int CNAME(BLASLONG m, FLOAT alpha_r, FLOAT *x, BLASLONG incx,
FLOAT *y, BLASLONG incy, FLOAT *a, FLOAT *buffer){
BLASLONG i;
FLOAT *X, *Y;
X = x;
Y = y;
if (incx != 1) {
COPY_K(m, x, incx, buffer, 1);
X = buffer;
}
if (incy != 1) {
COPY_K(m, y, incy, (FLOAT *)((BLASLONG)buffer + (BUFFER_SIZE / 2)), 1);
Y = (FLOAT *)((BLASLONG)buffer + (BUFFER_SIZE / 2));
}
for (i = 0; i < m; i++){
#ifndef LOWER
AXPYU_K(i + 1, 0, 0, alpha_r * X[i], Y, 1, a, 1, NULL, 0);
AXPYU_K(i + 1, 0, 0, alpha_r * Y[i], X, 1, a, 1, NULL, 0);
a += i + 1;
#else
AXPYU_K(m - i, 0, 0, alpha_r * X[i], Y + i, 1, a, 1, NULL, 0);
AXPYU_K(m - i, 0, 0, alpha_r * Y[i], X + i, 1, a, 1, NULL, 0);
a += m - i;
#endif
}
return 0;
}
int CNAME(BLASLONG m, FLOAT alpha_r,
FLOAT *x, BLASLONG incx, FLOAT *a, FLOAT *buffer){
BLASLONG i;
FLOAT *X;
X = x;
if (incx != 1) {
COPY_K(m, x, incx, buffer, 1);
X = buffer;
}
for (i = 0; i < m; i++){
#ifndef LOWER
if (X[i] != ZERO) {
AXPYU_K(i + 1, 0, 0, alpha_r * X[i], X, 1, a, 1, NULL, 0);
}
a += i + 1;
#else
if (X[i] != ZERO) {
AXPYU_K(m - i, 0, 0, alpha_r * X[i], X + i, 1, a, 1, NULL, 0);
}
a += m - i;
#endif
}
return 0;
}
int CNAME(BLASLONG m, FLOAT alpha_r, FLOAT alpha_i,
FLOAT *x, BLASLONG incx, FLOAT *a, FLOAT *buffer){
BLASLONG i;
FLOAT *X;
X = x;
if (incx != 1) {
COPY_K(m, x, incx, buffer, 1);
X = buffer;
}
for (i = 0; i < m; i++){
#ifndef LOWER
if ((X[i * 2 + 0] != ZERO) && (X[i * 2 + 1] != ZERO)) {
AXPYU_K(i + 1, 0, 0,
alpha_r * X[i * 2 + 0] - alpha_i * X[i * 2 + 1],
alpha_i * X[i * 2 + 0] + alpha_r * X[i * 2 + 1],
X, 1, a, 1, NULL, 0);
}
a += (i + 1) * 2;
#else
if ((X[i * 2 + 0] != ZERO) && (X[i * 2 + 1] != ZERO)) {
AXPYU_K(m - i, 0, 0,
alpha_r * X[i * 2 + 0] - alpha_i * X[i * 2 + 1],
alpha_i * X[i * 2 + 0] + alpha_r * X[i * 2 + 1],
X + i * 2, 1, a, 1, NULL, 0);
}
a += (m - i) * 2;
#endif
}
return 0;
}
int CNAME(BLASLONG m, FLOAT *a, BLASLONG lda, FLOAT *b, BLASLONG incb, void *buffer){
BLASLONG i, is, min_i;
FLOAT *gemvbuffer = (FLOAT *)buffer;
FLOAT *B = b;
if (incb != 1) {
B = buffer;
gemvbuffer = (FLOAT *)(((BLASLONG)buffer + m * sizeof(FLOAT) + 4095) & ~4095);
COPY_K(m, b, incb, buffer, 1);
}
for (is = m; is > 0; is -= DTB_ENTRIES){
min_i = MIN(is, DTB_ENTRIES);
#ifdef TRANSA
if (m - is > 0){
GEMV_T(m - is, min_i, 0, dm1,
a + is + (is - min_i) * lda, lda,
B + is, 1,
B + is - min_i, 1, gemvbuffer);
}
#endif
for (i = 0; i < min_i; i++) {
FLOAT *AA = a + (is - i - 1) + (is - i - 1) * lda;
FLOAT *BB = B + (is - i - 1);
#ifdef TRANSA
if (i > 0) BB[0] -= DOTU_K(i, AA + 1, 1, BB + 1, 1);
#endif
#ifndef UNIT
BB[0] /= AA[0];
#endif
#ifndef TRANSA
if (i < min_i - 1) AXPYU_K(min_i - i - 1, 0, 0, -BB[0], AA - (min_i - i - 1), 1, BB - (min_i - i - 1), 1, NULL, 0);
#endif
}
#ifndef TRANSA
if (is - min_i > 0){
GEMV_N(is - min_i, min_i, 0, dm1,
a + (is - min_i) * lda, lda,
B + is - min_i, 1,
B, 1, gemvbuffer);
}
#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, FLOAT *buffer){
BLASLONG i, is, min_i;
FLOAT *gemvbuffer = (FLOAT *)buffer;
FLOAT *B = b;
if (incb != 1) {
B = buffer;
gemvbuffer = (FLOAT *)(((BLASLONG)buffer + m * sizeof(FLOAT) + 4095) & ~4095);
COPY_K(m, b, incb, buffer, 1);
}
for (is = 0; is < m; is += DTB_ENTRIES){
min_i = MIN(m - is, DTB_ENTRIES);
#ifndef TRANSA
if (is > 0){
GEMV_N(is, min_i, 0, dp1,
a + is * lda, lda,
B + is, 1,
B, 1, gemvbuffer);
}
#endif
for (i = 0; i < min_i; i++) {
FLOAT *AA = a + is + (i + is) * lda;
FLOAT *BB = B + is;
#ifndef TRANSA
if (i > 0) AXPYU_K(i, 0, 0, BB[i], AA, 1, BB, 1, NULL, 0);
#endif
#ifndef UNIT
BB[i] *= AA[i];
#endif
#ifdef TRANSA
if (i < min_i - 1) BB[i] += DOTU_K(min_i - i - 1, AA + i + 1, 1, BB + i + 1, 1);
#endif
}
#ifdef TRANSA
if (m - is > min_i){
GEMV_T(m - is - min_i, min_i, 0, dp1,
a + is + min_i + is * lda, lda,
B + is + min_i, 1,
B + is, 1, gemvbuffer);
}
#endif
}
if (incb != 1) {
COPY_K(m, buffer, 1, b, incb);
}
return 0;
}
int CNAME(BLASLONG n, BLASLONG k, FLOAT *a, BLASLONG lda, FLOAT *b, BLASLONG incb, void *buffer){
BLASLONG i;
FLOAT *gemvbuffer = (FLOAT *)buffer;
FLOAT *B = b;
BLASLONG length;
if (incb != 1) {
B = buffer;
gemvbuffer = (FLOAT *)(((BLASLONG)buffer + n * sizeof(FLOAT) + 4095) & ~4095);
COPY_K(n, b, incb, buffer, 1);
}
a += (n - 1) * lda;
for (i = n - 1; i >= 0; i--) {
#ifdef TRANSA
length = n - i - 1;
if (length > k) length = k;
if (length > 0) {
B[i] -= DOTU_K(length, a + 1, 1, B + i + 1, 1);
}
#endif
#ifndef UNIT
#ifdef TRANSA
B[i] /= a[0];
#else
B[i] /= a[k];
#endif
#endif
#ifndef TRANSA
length = i;
if (length > k) length = k;
if (length > 0) {
AXPYU_K(length, 0, 0,
- B[i],
a + k - length, 1, B + i - length, 1, NULL, 0);
}
#endif
a -= lda;
}
if (incb != 1) {
COPY_K(n, buffer, 1, b, incb);
}
return 0;
}
int CNAME(BLASLONG m, FLOAT alpha, FLOAT *x,
BLASLONG incx, FLOAT *a, FLOAT *buffer){
BLASLONG i;
FLOAT *X;
X = x;
if (incx != 1) {
COPY_K(m, x, incx, buffer, 1);
X = buffer;
}
for (i = 0; i < m; i++){
#ifndef HEMVREV
#ifndef LOWER
AXPYU_K(i + 1, 0, 0, alpha * X[i * 2 + 0], -alpha * X[i * 2 + 1], X, 1, a, 1, NULL, 0);
a[i * 2 + 1] = ZERO;
a += (i + 1) * 2;
#else
AXPYU_K(m - i, 0, 0, alpha * X[i * 2 + 0], -alpha * X[i * 2 + 1], X + i * 2, 1, a, 1, NULL, 0);
a[1] = ZERO;
a += (m - i) * 2;
#endif
#else
#ifndef LOWER
AXPYC_K(i + 1, 0, 0, alpha * X[i * 2 + 0], alpha * X[i * 2 + 1], X, 1, a, 1, NULL, 0);
a[i * 2 + 1] = ZERO;
a += (i + 1) * 2;
#else
AXPYC_K(m - i, 0, 0, alpha * X[i * 2 + 0], alpha * X[i * 2 + 1], X + i * 2, 1, a, 1, NULL, 0);
a[1] = ZERO;
a += (m - i) * 2;
#endif
#endif
}
return 0;
}
int CNAME(BLASLONG n, BLASLONG k, FLOAT *a, BLASLONG lda, FLOAT *b, BLASLONG incb, void *buffer){
BLASLONG i;
FLOAT *B = b;
BLASLONG length;
if (incb != 1) {
B = buffer;
COPY_K(n, b, incb, buffer, 1);
}
for (i = 0; i < n; i++) {
#ifndef TRANSA
length = i;
if (length > k) length = k;
if (length > 0) {
AXPYU_K(length, 0, 0,
B[i],
a + k - length, 1, B + i - length, 1, NULL, 0);
}
#endif
#ifndef UNIT
#ifndef TRANSA
B[i] *= a[k];
#else
B[i] *= a[0];
#endif
#endif
#ifdef TRANSA
length = n - i - 1;
if (length > k) length = k;
if (length > 0) {
B[i] += DOTU_K(length, a + 1, 1, B + i + 1, 1);
}
#endif
a += lda;
}
if (incb != 1) {
COPY_K(n, buffer, 1, b, incb);
}
return 0;
}
int CNAME(BLASLONG m, FLOAT *a, FLOAT *b, BLASLONG incb, void *buffer){
BLASLONG i;
FLOAT *B = b;
if (incb != 1) {
B = buffer;
COPY_K(m, b, incb, buffer, 1);
}
a += (m + 1) * m / 2 - 1;
for (i = 0; i < m; i++) {
#ifndef TRANSA
if (i > 0) AXPYU_K(i, 0, 0, B[m - i - 1], a + 1, 1, B + m - i, 1, NULL, 0);
#endif
#ifndef UNIT
B[m - i - 1] *= a[0];
#endif
#ifdef TRANSA
if (i < m - 1) B[m - i - 1] += DOTU_K(m - i - 1, a - (m - i - 1), 1, B, 1);
#endif
#ifndef TRANSA
a -= (i + 2);
#else
a -= (m - i);
#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, 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, FLOAT *b, BLASLONG incb, void *buffer){
BLASLONG i;
#if (TRANSA == 2) || (TRANSA == 4)
FLOAT _Complex 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);
}
a += (m + 1) * m - 2;
for (i = 0; i < m; i++) {
#if (TRANSA == 1) || (TRANSA == 3)
#if TRANSA == 1
if (i > 0) AXPYU_K (i, 0, 0,
B[(m - i - 1) * 2 + 0], B[(m - i - 1) * 2 + 1],
a + 2, 1, B + (m - i) * 2, 1, NULL, 0);
#else
if (i > 0) AXPYC_K(i, 0, 0,
B[(m - i - 1) * 2 + 0], B[(m - i - 1) * 2 + 1],
a + 2, 1, B + (m - i) * 2, 1, NULL, 0);
#endif
#endif
#ifndef UNIT
atemp1 = a[0];
atemp2 = a[1];
btemp1 = B[(m - i - 1) * 2 + 0];
btemp2 = B[(m - i - 1) * 2 + 1];
#if (TRANSA == 1) || (TRANSA == 2)
B[(m - i - 1) * 2 + 0] = atemp1 * btemp1 - atemp2 * btemp2;
B[(m - i - 1) * 2 + 1] = atemp1 * btemp2 + atemp2 * btemp1;
#else
B[(m - i - 1) * 2 + 0] = atemp1 * btemp1 + atemp2 * btemp2;
B[(m - i - 1) * 2 + 1] = atemp1 * btemp2 - atemp2 * btemp1;
#endif
#endif
#if (TRANSA == 2) || (TRANSA == 4)
if (i < m - 1) {
#if TRANSA == 2
temp = DOTU_K(m - i - 1, a - (m - i - 1) * 2, 1, B, 1);
#else
temp = DOTC_K(m - i - 1, a - (m - i - 1) * 2, 1, B, 1);
#endif
B[(m - i - 1) * 2 + 0] += CREAL(temp);
B[(m - i - 1) * 2 + 1] += CIMAG(temp);
}
#endif
#if (TRANSA == 1) || (TRANSA == 3)
a -= (i + 2) * 2;
#else
a -= (m - i) * 2;
#endif
}
if (incb != 1) {
COPY_K(m, buffer, 1, b, incb);
}
return 0;
}
int CNAME(BLASLONG m, FLOAT alpha_r, FLOAT alpha_i,
FLOAT *a, FLOAT *x, BLASLONG incx, FLOAT *y, BLASLONG incy, void *buffer){
BLASLONG i;
FLOAT *X = x;
FLOAT *Y = y;
FLOAT *gemvbuffer = (FLOAT *)buffer;
FLOAT *bufferY = gemvbuffer;
FLOAT *bufferX = gemvbuffer;
FLOAT temp[2];
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);
}
for (i = 0; i < m; i++) {
#ifndef HEMVREV
#ifndef LOWER
if (i > 0) {
FLOAT _Complex result = DOTC_K(i, a, 1, X, 1);
Y[i * 2 + 0] += alpha_r * CREAL(result) - alpha_i * CIMAG(result);
Y[i * 2 + 1] += alpha_r * CIMAG(result) + alpha_i * CREAL(result);
}
temp[0] = a[i * 2 + 0] * X[i * 2 + 0];
temp[1] = a[i * 2 + 0] * X[i * 2 + 1];
Y[i * 2 + 0] += alpha_r * temp[0] - alpha_i * temp[1];
Y[i * 2 + 1] += alpha_r * temp[1] + alpha_i * temp[0];
if (i > 0) {
AXPYU_K(i, 0, 0,
alpha_r * X[i * 2 + 0] - alpha_i * X[i * 2 + 1],
alpha_r * X[i * 2 + 1] + alpha_i * X[i * 2 + 0],
a, 1, Y, 1, NULL, 0);
}
a += (i + 1) * 2;
#else
if (m - i > 1) {
FLOAT _Complex result = DOTC_K(m - i - 1, a + (i + 1) * 2, 1, X + (i + 1) * 2, 1);
Y[i * 2 + 0] += alpha_r * CREAL(result) - alpha_i * CIMAG(result);
Y[i * 2 + 1] += alpha_r * CIMAG(result) + alpha_i * CREAL(result);
}
temp[0] = a[i * 2 + 0] * X[i * 2 + 0];
temp[1] = a[i * 2 + 0] * X[i * 2 + 1];
Y[i * 2 + 0] += alpha_r * temp[0] - alpha_i * temp[1];
Y[i * 2 + 1] += alpha_r * temp[1] + alpha_i * temp[0];
if (m - i > 1) {
AXPYU_K(m - i - 1, 0, 0,
alpha_r * X[i * 2 + 0] - alpha_i * X[i * 2 + 1],
alpha_r * X[i * 2 + 1] + alpha_i * X[i * 2 + 0],
a + (i + 1) * 2, 1, Y + (i + 1) * 2, 1, NULL, 0);
}
a += (m - i - 1) * 2;
#endif
#else
#ifndef LOWER
if (i > 0) {
FLOAT _Complex result = DOTU_K(i, a, 1, X, 1);
Y[i * 2 + 0] += alpha_r * CREAL(result) - alpha_i * CIMAG(result);
Y[i * 2 + 1] += alpha_r * CIMAG(result) + alpha_i * CREAL(result);
}
temp[0] = a[i * 2 + 0] * X[i * 2 + 0];
temp[1] = a[i * 2 + 0] * X[i * 2 + 1];
Y[i * 2 + 0] += alpha_r * temp[0] - alpha_i * temp[1];
Y[i * 2 + 1] += alpha_r * temp[1] + alpha_i * temp[0];
if (i > 0) {
AXPYC_K(i, 0, 0,
alpha_r * X[i * 2 + 0] - alpha_i * X[i * 2 + 1],
alpha_r * X[i * 2 + 1] + alpha_i * X[i * 2 + 0],
a, 1, Y, 1, NULL, 0);
}
a += (i + 1) * 2;
#else
if (m - i > 1) {
FLOAT _Complex result = DOTU_K(m - i - 1, a + (i + 1) * 2, 1, X + (i + 1) * 2, 1);
Y[i * 2 + 0] += alpha_r * CREAL(result) - alpha_i * CIMAG(result);
Y[i * 2 + 1] += alpha_r * CIMAG(result) + alpha_i * CREAL(result);
}
temp[0] = a[i * 2 + 0] * X[i * 2 + 0];
temp[1] = a[i * 2 + 0] * X[i * 2 + 1];
Y[i * 2 + 0] += alpha_r * temp[0] - alpha_i * temp[1];
Y[i * 2 + 1] += alpha_r * temp[1] + alpha_i * temp[0];
if (m - i > 1) {
AXPYC_K(m - i - 1, 0, 0,
alpha_r * X[i * 2 + 0] - alpha_i * X[i * 2 + 1],
alpha_r * X[i * 2 + 1] + alpha_i * X[i * 2 + 0],
a + (i + 1) * 2, 1, Y + (i + 1) * 2, 1, NULL, 0);
}
a += (m - i - 1) * 2;
#endif
#endif
}
if (incy != 1) {
COPY_K(m, Y, 1, y, incy);
}
return 0;
}
int CNAME(BLASLONG n, BLASLONG k, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG incx, FLOAT *y, BLASLONG incy, FLOAT *buffer, int nthreads){
#else
int CNAME(BLASLONG n, BLASLONG k, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG incx, FLOAT *y, BLASLONG incy, FLOAT *buffer, int nthreads){
#endif
blas_arg_t args;
blas_queue_t queue[MAX_CPU_NUMBER];
BLASLONG range_m[MAX_CPU_NUMBER + 1];
BLASLONG range_n[MAX_CPU_NUMBER];
BLASLONG width, i, num_cpu;
double dnum;
int mask = 7;
#ifdef SMP
#ifndef COMPLEX
#ifdef XDOUBLE
int mode = BLAS_XDOUBLE | BLAS_REAL;
#elif defined(DOUBLE)
int mode = BLAS_DOUBLE | BLAS_REAL;
#else
int mode = BLAS_SINGLE | BLAS_REAL;
#endif
#else
#ifdef XDOUBLE
int mode = BLAS_XDOUBLE | BLAS_COMPLEX;
#elif defined(DOUBLE)
int mode = BLAS_DOUBLE | BLAS_COMPLEX;
#else
int mode = BLAS_SINGLE | BLAS_COMPLEX;
#endif
#endif
#endif
args.n = n;
args.k = k;
args.a = (void *)a;
args.b = (void *)x;
args.c = (void *)buffer;
args.lda = lda;
args.ldb = incx;
args.ldc = incy;
dnum = (double)n * (double)n / (double)nthreads;
num_cpu = 0;
if (n < 2 * k) {
#ifndef LOWER
range_m[MAX_CPU_NUMBER] = n;
i = 0;
while (i < n){
if (nthreads - num_cpu > 1) {
double di = (double)(n - i);
if (di * di - dnum > 0) {
width = ((BLASLONG)(-sqrt(di * di - dnum) + di) + mask) & ~mask;
} else {
width = n - i;
}
if (width < 16) width = 16;
if (width > n - i) width = n - i;
} else {
width = n - i;
}
range_m[MAX_CPU_NUMBER - num_cpu - 1] = range_m[MAX_CPU_NUMBER - num_cpu] - width;
range_n[num_cpu] = num_cpu * (((n + 15) & ~15) + 16);
queue[num_cpu].mode = mode;
queue[num_cpu].routine = sbmv_kernel;
queue[num_cpu].args = &args;
queue[num_cpu].range_m = &range_m[MAX_CPU_NUMBER - num_cpu - 1];
queue[num_cpu].range_n = &range_n[num_cpu];
queue[num_cpu].sa = NULL;
queue[num_cpu].sb = NULL;
queue[num_cpu].next = &queue[num_cpu + 1];
num_cpu ++;
i += width;
}
#else
range_m[0] = 0;
i = 0;
while (i < n){
if (nthreads - num_cpu > 1) {
double di = (double)(n - i);
if (di * di - dnum > 0) {
width = ((BLASLONG)(-sqrt(di * di - dnum) + di) + mask) & ~mask;
} else {
width = n - i;
}
if (width < 16) width = 16;
if (width > n - i) width = n - i;
} else {
width = n - i;
}
range_m[num_cpu + 1] = range_m[num_cpu] + width;
range_n[num_cpu] = num_cpu * (((n + 15) & ~15) + 16);
queue[num_cpu].mode = mode;
queue[num_cpu].routine = sbmv_kernel;
queue[num_cpu].args = &args;
queue[num_cpu].range_m = &range_m[num_cpu];
queue[num_cpu].range_n = &range_n[num_cpu];
queue[num_cpu].sa = NULL;
queue[num_cpu].sb = NULL;
queue[num_cpu].next = &queue[num_cpu + 1];
num_cpu ++;
i += width;
}
#endif
} else {
range_m[0] = 0;
i = n;
while (i > 0){
width = blas_quickdivide(i + nthreads - num_cpu - 1, nthreads - num_cpu);
if (width < 4) width = 4;
if (i < width) width = i;
range_m[num_cpu + 1] = range_m[num_cpu] + width;
range_n[num_cpu] = num_cpu * ((n + 15) & ~15);
queue[num_cpu].mode = mode;
queue[num_cpu].routine = sbmv_kernel;
queue[num_cpu].args = &args;
queue[num_cpu].range_m = &range_m[num_cpu];
queue[num_cpu].range_n = &range_n[num_cpu];
queue[num_cpu].sa = NULL;
queue[num_cpu].sb = NULL;
queue[num_cpu].next = &queue[num_cpu + 1];
num_cpu ++;
i -= width;
}
}
if (num_cpu) {
queue[0].sa = NULL;
queue[0].sb = buffer;
queue[num_cpu - 1].next = NULL;
exec_blas(num_cpu, queue);
}
for (i = 1; i < num_cpu; i ++) {
AXPYU_K(n, 0, 0,
#ifndef COMPLEX
ONE,
#else
ONE, ZERO,
#endif
(FLOAT*)(queue[i].sb), 1, buffer, 1, NULL, 0);
}
AXPYU_K(n, 0, 0,
#ifndef COMPLEX
alpha,
#else
alpha[0], alpha[1],
#endif
buffer, 1, y, incy, NULL, 0);
return 0;
}
void NAME(blasint *N, FLOAT *ALPHA, FLOAT *x, blasint *INCX, FLOAT *y, blasint *INCY){
blasint n = *N;
blasint incx = *INCX;
blasint incy = *INCY;
#else
void CNAME(blasint n, FLOAT *ALPHA, FLOAT *x, blasint incx, FLOAT *y, blasint incy){
#endif
FLOAT alpha_r = *(ALPHA + 0);
FLOAT alpha_i = *(ALPHA + 1);
#ifdef SMP
int mode, nthreads;
#endif
#ifndef CBLAS
PRINT_DEBUG_CNAME;
#else
PRINT_DEBUG_CNAME;
#endif
if (n <= 0) return;
if ((alpha_r == ZERO) && (alpha_i == ZERO)) return;
IDEBUG_START;
FUNCTION_PROFILE_START();
if (incx < 0) x -= (n - 1) * incx * 2;
if (incy < 0) y -= (n - 1) * incy * 2;
#ifdef SMP
nthreads = num_cpu_avail(1);
if (nthreads == 1) {
#endif
#ifndef CONJ
AXPYU_K (n, 0, 0, alpha_r, alpha_i, x, incx, y, incy, NULL, 0);
#else
AXPYC_K(n, 0, 0, alpha_r, alpha_i, x, incx, y, incy, NULL, 0);
#endif
#ifdef SMP
} else {
#ifdef XDOUBLE
mode = BLAS_XDOUBLE | BLAS_COMPLEX;
#elif defined(DOUBLE)
mode = BLAS_DOUBLE | BLAS_COMPLEX;
#else
mode = BLAS_SINGLE | BLAS_COMPLEX;
#endif
blas_level1_thread(mode, n, 0, 0, ALPHA, x, incx, y, incy, NULL, 0,
#ifndef CONJ
(void *)AXPYU_K,
#else
(void *)AXPYC_K,
#endif
nthreads);
}
#endif
FUNCTION_PROFILE_END(4, 2 * n, 2 * n);
IDEBUG_END;
return;
}
int CNAME(BLASLONG m, FLOAT *a, FLOAT *x, BLASLONG incx, FLOAT *buffer, int nthreads){
#else
int CNAME(BLASLONG m, FLOAT *a, FLOAT *x, BLASLONG incx, FLOAT *buffer, int nthreads){
#endif
blas_arg_t args;
blas_queue_t queue[MAX_CPU_NUMBER];
BLASLONG range_m[MAX_CPU_NUMBER + 1];
BLASLONG range_n[MAX_CPU_NUMBER];
BLASLONG width, i, num_cpu;
double dnum;
int mask = 7;
#ifdef SMP
#ifndef COMPLEX
#ifdef XDOUBLE
int mode = BLAS_XDOUBLE | BLAS_REAL;
#elif defined(DOUBLE)
int mode = BLAS_DOUBLE | BLAS_REAL;
#else
int mode = BLAS_SINGLE | BLAS_REAL;
#endif
#else
#ifdef XDOUBLE
int mode = BLAS_XDOUBLE | BLAS_COMPLEX;
#elif defined(DOUBLE)
int mode = BLAS_DOUBLE | BLAS_COMPLEX;
#else
int mode = BLAS_SINGLE | BLAS_COMPLEX;
#endif
#endif
#endif
args.m = m;
args.a = (void *)a;
args.b = (void *)x;
args.c = (void *)(buffer);
args.ldb = incx;
args.ldc = incx;
dnum = (double)m * (double)m / (double)nthreads;
num_cpu = 0;
#ifndef LOWER
range_m[MAX_CPU_NUMBER] = m;
i = 0;
while (i < m){
if (nthreads - num_cpu > 1) {
double di = (double)(m - i);
if (di * di - dnum > 0) {
width = ((BLASLONG)(-sqrt(di * di - dnum) + di) + mask) & ~mask;
} else {
width = m - i;
}
if (width < 16) width = 16;
if (width > m - i) width = m - i;
} else {
width = m - i;
}
range_m[MAX_CPU_NUMBER - num_cpu - 1] = range_m[MAX_CPU_NUMBER - num_cpu] - width;
range_n[num_cpu] = num_cpu * (((m + 15) & ~15) + 16);
queue[num_cpu].mode = mode;
queue[num_cpu].routine = tpmv_kernel;
queue[num_cpu].args = &args;
queue[num_cpu].range_m = &range_m[MAX_CPU_NUMBER - num_cpu - 1];
queue[num_cpu].range_n = &range_n[num_cpu];
queue[num_cpu].sa = NULL;
queue[num_cpu].sb = NULL;
queue[num_cpu].next = &queue[num_cpu + 1];
num_cpu ++;
i += width;
}
#else
range_m[0] = 0;
i = 0;
while (i < m){
if (nthreads - num_cpu > 1) {
double di = (double)(m - i);
if (di * di - dnum > 0) {
width = ((BLASLONG)(-sqrt(di * di - dnum) + di) + mask) & ~mask;
} else {
width = m - i;
}
if (width < 16) width = 16;
if (width > m - i) width = m - i;
} else {
width = m - i;
}
range_m[num_cpu + 1] = range_m[num_cpu] + width;
range_n[num_cpu] = num_cpu * (((m + 15) & ~15) + 16);
queue[num_cpu].mode = mode;
queue[num_cpu].routine = tpmv_kernel;
queue[num_cpu].args = &args;
queue[num_cpu].range_m = &range_m[num_cpu];
queue[num_cpu].range_n = &range_n[num_cpu];
queue[num_cpu].sa = NULL;
queue[num_cpu].sb = NULL;
queue[num_cpu].next = &queue[num_cpu + 1];
num_cpu ++;
i += width;
}
#endif
if (num_cpu) {
queue[0].sa = NULL;
queue[0].sb = buffer + num_cpu * (((m + 255) & ~255) + 16) * COMPSIZE;
queue[num_cpu - 1].next = NULL;
exec_blas(num_cpu, queue);
}
#ifndef TRANS
for (i = 1; i < num_cpu; i ++) {
#ifndef LOWER
AXPYU_K(range_m[MAX_CPU_NUMBER - i], 0, 0, ONE,
#ifdef COMPLEX
ZERO,
#endif
buffer + range_n[i] * COMPSIZE, 1, buffer, 1, NULL, 0);
#else
AXPYU_K(m - range_m[i], 0, 0, ONE,
#ifdef COMPLEX
ZERO,
#endif
buffer + (range_n[i] + range_m[i]) * COMPSIZE, 1, buffer + range_m[i] * COMPSIZE, 1, NULL, 0);
#endif
}
#endif
COPY_K(m, buffer, 1, x, incx);
return 0;
}
int CNAME(BLASLONG n, BLASLONG k, FLOAT alpha_r, FLOAT alpha_i,
FLOAT *a, BLASLONG lda,
FLOAT *x, BLASLONG incx, FLOAT *y, BLASLONG incy, void *buffer){
BLASLONG i, length;
#ifndef LOWER
BLASLONG offset;
#endif
FLOAT *X = x;
FLOAT *Y = y;
FLOAT *sbmvbuffer = (FLOAT *)buffer;
FLOAT *bufferY = sbmvbuffer;
FLOAT *bufferX = sbmvbuffer;
if (incy != 1) {
Y = bufferY;
bufferX = (FLOAT *)(((BLASLONG)bufferY + n * sizeof(FLOAT) * COMPSIZE + 4095) & ~4095);
sbmvbuffer = bufferX;
COPY_K(n, y, incy, Y, 1);
}
if (incx != 1) {
X = bufferX;
sbmvbuffer = (FLOAT *)(((BLASLONG)bufferX + n * sizeof(FLOAT) * COMPSIZE + 4095) & ~4095);
COPY_K(n, x, incx, X, 1);
}
#ifndef LOWER
offset = k;
#endif
for (i = 0; i < n; i++) {
#ifndef LOWER
length = k - offset;
AXPYU_K(length + 1, 0, 0,
alpha_r * X[i * 2 + 0] - alpha_i * X[i * 2 + 1],
alpha_r * X[i * 2 + 1] + alpha_i * X[i * 2 + 0],
a + offset * COMPSIZE, 1, Y + (i - length) * COMPSIZE, 1, NULL, 0);
if (length > 0) {
FLOAT _Complex result = DOTU_K(length, a + offset * COMPSIZE, 1, X + (i - length) * COMPSIZE, 1);
Y[i * 2 + 0] += alpha_r * CREAL(result) - alpha_i * CIMAG(result);
Y[i * 2 + 1] += alpha_r * CIMAG(result) + alpha_i * CREAL(result);
}
if (offset > 0) offset --;
#else
length = k;
if (n - i - 1 < k) length = n - i - 1;
AXPYU_K(length + 1, 0, 0,
alpha_r * X[i * 2 + 0] - alpha_i * X[i * 2 + 1],
alpha_r * X[i * 2 + 1] + alpha_i * X[i * 2 + 0],
a, 1, Y + i * COMPSIZE, 1, NULL, 0);
if (length > 0) {
FLOAT _Complex result = DOTU_K(length, a + COMPSIZE, 1, X + (i + 1) * COMPSIZE, 1);
Y[i * 2 + 0] += alpha_r * CREAL(result) - alpha_i * CIMAG(result);
Y[i * 2 + 1] += alpha_r * CIMAG(result) + alpha_i * CREAL(result);
}
#endif
a += lda * 2;
}
if (incy != 1) {
COPY_K(n, Y, 1, y, incy);
}
return 0;
}
int CNAME(BLASLONG m, BLASLONG n, BLASLONG ku, BLASLONG kl, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG incx, FLOAT *y, BLASLONG incy, FLOAT *buffer, int nthreads){
#else
int CNAME(BLASLONG m, BLASLONG n, BLASLONG ku, BLASLONG kl, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG incx, FLOAT *y, BLASLONG incy, FLOAT *buffer, int nthreads){
#endif
blas_arg_t args;
blas_queue_t queue[MAX_CPU_NUMBER];
BLASLONG range_m[MAX_CPU_NUMBER];
BLASLONG range_n[MAX_CPU_NUMBER + 1];
BLASLONG width, i, num_cpu;
#ifdef SMP
#ifndef COMPLEX
#ifdef XDOUBLE
int mode = BLAS_XDOUBLE | BLAS_REAL;
#elif defined(DOUBLE)
int mode = BLAS_DOUBLE | BLAS_REAL;
#else
int mode = BLAS_SINGLE | BLAS_REAL;
#endif
#else
#ifdef XDOUBLE
int mode = BLAS_XDOUBLE | BLAS_COMPLEX;
#elif defined(DOUBLE)
int mode = BLAS_DOUBLE | BLAS_COMPLEX;
#else
int mode = BLAS_SINGLE | BLAS_COMPLEX;
#endif
#endif
#endif
args.m = m;
args.n = n;
args.a = (void *)a;
args.b = (void *)x;
args.c = (void *)buffer;
args.lda = lda;
args.ldb = incx;
args.ldc = ku;
args.ldd = kl;
num_cpu = 0;
range_n[0] = 0;
i = n;
while (i > 0){
width = blas_quickdivide(i + nthreads - num_cpu - 1, nthreads - num_cpu);
if (width < 4) width = 4;
if (i < width) width = i;
range_n[num_cpu + 1] = range_n[num_cpu] + width;
#ifndef TRANSA
range_m[num_cpu] = num_cpu * ((m + 15) & ~15);
#else
range_m[num_cpu] = num_cpu * ((n + 15) & ~15);
#endif
queue[num_cpu].mode = mode;
queue[num_cpu].routine = gbmv_kernel;
queue[num_cpu].args = &args;
queue[num_cpu].range_m = &range_m[num_cpu];
queue[num_cpu].range_n = &range_n[num_cpu];
queue[num_cpu].sa = NULL;
queue[num_cpu].sb = NULL;
queue[num_cpu].next = &queue[num_cpu + 1];
num_cpu ++;
i -= width;
}
if (num_cpu) {
queue[0].sa = NULL;
#ifndef TRANSA
queue[0].sb = buffer + num_cpu * (((m + 255) & ~255) + 16) * COMPSIZE;
#else
queue[0].sb = buffer + num_cpu * (((n + 255) & ~255) + 16) * COMPSIZE;
#endif
queue[num_cpu - 1].next = NULL;
exec_blas(num_cpu, queue);
}
for (i = 1; i < num_cpu; i ++) {
AXPYU_K(
#ifndef TRANSA
m,
#else
n,
#endif
0, 0,
#ifndef COMPLEX
ONE,
#else
ONE, ZERO,
#endif
buffer + range_m[i] * COMPSIZE, 1, buffer, 1, NULL, 0);
}
AXPYU_K(
#ifndef TRANSA
m,
#else
n,
#endif
0, 0,
#ifndef COMPLEX
alpha,
#else
alpha[0], alpha[1],
#endif
buffer, 1, y, incy, NULL, 0);
return 0;
}
int CNAME(BLASLONG m, FLOAT alpha_r, FLOAT alpha_i,
FLOAT *x, BLASLONG incx,
FLOAT *y, BLASLONG incy, FLOAT *a, BLASLONG lda, FLOAT *buffer){
BLASLONG i;
FLOAT *X, *Y;
X = x;
Y = y;
lda *= 2;
if (incx != 1) {
COPY_K(m, x, incx, buffer, 1);
X = buffer;
}
if (incy != 1) {
COPY_K(m, y, incy, (FLOAT *)((BLASLONG)buffer + (BUFFER_SIZE / 2)), 1);
Y = (FLOAT *)((BLASLONG)buffer + (BUFFER_SIZE / 2));
}
for (i = 0; i < m; i++){
#ifndef HEMVREV
#ifndef LOWER
AXPYU_K(i + 1, 0, 0,
alpha_r * X[i * 2 + 0] - alpha_i * X[i * 2 + 1],
- alpha_i * X[i * 2 + 0] - alpha_r * X[i * 2 + 1],
Y, 1, a, 1, NULL, 0);
AXPYU_K(i + 1, 0, 0,
alpha_r * Y[i * 2 + 0] + alpha_i * Y[i * 2 + 1],
alpha_i * Y[i * 2 + 0] - alpha_r * Y[i * 2 + 1],
X, 1, a, 1, NULL, 0);
a[i * 2 + 1] = ZERO;
a += lda;
#else
AXPYU_K(m - i, 0, 0,
alpha_r * X[i * 2 + 0] - alpha_i * X[i * 2 + 1],
- alpha_i * X[i * 2 + 0] - alpha_r * X[i * 2 + 1],
Y + i * 2, 1, a, 1, NULL, 0);
AXPYU_K(m - i, 0, 0,
alpha_r * Y[i * 2 + 0] + alpha_i * Y[i * 2 + 1],
alpha_i * Y[i * 2 + 0] - alpha_r * Y[i * 2 + 1],
X + i * 2, 1, a, 1, NULL, 0);
a[1] = ZERO;
a += 2 + lda;
#endif
#else
#ifndef LOWER
AXPYC_K(i + 1, 0, 0,
alpha_r * X[i * 2 + 0] - alpha_i * X[i * 2 + 1],
alpha_i * X[i * 2 + 0] + alpha_r * X[i * 2 + 1],
Y, 1, a, 1, NULL, 0);
AXPYC_K(i + 1, 0, 0,
alpha_r * Y[i * 2 + 0] + alpha_i * Y[i * 2 + 1],
- alpha_i * Y[i * 2 + 0] + alpha_r * Y[i * 2 + 1],
X, 1, a, 1, NULL, 0);
a[i * 2 + 1] = ZERO;
a += lda;
#else
AXPYC_K(m - i, 0, 0,
alpha_r * X[i * 2 + 0] - alpha_i * X[i * 2 + 1],
alpha_i * X[i * 2 + 0] + alpha_r * X[i * 2 + 1],
Y + i * 2, 1, a, 1, NULL, 0);
AXPYC_K(m - i, 0, 0,
alpha_r * Y[i * 2 + 0] + alpha_i * Y[i * 2 + 1],
- alpha_i * Y[i * 2 + 0] + alpha_r * Y[i * 2 + 1],
X + i * 2, 1, a, 1, NULL, 0);
a[1] = ZERO;
a += 2 + lda;
#endif
#endif
}
return 0;
}
void NAME(blasint *N, FLOAT *ALPHA, FLOAT *x, blasint *INCX, FLOAT *y, blasint *INCY){
BLASLONG n = *N;
BLASLONG incx = *INCX;
BLASLONG incy = *INCY;
FLOAT alpha = *ALPHA;
#else
void CNAME(blasint n, FLOAT alpha, FLOAT *x, blasint incx, FLOAT *y, blasint incy){
#endif
#ifdef SMP
int mode, nthreads;
#endif
#ifndef CBLAS
PRINT_DEBUG_NAME;
#else
PRINT_DEBUG_CNAME;
#endif
if (n <= 0) return;
if (alpha == ZERO) return;
IDEBUG_START;
FUNCTION_PROFILE_START();
if (incx < 0) x -= (n - 1) * incx;
if (incy < 0) y -= (n - 1) * incy;
#ifdef SMP
nthreads = num_cpu_avail(1);
//disable multi-thread when incx==0 or incy==0
//In that case, the threads would be dependent.
if (incx == 0 || incy == 0)
nthreads = 1;
//Temporarily work-around the low performance issue with small imput size &
//multithreads.
if (n <= 10000)
nthreads = 1;
if (nthreads == 1) {
#endif
AXPYU_K(n, 0, 0, alpha, x, incx, y, incy, NULL, 0);
#ifdef SMP
} else {
#ifdef XDOUBLE
mode = BLAS_XDOUBLE | BLAS_REAL;
#elif defined(DOUBLE)
mode = BLAS_DOUBLE | BLAS_REAL;
#else
mode = BLAS_SINGLE | BLAS_REAL;
#endif
blas_level1_thread(mode, n, 0, 0, &alpha,
x, incx, y, incy, NULL, 0, (void *)AXPYU_K, nthreads);
}
#endif
FUNCTION_PROFILE_END(1, 2 * n, 2 * n);
IDEBUG_END;
return;
}
int CNAME(BLASLONG n, BLASLONG k, FLOAT *a, BLASLONG lda, FLOAT *b, BLASLONG incb, void *buffer){
BLASLONG i;
FLOAT *gemvbuffer = (FLOAT *)buffer;
FLOAT *B = b;
BLASLONG length;
#if (TRANSA == 2) || (TRANSA == 4)
OPENBLAS_COMPLEX_FLOAT temp;
#endif
#ifndef UNIT
FLOAT atemp1, atemp2, btemp1, btemp2;
#endif
if (incb != 1) {
B = buffer;
gemvbuffer = (FLOAT *)(((BLASLONG)buffer + n * sizeof(FLOAT) * COMPSIZE + 4095) & ~4095);
COPY_K(n, b, incb, buffer, 1);
}
for (i = 0; i < n; i++) {
#if (TRANSA == 1) || (TRANSA == 3)
length = i;
if (length > k) length = k;
if (length > 0) {
#if TRANSA == 1
AXPYU_K(length, 0, 0,
B[i * 2 + 0], B[i * 2 + 1],
a + (k - length) * COMPSIZE, 1, B + (i - length) * COMPSIZE, 1, NULL, 0);
#else
AXPYC_K(length, 0, 0,
B[i * 2 + 0], B[i * 2 + 1],
a + (k - length) * COMPSIZE, 1, B + (i - length) * COMPSIZE, 1, NULL, 0);
#endif
}
#endif
#ifndef UNIT
#if (TRANSA == 1) || (TRANSA == 3)
atemp1 = a[k * 2 + 0];
atemp2 = a[k * 2 + 1];
#else
atemp1 = a[0];
atemp2 = a[1];
#endif
btemp1 = B[i * 2 + 0];
btemp2 = B[i * 2 + 1];
#if (TRANSA == 1) || (TRANSA == 2)
B[i * 2 + 0] = atemp1 * btemp1 - atemp2 * btemp2;
B[i * 2 + 1] = atemp1 * btemp2 + atemp2 * btemp1;
#else
B[i * 2 + 0] = atemp1 * btemp1 + atemp2 * btemp2;
B[i * 2 + 1] = atemp1 * btemp2 - atemp2 * btemp1;
#endif
#endif
#if (TRANSA == 2) || (TRANSA == 4)
length = n - i - 1;
if (length > k) length = k;
if (length > 0) {
#if TRANSA == 2
temp = DOTU_K(length, a + COMPSIZE, 1, B + (i + 1) * COMPSIZE, 1);
#else
temp = DOTC_K(length, a + COMPSIZE, 1, B + (i + 1) * COMPSIZE, 1);
#endif
B[i * 2 + 0] += CREAL(temp);
B[i * 2 + 1] += CIMAG(temp);
}
#endif
a += lda * COMPSIZE;
}
if (incb != 1) {
COPY_K(n, buffer, 1, b, incb);
}
return 0;
}
static int syr_kernel(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *dummy1, FLOAT *buffer, BLASLONG pos){
FLOAT *a, *x, *y;
BLASLONG lda, incx, incy;
BLASLONG i, m_from, m_to;
FLOAT alpha_r;
#ifdef COMPLEX
FLOAT alpha_i;
#endif
x = (FLOAT *)args -> a;
y = (FLOAT *)args -> b;
a = (FLOAT *)args -> c;
incx = args -> lda;
incy = args -> ldb;
lda = args -> ldc;
alpha_r = *((FLOAT *)args -> alpha + 0);
#ifdef COMPLEX
alpha_i = *((FLOAT *)args -> alpha + 1);
#endif
m_from = 0;
m_to = args -> m;
if (range_m) {
m_from = *(range_m + 0);
m_to = *(range_m + 1);
}
if (incx != 1) {
#ifndef LOWER
COPY_K(m_to, x, incx, buffer, 1);
#else
COPY_K(args -> m - m_from, x + m_from * incx * COMPSIZE, incx, buffer + m_from * COMPSIZE, 1);
#endif
x = buffer;
buffer += ((COMPSIZE * args -> m + 1023) & ~1023);
}
if (incy != 1) {
#ifndef LOWER
COPY_K(m_to, y, incy, buffer, 1);
#else
COPY_K(args -> m - m_from, y + m_from * incy * COMPSIZE, incy, buffer + m_from * COMPSIZE, 1);
#endif
y = buffer;
}
#ifndef LOWER
a += (m_from + 1) * m_from / 2 * COMPSIZE;
#else
a += (2 * args -> m - m_from + 1) * m_from / 2 * COMPSIZE;
#endif
for (i = m_from; i < m_to; i++){
#if !defined(HEMV) && !defined(HEMVREV)
#ifndef COMPLEX
if (x[i] != ZERO) {
#ifndef LOWER
AXPYU_K(i + 1, 0, 0, alpha_r * x[i], y, 1, a, 1, NULL, 0);
#else
AXPYU_K(args -> m - i, 0, 0, alpha_r * x[i], y + i, 1, a, 1, NULL, 0);
#endif
}
if (y[i] != ZERO) {
#ifndef LOWER
AXPYU_K(i + 1, 0, 0, alpha_r * y[i], x, 1, a, 1, NULL, 0);
#else
AXPYU_K(args -> m - i, 0, 0, alpha_r * y[i], x + i, 1, a, 1, NULL, 0);
#endif
}
#else
if ((x[i * COMPSIZE + 0] != ZERO) || (x[i * COMPSIZE + 1] != ZERO)) {
#ifndef LOWER
AXPYU_K(i + 1, 0, 0,
alpha_r * x[i * COMPSIZE + 0] - alpha_i * x[i * COMPSIZE + 1],
alpha_i * x[i * COMPSIZE + 0] + alpha_r * x[i * COMPSIZE + 1],
y, 1, a, 1, NULL, 0);
#else
AXPYU_K(args -> m - i, 0, 0,
alpha_r * x[i * COMPSIZE + 0] - alpha_i * x[i * COMPSIZE + 1],
alpha_i * x[i * COMPSIZE + 0] + alpha_r * x[i * COMPSIZE + 1],
y + i * COMPSIZE, 1, a, 1, NULL, 0);
#endif
}
if ((y[i * COMPSIZE + 0] != ZERO) || (y[i * COMPSIZE + 1] != ZERO)) {
#ifndef LOWER
AXPYU_K(i + 1, 0, 0,
alpha_r * y[i * COMPSIZE + 0] - alpha_i * y[i * COMPSIZE + 1],
alpha_i * y[i * COMPSIZE + 0] + alpha_r * y[i * COMPSIZE + 1],
x, 1, a, 1, NULL, 0);
#else
AXPYU_K(args -> m - i, 0, 0,
alpha_r * y[i * COMPSIZE + 0] - alpha_i * y[i * COMPSIZE + 1],
alpha_i * y[i * COMPSIZE + 0] + alpha_r * y[i * COMPSIZE + 1],
x + i * COMPSIZE, 1, a, 1, NULL, 0);
#endif
}
#endif
#else
if ((x[i * COMPSIZE + 0] != ZERO) || (x[i * COMPSIZE + 1] != ZERO)) {
#ifndef HEMVREV
#ifndef LOWER
AXPYU_K(i + 1, 0, 0,
alpha_r * x[i * COMPSIZE + 0] - alpha_i * x[i * COMPSIZE + 1],
- alpha_i * x[i * COMPSIZE + 0] - alpha_r * x[i * COMPSIZE + 1],
y, 1, a, 1, NULL, 0);
#else
AXPYU_K(args -> m - i, 0, 0,
alpha_r * x[i * COMPSIZE + 0] - alpha_i * x[i * COMPSIZE + 1],
- alpha_i * x[i * COMPSIZE + 0] - alpha_r * x[i * COMPSIZE + 1],
y + i * COMPSIZE, 1, a, 1, NULL, 0);
#endif
#else
#ifndef LOWER
AXPYC_K(i + 1, 0, 0,
alpha_r * x[i * COMPSIZE + 0] - alpha_i * x[i * COMPSIZE + 1],
alpha_i * x[i * COMPSIZE + 0] + alpha_r * x[i * COMPSIZE + 1],
y, 1, a, 1, NULL, 0);
#else
AXPYC_K(args -> m - i, 0, 0,
alpha_r * x[i * COMPSIZE + 0] - alpha_i * x[i * COMPSIZE + 1],
alpha_i * x[i * COMPSIZE + 0] + alpha_r * x[i * COMPSIZE + 1],
y + i * COMPSIZE, 1, a, 1, NULL, 0);
#endif
#endif
}
if ((y[i * COMPSIZE + 0] != ZERO) || (y[i * COMPSIZE + 1] != ZERO)) {
#ifndef HEMVREV
#ifndef LOWER
AXPYU_K(i + 1, 0, 0,
alpha_r * y[i * COMPSIZE + 0] + alpha_i * y[i * COMPSIZE + 1],
alpha_i * y[i * COMPSIZE + 0] - alpha_r * y[i * COMPSIZE + 1],
x, 1, a, 1, NULL, 0);
#else
AXPYU_K(args -> m - i, 0, 0,
alpha_r * y[i * COMPSIZE + 0] + alpha_i * y[i * COMPSIZE + 1],
alpha_i * y[i * COMPSIZE + 0] - alpha_r * y[i * COMPSIZE + 1],
x + i * COMPSIZE, 1, a, 1, NULL, 0);
#endif
#else
#ifndef LOWER
AXPYC_K(i + 1, 0, 0,
alpha_r * y[i * COMPSIZE + 0] + alpha_i * y[i * COMPSIZE + 1],
- alpha_i * y[i * COMPSIZE + 0] + alpha_r * y[i * COMPSIZE + 1],
x, 1, a, 1, NULL, 0);
#else
AXPYC_K(args -> m - i, 0, 0,
alpha_r * y[i * COMPSIZE + 0] + alpha_i * y[i * COMPSIZE + 1],
- alpha_i * y[i * COMPSIZE + 0] + alpha_r * y[i * COMPSIZE + 1],
x + i * COMPSIZE, 1, a, 1, NULL, 0);
#endif
#endif
}
#ifndef LOWER
a[i * COMPSIZE + 1] = ZERO;
#else
a[ 1] = ZERO;
#endif
#endif
#ifndef LOWER
a += (i + 1) * COMPSIZE;
#else
a += (args -> m - i) * COMPSIZE;
#endif
}
return 0;
}
int CNAME(BLASLONG n, BLASLONG k, FLOAT *a, BLASLONG lda, FLOAT *b, BLASLONG incb, void *buffer){
BLASLONG i;
FLOAT *B = b;
BLASLONG length;
#if (TRANSA == 2) || (TRANSA == 4)
OPENBLAS_COMPLEX_FLOAT temp;
#endif
#ifndef UNIT
FLOAT ar, ai, br, bi, ratio, den;
#endif
if (incb != 1) {
B = buffer;
COPY_K(n, b, incb, buffer, 1);
}
for (i = 0; i < n; i++) {
#if (TRANSA == 2) || (TRANSA == 4)
length = i;
if (length > k) length = k;
if (length > 0) {
#if TRANSA == 2
temp = DOTU_K(length, a + (k - length) * COMPSIZE, 1, B + (i - length) * COMPSIZE, 1);
#else
temp = DOTC_K(length, a + (k - length) * COMPSIZE, 1, B + (i - length) * COMPSIZE, 1);
#endif
B[i * 2 + 0] -= CREAL(temp);
B[i * 2 + 1] -= CIMAG(temp);
}
#endif
#ifndef UNIT
#if (TRANSA == 1) || (TRANSA == 3)
ar = a[0];
ai = a[1];
#else
ar = a[k * 2 + 0];
ai = a[k * 2 + 1];
#endif
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 = B[i * 2 + 0];
bi = B[i * 2 + 1];
B[i * 2 + 0] = ar*br - ai*bi;
B[i * 2 + 1] = ar*bi + ai*br;
#endif
#if (TRANSA == 1) || (TRANSA == 3)
length = n - i - 1;
if (length > k) length = k;
if (length > 0) {
#if TRANSA == 1
AXPYU_K(length, 0, 0,
-B[i * 2 + 0], -B[i * 2 + 1],
a + COMPSIZE, 1, B + (i + 1) * COMPSIZE, 1, NULL, 0);
#else
AXPYC_K(length, 0, 0,
-B[i * 2 + 0], -B[i * 2 + 1],
a + COMPSIZE, 1, B + (i + 1) * COMPSIZE, 1, NULL, 0);
#endif
}
#endif
a += lda * COMPSIZE;
}
if (incb != 1) {
COPY_K(n, buffer, 1, b, incb);
}
return 0;
}