floraachy
/
OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fixes RVV masked intrinsics for iamax/iamin/imax/imin kernels 

Changes masked intrinsics from _m to _mu and reintroduces maskedoff argument.
This commit is contained in:
Octavian Maghiar 2023-07-05 11:34:00 +01:00
parent 62f0f506ec
commit e1958eb705
4 changed files with 32 additions and 32 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
kernel/riscv64/iamax_rvv.c
View File

@ -45,9 +45,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define VFMAXVV_FLOAT __riscv_vfmax_vv_f64m8
#define VFIRSTM __riscv_vfirst_m_b8
#define UINT_V_T vuint64m8_t
#define VIDV_MASK_UINT __riscv_vid_v_u64m8_m
#define VIDV_MASK_UINT __riscv_vid_v_u64m8_mu
#define VIDV_UINT __riscv_vid_v_u64m8
#define VADDVX_MASK_UINT __riscv_vadd_vx_u64m8_m
#define VADDVX_MASK_UINT __riscv_vadd_vx_u64m8_mu
#define VADDVX_UINT __riscv_vadd_vx_u64m8
#define VMVVX_UINT __riscv_vmv_v_x_u64m8
#define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
@ -71,9 +71,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define VFMAXVV_FLOAT __riscv_vfmax_vv_f32m8
#define VFIRSTM __riscv_vfirst_m_b4
#define UINT_V_T vuint32m8_t
#define VIDV_MASK_UINT __riscv_vid_v_u32m8_m
#define VIDV_MASK_UINT __riscv_vid_v_u32m8_mu
#define VIDV_UINT __riscv_vid_v_u32m8
#define VADDVX_MASK_UINT __riscv_vadd_vx_u32m8_m
#define VADDVX_MASK_UINT __riscv_vadd_vx_u32m8_mu
#define VADDVX_UINT __riscv_vadd_vx_u32m8
#define VMVVX_UINT __riscv_vmv_v_x_u32m8
#define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
@ -106,8 +106,8 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
//index where element greater than v_max
mask = VMFLTVV_FLOAT(v_max, vx, vl);
v_max_index = VIDV_MASK_UINT(mask, vl);
v_max_index = VADDVX_MASK_UINT(mask, v_max_index, j, vl);
v_max_index = VIDV_MASK_UINT(mask, v_max_index, vl);
v_max_index = VADDVX_MASK_UINT(mask, v_max_index, v_max_index, j, vl);
//update v_max
v_max = VFMAXVV_FLOAT(v_max, vx, vl);
@ -125,8 +125,8 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
//index where element greater than v_max
mask = VMFLTVV_FLOAT(v_max, vx, vl);
v_max_index = VIDV_MASK_UINT(mask, vl);
v_max_index = VADDVX_MASK_UINT(mask, v_max_index, j, vl);
v_max_index = VIDV_MASK_UINT(mask, v_max_index, vl);
v_max_index = VADDVX_MASK_UINT(mask, v_max_index, v_max_index, j, vl);
//update v_max
v_max = VFMAXVV_FLOAT(v_max, vx, vl);

16
kernel/riscv64/iamin_rvv.c
View File

@ -46,9 +46,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define VFMINVV_FLOAT __riscv_vfmin_vv_f64m8
#define VFIRSTM __riscv_vfirst_m_b8
#define UINT_V_T vuint64m8_t
#define VIDV_MASK_UINT __riscv_vid_v_u64m8_m
#define VIDV_MASK_UINT __riscv_vid_v_u64m8_mu
#define VIDV_UINT __riscv_vid_v_u64m8
#define VADDVX_MASK_UINT __riscv_vadd_vx_u64m8_m
#define VADDVX_MASK_UINT __riscv_vadd_vx_u64m8_mu
#define VADDVX_UINT __riscv_vadd_vx_u64m8
#define VMVVX_UINT __riscv_vmv_v_x_u64m8
#define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
@ -72,9 +72,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define VFMINVV_FLOAT __riscv_vfmin_vv_f32m8
#define VFIRSTM __riscv_vfirst_m_b4
#define UINT_V_T vuint32m8_t
#define VIDV_MASK_UINT __riscv_vid_v_u32m8_m
#define VIDV_MASK_UINT __riscv_vid_v_u32m8_mu
#define VIDV_UINT __riscv_vid_v_u32m8
#define VADDVX_MASK_UINT __riscv_vadd_vx_u32m8_m
#define VADDVX_MASK_UINT __riscv_vadd_vx_u32m8_mu
#define VADDVX_UINT __riscv_vadd_vx_u32m8
#define VMVVX_UINT __riscv_vmv_v_x_u32m8
#define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
@ -107,8 +107,8 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
// index where element less than v_min
mask = VMFLTVV_FLOAT(vx, v_min, vl);
v_min_index = VIDV_MASK_UINT(mask, vl);
v_min_index = VADDVX_MASK_UINT(mask, v_min_index, j, vl);
v_min_index = VIDV_MASK_UINT(mask, v_min_index, vl);
v_min_index = VADDVX_MASK_UINT(mask, v_min_index, v_min_index, j, vl);
//update v_min and start_index j
v_min = VFMINVV_FLOAT(v_min, vx, vl);
@ -126,8 +126,8 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
// index where element less than v_min
mask = VMFLTVV_FLOAT(vx, v_min, vl);
v_min_index = VIDV_MASK_UINT(mask, vl);
v_min_index = VADDVX_MASK_UINT(mask, v_min_index, j, vl);
v_min_index = VIDV_MASK_UINT(mask, v_min_index, vl);
v_min_index = VADDVX_MASK_UINT(mask, v_min_index, v_min_index, j, vl);
//update v_min and start_index j
v_min = VFMINVV_FLOAT(v_min, vx, vl);

16
kernel/riscv64/imax_rvv.c
View File

@ -45,9 +45,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define VFMAXVV_FLOAT __riscv_vfmax_vv_f64m8
#define VFIRSTM __riscv_vfirst_m_b8
#define UINT_V_T vuint64m8_t
#define VIDV_MASK_UINT __riscv_vid_v_u64m8_m
#define VIDV_MASK_UINT __riscv_vid_v_u64m8_mu
#define VIDV_UINT __riscv_vid_v_u64m8
#define VADDVX_MASK_UINT __riscv_vadd_vx_u64m8_m
#define VADDVX_MASK_UINT __riscv_vadd_vx_u64m8_mu
#define VADDVX_UINT __riscv_vadd_vx_u64m8
#define VMVVX_UINT __riscv_vmv_v_x_u64m8
#define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
@ -70,9 +70,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define VFMAXVV_FLOAT __riscv_vfmax_vv_f32m8
#define VFIRSTM __riscv_vfirst_m_b4
#define UINT_V_T vuint32m8_t
#define VIDV_MASK_UINT __riscv_vid_v_u32m8_m
#define VIDV_MASK_UINT __riscv_vid_v_u32m8_mu
#define VIDV_UINT __riscv_vid_v_u32m8
#define VADDVX_MASK_UINT __riscv_vadd_vx_u32m8_m
#define VADDVX_MASK_UINT __riscv_vadd_vx_u32m8_mu
#define VADDVX_UINT __riscv_vadd_vx_u32m8
#define VMVVX_UINT __riscv_vmv_v_x_u32m8
#define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
@ -104,8 +104,8 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
//index where element greater than v_max
mask = VMFLTVV_FLOAT(v_max, vx, vl);
v_max_index = VIDV_MASK_UINT(mask, vl);
v_max_index = VADDVX_MASK_UINT(mask, v_max_index, j, vl);
v_max_index = VIDV_MASK_UINT(mask, v_max_index, vl);
v_max_index = VADDVX_MASK_UINT(mask, v_max_index, v_max_index, j, vl);
//update v_max and start_index j
v_max = VFMAXVV_FLOAT(v_max, vx, vl);
@ -122,8 +122,8 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
//index where element greater than v_max
mask = VMFLTVV_FLOAT(v_max, vx, vl);
v_max_index = VIDV_MASK_UINT(mask, vl);
v_max_index = VADDVX_MASK_UINT(mask, v_max_index, j, vl);
v_max_index = VIDV_MASK_UINT(mask, v_max_index, vl);
v_max_index = VADDVX_MASK_UINT(mask, v_max_index, v_max_index, j, vl);
//update v_max and start_index j
v_max = VFMAXVV_FLOAT(v_max, vx, vl);

16
kernel/riscv64/imin_rvv.c
View File

@ -45,9 +45,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define VFMINVV_FLOAT __riscv_vfmin_vv_f64m8
#define VFIRSTM __riscv_vfirst_m_b8
#define UINT_V_T vuint64m8_t
#define VIDV_MASK_UINT __riscv_vid_v_u64m8_m
#define VIDV_MASK_UINT __riscv_vid_v_u64m8_mu
#define VIDV_UINT __riscv_vid_v_u64m8
#define VADDVX_MASK_UINT __riscv_vadd_vx_u64m8_m
#define VADDVX_MASK_UINT __riscv_vadd_vx_u64m8_mu
#define VADDVX_UINT __riscv_vadd_vx_u64m8
#define VMVVX_UINT __riscv_vmv_v_x_u64m8
#define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f64m1_f64
@ -70,9 +70,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define VFMINVV_FLOAT __riscv_vfmin_vv_f32m8
#define VFIRSTM __riscv_vfirst_m_b4
#define UINT_V_T vuint32m8_t
#define VIDV_MASK_UINT __riscv_vid_v_u32m8_m
#define VIDV_MASK_UINT __riscv_vid_v_u32m8_mu
#define VIDV_UINT __riscv_vid_v_u32m8
#define VADDVX_MASK_UINT __riscv_vadd_vx_u32m8_m
#define VADDVX_MASK_UINT __riscv_vadd_vx_u32m8_mu
#define VADDVX_UINT __riscv_vadd_vx_u32m8
#define VMVVX_UINT __riscv_vmv_v_x_u32m8
#define VFMVFS_FLOAT_M1 __riscv_vfmv_f_s_f32m1_f32
@ -104,8 +104,8 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
// index where element less than v_min
mask = VMFLTVV_FLOAT(vx, v_min, vl);
v_min_index = VIDV_MASK_UINT(mask, vl);
v_min_index = VADDVX_MASK_UINT(mask, v_min_index, j, vl);
v_min_index = VIDV_MASK_UINT(mask, v_min_index, vl);
v_min_index = VADDVX_MASK_UINT(mask, v_min_index, v_min_index, j, vl);
//update v_min and start_index j
v_min = VFMINVV_FLOAT(v_min, vx, vl);
@ -122,8 +122,8 @@ BLASLONG CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
// index where element less than v_min
mask = VMFLTVV_FLOAT(vx, v_min, vl);
v_min_index = VIDV_MASK_UINT(mask, vl);
v_min_index = VADDVX_MASK_UINT(mask, v_min_index, j, vl);
v_min_index = VIDV_MASK_UINT(mask, v_min_index, vl);
v_min_index = VADDVX_MASK_UINT(mask, v_min_index, v_min_index, j, vl);
//update v_min and start_index j
v_min = VFMINVV_FLOAT(v_min, vx, vl);