[Loop Specialization]: Specialize loops containing masked operations with loop invariant mask (#3586)

Version loops containing masked operations (e.g. tt.load with a mask)
where the mask is loop invariant.

---------

Signed-off-by: Tiotto, Ettore <[email protected]>

Author: etiotto

test/Triton/Intel/RaiseToBlockPointers/kernel-03-matrix-multiplication.mlir

@@ -1,6 +1,7 @@
 // RUN: triton-opt %s -triton-intel-remove-masks -triton-raise-block-pointer -canonicalize | FileCheck %s
 
 module {
+  // COM: Derived from tutorial 03-matrix-multiplication.
   tt.func public @matmul_kernel(%arg0: !tt.ptr<f16> {tt.divisibility = 16 : i32}, %arg1: !tt.ptr<f16> {tt.divisibility = 16 : i32}, %arg2: !tt.ptr<f16> {tt.divisibility = 16 : i32}, %arg3: i32 {tt.divisibility = 16 : i32}, %arg4: i32 {tt.divisibility = 16 : i32}, %arg5: i32 {tt.divisibility = 16 : i32}, %arg6: i32 {tt.divisibility = 16 : i32}, %arg7: i32 {tt.divisibility = 16 : i32}, %arg8: i32 {tt.divisibility = 16 : i32}) {
     %c31_i32 = arith.constant 31 : i32
     %cst = arith.constant dense<0.000000e+00> : tensor<64x128xf32>

third_party/intel/backend/compiler.py

@@ -224,6 +224,8 @@ def make_ttir(mod, metadata, opt):
         pm.enable_debug()
         passes.common.add_inliner(pm)
         passes.ttir.add_combine(pm)
+        passes.common.add_cse(pm)
+        passes.common.add_licm(pm)
         intel.passes.ttir.add_remove_masks(pm)
         if raise_block_ptr_flags['enabled']:
             ignore_masks = True if raise_block_ptr_flags['ignore-masks'] else False