Skip to content

Zeros-init + well-conditioned Gram-Schmidt; add dace_canonicalize_{cpu,gpu} frameworks #47

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
ThrudPrimrose wants to merge 4 commits into
base: main
Choose a base branch
from
Open

Zeros-init + well-conditioned Gram-Schmidt; add dace_canonicalize_{cpu,gpu} frameworks #47

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
0a58d76
Zero-initialize output buffers; make Gram-Schmidt input well-conditioned
ThrudPrimrose Jul 11, 2026
9febb1d
Add dace_canonicalize_cpu / dace_canonicalize_gpu frameworks
ThrudPrimrose Jul 11, 2026
c77d0f4
Draw mlp's input from the seeded generator like the rest of it
ThrudPrimrose Aug 6, 2026
3944369
Name each kernel what the runner looks up
ThrudPrimrose Aug 6, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 5 additions & 1 deletion README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -58,7 +58,11 @@ python -m pip install dace
```
However, you may want to install the latest version from the [GitHub repository](https://github.com/spcl/dace).
To run NPBench with DaCe, you have to select as framework (see details below)
either `dace_cpu` or `dace_gpu`.
either `dace_cpu` or `dace_gpu` (which build the SDFG and `auto_optimize` it), or
`dace_canonicalize_cpu` / `dace_canonicalize_gpu` (which instead run DaCe's
`canonicalize` pipeline, leaving `sdfg.openmp_array_reductions` on so whole-buffer
WCR accumulators lower to OpenMP array-section reductions rather than per-element
atomics).

### DPNP

Expand Down
10 changes: 10 additions & 0 deletions framework_info/dace_canonicalize_cpu.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,10 @@
{
"framework": {
"simple_name": "dace_canonicalize_cpu",
"full_name": "DaCe Canonicalize CPU",
"prefix": "dc",
"postfix": "dace",
"class": "DaceCanonicalizeFramework",
"arch": "cpu"
}
}
10 changes: 10 additions & 0 deletions framework_info/dace_canonicalize_gpu.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,10 @@
{
"framework": {
"simple_name": "dace_canonicalize_gpu",
"full_name": "DaCe Canonicalize GPU",
"prefix": "dc",
"postfix": "dace",
"class": "DaceCanonicalizeFramework",
"arch": "gpu"
}
}
Binary file added npbench.db
View file
Open in desktop
Binary file not shown.
6 changes: 3 additions & 3 deletions npbench/benchmarks/cavity_flow/cavity_flow_cupy.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,7 +22,7 @@ def build_up_b(b, rho, dt, u, v, dx, dy):


def pressure_poisson(nit, p, dx, dy, b):
pn = np.empty_like(p)
pn = np.zeros_like(p)
pn = p.copy()

for q in range(nit):
Expand All @@ -39,8 +39,8 @@ def pressure_poisson(nit, p, dx, dy, b):


def cavity_flow(nx, ny, nt, nit, u, v, dt, dx, dy, p, rho, nu):
un = np.empty_like(u)
vn = np.empty_like(v)
un = np.zeros_like(u)
vn = np.zeros_like(v)
b = np.zeros((ny, nx))

for n in range(nt):
Expand Down
6 changes: 3 additions & 3 deletions npbench/benchmarks/cavity_flow/cavity_flow_dace.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -30,7 +30,7 @@ def build_up_b(b: dc.float64[ny, nx], rho: dc.float64, dt: dc.float64,
@dc.program
def pressure_poisson(p: dc.float64[ny, nx], dx: dc.float64, dy: dc.float64,
b: dc.float64[ny, nx]):
pn = np.empty_like(p)
pn = np.zeros_like(p)
pn[:] = p.copy()

for q in range(nit):
Expand All @@ -51,8 +51,8 @@ def cavity_flow(nt: dc.int64, nit: dc.int64, u: dc.float64[ny, nx],
v: dc.float64[ny, nx], dt: dc.float64, dx: dc.float64,
dy: dc.float64, p: dc.float64[ny, nx], rho: dc.float64,
nu: dc.float64):
un = np.empty_like(u)
vn = np.empty_like(v)
un = np.zeros_like(u)
vn = np.zeros_like(v)
b = np.zeros((ny, nx))

for n in range(nt):
Expand Down
148 changes: 74 additions & 74 deletions npbench/benchmarks/cavity_flow/cavity_flow_dpnp.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,74 +1,74 @@
import dpnp as np
def build_up_b(b, rho, dt, u, v, dx, dy):
b[1:-1, 1:-1] = (
rho * (
(1 / dt * ((u[1:-1, 2:] - u[1:-1, 0:-2]) / (2 * dx) +
(v[2:, 1:-1] - v[0:-2, 1:-1]) / (2 * dy))) -
((u[1:-1, 2:] - u[1:-1, 0:-2]) / (2 * dx))**2 -
2 * ((u[2:, 1:-1] - u[0:-2, 1:-1]) / (2 * dy) *
(v[1:-1, 2:] - v[1:-1, 0:-2]) / (2 * dx)) -
((v[2:, 1:-1] - v[0:-2, 1:-1]) / (2 * dy))**2
)
)
def pressure_poisson(nit, p, dx, dy, b):
pn = np.empty_like(p)
for q in range(nit):
pn[:] = p.copy()
p[1:-1, 1:-1] = (
((pn[1:-1, 2:] + pn[1:-1, 0:-2]) * dy**2 +
(pn[2:, 1:-1] + pn[0:-2, 1:-1]) * dx**2) /
(2 * (dx**2 + dy**2)) - dx**2 * dy**2 /
(2 * (dx**2 + dy**2)) * b[1:-1, 1:-1]
)
p[:, -1] = p[:, -2] # dp/dx = 0 at x = 2
p[0, :] = p[1, :] # dp/dy = 0 at y = 0
p[:, 0] = p[:, 1] # dp/dx = 0 at x = 0
p[-1, :] = 0 # p = 0 at y = 2
def cavity_flow(nx, ny, nt, nit, u, v, dt, dx, dy, p, rho, nu):
un = np.empty_like(u)
vn = np.empty_like(v)
b = np.zeros((ny, nx))
for n in range(nt):
un[:] = u.copy()
vn[:] = v.copy()
build_up_b(b, rho, dt, u, v, dx, dy)
pressure_poisson(nit, p, dx, dy, b)
u[1:-1, 1:-1] = (
un[1:-1, 1:-1] - un[1:-1, 1:-1] * dt / dx *
(un[1:-1, 1:-1] - un[1:-1, 0:-2]) -
vn[1:-1, 1:-1] * dt / dy *
(un[1:-1, 1:-1] - un[0:-2, 1:-1]) - dt / (2 * rho * dx) *
(p[1:-1, 2:] - p[1:-1, 0:-2]) + nu *
(dt / dx**2 *
(un[1:-1, 2:] - 2 * un[1:-1, 1:-1] + un[1:-1, 0:-2]) +
dt / dy**2 *
(un[2:, 1:-1] - 2 * un[1:-1, 1:-1] + un[0:-2, 1:-1]))
)
v[1:-1, 1:-1] = (
vn[1:-1, 1:-1] - un[1:-1, 1:-1] * dt / dx *
(vn[1:-1, 1:-1] - vn[1:-1, 0:-2]) -
vn[1:-1, 1:-1] * dt / dy *
(vn[1:-1, 1:-1] - vn[0:-2, 1:-1]) - dt / (2 * rho * dy) *
(p[2:, 1:-1] - p[0:-2, 1:-1]) + nu *
(dt / dx**2 *
(vn[1:-1, 2:] - 2 * vn[1:-1, 1:-1] + vn[1:-1, 0:-2]) +
dt / dy**2 *
(vn[2:, 1:-1] - 2 * vn[1:-1, 1:-1] + vn[0:-2, 1:-1]))
)
u[0, :] = 0
u[:, 0] = 0
u[:, -1] = 0
u[-1, :] = 1 # set velocity on cavity lid equal to 1
v[0, :] = 0
v[-1, :] = 0
v[:, 0] = 0
v[:, -1] = 0
import dpnp as np

def build_up_b(b, rho, dt, u, v, dx, dy):
b[1:-1, 1:-1] = (
rho * (
(1 / dt * ((u[1:-1, 2:] - u[1:-1, 0:-2]) / (2 * dx) +
(v[2:, 1:-1] - v[0:-2, 1:-1]) / (2 * dy))) -
((u[1:-1, 2:] - u[1:-1, 0:-2]) / (2 * dx))**2 -
2 * ((u[2:, 1:-1] - u[0:-2, 1:-1]) / (2 * dy) *
(v[1:-1, 2:] - v[1:-1, 0:-2]) / (2 * dx)) -
((v[2:, 1:-1] - v[0:-2, 1:-1]) / (2 * dy))**2
)
)

def pressure_poisson(nit, p, dx, dy, b):
pn = np.zeros_like(p)
for q in range(nit):
pn[:] = p.copy()
p[1:-1, 1:-1] = (
((pn[1:-1, 2:] + pn[1:-1, 0:-2]) * dy**2 +
(pn[2:, 1:-1] + pn[0:-2, 1:-1]) * dx**2) /
(2 * (dx**2 + dy**2)) - dx**2 * dy**2 /
(2 * (dx**2 + dy**2)) * b[1:-1, 1:-1]
)

p[:, -1] = p[:, -2] # dp/dx = 0 at x = 2
p[0, :] = p[1, :] # dp/dy = 0 at y = 0
p[:, 0] = p[:, 1] # dp/dx = 0 at x = 0
p[-1, :] = 0 # p = 0 at y = 2

def cavity_flow(nx, ny, nt, nit, u, v, dt, dx, dy, p, rho, nu):
un = np.zeros_like(u)
vn = np.zeros_like(v)
b = np.zeros((ny, nx))

for n in range(nt):
un[:] = u.copy()
vn[:] = v.copy()

build_up_b(b, rho, dt, u, v, dx, dy)
pressure_poisson(nit, p, dx, dy, b)

u[1:-1, 1:-1] = (
un[1:-1, 1:-1] - un[1:-1, 1:-1] * dt / dx *
(un[1:-1, 1:-1] - un[1:-1, 0:-2]) -
vn[1:-1, 1:-1] * dt / dy *
(un[1:-1, 1:-1] - un[0:-2, 1:-1]) - dt / (2 * rho * dx) *
(p[1:-1, 2:] - p[1:-1, 0:-2]) + nu *
(dt / dx**2 *
(un[1:-1, 2:] - 2 * un[1:-1, 1:-1] + un[1:-1, 0:-2]) +
dt / dy**2 *
(un[2:, 1:-1] - 2 * un[1:-1, 1:-1] + un[0:-2, 1:-1]))
)

v[1:-1, 1:-1] = (
vn[1:-1, 1:-1] - un[1:-1, 1:-1] * dt / dx *
(vn[1:-1, 1:-1] - vn[1:-1, 0:-2]) -
vn[1:-1, 1:-1] * dt / dy *
(vn[1:-1, 1:-1] - vn[0:-2, 1:-1]) - dt / (2 * rho * dy) *
(p[2:, 1:-1] - p[0:-2, 1:-1]) + nu *
(dt / dx**2 *
(vn[1:-1, 2:] - 2 * vn[1:-1, 1:-1] + vn[1:-1, 0:-2]) +
dt / dy**2 *
(vn[2:, 1:-1] - 2 * vn[1:-1, 1:-1] + vn[0:-2, 1:-1]))
)

u[0, :] = 0
u[:, 0] = 0
u[:, -1] = 0
u[-1, :] = 1 # set velocity on cavity lid equal to 1
v[0, :] = 0
v[-1, :] = 0
v[:, 0] = 0
v[:, -1] = 0
6 changes: 3 additions & 3 deletions npbench/benchmarks/cavity_flow/cavity_flow_legate.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,7 +22,7 @@ def build_up_b(b, rho, dt, u, v, dx, dy):


def pressure_poisson(nit, p, dx, dy, b):
pn = np.empty_like(p)
pn = np.zeros_like(p)
pn = p.copy()

for q in range(nit):
Expand All @@ -39,8 +39,8 @@ def pressure_poisson(nit, p, dx, dy, b):


def cavity_flow(nx, ny, nt, nit, u, v, dt, dx, dy, p, rho, nu):
un = np.empty_like(u)
vn = np.empty_like(v)
un = np.zeros_like(u)
vn = np.zeros_like(v)
b = np.zeros((ny, nx))

for n in range(nt):
Expand Down
6 changes: 3 additions & 3 deletions npbench/benchmarks/cavity_flow/cavity_flow_numba_n.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,7 @@ def build_up_b(b, rho, dt, u, v, dx, dy):

@nb.jit(nopython=True, parallel=False, fastmath=True)
def pressure_poisson(nit, p, dx, dy, b):
pn = np.empty_like(p)
pn = np.zeros_like(p)
pn = p.copy()

for q in range(nit):
Expand All @@ -34,8 +34,8 @@ def pressure_poisson(nit, p, dx, dy, b):

@nb.jit(nopython=True, parallel=False, fastmath=True)
def cavity_flow(nx, ny, nt, nit, u, v, dt, dx, dy, p, rho, nu):
un = np.empty_like(u)
vn = np.empty_like(v)
un = np.zeros_like(u)
vn = np.zeros_like(v)
b = np.zeros((ny, nx))

for n in range(nt):
Expand Down
8 changes: 4 additions & 4 deletions npbench/benchmarks/cavity_flow/cavity_flow_numba_np.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,7 @@ def build_up_b(b, rho, dt, u, v, dx, dy):

@nb.jit(nopython=True, parallel=True, fastmath=True)
def pressure_poisson(nit, p, dx, dy, b):
pn = np.empty_like(p)
pn = np.zeros_like(p)
pn = p.copy()

for q in range(nit):
Expand All @@ -33,9 +33,9 @@ def pressure_poisson(nit, p, dx, dy, b):


@nb.jit(nopython=True, parallel=True, fastmath=True)
def nopython_mode(nx, ny, nt, nit, u, v, dt, dx, dy, p, rho, nu):
un = np.empty_like(u)
vn = np.empty_like(v)
def cavity_flow(nx, ny, nt, nit, u, v, dt, dx, dy, p, rho, nu):
un = np.zeros_like(u)
vn = np.zeros_like(v)
b = np.zeros((ny, nx))

for n in range(nt):
Expand Down
6 changes: 3 additions & 3 deletions npbench/benchmarks/cavity_flow/cavity_flow_numpy.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,7 +22,7 @@ def build_up_b(b, rho, dt, u, v, dx, dy):


def pressure_poisson(nit, p, dx, dy, b):
pn = np.empty_like(p)
pn = np.zeros_like(p)
pn = p.copy()

for q in range(nit):
Expand All @@ -39,8 +39,8 @@ def pressure_poisson(nit, p, dx, dy, b):


def cavity_flow(nx, ny, nt, nit, u, v, dt, dx, dy, p, rho, nu):
un = np.empty_like(u)
vn = np.empty_like(v)
un = np.zeros_like(u)
vn = np.zeros_like(v)
b = np.zeros((ny, nx))

for n in range(nt):
Expand Down
6 changes: 3 additions & 3 deletions npbench/benchmarks/cavity_flow/cavity_flow_pythran.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,7 +26,7 @@ def build_up_b(b, rho, dt, u, v, dx, dy):
# pythran export pressure_poisson(int64, float64[:,:], float64, float64,
# float64[:,:])
def pressure_poisson(nit, p, dx, dy, b):
pn = np.empty_like(p)
pn = np.zeros_like(p)
pn = p.copy()

for q in range(nit):
Expand All @@ -46,8 +46,8 @@ def pressure_poisson(nit, p, dx, dy, b):
# float64[:,:], float64, float64, float64,
# float64[:,:], float64, float64)
def cavity_flow(nx, ny, nt, nit, u, v, dt, dx, dy, p, rho, nu):
un = np.empty_like(u)
vn = np.empty_like(v)
un = np.zeros_like(u)
vn = np.zeros_like(v)
b = np.zeros((ny, nx))

for n in range(nt):
Expand Down
2 changes: 1 addition & 1 deletion npbench/benchmarks/channel_flow/channel_flow_cupy.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,7 +40,7 @@ def build_up_b(rho, dt, dx, dy, u, v):


def pressure_poisson_periodic(nit, p, dx, dy, b):
pn = np.empty_like(p)
pn = np.zeros_like(p)

for q in range(nit):
pn = p.copy()
Expand Down
2 changes: 1 addition & 1 deletion npbench/benchmarks/channel_flow/channel_flow_dace.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,7 +47,7 @@ def build_up_b(rho: dc.float64, dt: dc.float64, dx: dc.float64, dy: dc.float64,
@dc.program
def pressure_poisson_periodic(p: dc.float64[ny, nx], dx: dc.float64,
dy: dc.float64, b: dc.float64[ny, nx]):
pn = np.empty_like(p)
pn = np.zeros_like(p)

for q in range(nit):
pn[:] = p.copy()
Expand Down
Loading
Loading