Change label syntax. Add syntax for forced iname deps.

MEMO

@@ -37,26 +37,30 @@ TODO: variable shuffle detection
 Things to consider
 ^^^^^^^^^^^^^^^^^^
 
-- implemented_domain may end up being smaller than requested in cse
-  evaluations--check that!
-
 - Depedencies are pointwise for shared loop dimensions
   and global over non-shared ones (between dependent and ancestor)
 
 - multiple insns could fight over which iname gets local axis 0
   -> complicated optimization problem
 
+- Every loop in loopy is opened at most once.
+  Too restrictive?
+
+- Loop bounds currently may not depend on parallel dimensions
+  Does it make sense to relax this?
+
 TODO
 ^^^^
 
+- implemented_domain may end up being smaller than requested in cse
+  evaluations--check that!
+
 - Parallel dimension splitting/merging via tags
 
 - FIXME: Deal with insns losing a seq iname dep in a CSE realization
 
   a <- cse(reduce(stuff))
 
-- Every loop in loopy is opened at most once.
-
 - reimplement add_prefetch
 
 - user interface for dim length prescription
@@ -71,13 +75,17 @@ TODO
 
 - Sharing of checks across ILP instances
 
-- Loop bounds currently may not depend on parallel dimensions
-  Does it make sense to relax this?
+- Some things involving CSEs might be impossible to schedule
+  a[i,j] = cse(b[i]) * cse(c[j])
+
+- Flag, exploit idempotence
 
 Dealt with
 ^^^^^^^^^^
 
-- make syntax for explicit loop dependencies
+- Make syntax for iname dependencies
+
+- make syntax for insn dependencies
 
 - Implement get_problems()
 

loopy/kernel.py

@@ -404,15 +404,17 @@ class LoopKernel(Record):
         """
         import re
         LABEL_DEP_RE = re.compile(
-                r"^(?:\{(?P<label>\w+)\})?"
+                r"^\s*(?:(?P<label>\w+):)?"
+                "\s*(?:\[(?P<iname_deps>[\s\w,]+)\])?"
                 "\s*(?P<lhs>.+)\s*=\s*(?P<rhs>.+?)\s*?"
-                "(?:\:\s*(?P<deps>[\s\w,]+))?$"
+                "(?:\:\s*(?P<insn_deps>[\s\w,]+))?$"
                 )
 
         def parse_if_necessary(insn):
             from pymbolic import parse
 
-            deps = []
+            insn_deps = []
+            forced_iname_deps = []
             label = "insn"
 
             if isinstance(insn, Instruction):
@@ -425,8 +427,10 @@ class LoopKernel(Record):
                 groups = label_dep_match.groupdict()
                 if groups["label"] is not None:
                     label = groups["label"]
-                if groups["deps"] is not None:
-                    deps = [dep.strip() for dep in groups["deps"].split(",")]
+                if groups["insn_deps"] is not None:
+                    insn_deps = [dep.strip() for dep in groups["insn_deps"].split(",")]
+                if groups["iname_deps"] is not None:
+                    forced_iname_deps = [dep.strip() for dep in groups["iname_deps"].split(",")]
 
                 lhs = parse(groups["lhs"])
                 from loopy.symbolic import FunctionToPrimitiveMapper
@@ -434,7 +438,8 @@ class LoopKernel(Record):
 
             return Instruction(
                     id=self.make_unique_instruction_id(insns, based_on=label),
-                    insn_deps=deps,
+                    insn_deps=insn_deps,
+                    forced_iname_deps=forced_iname_deps,
                     assignee=lhs, expression=rhs)
 
         if isinstance(domain, str):

test/test_matmul.py

@@ -203,7 +203,7 @@ def test_plain_matrix_mul_new_ui(ctx_factory):
     knl = lp.LoopKernel(ctx.devices[0],
             "[n] -> {[i,j,k]: 0<=i,j,k<n}",
             [
-                "{label} c[i, j] = sum_float32(k, cse(a[i, k], lhsmat)*cse(b[k, j], rhsmat))"
+                "label: c[i, j] = sum_float32(k, cse(a[i, k], lhsmat)*cse(b[k, j], rhsmat))"
                 ],
             [
                 lp.ArrayArg("a", dtype, shape=(n, n), order=order),