[X86] Remove USER_MSR from DMR (#164232)

Per Intel Architecture Instruction Set Extensions Programming Reference
rev. 59 (https://cdrdv2.intel.com/v1/dl/getContent/671368), table 1-2,
DMR doesn't support USER_MSR (URDMSR and UWRMSR instructions)

Author: mikolaj-pirog

clang/test/Preprocessor/predefined-arch-macros.c

@@ -1911,7 +1911,6 @@
 // CHECK_GNR_M32: #define __TSXLDTRK__ 1
 // CHECK_GNR_M32: #define __UINTR__ 1
 // CHECK_GNR_M32-NOT: #define __USERMSR__ 1
-// CHECK_DMR_M32: #define __USERMSR__ 1
 // CHECK_GNR_M32: #define __VAES__ 1
 // CHECK_GNR_M32: #define __VPCLMULQDQ__ 1
 // CHECK_GNR_M32: #define __WAITPKG__ 1
@@ -2018,7 +2017,6 @@
 // CHECK_GNR_M64: #define __TSXLDTRK__ 1
 // CHECK_GNR_M64: #define __UINTR__ 1
 // CHECK_GNR_M64-NOT: #define __USERMSR__ 1
-// CHECK_DMR_M64: #define __USERMSR__ 1
 // CHECK_GNR_M64: #define __VAES__ 1
 // CHECK_GNR_M64: #define __VPCLMULQDQ__ 1
 // CHECK_GNR_M64: #define __WAITPKG__ 1

llvm/lib/Target/X86/X86.td

@@ -1164,7 +1164,6 @@ def ProcessorFeatures {
                                                   FeatureAVXNECONVERT,
                                                   FeatureAVXVNNIINT8,
                                                   FeatureAVXVNNIINT16,
-                                                  FeatureUSERMSR,
                                                   FeatureSHA512,
                                                   FeatureSM3,
                                                   FeatureEGPR,

llvm/lib/TargetParser/X86TargetParser.cpp

@@ -143,8 +143,7 @@ constexpr FeatureBitset FeaturesDiamondRapids =
     FeatureAVXVNNIINT8 | FeatureAVXVNNIINT16 | FeatureSHA512 | FeatureSM3 |
     FeatureSM4 | FeatureEGPR | FeatureZU | FeatureCCMP | FeaturePush2Pop2 |
     FeaturePPX | FeatureNDD | FeatureNF | FeatureMOVRS | FeatureAMX_MOVRS |
-    FeatureAMX_AVX512 | FeatureAMX_FP8 | FeatureAMX_TF32 |
-    FeatureAMX_TRANSPOSE | FeatureUSERMSR;
+    FeatureAMX_AVX512 | FeatureAMX_FP8 | FeatureAMX_TF32 | FeatureAMX_TRANSPOSE;
 
 // Intel Atom processors.
 // Bonnell has feature parity with Core2 and adds MOVBE.