cumo 0.1.0

data/ext/cumo/include/cumo/template_kernel.h

@@ -0,0 +1,77 @@
+#ifndef CUMO_TEMPLATE_KERNEL_H
+#define CUMO_TEMPLATE_KERNEL_H
+
+#define LOAD_BIT( adr, pos, val )                       \
+    {                                                   \
+        size_t dig = (size_t)(pos) / NB;                \
+        int    bit = (size_t)(pos) % NB;                \
+        val = (((BIT_DIGIT*)(adr))[dig]>>(bit)) & 1u;   \
+    }
+
+#define LOAD_BIT_STEP( adr, pos, step, idx, val )       \
+    {                                                   \
+        size_t dig; int bit;                            \
+        if (idx) {                                      \
+            dig = (size_t)((pos) + *(idx)) / NB;        \
+            bit = (size_t)((pos) + *(idx)) % NB;        \
+            idx++;                                      \
+        } else {                                        \
+            dig = (size_t)(pos) / NB;                   \
+            bit = (size_t)(pos) % NB;                   \
+            pos += step;                                \
+        }                                               \
+        val = (((BIT_DIGIT*)(adr))[dig]>>bit) & 1u;     \
+    }
+
+#define STORE_BIT(adr,pos,val)                                     \
+    {                                                              \
+        size_t dig = (size_t)(pos) / NB;                           \
+        int    bit = (size_t)(pos) % NB;                           \
+        if (val) {                                                 \
+            atomicOr((BIT_DIGIT*)(adr) + (dig), (val)<<(bit));     \
+        } else {                                                   \
+            atomicAnd((BIT_DIGIT*)(adr) + (dig), ~(1u<<(bit)));    \
+        }                                                          \
+    }
+// val -> val&1 ??
+
+#define STORE_BIT_STEP( adr, pos, step, idx, val )                 \
+    {                                                              \
+        size_t dig; int bit;                                       \
+        if (idx) {                                                 \
+            dig = (size_t)((pos) + *(idx)) / NB;                   \
+            bit = (size_t)((pos) + *(idx)) % NB;                   \
+            idx++;                                                 \
+        } else {                                                   \
+            dig = (size_t)(pos) / NB;                              \
+            bit = (size_t)(pos) % NB;                              \
+            pos += step;                                           \
+        }                                                          \
+        if (val) {                                                 \
+            atomicOr((BIT_DIGIT*)(adr) + (dig), (val)<<(bit));     \
+        } else {                                                   \
+            atomicAnd((BIT_DIGIT*)(adr) + (dig), ~((1u)<<(bit)));  \
+        }                                                          \
+    }
+// val -> val&1 ??
+
+#define MAX_BLOCK_DIM 128
+#define MAX_GRID_DIM 2147483647 // ref. http://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#compute-capabilities
+
+static inline size_t
+get_gridDim(size_t n)
+{
+    size_t gridDim = (n / MAX_BLOCK_DIM) + 1;
+    if (gridDim > MAX_GRID_DIM) gridDim = MAX_GRID_DIM;
+    return gridDim;
+}
+
+static inline size_t
+get_blockDim(size_t n)
+{
+    size_t blockDim = (n > MAX_BLOCK_DIM) ? MAX_BLOCK_DIM : n;
+    return blockDim;
+}
+
+
+#endif /* ifndef CUMO_TEMPLATE_KERNEL_H */

data/ext/cumo/include/cumo/types/bit.h

@@ -0,0 +1,40 @@
+typedef BIT_DIGIT dtype;
+typedef BIT_DIGIT rtype;
+#define cT  cumo_cBit
+#define cRT cT
+
+#define m_zero 0
+#define m_one  1
+
+#define m_abs(x)     (x)
+#define m_sign(x)    (((x)==0) ? 0:1)
+
+#define m_from_double(x) (((x)==0) ? 0 : 1)
+#define m_from_real(x) (((x)==0) ? 0 : 1)
+#define m_from_sint(x) (((x)==0) ? 0 : 1)
+#define m_from_int32(x) (((x)==0) ? 0 : 1)
+#define m_from_int64(x) (((x)==0) ? 0 : 1)
+#define m_from_uint32(x) (((x)==0) ? 0 : 1)
+#define m_from_uint64(x) (((x)==0) ? 0 : 1)
+#define m_data_to_num(x) INT2FIX(x)
+#define m_sprintf(s,x)   sprintf(s,"%1d",(int)(x))
+
+#define m_copy(x)  (x)
+#define m_not(x)   (~(x))
+#define m_and(x,y) ((x)&(y))
+#define m_or(x,y)  ((x)|(y))
+#define m_xor(x,y) ((x)^(y))
+#define m_eq(x,y)  (~((x)^(y)))
+#define m_count_true(x)  ((x)!=0)
+#define m_count_true_cpu(x)  ((x)!=0)
+#define m_count_false(x) ((x)==0)
+#define m_count_false_cpu(x) ((x)==0)
+
+static inline BIT_DIGIT m_num_to_data(VALUE num) {
+    if (RTEST(num)) {
+        if (!RTEST(rb_equal(num,INT2FIX(0)))) {
+            return 1;
+        }
+    }
+    return 0;
+}

data/ext/cumo/include/cumo/types/bit_kernel.h

@@ -0,0 +1,34 @@
+#ifndef CUMO_BIT_KERNEL_H
+#define CUMO_BIT_KERNEL_H
+
+typedef BIT_DIGIT dtype;
+typedef BIT_DIGIT rtype;
+
+#define m_zero 0
+#define m_one  1
+
+#define m_abs(x)     (x)
+#define m_sign(x)    (((x)==0) ? 0:1)
+
+#define m_from_double(x) (((x)==0) ? 0 : 1)
+#define m_from_real(x) (((x)==0) ? 0 : 1)
+#define m_from_sint(x) (((x)==0) ? 0 : 1)
+#define m_from_int32(x) (((x)==0) ? 0 : 1)
+#define m_from_int64(x) (((x)==0) ? 0 : 1)
+#define m_from_uint32(x) (((x)==0) ? 0 : 1)
+#define m_from_uint64(x) (((x)==0) ? 0 : 1)
+#define m_data_to_num(x) INT2FIX(x)
+#define m_sprintf(s,x)   sprintf(s,"%1d",(int)(x))
+
+#define m_copy(x)  (x)
+#define m_not(x)   (~(x))
+#define m_and(x,y) ((x)&(y))
+#define m_or(x,y)  ((x)|(y))
+#define m_xor(x,y) ((x)^(y))
+#define m_eq(x,y)  (~((x)^(y)))
+#define m_count_true(x)  ((x)!=0)
+#define m_count_true_cpu(x)  m_count_true(x)
+#define m_count_false(x) ((x)==0)
+#define m_count_false_cpu(x) m_count_false(x)
+
+#endif // CUMO_BIT_KERNEL_H

data/ext/cumo/include/cumo/types/complex.h

@@ -0,0 +1,402 @@
+static inline dtype c_new(rtype r, rtype i) {
+    dtype z;
+    REAL(z) = r;
+    IMAG(z) = i;
+    return z;
+}
+
+static inline dtype c_set_real(dtype x, rtype r) {
+    REAL(x)=r;
+    return x;
+}
+
+static inline dtype c_set_imag(dtype x, rtype i) {
+    IMAG(x)=i;
+    return x;
+}
+
+static inline VALUE COMP2NUM(dtype x) {
+    VALUE v;
+    v = rb_funcall(rb_intern("Kernel"), rb_intern("Complex"), 2,
+                   rb_float_new(REAL(x)), rb_float_new(IMAG(x)));
+    return v;
+}
+
+static inline dtype NUM2COMP(VALUE v) {
+    dtype z;
+    REAL(z) = NUM2DBL(rb_funcall(v,id_real,0));
+    IMAG(z) = NUM2DBL(rb_funcall(v,id_imag,0));
+    return z;
+}
+
+#define c_is_zero(x) (REAL(x)==0 && IMAG(x)==0)
+#define c_eq(x,y) (REAL(x)==REAL(y) && IMAG(x)==IMAG(y))
+#define c_ne(x,y) (REAL(x)!=REAL(y) || IMAG(x)!=IMAG(y))
+#define c_isnan(x) (isnan(REAL(x)) || isnan(IMAG(x)))
+#define c_isinf(x) (isinf(REAL(x)) || isinf(IMAG(x)))
+#define c_isposinf(x) ((isinf(REAL(x)) && signbit(REAL(x))==0) || \
+                       (isinf(IMAG(x)) && signbit(IMAG(x))==0))
+#define c_isneginf(x) ((isinf(REAL(x)) && signbit(REAL(x))) || \
+                       (isinf(IMAG(x)) && signbit(IMAG(x))))
+#define c_isfinite(x) (isfinite(REAL(x)) && isfinite(IMAG(x)))
+
+static inline dtype c_zero() {
+    dtype z;
+    REAL(z) = 0;
+    IMAG(z) = 0;
+    return z;
+}
+
+static inline dtype c_one() {
+    dtype z;
+    REAL(z) = 1;
+    IMAG(z) = 0;
+    return z;
+}
+
+static inline dtype c_minus(dtype x) {
+    dtype z;
+    REAL(z) = -REAL(x);
+    IMAG(z) = -IMAG(x);
+    return z;
+}
+
+static inline dtype c_im(dtype x) {
+    dtype z;
+    REAL(z) = -IMAG(x);
+    IMAG(z) = REAL(x);
+    return z;
+}
+
+static inline dtype c_add(dtype x, dtype y) {
+    dtype z;
+    REAL(z) = REAL(x)+REAL(y);
+    IMAG(z) = IMAG(x)+IMAG(y);
+    return z;
+}
+
+static inline dtype c_sub(dtype x, dtype y) {
+    dtype z;
+    REAL(z) = REAL(x)-REAL(y);
+    IMAG(z) = IMAG(x)-IMAG(y);
+    return z;
+}
+
+
+static inline dtype c_mul(dtype x, dtype y) {
+    dtype z;
+    REAL(z) = REAL(x)*REAL(y)-IMAG(x)*IMAG(y);
+    IMAG(z) = REAL(x)*IMAG(y)+IMAG(x)*REAL(y);
+    return z;
+}
+
+static inline dtype c_mul_r(dtype x, rtype y) {
+    dtype z;
+    REAL(z) = REAL(x)*y;
+    IMAG(z) = IMAG(x)*y;
+    return z;
+}
+
+static inline dtype c_div(dtype x, dtype y) {
+    dtype z;
+    rtype s,yr,yi;
+    s  = r_hypot(REAL(y),IMAG(y));
+    yr = REAL(y)/s;
+    yi = IMAG(y)/s;
+    REAL(z) = (REAL(x)*yr+IMAG(x)*yi)/s;
+    IMAG(z) = (IMAG(x)*yr-REAL(x)*yi)/s;
+    return z;
+}
+
+static inline dtype c_div_r(dtype x, rtype y) {
+    dtype z;
+    REAL(z) = REAL(x)/y;
+    IMAG(z) = IMAG(x)/y;
+    return z;
+}
+
+static inline dtype c_reciprocal(dtype x) {
+    dtype z;
+    if ( r_abs(REAL(x)) > r_abs(IMAG(x)) ) {
+        IMAG(z) = IMAG(x)/REAL(x);
+        REAL(z) = (1+IMAG(z)*IMAG(z))*REAL(x);
+        IMAG(z) /= -REAL(z);
+        REAL(z) = 1/REAL(z);
+    } else {
+        REAL(z) = REAL(x)/IMAG(x);
+        IMAG(z) = (1+REAL(z)*REAL(z))*IMAG(x);
+        REAL(z) /= IMAG(z);
+        IMAG(z) = -1/IMAG(z);
+    }
+    return z;
+}
+
+static inline dtype c_square(dtype x) {
+    dtype z;
+    REAL(z) = REAL(x)*REAL(x)-IMAG(x)*IMAG(x);
+    IMAG(z) = 2*REAL(x)*IMAG(x);
+    return z;
+}
+
+static inline dtype c_sqrt(dtype x) {
+    dtype z;
+    rtype xr, xi, r;
+    xr = REAL(x)/2;
+    xi = IMAG(x)/2;
+    r  = r_hypot(xr,xi);
+    if (xr>0) {
+        REAL(z) = sqrt(r+xr);
+        IMAG(z) = xi/REAL(z);
+    } else if ( (r-=xr)!=0 ) {
+        IMAG(z) = (xi>=0) ? sqrt(r):-sqrt(r);
+        REAL(z) = xi/IMAG(z);
+    } else {
+        REAL(z) = IMAG(z) = 0;
+    }
+    return z;
+}
+
+static inline dtype c_log(dtype x) {
+    dtype z;
+    REAL(z) = r_log(r_hypot(REAL(x),IMAG(x)));
+    IMAG(z) = r_atan2(IMAG(x),REAL(x));
+    return z;
+}
+
+static inline dtype c_log2(dtype x) {
+    dtype z;
+    z = c_log(x);
+    z = c_mul_r(x,M_LOG2E);
+    return z;
+}
+
+static inline dtype c_log10(dtype x) {
+    dtype z;
+    z = c_log(x);
+    z = c_mul_r(x,M_LOG10E);
+    return z;
+}
+
+static inline dtype c_exp(dtype x) {
+    dtype z;
+    rtype a = r_exp(REAL(x));
+    REAL(z) = a*r_cos(IMAG(x));
+    IMAG(z) = a*r_sin(IMAG(x));
+    return z;
+}
+
+static inline dtype c_exp2(dtype x) {
+    dtype z;
+    rtype a = r_exp(REAL(x)*M_LN2);
+    REAL(z) = a*r_cos(IMAG(x));
+    IMAG(z) = a*r_sin(IMAG(x));
+    return z;
+}
+
+static inline dtype c_exp10(dtype x) {
+    dtype z;
+    rtype a = r_exp(REAL(x)*M_LN10);
+    REAL(z) = a*r_cos(IMAG(x));
+    IMAG(z) = a*r_sin(IMAG(x));
+    return z;
+}
+
+static inline dtype c_sin(dtype x) {
+    dtype z;
+    REAL(z) = r_sin(REAL(x))*r_cosh(IMAG(x));
+    IMAG(z) = r_cos(REAL(x))*r_sinh(IMAG(x));
+    return z;
+}
+
+static inline dtype c_sinh(dtype x) {
+    dtype z;
+    REAL(z) = r_sinh(REAL(x))*r_cos(IMAG(x));
+    IMAG(z) = r_cosh(REAL(x))*r_sin(IMAG(x));
+    return z;
+}
+
+static inline dtype c_cos(dtype x) {
+    dtype z;
+    REAL(z) = r_cos(REAL(x))*r_cosh(IMAG(x));
+    IMAG(z) = -r_sin(REAL(x))*r_sinh(IMAG(x));
+    return z;
+}
+
+static inline dtype c_cosh(dtype x) {
+    dtype z;
+    REAL(z) = r_cosh(REAL(x))*r_cos(IMAG(x));
+    IMAG(z) = r_sinh(REAL(x))*r_sin(IMAG(x));
+    return z;
+}
+
+static inline dtype c_tan(dtype x) {
+    dtype z;
+    rtype c, d;
+    if (r_abs(IMAG(x))<1) {
+        c = r_cos(REAL(x));
+        d = r_sinh(IMAG(x));
+        d = c*c + d*d;
+        REAL(z) = 0.5*r_sin(2*REAL(x))/d;
+        IMAG(z) = 0.5*r_sinh(2*IMAG(x))/d;
+    } else {
+        d = r_exp(-IMAG(x));
+        c = 2*d/(1-d*d);
+        c = c*c;
+        d = r_cos(REAL(x));
+        d = 1.0 + d*d*c;
+        REAL(z) = 0.5*r_sin(2*REAL(x))*c/d;
+        IMAG(z) = 1/r_tanh(IMAG(x))/d;
+    }
+    return z;
+}
+
+static inline dtype c_tanh(dtype x) {
+    dtype z;
+    rtype c, d, s;
+    c = r_cos(IMAG(x));
+    s = r_sinh(REAL(x));
+    d = c*c + s*s;
+    if (r_abs(REAL(x))<1) {
+        REAL(z) = s*r_cosh(REAL(x))/d;
+        IMAG(z) = 0.5*r_sin(2*IMAG(x))/d;
+    } else {
+        c = c / s;
+        c = 1 + c*c;
+        REAL(z) = 1/(r_tanh(REAL(x))*c);
+        IMAG(z) = 0.5*r_sin(2*IMAG(x))/d;
+    }
+    return z;
+}
+
+static inline dtype c_asin(dtype x) {
+    dtype z, y;
+    y = c_square(x);
+    REAL(y) = 1-REAL(y);
+    IMAG(y) = -IMAG(y);
+    y = c_sqrt(y);
+    REAL(y) -= IMAG(x);
+    IMAG(y) += REAL(x);
+    y = c_log(y);
+    REAL(z) = IMAG(y);
+    IMAG(z) = -REAL(y);
+    return z;
+}
+
+static inline dtype c_asinh(dtype x) {
+    dtype z, y;
+    y = c_square(x);
+    REAL(y) += 1;
+    y = c_sqrt(y);
+    REAL(y) += REAL(x);
+    IMAG(y) += IMAG(x);
+    z = c_log(y);
+    return z;
+}
+
+static inline dtype c_acos(dtype x) {
+    dtype z, y;
+    y = c_square(x);
+    REAL(y) = 1-REAL(y);
+    IMAG(y) = -IMAG(y);
+    y = c_sqrt(y);
+    REAL(z) = REAL(x)-IMAG(y);
+    IMAG(z) = IMAG(x)+REAL(y);
+    y = c_log(z);
+    REAL(z) = IMAG(y);
+    IMAG(z) = -REAL(y);
+    return z;
+}
+
+static inline dtype c_acosh(dtype x) {
+    dtype z, y;
+    y = c_square(x);
+    REAL(y) -= 1;
+    y = c_sqrt(y);
+    REAL(y) += REAL(x);
+    IMAG(y) += IMAG(x);
+    z = c_log(y);
+    return z;
+}
+
+static inline dtype c_atan(dtype x) {
+    dtype z, y;
+    REAL(y) = -REAL(x);
+    IMAG(y) = 1-IMAG(x);
+    REAL(z) = REAL(x);
+    IMAG(z) = 1+IMAG(x);
+    y = c_div(z,y);
+    y = c_log(y);
+    REAL(z) = -IMAG(y)/2;
+    IMAG(z) = REAL(y)/2;
+    return z;
+}
+
+static inline dtype c_atanh(dtype x) {
+    dtype z, y;
+    REAL(y) = 1-REAL(x);
+    IMAG(y) = -IMAG(x);
+    REAL(z) = 1+REAL(x);
+    IMAG(z) = IMAG(x);
+    y = c_div(z,y);
+    y = c_log(y);
+    REAL(z) = REAL(y)/2;
+    IMAG(z) = IMAG(y)/2;
+    return z;
+}
+
+static inline dtype c_pow(dtype x, dtype y)
+{
+    dtype z;
+    if (c_is_zero(y)) {
+        z = c_one();
+    } else if (c_is_zero(x) && REAL(y)>0 && IMAG(y)==0) {
+        z = c_zero();
+    } else {
+        z = c_log(x);
+        z = c_mul(y,z);
+        z = c_exp(z);
+    }
+    return z;
+}
+
+static inline dtype c_pow_int(dtype x, int p)
+{
+    dtype z = c_one();
+    if (p<0) {
+	x = c_pow_int(x,-p);
+	return c_reciprocal(x);
+    }
+    if (p==2) {return c_square(x);}
+    if (p&1) {z = x;}
+    p >>= 1;
+    while (p) {
+	x = c_square(x);
+	if (p&1) z = c_mul(z,x);
+	p >>= 1;
+    }
+    return z;
+}
+
+static inline dtype c_cbrt(dtype x) {
+    dtype z;
+    z = c_log(x);
+    z = c_div_r(z,3);
+    z = c_exp(z);
+    return z;
+}
+
+static inline rtype c_abs(dtype x) {
+    return r_hypot(REAL(x),IMAG(x));
+}
+
+static inline rtype c_abs_square(dtype x) {
+    return REAL(x)*REAL(x)+IMAG(x)*IMAG(x);
+}
+
+
+
+/*
+static inline rtype c_hypot(dtype x, dtype y) {
+    return r_hypot(c_abs(x),c_abs(y));
+}
+*/