int CNAME(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b, BLASLONG incb, FLOAT *buffer) {
BLASLONG i, is, min_i;
#if (TRANSA == 2) || (TRANSA == 4)
OPENBLAS_COMPLEX_FLOAT temp;
#endif
#ifndef UNIT
FLOAT atemp1, atemp2, btemp1, btemp2;
#endif
FLOAT *gemvbuffer = (FLOAT *)buffer;
FLOAT *B = b;
if (incb != 1) {
B = buffer;
gemvbuffer = (FLOAT *)(((BLASLONG)buffer + m * sizeof(FLOAT) * 2 + 4095) & ~4095);
COPY_K(m, b, incb, buffer, 1);
}
for (is =0; is < m; is += DTB_ENTRIES) {
min_i = MIN(m - is, DTB_ENTRIES);
#if (TRANSA) == 1 || (TRANSA == 3)
if (is > 0) {
#if TRANSA == 1
GEMV_N(is, min_i, 0, dp1, ZERO,
a + is * lda * 2, lda,
B + is * 2, 1,
B, 1, gemvbuffer);
#else
GEMV_R(is, min_i, 0, dp1, ZERO,
a + is * lda * 2, lda,
B + is * 2, 1,
B, 1, gemvbuffer);
#endif
}
#endif
for (i = 0; i < min_i; i++) {
FLOAT *AA = a + (is + (i + is) * lda) * 2;
FLOAT *BB = B + is * 2;
#if (TRANSA == 1) || (TRANSA == 3)
#if TRANSA == 1
if (i > 0) AXPYU_K (i, 0, 0, BB[i * 2 + 0], BB[i * 2 + 1],
AA, 1, BB, 1, NULL, 0);
#else
if (i > 0) AXPYC_K(i, 0, 0, BB[i * 2 + 0], BB[i * 2 + 1],
AA, 1, BB, 1, NULL, 0);
#endif
#endif
#ifndef UNIT
atemp1 = AA[i * 2 + 0];
atemp2 = AA[i * 2 + 1];
btemp1 = BB[i * 2 + 0];
btemp2 = BB[i * 2 + 1];
#if (TRANSA == 1) || (TRANSA == 2)
BB[i * 2 + 0] = atemp1 * btemp1 - atemp2 * btemp2;
BB[i * 2 + 1] = atemp1 * btemp2 + atemp2 * btemp1;
#else
BB[i * 2 + 0] = atemp1 * btemp1 + atemp2 * btemp2;
BB[i * 2 + 1] = atemp1 * btemp2 - atemp2 * btemp1;
#endif
#endif
#if (TRANSA == 2) || (TRANSA == 4)
if (i < min_i - 1) {
#if TRANSA == 2
temp = DOTU_K(min_i - i - 1,
AA + (i + 1) * 2, 1,
BB + (i + 1) * 2, 1);
#else
temp = DOTC_K(min_i - i - 1,
AA + (i + 1) * 2, 1,
BB + (i + 1) * 2, 1);
#endif
BB[i * 2 + 0] += CREAL(temp);
BB[i * 2 + 1] += CIMAG(temp);
}
#endif
}
#if (TRANSA) == 2 || (TRANSA == 4)
if (m - is > min_i) {
#if TRANSA == 2
GEMV_T(m - is - min_i, min_i, 0, dp1, ZERO,
a + (is + min_i + is * lda) * 2, lda,
B + (is + min_i) * 2, 1,
B + is * 2, 1, gemvbuffer);
#else
GEMV_C(m - is - min_i, min_i, 0, dp1, ZERO,
a + (is + min_i + is * lda) * 2, lda,
B + (is + min_i) * 2, 1,
B + is * 2, 1, gemvbuffer);
#endif
}
#endif
}
if (incb != 1) {
COPY_K(m, buffer, 1, b, incb);
}
return 0;
}
int CNAME(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b, BLASLONG incb, void *buffer){
BLASLONG i, is, min_i;
#if (TRANSA == 2) || (TRANSA == 4)
FLOAT _Complex result;
#endif
#ifndef UNIT
FLOAT ar, ai, br, bi, ratio, den;
#endif
FLOAT *gemvbuffer = (FLOAT *)buffer;
FLOAT *B = b;
if (incb != 1) {
B = buffer;
gemvbuffer = (FLOAT *)(((BLASLONG)buffer + m * sizeof(FLOAT) * 2 + 4095) & ~4095);
COPY_K(m, b, incb, buffer, 1);
}
for (is = m; is > 0; is -= DTB_ENTRIES){
min_i = MIN(is, DTB_ENTRIES);
#if (TRANSA == 2) || (TRANSA == 4)
if (m - is > 0){
#if TRANSA == 2
GEMV_T(m - is, min_i, 0, dm1, ZERO,
a + (is + (is - min_i) * lda) * COMPSIZE, lda,
B + is * COMPSIZE, 1,
B + (is - min_i) * COMPSIZE, 1, gemvbuffer);
#else
GEMV_C(m - is, min_i, 0, dm1, ZERO,
a + (is + (is - min_i) * lda) * COMPSIZE, lda,
B + is * COMPSIZE, 1,
B + (is - min_i) * COMPSIZE, 1, gemvbuffer);
#endif
}
#endif
for (i = 0; i < min_i; i++) {
FLOAT *AA = a + ((is - i - 1) + (is - i - 1) * lda) * COMPSIZE;
FLOAT *BB = B + (is - i - 1) * COMPSIZE;
#if (TRANSA == 2) || (TRANSA == 4)
if (i > 0) {
#if TRANSA == 2
result = DOTU_K(i, AA + 2, 1, BB + 2, 1);
#else
result = DOTC_K(i, AA + 2, 1, BB + 2, 1);
#endif
BB[0] -= CREAL(result);
BB[1] -= CIMAG(result);
}
#endif
#ifndef UNIT
ar = AA[0];
ai = AA[1];
if (fabs(ar) >= fabs(ai)){
ratio = ai / ar;
den = 1./(ar * ( 1 + ratio * ratio));
ar = den;
#if TRANSA < 3
ai = -ratio * den;
#else
ai = ratio * den;
#endif
} else {
ratio = ar / ai;
den = 1./(ai * ( 1 + ratio * ratio));
ar = ratio * den;
#if TRANSA < 3
ai = -den;
#else
ai = den;
#endif
}
br = BB[0];
bi = BB[1];
BB[0] = ar*br - ai*bi;
BB[1] = ar*bi + ai*br;
#endif
#if (TRANSA == 1) || (TRANSA == 3)
if (i < min_i - 1) {
#if TRANSA == 1
AXPYU_K (min_i - i - 1, 0, 0, - BB[0], -BB[1],
AA - (min_i - i - 1) * COMPSIZE, 1, BB - (min_i - i - 1) * COMPSIZE, 1, NULL, 0);
#else
AXPYC_K(min_i - i - 1, 0, 0, - BB[0], -BB[1],
AA - (min_i - i - 1) * COMPSIZE, 1, BB - (min_i - i - 1) * COMPSIZE, 1, NULL, 0);
#endif
}
#endif
}
#if (TRANSA == 1) || (TRANSA == 3)
if (is - min_i > 0){
#if TRANSA == 1
GEMV_N(is - min_i, min_i, 0, dm1, ZERO,
a + (is - min_i) * lda * COMPSIZE, lda,
B + (is - min_i) * COMPSIZE, 1,
B, 1, gemvbuffer);
#else
GEMV_R(is - min_i, min_i, 0, dm1, ZERO,
a + (is - min_i) * lda * COMPSIZE, lda,
B + (is - min_i) * COMPSIZE, 1,
B, 1, gemvbuffer);
#endif
}
#endif
}
if (incb != 1) {
COPY_K(m, buffer, 1, b, incb);
}
return 0;
}
int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha_r, FLOAT alpha_i,
FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG incx, FLOAT *y, BLASLONG incy, FLOAT *buffer){
BLASLONG is, min_i;
FLOAT *X = x;
FLOAT *Y = y;
FLOAT *symbuffer = buffer;
FLOAT *gemvbuffer = (FLOAT *)(((BLASLONG)buffer + SYMV_P * SYMV_P * sizeof(FLOAT) * 2 + 4095) & ~4095);
FLOAT *bufferY = gemvbuffer;
FLOAT *bufferX = gemvbuffer;
if (incy != 1) {
Y = bufferY;
bufferX = (FLOAT *)(((BLASLONG)bufferY + m * sizeof(FLOAT) * 2 + 4095) & ~4095);
gemvbuffer = bufferX;
COPY_K(m, y, incy, Y, 1);
}
if (incx != 1) {
X = bufferX;
gemvbuffer = (FLOAT *)(((BLASLONG)bufferX + m * sizeof(FLOAT) * 2 + 4095) & ~4095);
COPY_K(m, x, incx, X, 1);
}
#ifndef LOWER
for(is = m - offset; is < m; is += SYMV_P){
min_i = MIN(m - is, SYMV_P);
#else
for(is = 0; is < offset; is += SYMV_P){
min_i = MIN(offset - is, SYMV_P);
#endif
#ifndef LOWER
if (is > 0){
#ifndef HEMVREV
GEMV_C(is, min_i, 0, alpha_r, alpha_i,
a + is * lda * 2, lda,
X, 1,
Y + is * 2, 1, gemvbuffer);
GEMV_N(is, min_i, 0, alpha_r, alpha_i,
a + is * lda * 2, lda,
X + is * 2, 1,
Y, 1, gemvbuffer);
#else
GEMV_T(is, min_i, 0, alpha_r, alpha_i,
a + is * lda * 2, lda,
X, 1,
Y + is * 2, 1, gemvbuffer);
GEMV_R(is, min_i, 0, alpha_r, alpha_i,
a + is * lda * 2, lda,
X + is * 2, 1,
Y, 1, gemvbuffer);
#endif
}
#endif
#ifndef HEMVREV
#ifdef LOWER
ZHEMCOPY_L(min_i, a + (is + is * lda) * 2, lda, symbuffer);
#else
ZHEMCOPY_U(min_i, a + (is + is * lda) * 2, lda, symbuffer);
#endif
#else
#ifdef LOWER
ZHEMCOPY_M(min_i, a + (is + is * lda) * 2, lda, symbuffer);
#else
ZHEMCOPY_V(min_i, a + (is + is * lda) * 2, lda, symbuffer);
#endif
#endif
GEMV_N(min_i, min_i, 0, alpha_r, alpha_i,
symbuffer, min_i,
X + is * 2, 1,
Y + is * 2, 1, gemvbuffer);
#ifdef LOWER
if (m - is - min_i > 0){
#ifndef HEMVREV
GEMV_C(m - is - min_i, min_i, 0, alpha_r, alpha_i,
a + ((is + min_i) + is * lda) * 2, lda,
X + (is + min_i) * 2, 1,
Y + is * 2, 1, gemvbuffer);
GEMV_N(m - is - min_i, min_i, 0, alpha_r, alpha_i,
a + ((is + min_i) + is * lda) * 2, lda,
X + is * 2, 1,
Y + (is + min_i) * 2, 1, gemvbuffer);
#else
GEMV_T(m - is - min_i, min_i, 0, alpha_r, alpha_i,
a + ((is + min_i) + is * lda) * 2, lda,
X + (is + min_i) * 2, 1,
Y + is * 2, 1, gemvbuffer);
GEMV_R(m - is - min_i, min_i, 0, alpha_r, alpha_i,
a + ((is + min_i) + is * lda) * 2, lda,
X + is * 2, 1,
Y + (is + min_i) * 2, 1, gemvbuffer);
#endif
}
#endif
} /* end of is */
if (incy != 1) {
COPY_K(m, Y, 1, y, incy);
}
return 0;
}
