diff --git a/examples/sem_reagan.py b/examples/sem_reagan.py
index 88dde57a6c5927ecb6f13bd0a6ce213008438fc9..42c3dcfec32907dd66cb125a0a51776636fcc720 100644
--- a/examples/sem_reagan.py
+++ b/examples/sem_reagan.py
@@ -35,129 +35,6 @@ from pyopencl.tools import pytest_generate_tests_for_pyopencl \
-def test_tim2d(ctx_factory):
- dtype = np.float32
- ctx = ctx_factory()
- order = "C"
-
- n = 8
-
- from pymbolic import var
- K_sym = var("K")
-
- field_shape = (K_sym, n, n)
-
- # K - run-time symbolic
- knl = lp.make_kernel(ctx.devices[0],
- "[K] -> {[i,j,e,m,o,gi]: 0<=i,j,m,o<%d and 0<=e<K and 0<=gi<3}" % n,
- [
- "ur(a,b) := sum(@o, D[a,o]*u[e,o,b])",
- "us(a,b) := sum(@o, D[b,o]*u[e,a,o])",
-
- #"Gu(mat_entry,a,b) := G[mat_entry,e,m,j]*ur(m,j)",
-
- "Gux(a,b) := G$x[0,e,a,b]*ur(a,b)+G$x[1,e,a,b]*us(a,b)",
- "Guy(a,b) := G$y[1,e,a,b]*ur(a,b)+G$y[2,e,a,b]*us(a,b)",
- "lap[e,i,j] = "
- " sum(m, D[m,i]*Gux(m,j))"
- "+ sum(m, D[m,j]*Guy(i,m))"
-
- ],
- [
- lp.GlobalArg("u", dtype, shape=field_shape, order=order),
- lp.GlobalArg("lap", dtype, shape=field_shape, order=order),
- lp.GlobalArg("G", dtype, shape=(3,)+field_shape, order=order),
- # lp.ConstantArrayArg("D", dtype, shape=(n, n), order=order),
- lp.GlobalArg("D", dtype, shape=(n, n), order=order),
- # lp.ImageArg("D", dtype, shape=(n, n)),
- lp.ValueArg("K", np.int32, approximately=1000),
- ],
- name="semlap2D", assumptions="K>=1")
-
- seq_knl = knl
-
- def variant_orig(knl):
- knl = lp.tag_inames(knl, dict(i="l.0", j="l.1", e="g.0"))
-
- knl = lp.add_prefetch(knl, "D[:,:]")
- knl = lp.add_prefetch(knl, "u[e, :, :]")
-
- knl = lp.precompute(knl, "ur(m,j)", np.float32, ["m", "j"])
- knl = lp.precompute(knl, "us(i,m)", np.float32, ["i", "m"])
-
- knl = lp.precompute(knl, "Gux(m,j)", np.float32, ["m", "j"])
- knl = lp.precompute(knl, "Guy(i,m)", np.float32, ["i", "m"])
-
- knl = lp.add_prefetch(knl, "G$x[:,e,:,:]")
- knl = lp.add_prefetch(knl, "G$y[:,e,:,:]")
-
- knl = lp.tag_inames(knl, dict(o="unr"))
- knl = lp.tag_inames(knl, dict(m="unr"))
-
- knl = lp.set_instruction_priority(knl, "D_fetch", 5)
- print knl
-
- return knl
-
- for variant in [variant_orig]:
- kernel_gen = lp.generate_loop_schedules(variant(knl))
- kernel_gen = lp.check_kernels(kernel_gen, dict(K=1000))
-
- K = 1000
- lp.auto_test_vs_ref(seq_knl, ctx, kernel_gen,
- op_count=[K*(n*n*n*2*2 + n*n*2*3 + n**3 * 2*2)/1e9],
- op_label=["GFlops"],
- parameters={"K": K})
-
-
-
-
-def make_me_a_test_test_tim3d_slab(ctx_factory):
- dtype = np.float32
- ctx = ctx_factory()
- order = "C"
-
- n = 8
-
- from pymbolic import var
-
- # K - run-time symbolic
- knl = lp.make_kernel(ctx.devices[0],
- "[E] -> {[i,j,k, o, e]: 0<=i,j,k,o < n and 0<=e<E }",
- """
- <> ur[i,j,k] = sum(@o, D[i,o]*u[e,o,j,k])
- <> us[i,j,k] = sum(@o, D[j,o]*u[e,i,o,k])
- <> ut[i,j,k] = sum(@o, D[k,o]*u[e,i,j,o])
- """,
- [
- lp.GlobalArg("u", dtype, shape="E,n,n,n", order=order),
- # lp.GlobalArg("G", dtype, shape=(3,)+field_shape, order=order),
- # lp.ConstantArrayArg("D", dtype, shape=(n, n), order=order),
- lp.GlobalArg("D", dtype, shape=(n, n), order=order),
- # lp.ImageArg("D", dtype, shape=(n, n)),
- lp.ValueArg("E", np.int32, approximately=1000),
- ],
- name="semdiff3D", assumptions="E>=1",
- defines={"n": n})
-
- seq_knl = knl
-
- def variant_orig(knl):
- return knl
-
- for variant in [variant_orig]:
- kernel_gen = lp.generate_loop_schedules(variant(knl))
- kernel_gen = lp.check_kernels(kernel_gen, dict(K=1000))
-
- E = 1000
- lp.auto_test_vs_ref(seq_knl, ctx, kernel_gen,
- op_count=[E*(n*n*n*2*2 + n*n*2*3 + n**3 * 2*2)/1e9],
- op_label=["GFlops"],
- parameters={"E": E})
-
-
-
-
def test_tim3d_slab(ctx_factory):
dtype = np.float32
ctx = ctx_factory()
@@ -165,9 +42,6 @@ def test_tim3d_slab(ctx_factory):
Nq = 8
- from pymbolic import var
-
- # K - run-time symbolic
knl = lp.make_kernel(ctx.devices[0],
"[E] -> {[i,j, k, o,m, e]: 0<=i,j,k,o,m < Nq and 0<=e<E }",
"""
@@ -179,9 +53,9 @@ def test_tim3d_slab(ctx_factory):
Gus(a,b,c) := G[1,e,a,b,c]*ur(a,b,c)+G[3,e,a,b,c]*us(a,b,c)+G[4,e,a,b,c]*ut(a,b,c)
Gut(a,b,c) := G[2,e,a,b,c]*ur(a,b,c)+G[4,e,a,b,c]*us(a,b,c)+G[5,e,a,b,c]*ut(a,b,c)
- lapr(a,b,c):= sum(m, D[m,a]*Gur(m,b,c))
- laps(a,b,c):= sum(m, D[m,b]*Gus(a,m,c))
- lapt(a,b,c):= sum(m, D[m,c]*Gut(a,b,m))
+ lapr(a,b,c):= sum(m, D[m,a]*Gur(m,b,c))
+ laps(a,b,c):= sum(m, D[m,b]*Gus(a,m,c))
+ lapt(a,b,c):= sum(m, D[m,c]*Gut(a,b,m))
lap[e,i,j,k] = lapr(i,j,k) + laps(i,j,k) + lapt(i,j,k)
""",
@@ -194,7 +68,7 @@ def test_tim3d_slab(ctx_factory):
lp.ValueArg("E", np.int32, approximately=1000),
],
name="semdiff3D", assumptions="E>=1",
- defines={"Nq": Nq})
+ defines={"Nq": Nq})
for derivative in "rst":
knl = lp.duplicate_inames(knl, "o", within="... < lap"+derivative, suffix="_"+derivative)
@@ -203,7 +77,7 @@ def test_tim3d_slab(ctx_factory):
def variant_orig(knl):
return knl
- knl = lp.tag_inames(knl, dict(e="g.0", i="l.0", j="l.1"), )
+ knl = lp.tag_inames(knl, dict(e="g.0", i="l.0", j="l.1"), )
for derivative, par_names in [
("r", ["j", "k"]),
@@ -214,7 +88,7 @@ def test_tim3d_slab(ctx_factory):
print knl
- return knl
+ return knl
#print lp.preprocess_kernel(knl)
#1/0