Unverified Commit fa24e548 authored by Andreas Marek's avatar Andreas Marek
Browse files

Solve problem with single precision AVX2 kernels

a preprocessor flag was missing
parent c4dfb99e
...@@ -73,7 +73,7 @@ static __forceinline void hh_trafo_complex_kernel_8_AVX_1hv_single(float comple ...@@ -73,7 +73,7 @@ static __forceinline void hh_trafo_complex_kernel_8_AVX_1hv_single(float comple
static __forceinline void hh_trafo_complex_kernel_4_AVX_1hv_single(float complex* q, float complex* hh, int nb, int ldq); static __forceinline void hh_trafo_complex_kernel_4_AVX_1hv_single(float complex* q, float complex* hh, int nb, int ldq);
/* /*
!f>#ifdef HAVE_AVX !f>#if defined(HAVE_AVX) || defined(HAVE_AVX2)
!f> interface !f> interface
!f> subroutine single_hh_trafo_complex_avx_avx2_1hv_single(q, hh, pnb, pnq, pldq) & !f> subroutine single_hh_trafo_complex_avx_avx2_1hv_single(q, hh, pnb, pnq, pldq) &
!f> bind(C, name="single_hh_trafo_complex_avx_avx2_1hv_single") !f> bind(C, name="single_hh_trafo_complex_avx_avx2_1hv_single")
......
...@@ -74,7 +74,7 @@ static __forceinline void hh_trafo_complex_kernel_4_AVX_2hv_single(float complex ...@@ -74,7 +74,7 @@ static __forceinline void hh_trafo_complex_kernel_4_AVX_2hv_single(float complex
//static __forceinline void hh_trafo_complex_kernel_2_AVX_2hv_single(float complex* q, float complex* hh, int nb, int ldq, int ldh, float complex s, float complex s1); //static __forceinline void hh_trafo_complex_kernel_2_AVX_2hv_single(float complex* q, float complex* hh, int nb, int ldq, int ldh, float complex s, float complex s1);
/* /*
!f>#ifdef HAVE_AVX !f>#if defined(HAVE_AVX) || defined(HAVE_AVX2)
!f> interface !f> interface
!f> subroutine double_hh_trafo_complex_avx_avx2_2hv_single(q, hh, pnb, pnq, pldq, pldh) & !f> subroutine double_hh_trafo_complex_avx_avx2_2hv_single(q, hh, pnb, pnq, pldq, pldh) &
!f> bind(C, name="double_hh_trafo_complex_avx_avx2_2hv_single") !f> bind(C, name="double_hh_trafo_complex_avx_avx2_2hv_single")
......
...@@ -74,7 +74,7 @@ __forceinline void hh_trafo_kernel_24_AVX_2hv_single(float* q, float* hh, int nb ...@@ -74,7 +74,7 @@ __forceinline void hh_trafo_kernel_24_AVX_2hv_single(float* q, float* hh, int nb
void double_hh_trafo_real_avx_avx2_2hv_single(float* q, float* hh, int* pnb, int* pnq, int* pldq, int* pldh); void double_hh_trafo_real_avx_avx2_2hv_single(float* q, float* hh, int* pnb, int* pnq, int* pldq, int* pldh);
/* /*
!f>#ifdef HAVE_AVX !f>#if defined(HAVE_AVX) || defined(HAVE_AVX2)
!f> interface !f> interface
!f> subroutine double_hh_trafo_real_avx_avx2_2hv_single(q, hh, pnb, pnq, pldq, pldh) & !f> subroutine double_hh_trafo_real_avx_avx2_2hv_single(q, hh, pnb, pnq, pldq, pldh) &
!f> bind(C, name="double_hh_trafo_real_avx_avx2_2hv_single") !f> bind(C, name="double_hh_trafo_real_avx_avx2_2hv_single")
......
...@@ -78,7 +78,7 @@ __forceinline void hh_trafo_kernel_24_AVX_4hv_single(float* q, float* hh, int nb ...@@ -78,7 +78,7 @@ __forceinline void hh_trafo_kernel_24_AVX_4hv_single(float* q, float* hh, int nb
void quad_hh_trafo_real_avx_avx2_4hv_single(float* q, float* hh, int* pnb, int* pnq, int* pldq, int* pldh); void quad_hh_trafo_real_avx_avx2_4hv_single(float* q, float* hh, int* pnb, int* pnq, int* pldq, int* pldh);
/* /*
!f>#ifdef HAVE_AVX !f>#if defined(HAVE_AVX) || defined(HAVE_AVX2)
!f> interface !f> interface
!f> subroutine quad_hh_trafo_real_avx_avx2_4hv_single(q, hh, pnb, pnq, pldq, pldh) & !f> subroutine quad_hh_trafo_real_avx_avx2_4hv_single(q, hh, pnb, pnq, pldq, pldh) &
!f> bind(C, name="quad_hh_trafo_real_avx_avx2_4hv_single") !f> bind(C, name="quad_hh_trafo_real_avx_avx2_4hv_single")
......
...@@ -75,7 +75,7 @@ static void hh_trafo_kernel_8_AVX_6hv_single(float* q, float* hh, int nb, int ld ...@@ -75,7 +75,7 @@ static void hh_trafo_kernel_8_AVX_6hv_single(float* q, float* hh, int nb, int ld
void hexa_hh_trafo_real_avx_avx2_6hv_single_(float* q, float* hh, int* pnb, int* pnq, int* pldq, int* pldh); void hexa_hh_trafo_real_avx_avx2_6hv_single_(float* q, float* hh, int* pnb, int* pnq, int* pldq, int* pldh);
/* /*
!f>#ifdef HAVE_AVX !f>#if defined(HAVE_AVX) || defined(HAVE_AVX2)
!f> interface !f> interface
!f> subroutine hexa_hh_trafo_real_avx_avx2_6hv_single(q, hh, pnb, pnq, pldq, pldh) & !f> subroutine hexa_hh_trafo_real_avx_avx2_6hv_single(q, hh, pnb, pnq, pldq, pldh) &
!f> bind(C, name="hexa_hh_trafo_real_avx_avx2_6hv_single") !f> bind(C, name="hexa_hh_trafo_real_avx_avx2_6hv_single")
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment