[XPU][TritonIntelGPUToLLVM] Add support for more transpose kinds (#2786)

Add support for layout conversion transposes in which rows managed by a
single thread are contiguous in the output matrix.

---------

Signed-off-by: victor-eds <[email protected]>
Co-authored-by: Ettore Tiotto <[email protected]>

Author: victor-eds

test/Conversion/intel/intel-allocate-shared-memory.mlir

@@ -81,3 +81,19 @@ module attributes {"triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 8 :
     tt.return %0 : tensor<128x64xf32, #blocked1>
   }
 }
+
+// -----
+
+#blocked = #triton_gpu.blocked<{sizePerThread = [1, 32], threadsPerWarp = [16, 1], warpsPerCTA = [2, 4], order = [0, 1]}>
+#blocked1 = #triton_gpu.blocked<{sizePerThread = [16, 2], threadsPerWarp = [1, 16], warpsPerCTA = [2, 4], order = [0, 1]}>
+
+// Check scracth memory configuration for different sub-group transpose-like layout conversions.
+
+// CHECK-LABEL: module attributes
+// CHECK-SAME: triton_gpu.shared = 17408 : i32
+module attributes {"triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 8 : i32, "triton_gpu.threads-per-warp" = 16 : i32} {
+  tt.func @test_contiguous(%arg0: tensor<32x128xf32, #blocked>) -> tensor<32x128xf32, #blocked1> {
+    %0 = triton_gpu.convert_layout %arg0 : tensor<32x128xf32, #blocked> -> tensor<32x128xf32, #blocked1>
+    tt.return %0 : tensor<32x128xf32, #blocked1>
+  }
+}