Use stride instead of order to determine block attr

[alexbaden]

Per the Triton slack, order is unused on architecture below Hopper. But more importantly, order provides information that stride already has. In fact, order can be completely different from stride (i.e. wrong) and we still generate correct code. I think it is better to use the stride assuming the logic I added here makes sense.

Note this depends on #2348, I'd like to land the debug logging separately, so we have it even if we decide to modify this approach. It was very useful in debugging this problem.

cc #2347

[whitneywhtsang]

can we add a lit test?

[alexbaden]

can we add a lit test?

The existing tests actually cover this scenario - because they change both the order and the stride.

    // CHECK: tt.load {{.*}} {boundaryCheck = array<i32: 1>, padding = 1 : i32, triton_intel_gpu.block_io = "row_major"}
    // CHECK: tt.load {{.*}} {boundaryCheck = array<i32: 0>, padding = 1 : i32, triton_intel_gpu.block_io = "row_major"}
    %3 = tt.make_tensor_ptr %arg0, [%c0_i64, %c0_i64], [%pitch, %c1_i64], [%c0_i32, %c0_i32] {order = array<i32: 1, 0>} : <tensor<64x32xf16, #dot_a>>
    %4 = tt.make_tensor_ptr %arg0, [%c0_i64, %c0_i64], [%pitch, %c1_i64], [%c0_i32, %c0_i32] {order = array<i32: 1, 0>} : <tensor<32x64xf16, #dot_b>>
    %5 = tt.load %3 {boundaryCheck = array<i32: 1>, cache = 1 : i32, evict = 1 : i32, isVolatile = false, padding = 1 : i32} : !tt.ptr<tensor<64x32xf16, #dot_a>>
    %6 = tt.load %4 {boundaryCheck = array<i32: 0>, cache = 1 : i32, evict = 1 : i32, isVolatile = false, padding = 1 : i32} : !tt.ptr<tensor<32x64xf16, #dot_b>>

    // CHECK: tt.load {{.*}} {boundaryCheck = array<i32: 1>, padding = 1 : i32, triton_intel_gpu.block_io = "column_major"}
    // CHECK: tt.load {{.*}} {boundaryCheck = array<i32: 0>, padding = 1 : i32, triton_intel_gpu.block_io = "column_major"}
    %7 = tt.make_tensor_ptr %arg0, [%c0_i64, %c0_i64], [%c1_i64, %pitch], [%c0_i32, %c0_i32] {order = array<i32: 0, 1>} : <tensor<64x32xf16, #dot_a>>
    %8 = tt.make_tensor_ptr %arg0, [%c0_i64, %c0_i64], [%c1_i64, %pitch], [%c0_i32, %c0_i32] {order = array<i32: 0, 1>} : <tensor<32x64xf16, #dot_b>>
    %9 = tt.load %7 {boundaryCheck = array<i32: 1>, cache = 1 : i32, evict = 1 : i32, isVolatile = false, padding = 1 : i32} : !tt.ptr<tensor<64x32xf16, #dot_a>>
    %10 = tt.load %8 {boundaryCheck = array<i32: 0>, cache = 1 : i32, evict = 1 : i32, isVolatile = false, padding = 1 : i32} : !tt.ptr<tensor<32x64xf16, #dot_b>>

But, I added one that covers this scenario + rewrite tensor pointer here: #2347