[nvq++] Removes MLIR's memref dialect. (#2438)

* [nvq++] Removes MLIR's `scf` dialect

We don't use this dialect and its presence is a historical artifact.
This change triggered the removal of two tests:

* `test/Quake/ghz.qke`
* `test/Quake/iqft.qke`

Both tests are a reminder of a past when we had to write quantum kernels
directly in MLIR because of a lack of frontend. Both no longer test
aything useful.

The commit modifies `test/Quake/canonical-2.qke`, which was only testing
the canonicalization of `cc.scope` operations. The new form is removes
the clutter, making the test more precise.

`test/Translate/ghz.qke` had to be modified because it uses MLIR's
`affined.for` and its conversion to LLVMDialect requires `scf.for`.

Signed-off-by: boschmitt <[email protected]>

* [nvq++] Remove MLIR's `affine` dialect.

We don't use this dialect. The removal triggered the removel of one
test:

* `test/Quake/ccnot.qke`

The test was not testing anything useful, only kernel inlining, which is
already covered in other tests. Furthermore, the test is misleading
because, contrary to what the kernel name might indicate, it is not
implementing a `ccnot`.

Signed-off-by: boschmitt <[email protected]>

* [nvq++] Removes MLIR's `memref` dialect.

We don't use this dialect. Most changes were to substitute `memref` ops
with `cc` ops in tests.

Signed-off-by: boschmitt <[email protected]>

---------

Signed-off-by: boschmitt <[email protected]>
Signed-off-by: Eric Schweitz <[email protected]>
Co-authored-by: Eric Schweitz <[email protected]>

include/cudaq/Optimizer/InitAllDialects.h

@@ -16,7 +16,6 @@
 #include "mlir/Dialect/Func/IR/FuncOps.h"
 #include "mlir/Dialect/LLVMIR/LLVMDialect.h"
 #include "mlir/Dialect/Math/IR/Math.h"
-#include "mlir/Dialect/MemRef/IR/MemRef.h"
 
 namespace cudaq {
 
@@ -31,7 +30,6 @@ inline void registerAllDialects(mlir::DialectRegistry &registry) {
     mlir::func::FuncDialect,
     mlir::LLVM::LLVMDialect,
     mlir::math::MathDialect,
-    mlir::memref::MemRefDialect,
 
     // CUDA-Q dialects
     cudaq::cc::CCDialect,

test/Quake/lambda_kernel_exec.qke

@@ -27,10 +27,10 @@ module attributes {quake.mangled_name_map = {__nvqpp__mlirgen__lambda.main.canHa
     %4 = arith.extsi %c0_i32_0 : i32 to i64
     %5 = quake.extract_ref %1[%4] : (!quake.veq<?>, i64) -> !quake.ref
     %16 = quake.mz %5 name "b" : (!quake.ref) -> !quake.measure
-    %alloca = memref.alloca() : memref<i1>
+    %alloca = cc.alloca i1
     %6 = quake.discriminate %16 : (!quake.measure) -> i1
-    memref.store %6, %alloca[] : memref<i1>
-    %7 = memref.load %alloca[] : memref<i1>
+    cc.store %6, %alloca : !cc.ptr<i1>
+    %7 = cc.load %alloca : !cc.ptr<i1>
     cc.if(%7) {
       cc.scope {
         %c1_i32_1 = arith.constant 1 : i32
@@ -67,9 +67,9 @@ module attributes {quake.mangled_name_map = {__nvqpp__mlirgen__lambda.main.canHa
     %5 = quake.extract_ref %1[%4] : (!quake.veq<?>,i64) -> !quake.ref
     %16 = quake.mz %5 name "b" : (!quake.ref) -> !quake.measure
     %6 = quake.discriminate %16 : (!quake.measure) -> i1
-    %alloca = memref.alloca() : memref<i1>
-    memref.store %6, %alloca[] : memref<i1>
-    %7 = memref.load %alloca[] : memref<i1>
+    %alloca = cc.alloca i1
+    cc.store %6, %alloca : !cc.ptr<i1>
+    %7 = cc.load %alloca : !cc.ptr<i1>
     cc.if(%7) {
       cc.scope {
         %c1_i32_1 = arith.constant 1 : i32