[mlir][tosa] Remove EXT_MXFP support for cast (#167301)

This was removed in the specification by:
arm/tosa-specification#11

Author: lhutton1

mlir/include/mlir/Dialect/Tosa/IR/TosaComplianceData.h.inc

@@ -866,15 +866,7 @@ extensionComplianceMap = {
        {{{fp8e5m2T, fp16T}, SpecificationVersion::V_1_0},
         {{fp8e5m2T, fp32T}, SpecificationVersion::V_1_0},
         {{fp16T, fp8e5m2T}, SpecificationVersion::V_1_0},
-        {{fp32T, fp8e5m2T}, SpecificationVersion::V_1_0}}},
-      {{Extension::bf16, Extension::mxfp},
-       {{{fp4e2m1T, bf16T}, SpecificationVersion::V_1_1_DRAFT},
-        {{fp6e3m2T, bf16T}, SpecificationVersion::V_1_1_DRAFT},
-        {{fp6e2m3T, bf16T}, SpecificationVersion::V_1_1_DRAFT},
-        {{bf16T, fp4e2m1T}, SpecificationVersion::V_1_1_DRAFT},
-        {{bf16T, fp6e3m2T}, SpecificationVersion::V_1_1_DRAFT},
-        {{bf16T, fp6e2m3T}, SpecificationVersion::V_1_1_DRAFT}},
-       allOf}}},
+        {{fp32T, fp8e5m2T}, SpecificationVersion::V_1_0}}}}},
     {"tosa.cast_from_block_scaled",
      {{{Extension::bf16, Extension::mxfp},
        {{{fp4e2m1T, fp8ue8m0T, bf16T}, SpecificationVersion::V_1_1_DRAFT},

mlir/include/mlir/Dialect/Tosa/IR/TosaOps.td

@@ -2464,7 +2464,7 @@ def Tosa_CastOp: Tosa_Op<"cast", [Pure, SameOperandsAndResultShape,
 
   list<Availability> availability = [
     Profile<[Tosa_PRO_INT, Tosa_PRO_FP]>,
-    Extension<[Tosa_EXT_FP8E4M3, Tosa_EXT_FP8E5M2, Tosa_EXT_BF16, Tosa_EXT_MXFP, Tosa_EXT_INT64]>,
+    Extension<[Tosa_EXT_FP8E4M3, Tosa_EXT_FP8E5M2, Tosa_EXT_BF16, Tosa_EXT_INT64]>,
   ];
 
   let assemblyFormat = "operands attr-dict `:` functional-type(operands, results)";

mlir/test/Dialect/Tosa/availability.mlir

@@ -606,7 +606,7 @@ func.func @test_resize(%arg0: tensor<1x32x32x8xf32>) -> tensor<1x64x64x8xf32> {
 // CHECK-LABEL: cast
 func.func @test_cast1(%arg0: tensor<13x21x3xi32>) -> tensor<13x21x3xf32> {
   // CHECK: profiles: [ [pro_int, pro_fp] ]
-  // CHECK: extensions: [ [fp8e4m3, fp8e5m2, bf16, mxfp, int64] ]
+  // CHECK: extensions: [ [fp8e4m3, fp8e5m2, bf16, int64] ]
   %0 = tosa.cast %arg0 : (tensor<13x21x3xi32>) -> tensor<13x21x3xf32>
   return %0 : tensor<13x21x3xf32>
 }

mlir/test/Dialect/Tosa/invalid_extension.mlir

@@ -562,13 +562,6 @@ func.func @test_const_fp6e3m2(%arg0 : index) -> tensor<4xf6E3M2FN> {
     return %0 : tensor<4xf6E3M2FN>
 }
 
-// -----
-func.func @test_cast_f4e2m1(%arg0: tensor<13x21x3xf4E2M1FN>) -> tensor<13x21x3xbf16> {
-  // expected-error@+1 {{'tosa.cast' op illegal: requires all of [bf16, mxfp] but not enabled in target}}
-  %0 = tosa.cast %arg0 : (tensor<13x21x3xf4E2M1FN>) -> tensor<13x21x3xbf16>
-  return %0 : tensor<13x21x3xbf16>
-}
-
 // -----
 
 func.func @test_cast_from_block_scaled_static(%arg0: tensor<4x32xf8E5M2>, %arg1: tensor<4x1xf8E8M0FNU>) -> tensor<4x32xf32> {

mlir/test/Dialect/Tosa/tosa-validation-version-1p1-valid.mlir

@@ -54,14 +54,6 @@ func.func @test_const_fp6e3m2() -> tensor<4xf6E3M2FN> {
 
 // -----
 
-// CHECK-LABEL: test_cast_f4e2m1
-func.func @test_cast_f4e2m1(%arg0: tensor<13x21x3xf4E2M1FN>) -> tensor<13x21x3xbf16> {
-  %0 = tosa.cast %arg0 : (tensor<13x21x3xf4E2M1FN>) -> tensor<13x21x3xbf16>
-  return %0 : tensor<13x21x3xbf16>
-}
-
-// -----
-
 // CHECK-LABEL: test_cast_from_block_scaled_fp8e5m2_fp32
 func.func @test_cast_from_block_scaled_fp8e5m2_fp32(%arg0: tensor<4x32xf8E5M2>, %arg1: tensor<4x1xf8E8M0FNU>) -> tensor<4x32xf32> {
   %0 = tosa.cast_from_block_scaled %arg0, %arg1 {block_size = #tosa.block_size<BLOCK_SIZE_32> : i32} : (tensor<4x32xf8E5M2>, tensor<4x1xf8E8M0FNU>) -> tensor<4x32xf32>
@@ -110,14 +102,6 @@ func.func @test_const_mxint8() -> tensor<2x!tosa.mxint8> {
 
 // -----
 
-// CHECK-LABEL: test_cast_f4e2m1
-func.func @test_cast_f4e2m1(%arg0: tensor<13x21x3xf4E2M1FN>) -> tensor<13x21x3xbf16> {
-  %0 = tosa.cast %arg0 : (tensor<13x21x3xf4E2M1FN>) -> tensor<13x21x3xbf16>
-  return %0 : tensor<13x21x3xbf16>
-}
-
-// -----
-
 // CHECK-LABEL: test_matmul_t_block_scaled_mxint8
 func.func @test_matmul_t_block_scaled_mxint8(%arg0: tensor<4x8x32x!tosa.mxint8>, %arg1: tensor<4x8x1xf8E8M0FNU>, %arg2: tensor<4x16x32x!tosa.mxint8>, %arg3: tensor<4x16x1xf8E8M0FNU>) -> tensor<4x8x16xf32> {
   %0 = tosa.matmul_t_block_scaled %arg0, %arg1, %arg2, %arg3 {block_size = #tosa.block_size<BLOCK_SIZE_32>} : (tensor<4x8x32x!tosa.mxint8>, tensor<4x8x1xf8E8M0FNU>, tensor<4x16x32x!tosa.mxint8>, tensor<4x16x1xf8E8M0FNU>) -> tensor<4x8x16xf32>