
     for (const float *ye = y + nc; y < ye; )
@@ -417,7 +417,7 @@ static void correlate(
  * x_max           Return the maximum value
  * return          Return the argument of the maximum
  */
-static int argmax(const float *x, int n, float *x_max)
+LC3_HOT static int argmax(const float *x, int n, float *x_max)
 {
     int arg = 0;
 
@@ -436,7 +436,7 @@ static int argmax(const float *x, int n, float *x_max)
  * x_max, xw_max   Return the maximum not weighted value
  * return          Return the argument of the weigthed maximum
  */
-static int argmax_weighted(
+LC3_HOT static int argmax_weighted(
     const float *x, int n, float w_incr, float *x_max)
 {
     int arg;
@@ -459,7 +459,7 @@ static int argmax_weighted(
  *
  * The size `n` of vectors must be multiple of 4
  */
-static void interpolate(const int16_t *x, int n, int d, int16_t *y)
+LC3_HOT static void interpolate(const int16_t *x, int n, int d, int16_t *y)
 {
     static const int16_t h4_q15[][4] = {
         { 6877, 19121,  6877,     0 }, { 3506, 18025, 11000,   220 },
@@ -492,7 +492,7 @@ static void interpolate(const int16_t *x, int n, int d, int16_t *y)
  * d               The phase of interpolation (-3 to 3)
  * return          The interpolated value
  */
-static float interpolate_corr(const float *x, int d)
+LC3_HOT static float interpolate_corr(const float *x, int d)
 {
     static const float h4[][8] = {
         {  1.53572770e-02, -4.72963246e-02,  8.35788573e-02,  8.98638285e-01,
@@ -689,8 +689,10 @@ bool lc3_ltpf_analyse(
  * c, w            Coefficients `den` then `num`, and width of filter
  * fade            Fading mode of filter  -1: Out  1: In  0: None
  */
-static inline void synthesize_template(const float *xh, int nh, int lag,
-    const float *x0, float *x, int n, const float *c, const int w, int fade)
+LC3_HOT static inline void synthesize_template(
+    const float *xh, int nh, int lag,
+    const float *x0, float *x, int n,
+    const float *c, const int w, int fade)
 {
     float g = (float)(fade <= 0);
     float g_incr = (float)((fade > 0) - (fade < 0)) / n;
@@ -742,25 +744,25 @@ static inline void synthesize_template(const float *xh, int nh, int lag,
  * Synthesis filter for each samplerates (width of filter)
  */
 
-static void synthesize_4(const float *xh, int nh, int lag,
+LC3_HOT static void synthesize_4(const float *xh, int nh, int lag,
     const float *x0, float *x, int n, const float *c, int fade)
 {
     synthesize_template(xh, nh, lag, x0, x, n, c, 4, fade);
 }
 
-static void synthesize_6(const float *xh, int nh, int lag,
+LC3_HOT static void synthesize_6(const float *xh, int nh, int lag,
     const float *x0, float *x, int n, const float *c, int fade)
 {
     synthesize_template(xh, nh, lag, x0, x, n, c, 6, fade);
 }
 
-static void synthesize_8(const float *xh, int nh, int lag,
+LC3_HOT static void synthesize_8(const float *xh, int nh, int lag,
     const float *x0, float *x, int n, const float *c, int fade)
 {
     synthesize_template(xh, nh, lag, x0, x, n, c, 8, fade);
 }
 
-static void synthesize_12(const float *xh, int nh, int lag,
+LC3_HOT static void synthesize_12(const float *xh, int nh, int lag,
     const float *x0, float *x, int n, const float *c, int fade)
 {
     synthesize_template(xh, nh, lag, x0, x, n, c, 12, fade);
diff --git a/src/ltpf_neon.h b/src/ltpf_neon.h
index 728ed1d..da60409 100644
--- a/src/ltpf_neon.h
+++ b/src/ltpf_neon.h
@@ -46,7 +46,7 @@ static inline int32_t filter_hp50(struct lc3_ltpf_hp50_state *, int32_t);
 /**
  * Resample from 16 Khz to 12.8 KHz
  */
-static void neon_resample_16k_12k8(
+LC3_HOT static void neon_resample_16k_12k8(
     struct lc3_ltpf_hp50_state *hp50, const int16_t *x, int16_t *y, int n)
 {
     static const int16_t h[4][20] = {
@@ -86,7 +86,7 @@ static void neon_resample_16k_12k8(
 /**
  * Resample from 32 Khz to 12.8 KHz
  */
-static void neon_resample_32k_12k8(
+LC3_HOT static void neon_resample_32k_12k8(
     struct lc3_ltpf_hp50_state *hp50, const int16_t *x, int16_t *y, int n)
 {
     x -= 40 - 1;
@@ -125,7 +125,7 @@ static void neon_resample_32k_12k8(
 /**
  * Resample from 48 Khz to 12.8 KHz
  */
-static void neon_resample_48k_12k8(
+LC3_HOT static void neon_resample_48k_12k8(
     struct lc3_ltpf_hp50_state *hp50, const int16_t *x, int16_t *y, int n)
 {
     static const int16_t alignas(16) h[4][64] = {
@@ -180,7 +180,7 @@ static void neon_resample_48k_12k8(
 /**
  * Return dot product of 2 vectors
  */
-static inline float neon_dot(const int16_t *a, const int16_t *b, int n)
+LC3_HOT static inline float neon_dot(const int16_t *a, const int16_t *b, int n)
 {
     int64x2_t v = vmovq_n_s64(0);
 
@@ -203,7 +203,7 @@ static inline float neon_dot(const int16_t *a, const int16_t *b, int n)
 /**
  * Return vector of correlations
  */
-static void neon_correlate(
+LC3_HOT static void neon_correlate(
     const int16_t *a, const int16_t *b, int n, float *y, int nc)
 {
     for ( ; nc >= 4; nc -= 4, b -= 4) {
diff --git a/src/mdct.c b/src/mdct.c
index 29bc161..fa03d5f 100644
--- a/src/mdct.c
+++ b/src/mdct.c
@@ -28,7 +28,7 @@
  * x, y            Input and output coefficients, of size 5xn
  * n               Number of interleaved transform to perform
  */
-static inline void fft_5(
+LC3_HOT static inline void fft_5(
     const struct lc3_complex *x, struct lc3_complex *y, int n)
 {
     static const float cos1 =  0.3090169944;  /* cos(-2Pi 1/5) */
@@ -84,7 +84,7 @@ static inline void fft_5(
  * twiddles        Twiddles factors, determine size of transform
  * n               Number of interleaved transforms
  */
-static inline void fft_bf3(
+LC3_HOT static inline void fft_bf3(
     const struct lc3_fft_bf3_twiddles *twiddles,
     const struct lc3_complex *x, struct lc3_complex *y, int n)
 {
@@ -126,7 +126,7 @@ static inline void fft_bf3(
  * x, y            Input and output coefficients
  * n               Number of interleaved transforms
  */
-static inline void fft_bf2(
+LC3_HOT static inline void fft_bf2(
     const struct lc3_fft_bf2_twiddles *twiddles,
     const struct lc3_complex *x, struct lc3_complex *y, int n)
 {
@@ -196,7 +196,7 @@ static struct lc3_complex *fft(const struct lc3_complex *x, int n,
  * x, y            Input current and delayed samples
  * y, d            Output windowed samples, and delayed ones
  */
-static void mdct_window(enum lc3_dt dt, enum lc3_srate sr,
+LC3_HOT static void mdct_window(enum lc3_dt dt, enum lc3_srate sr,
     const float *x, float *d, float *y)
 {
     int ns = LC3_NS(dt, sr), nd = LC3_ND(dt, sr);
@@ -232,7 +232,7 @@ static void mdct_window(enum lc3_dt dt, enum lc3_srate sr,
  *
  * `x` and y` can be the same buffer
  */
-static void mdct_pre_fft(const struct lc3_mdct_rot_def *def,
+LC3_HOT static void mdct_pre_fft(const struct lc3_mdct_rot_def *def,
     const float *x, struct lc3_complex *y)
 {
     int n4 = def->n4;
@@ -263,7 +263,7 @@ static void mdct_pre_fft(const struct lc3_mdct_rot_def *def,
  *
  * `x` and y` can be the same buffer
  */
-static void mdct_post_fft(const struct lc3_mdct_rot_def *def,
+LC3_HOT static void mdct_post_fft(const struct lc3_mdct_rot_def *def,
     const struct lc3_complex *x, float *y, float scale)
 {
     int n4 = def->n4, n8 = n4 >> 1;
@@ -295,7 +295,7 @@ static void mdct_post_fft(const struct lc3_mdct_rot_def *def,
  * The real and imaginary parts of `y` are swapped,
  * to operate on FFT instead of IFFT
  */
-static void imdct_pre_fft(const struct lc3_mdct_rot_def *def,
+LC3_HOT static void imdct_pre_fft(const struct lc3_mdct_rot_def *def,
     const float *x, struct lc3_complex *y)
 {
     int n4 = def->n4;
@@ -328,7 +328,7 @@ static void imdct_pre_fft(const struct lc3_mdct_rot_def *def,
  * The real and imaginary parts of `x` are swapped,
  * to operate on FFT instead of IFFT
  */
-static void imdct_post_fft(const struct lc3_mdct_rot_def *def,
+LC3_HOT static void imdct_post_fft(const struct lc3_mdct_rot_def *def,
     const struct lc3_complex *x, float *y, float scale)
 {
     int n4 = def->n4;
@@ -356,7 +356,7 @@ static void imdct_post_fft(const struct lc3_mdct_rot_def *def,
  * x, d            Middle half of IMDCT coefficients and delayed samples
  * y, d            Output samples and delayed ones
  */
-static void imdct_window(enum lc3_dt dt, enum lc3_srate sr,
+LC3_HOT static void imdct_window(enum lc3_dt dt, enum lc3_srate sr,
     const float *x, float *d, float *y)
 {
     /* The full MDCT coefficients is given by symmetry :
diff --git a/src/sns.c b/src/sns.c
index b06bf60..56a893c 100644
--- a/src/sns.c
+++ b/src/sns.c
@@ -120,7 +120,7 @@ static const float dct16_m[16][16] = {
  * Forward DCT-16 transformation
  * x, y            Input and output 16 values
  */
-static void dct16_forward(const float *x, float *y)
+LC3_HOT static void dct16_forward(const float *x, float *y)
 {
     for (int i = 0, j; i < 16; i++)
         for (y[i] = 0, j = 0; j < 16; j++)
@@ -131,7 +131,7 @@ static void dct16_forward(const float *x, float *y)
  * Inverse DCT-16 transformation
  * x, y            Input and output 16 values
  */
-static void dct16_inverse(const float *x, float *y)
+LC3_HOT static void dct16_inverse(const float *x, float *y)
 {
     for (int i = 0, j; i < 16; i++)
         for (y[i] = 0, j = 0; j < 16; j++)
@@ -150,7 +150,8 @@ static void dct16_inverse(const float *x, float *y)
  * att             1: Attack detected  0: Otherwise
  * scf             Output 16 scale factors
  */
-static void compute_scale_factors(enum lc3_dt dt, enum lc3_srate sr,
+LC3_HOT static void compute_scale_factors(
+    enum lc3_dt dt, enum lc3_srate sr,
     const float *eb, bool att, float *scf)
 {
     /* Pre-emphasis gain table :
@@ -343,7 +344,8 @@ static void compute_scale_factors(enum lc3_dt dt, enum lc3_srate sr,
  * scf             Input 16 scale factors
  * lf/hfcb_idx     Output the low and high frequency codebooks index
  */
-static void resolve_codebooks(const float *scf, int *lfcb_idx, int *hfcb_idx)
+LC3_HOT static void resolve_codebooks(
+    const float *scf, int *lfcb_idx, int *hfcb_idx)
 {
     float dlfcb_max = 0, dhfcb_max = 0;
     *lfcb_idx = *hfcb_idx = 0;
@@ -371,7 +373,7 @@ static void resolve_codebooks(const float *scf, int *lfcb_idx, int *hfcb_idx)
  * c               Pulse configuration
  * cn              Normalized pulse configuration
  */
-static void normalize(const int *c, float *cn)
+LC3_HOT static void normalize(const int *c, float *cn)
 {
     int c2_sum = 0;
     for (int i = 0; i < 16; i++)
@@ -389,7 +391,7 @@ static void normalize(const int *c, float *cn)
  * start, end      Current number of pulses, limit to reach
  * corr, energy    Correlation (x,y) and y energy, updated at output
  */
-static void add_pulse(const float *x, int *y, int n,
+LC3_HOT static void add_pulse(const float *x, int *y, int n,
     int start, int end, float *corr, float *energy)
 {
     for (int k = start; k < end; k++) {
@@ -418,7 +420,7 @@ static void add_pulse(const float *x, int *y, int n,
  * c, cn           Output 4 pulse configurations candidates, normalized
  * shape/gain_idx  Output selected shape/gain indexes
  */
-static void quantize(const float *scf, int lfcb_idx, int hfcb_idx,
+LC3_HOT static void quantize(const float *scf, int lfcb_idx, int hfcb_idx,
     int (*c)[16], float (*cn)[16], int *shape_idx, int *gain_idx)
 {
     /* --- Residual --- */
@@ -539,7 +541,7 @@ static void quantize(const float *scf, int lfcb_idx, int hfcb_idx,
  * shape/gain      Selected shape/gain indexes
  * scf             Return unquantized scale factors
  */
-static void unquantize(int lfcb_idx, int hfcb_idx,
+LC3_HOT static void unquantize(int lfcb_idx, int hfcb_idx,
     const float *c, int shape, int gain, float *scf)
 {
     const float *lfcb = lc3_sns_lfcb[lfcb_idx];
@@ -673,7 +675,7 @@ static void deenumerate(int shape,
  *
  * `x` and `y` can be the same buffer
  */
-static void spectral_shaping(enum lc3_dt dt, enum lc3_srate sr,
+LC3_HOT static void spectral_shaping(enum lc3_dt dt, enum lc3_srate sr,
     const float *scf_q, bool inv, const float *x, float *y)
 {
     /* --- Interpolate scale factors --- */
diff --git a/src/spec.c b/src/spec.c
index 67cd4ac..f650156 100644
--- a/src/spec.c
+++ b/src/spec.c
@@ -46,7 +46,7 @@ static int resolve_gain_offset(enum lc3_srate sr, int nbytes)
  * reset_off       Return True when the nbits_off must be reset
  * return          The quantized gain value
  */
-static int estimate_gain(
+LC3_HOT static int estimate_gain(
     enum lc3_dt dt, enum lc3_srate sr, const float *x,
     int nbits_budget, float nbits_off, int g_off, bool *reset_off)
 {
@@ -121,8 +121,8 @@ static int estimate_gain(
  * nbits_budget    Number of bits available for coding the spectrum
  * return          Gain adjust value (-1 to 2)
  */
-static int adjust_gain(enum lc3_srate sr,
-    int g_idx, int nbits, int nbits_budget)
+LC3_HOT static int adjust_gain(
+    enum lc3_srate sr, int g_idx, int nbits, int nbits_budget)
 {
     /* --- Compute delta threshold --- */
 
@@ -202,7 +202,7 @@ static float unquantize_gain(int g_int)
  * x               Spectral coefficients, scaled as output
  * xq, nq          Output spectral quantized coefficients, and count
  */
-static void quantize(enum lc3_dt dt, enum lc3_srate sr,
+LC3_HOT static void quantize(enum lc3_dt dt, enum lc3_srate sr,
     int g_int, float *x, int16_t *xq, int *nq)
 {
     float g_inv = 1 / unquantize_gain(g_int);
@@ -232,7 +232,7 @@ static void quantize(enum lc3_dt dt, enum lc3_srate sr,
  * x, nq           Spectral quantized, and count of significants
  * return          Unquantized gain value
  */
-static float unquantize(enum lc3_dt dt, enum lc3_srate sr,
+LC3_HOT static float unquantize(enum lc3_dt dt, enum lc3_srate sr,
     int g_int, float *x, int nq)
 {
     float g = unquantize_gain(g_int);
@@ -271,7 +271,7 @@ static int resolve_high_rate(enum lc3_srate sr, int nbytes)
  * p_lsb_mode      Return True when LSB's are not AC coded, or NULL
  * return          The number of bits coding the spectrum
  */
-static int compute_nbits(
+LC3_HOT static int compute_nbits(
     enum lc3_dt dt, enum lc3_srate sr, int nbytes,
     const int16_t *x, int *n, int nbits_budget, bool *p_lsb_mode)
 {
@@ -367,7 +367,7 @@ static int compute_nbits(
  * x               Spectral quantized
  * nq, lsb_mode    Count of significants, and LSB discard indication
  */
-static void put_quantized(lc3_bits_t *bits,
+LC3_HOT static void put_quantized(lc3_bits_t *bits,
     enum lc3_dt dt, enum lc3_srate sr, int nbytes,
     const int16_t *x, int nq, bool lsb_mode)
 {
@@ -443,7 +443,7 @@ static void put_quantized(lc3_bits_t *bits,
  * nf_seed         Return the noise factor seed associated
  * return          0: Ok  -1: Invalid bitstream data
  */
-static int get_quantized(lc3_bits_t *bits,
+LC3_HOT static int get_quantized(lc3_bits_t *bits,
     enum lc3_dt dt, enum lc3_srate sr, int nbytes,
     int nq, bool lsb_mode, float *xq, uint16_t *nf_seed)
 {
@@ -518,8 +518,8 @@ static int get_quantized(lc3_bits_t *bits,
  * xq, n           Spectral quantized, and count of significants
  * xf              Scaled spectral coefficients
  */
-static void put_residual(lc3_bits_t *bits, int nbits,
-    const int16_t *xq, int n, const float *xf)
+LC3_HOT static void put_residual(
+    lc3_bits_t *bits, int nbits, const int16_t *xq, int n, const float *xf)
 {
     for (int i = 0; i < n && nbits > 0; i++) {
 
@@ -537,7 +537,8 @@ static void put_residual(lc3_bits_t *bits, int nbits,
  * nbits           Maximum number of bits to output
  * x, nq           Spectral quantized, and count of significants
  */
-static void get_residual(lc3_bits_t *bits, int nbits, float *x, int nq)
+LC3_HOT static void get_residual(
+    lc3_bits_t *bits, int nbits, float *x, int nq)
 {
     for (int i = 0; i < nq && nbits > 0; i++) {
 
@@ -559,7 +560,8 @@ static void get_residual(lc3_bits_t *bits, int nbits, float *x, int nq)
  * nbits           Maximum number of bits to output
  * x, n            Spectral quantized, and count of significants
  */
-static void put_lsb(lc3_bits_t *bits, int nbits, const int16_t *x, int n)
+LC3_HOT static void put_lsb(
+    lc3_bits_t *bits, int nbits, const int16_t *x, int n)
 {
     for (int i = 0; i < n && nbits > 0; i += 2) {
 
@@ -590,7 +592,7 @@ static void put_lsb(lc3_bits_t *bits, int nbits, const int16_t *x, int n)
  * x, nq           Spectral quantized, and count of significants
  * nf_seed         Update the noise factor seed according
  */
-static void get_lsb(lc3_bits_t *bits,
+LC3_HOT static void get_lsb(lc3_bits_t *bits,
     int nbits, float *x, int nq, uint16_t *nf_seed)
 {
     for (int i = 0; i < nq && nbits > 0; i += 2) {
@@ -634,7 +636,7 @@ static void get_lsb(lc3_bits_t *bits,
  * x               Quantization scaled spectrum coefficients
  * return          Noise factor (0 to 7)
  */
-static int estimate_noise(enum lc3_dt dt, enum lc3_bandwidth bw,
+LC3_HOT static int estimate_noise(enum lc3_dt dt, enum lc3_bandwidth bw,
     const int16_t *xq, int nq, const float *x)
 {
     int bw_stop = (dt == LC3_DT_7M5 ? 60 : 80) * (1 + bw);
@@ -665,7 +667,7 @@ static int estimate_noise(enum lc3_dt dt, enum lc3_bandwidth bw,
  * g               Quantization gain
  * x, nq           Spectral quantized, and count of significants
  */
-static void fill_noise(enum lc3_dt dt, enum lc3_bandwidth bw,
+LC3_HOT static void fill_noise(enum lc3_dt dt, enum lc3_bandwidth bw,
     int nf, uint16_t nf_seed, float g, float *x, int nq)
 {
     int bw_stop = (dt == LC3_DT_7M5 ? 60 : 80) * (1 + bw);
diff --git a/src/tns.c b/src/tns.c
index fb7dc9c..0032b4f 100644
--- a/src/tns.c
+++ b/src/tns.c
@@ -39,7 +39,7 @@ static bool resolve_lpc_weighting(enum lc3_dt dt, int nbytes)
  * a, b, n         The 2 vectors of size `n`
  * return          sum( a[i] * b[i] ), i = [0..n-1]
  */
-static inline float dot(const float *a, const float *b, int n)
+LC3_HOT static inline float dot(const float *a, const float *b, int n)
 {
     float v = 0;
 
@@ -55,7 +55,8 @@ static inline float dot(const float *a, const float *b, int n)
  * x               Spectral coefficients
  * gain, a         Output the prediction gains and LPC coefficients
  */
-static void compute_lpc_coeffs(enum lc3_dt dt, enum lc3_bandwidth bw,
+LC3_HOT static void compute_lpc_coeffs(
+    enum lc3_dt dt, enum lc3_bandwidth bw,
     const float *x, float *gain, float (*a)[9])
 {
     static const int sub_7m5_nb[]   = {  9, 26,  43,  60 };
@@ -148,7 +149,7 @@ static void compute_lpc_coeffs(enum lc3_dt dt, enum lc3_bandwidth bw,
  * LPC Weighting
  * gain, a         Prediction gain and LPC coefficients, weighted as output
  */
-static void lpc_weighting(float pred_gain, float *a)
+LC3_HOT static void lpc_weighting(float pred_gain, float *a)
 {
     float gamma = 1.f - (1.f - 0.85f) * (2.f - pred_gain) / (2.f - 1.5f);
     float g = 1.f;
@@ -162,7 +163,7 @@ static void lpc_weighting(float pred_gain, float *a)
  * a               LPC coefficients
  * rc              Output refelection coefficients
  */
-static void lpc_reflection(const float *a, float *rc)
+LC3_HOT static void lpc_reflection(const float *a, float *rc)
 {
     float e, b[2][7], *b0, *b1;
 
@@ -251,7 +252,7 @@ static void unquantize_rc(const int *rc_q, int rc_order, float rc[8])
  * rc_order, rc    Order of coefficients, and coefficients
  * x               Spectral coefficients, filtered as output
  */
-static void forward_filtering(
+LC3_HOT static void forward_filtering(
     enum lc3_dt dt, enum lc3_bandwidth bw,
     const int rc_order[2], const float rc[2][8], float *x)
 {
@@ -292,7 +293,7 @@ static void forward_filtering(
  * rc_order, rc    Order of coefficients, and unquantized coefficients
  * x               Spectral coefficients, filtered as output
  */
-static void inverse_filtering(
+LC3_HOT static void inverse_filtering(
     enum lc3_dt dt, enum lc3_bandwidth bw,
     const int rc_order[2], const float rc[2][8], float *x)
 {