Revert semantic changes to tests and add new ones

Author: victor-eds

test/Triton/invalid.mlir

@@ -38,7 +38,7 @@ tt.func public @fn(%arg0: tensor<128xf32>, %arg1: tensor<64xf32>) {
 
 tt.func public @reshape_different_num_elements(%arg0: tensor<32x128xf16>) {
     // expected-error @+1 {{number of src and dst elements of reshape must be the same}}
-    %a = tt.reshape %arg0 allow_reorder : tensor<32x128xf16> -> tensor<64x32xf16>
+    %a = tt.reshape %arg0 : tensor<32x128xf16> -> tensor<64x32xf16>
     tt.return
 }
 

test/Triton/ops.mlir

@@ -225,11 +225,32 @@ tt.func @inline_asm_scalar(%0: i32) {
 
 // CHECK-LABEL: reshape
 tt.func @reshape(%0: tensor<512xi32>) {
+  // CHECK: tt.reshape %{{.+}} : tensor<512xi32> -> tensor<16x32xi32>
+  %1 = tt.reshape %0 : tensor<512xi32> -> tensor<16x32xi32>
+  tt.return
+}
+
+// CHECK-LABEL: reshape_allow_reorder
+tt.func @reshape_allow_reorder(%0: tensor<512xi32>) {
   // CHECK: tt.reshape %{{.+}} allow_reorder : tensor<512xi32> -> tensor<16x32xi32>
   %1 = tt.reshape %0 allow_reorder : tensor<512xi32> -> tensor<16x32xi32>
   tt.return
 }
 
+// CHECK-LABEL: reshape_efficient_layout
+tt.func @reshape_efficient_layout(%0: tensor<512xi32>) {
+  // CHECK: tt.reshape %{{.+}} efficient_layout : tensor<512xi32> -> tensor<16x32xi32>
+  %1 = tt.reshape %0 efficient_layout : tensor<512xi32> -> tensor<16x32xi32>
+  tt.return
+}
+
+// CHECK-LABEL: reshape_allow_reorder_efficient_layout
+tt.func @reshape_allow_reorder_efficient_layout(%0: tensor<512xi32>) {
+  // CHECK: tt.reshape %{{.+}} allow_reorder efficient_layout : tensor<512xi32> -> tensor<16x32xi32>
+  %1 = tt.reshape %0 allow_reorder efficient_layout : tensor<512xi32> -> tensor<16x32xi32>
+  tt.return
+}
+
 // CHECK-LABEL: histogram
 tt.func @histogram(%0: tensor<512xi32>) {
   // CHECK: tt.histogram %{{.+}} : tensor<512xi32> -> tensor<16xi32>

test/TritonGPU/loop-pipeline.mlir

@@ -1341,7 +1341,7 @@ module attributes {"triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 8 :
       %84 = arith.sitofp %82 : tensor<64x256xi8, #blocked> to tensor<64x256xf16, #blocked>
       %85 = tt.join %83, %84 : tensor<64x256xf16, #blocked> -> tensor<64x256x2xf16, #blocked3>
       %86 = tt.trans %85 {order = array<i32: 0, 2, 1>} : tensor<64x256x2xf16, #blocked3> -> tensor<64x2x256xf16, #blocked4>
-      %87 = tt.reshape %86 allow_reorder : tensor<64x2x256xf16, #blocked4> -> tensor<128x256xf16, #blocked5>
+      %87 = tt.reshape %86 : tensor<64x2x256xf16, #blocked4> -> tensor<128x256xf16, #blocked5>
       %88 = triton_gpu.convert_layout %78 : tensor<16x128xf16, #blocked1> -> tensor<16x128xf16, #triton_gpu.dot_op<{opIdx = 0, parent = #mma, kWidth = 2}>>
       %89 = triton_gpu.convert_layout %87 : tensor<128x256xf16, #blocked5> -> tensor<128x256xf16, #triton_gpu.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>>
       %90 = tt.dot %88, %89, %arg10 : tensor<16x128xf16, #triton_gpu.dot_op<{opIdx = 0, parent = #mma, kWidth = 2}>> * tensor<128x256xf16, #triton_gpu.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> -> tensor<16x256xf32, #mma>