[RISCV] Re-separate unaligned scalar and vector memory features in the backend.

clang/lib/Basic/Targets/RISCV.cpp

@@ -353,7 +353,8 @@ bool RISCVTargetInfo::handleTargetFeatures(std::vector<std::string> &Features,
   if (ISAInfo->hasExtension("zfh") || ISAInfo->hasExtension("zhinx"))
     HasLegalHalfType = true;
 
-  FastUnalignedAccess = llvm::is_contained(Features, "+fast-unaligned-access");
+  FastUnalignedAccess = llvm::is_contained(Features, "+unaligned-scalar-mem") &&
+                        llvm::is_contained(Features, "+unaligned-vector-mem");
 
   if (llvm::is_contained(Features, "+experimental"))
     HasExperimental = true;

clang/lib/Driver/ToolChains/Arch/RISCV.cpp

@@ -68,8 +68,10 @@ static void getRISCFeaturesFromMcpu(const Driver &D, const Arg *A,
           << A->getSpelling() << Mcpu;
   }
 
-  if (llvm::RISCV::hasFastUnalignedAccess(Mcpu))
-    Features.push_back("+fast-unaligned-access");
+  if (llvm::RISCV::hasFastUnalignedAccess(Mcpu)) {
+    Features.push_back("+unaligned-scalar-mem");
+    Features.push_back("+unaligned-vector-mem");
+  }
 }
 
 void riscv::getRISCVTargetFeatures(const Driver &D, const llvm::Triple &Triple,
@@ -168,12 +170,16 @@ void riscv::getRISCVTargetFeatures(const Driver &D, const llvm::Triple &Triple,
   }
 
   // Android requires fast unaligned access on RISCV64.
-  if (Triple.isAndroid())
-    Features.push_back("+fast-unaligned-access");
+  if (Triple.isAndroid()) {
+    Features.push_back("+unaligned-scalar-mem");
+    Features.push_back("+unaligned-vector-mem");
+  }
 
   // -mstrict-align is default, unless -mno-strict-align is specified.
   AddTargetFeature(Args, Features, options::OPT_mno_strict_align,
-                   options::OPT_mstrict_align, "fast-unaligned-access");
+                   options::OPT_mstrict_align, "unaligned-scalar-mem");
+  AddTargetFeature(Args, Features, options::OPT_mno_strict_align,
+                   options::OPT_mstrict_align, "unaligned-vector-mem");
 
   // Now add any that the user explicitly requested on the command line,
   // which may override the defaults.

clang/test/Driver/riscv-features.c

@@ -38,8 +38,8 @@
 // RUN: %clang --target=riscv32-unknown-elf -### %s -mno-strict-align 2>&1 | FileCheck %s -check-prefix=FAST-UNALIGNED-ACCESS
 // RUN: %clang --target=riscv32-unknown-elf -### %s -mstrict-align 2>&1 | FileCheck %s -check-prefix=NO-FAST-UNALIGNED-ACCESS
 
-// FAST-UNALIGNED-ACCESS: "-target-feature" "+fast-unaligned-access"
-// NO-FAST-UNALIGNED-ACCESS: "-target-feature" "-fast-unaligned-access"
+// FAST-UNALIGNED-ACCESS: "-target-feature" "+unaligned-scalar-mem" "-target-feature" "+unaligned-vector-mem"
+// NO-FAST-UNALIGNED-ACCESS: "-target-feature" "-unaligned-scalar-mem" "-target-feature" "-unaligned-vector-mem"
 
 // RUN: %clang --target=riscv32-unknown-elf -### %s 2>&1 | FileCheck %s -check-prefix=NOUWTABLE
 // RUN: %clang --target=riscv32-unknown-elf -fasynchronous-unwind-tables -### %s 2>&1 | FileCheck %s -check-prefix=UWTABLE

llvm/lib/Target/RISCV/RISCVExpandPseudoInsts.cpp

@@ -326,8 +326,8 @@ bool RISCVExpandPseudo::expandRV32ZdinxStore(MachineBasicBlock &MBB,
       .setMemRefs(MMOLo);
 
   if (MBBI->getOperand(2).isGlobal() || MBBI->getOperand(2).isCPI()) {
-    // FIXME: Zdinx RV32 can not work on unaligned memory.
-    assert(!STI->hasFastUnalignedAccess());
+    // FIXME: Zdinx RV32 can not work on unaligned scalar memory.
+    assert(!STI->enableUnalignedScalarMem());
 
     assert(MBBI->getOperand(2).getOffset() % 8 == 0);
     MBBI->getOperand(2).setOffset(MBBI->getOperand(2).getOffset() + 4);

llvm/lib/Target/RISCV/RISCVFeatures.td

@@ -1183,10 +1183,15 @@ def FeatureTrailingSeqCstFence : SubtargetFeature<"seq-cst-trailing-fence",
                                           "true",
                                           "Enable trailing fence for seq-cst store.">;
 
-def FeatureFastUnalignedAccess
-   : SubtargetFeature<"fast-unaligned-access", "HasFastUnalignedAccess",
-                      "true", "Has reasonably performant unaligned "
-                      "loads and stores (both scalar and vector)">;
+def FeatureUnalignedScalarMem
+   : SubtargetFeature<"unaligned-scalar-mem", "EnableUnalignedScalarMem",
+                      "true", "Has reasonably performant unaligned scalar "
+                      "loads and stores">;
+
+def FeatureUnalignedVectorMem
+   : SubtargetFeature<"unaligned-vector-mem", "EnableUnalignedVectorMem",
+                      "true", "Has reasonably performant unaligned vector "
+                      "loads and stores">;
 
 def FeaturePostRAScheduler : SubtargetFeature<"use-postra-scheduler",
     "UsePostRAScheduler", "true", "Schedule again after register allocation">;

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

@@ -1924,7 +1924,7 @@ bool RISCVTargetLowering::shouldConvertConstantLoadToIntImm(const APInt &Imm,
   // replace. If we don't support unaligned scalar mem, prefer the constant
   // pool.
   // TODO: Can the caller pass down the alignment?
-  if (!Subtarget.hasFastUnalignedAccess())
+  if (!Subtarget.enableUnalignedScalarMem())
     return true;
 
   // Prefer to keep the load if it would require many instructions.
@@ -15794,7 +15794,7 @@ static bool matchIndexAsWiderOp(EVT VT, SDValue Index, SDValue Mask,
   if (WiderElementSize > ST.getELen()/8)
     return false;
 
-  if (!ST.hasFastUnalignedAccess() && BaseAlign < WiderElementSize)
+  if (!ST.enableUnalignedVectorMem() && BaseAlign < WiderElementSize)
     return false;
 
   for (unsigned i = 0; i < Index->getNumOperands(); i++) {
@@ -20620,8 +20620,8 @@ bool RISCVTargetLowering::allowsMisalignedMemoryAccesses(
     unsigned *Fast) const {
   if (!VT.isVector()) {
     if (Fast)
-      *Fast = Subtarget.hasFastUnalignedAccess();
-    return Subtarget.hasFastUnalignedAccess();
+      *Fast = Subtarget.enableUnalignedScalarMem();
+    return Subtarget.enableUnalignedScalarMem();
   }
 
   // All vector implementations must support element alignment
@@ -20637,8 +20637,8 @@ bool RISCVTargetLowering::allowsMisalignedMemoryAccesses(
   // misaligned accesses.  TODO: Work through the codegen implications of
   // allowing such accesses to be formed, and considered fast.
   if (Fast)
-    *Fast = Subtarget.hasFastUnalignedAccess();
-  return Subtarget.hasFastUnalignedAccess();
+    *Fast = Subtarget.enableUnalignedVectorMem();
+  return Subtarget.enableUnalignedVectorMem();
 }
 
 
@@ -20673,7 +20673,7 @@ EVT RISCVTargetLowering::getOptimalMemOpType(const MemOp &Op,
 
   // Do we have sufficient alignment for our preferred VT?  If not, revert
   // to largest size allowed by our alignment criteria.
-  if (PreferredVT != MVT::i8 && !Subtarget.hasFastUnalignedAccess()) {
+  if (PreferredVT != MVT::i8 && !Subtarget.enableUnalignedVectorMem()) {
     Align RequiredAlign(PreferredVT.getStoreSize());
     if (Op.isFixedDstAlign())
       RequiredAlign = std::min(RequiredAlign, Op.getDstAlign());
@@ -20865,7 +20865,7 @@ bool RISCVTargetLowering::isLegalStridedLoadStore(EVT DataType,
   if (!isLegalElementTypeForRVV(ScalarType))
     return false;
 
-  if (!Subtarget.hasFastUnalignedAccess() &&
+  if (!Subtarget.enableUnalignedVectorMem() &&
       Alignment < ScalarType.getStoreSize())
     return false;
 

llvm/lib/Target/RISCV/RISCVProcessors.td

@@ -257,7 +257,8 @@ def SIFIVE_P450 : RISCVProcessorModel<"sifive-p450", SiFiveP400Model,
                                        FeatureStdExtZbb,
                                        FeatureStdExtZbs,
                                        FeatureStdExtZfhmin,
-                                       FeatureFastUnalignedAccess],
+                                       FeatureUnalignedScalarMem,
+                                       FeatureUnalignedVectorMem],
                                       [TuneNoDefaultUnroll,
                                        TuneConditionalCompressedMoveFusion,
                                        TuneLUIADDIFusion,
@@ -295,7 +296,8 @@ def SIFIVE_P670 : RISCVProcessorModel<"sifive-p670", SiFiveP600Model,
                                        FeatureStdExtZvkng,
                                        FeatureStdExtZvksc,
                                        FeatureStdExtZvksg,
-                                       FeatureFastUnalignedAccess],
+                                       FeatureUnalignedScalarMem,
+                                       FeatureUnalignedVectorMem],
                                       [TuneNoDefaultUnroll,
                                        TuneConditionalCompressedMoveFusion,
                                        TuneLUIADDIFusion,

llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h

@@ -228,7 +228,7 @@ class RISCVTTIImpl : public BasicTTIImplBase<RISCVTTIImpl> {
       return false;
 
     EVT ElemType = DataTypeVT.getScalarType();
-    if (!ST->hasFastUnalignedAccess() && Alignment < ElemType.getStoreSize())
+    if (!ST->enableUnalignedVectorMem() && Alignment < ElemType.getStoreSize())
       return false;
 
     return TLI->isLegalElementTypeForRVV(ElemType);
@@ -253,7 +253,7 @@ class RISCVTTIImpl : public BasicTTIImplBase<RISCVTTIImpl> {
       return false;
 
     EVT ElemType = DataTypeVT.getScalarType();
-    if (!ST->hasFastUnalignedAccess() && Alignment < ElemType.getStoreSize())
+    if (!ST->enableUnalignedVectorMem() && Alignment < ElemType.getStoreSize())
       return false;
 
     return TLI->isLegalElementTypeForRVV(ElemType);

llvm/test/CodeGen/RISCV/memcpy-inline.ll

@@ -3,9 +3,9 @@
 ; RUN:   | FileCheck %s --check-prefixes=RV32-BOTH,RV32
 ; RUN: llc < %s -mtriple=riscv64 \
 ; RUN:   | FileCheck %s --check-prefixes=RV64-BOTH,RV64
-; RUN: llc < %s -mtriple=riscv32 -mattr=+fast-unaligned-access \
+; RUN: llc < %s -mtriple=riscv32 -mattr=+unaligned-scalar-mem \
 ; RUN:   | FileCheck %s --check-prefixes=RV32-BOTH,RV32-FAST
-; RUN: llc < %s -mtriple=riscv64 -mattr=+fast-unaligned-access \
+; RUN: llc < %s -mtriple=riscv64 -mattr=+unaligned-scalar-mem \
 ; RUN:   | FileCheck %s --check-prefixes=RV64-BOTH,RV64-FAST
 
 ; ----------------------------------------------------------------------

llvm/test/CodeGen/RISCV/memcpy.ll

@@ -3,9 +3,9 @@
 ; RUN:   | FileCheck %s --check-prefixes=RV32-BOTH,RV32
 ; RUN: llc < %s -mtriple=riscv64 \
 ; RUN:   | FileCheck %s --check-prefixes=RV64-BOTH,RV64
-; RUN: llc < %s -mtriple=riscv32 -mattr=+fast-unaligned-access \
+; RUN: llc < %s -mtriple=riscv32 -mattr=+unaligned-scalar-mem \
 ; RUN:   | FileCheck %s --check-prefixes=RV32-BOTH,RV32-FAST
-; RUN: llc < %s -mtriple=riscv64 -mattr=+fast-unaligned-access \
+; RUN: llc < %s -mtriple=riscv64 -mattr=+unaligned-scalar-mem \
 ; RUN:   | FileCheck %s --check-prefixes=RV64-BOTH,RV64-FAST
 %struct.x = type { i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8 }
 

llvm/test/CodeGen/RISCV/memset-inline.ll

@@ -3,9 +3,9 @@
 ; RUN:   | FileCheck %s --check-prefixes=RV32-BOTH,RV32
 ; RUN: llc < %s -mtriple=riscv64 -mattr=+m \
 ; RUN:   | FileCheck %s --check-prefixes=RV64-BOTH,RV64
-; RUN: llc < %s -mtriple=riscv32 -mattr=+m,+fast-unaligned-access \
+; RUN: llc < %s -mtriple=riscv32 -mattr=+m,+unaligned-scalar-mem \
 ; RUN:   | FileCheck %s --check-prefixes=RV32-BOTH,RV32-FAST
-; RUN: llc < %s -mtriple=riscv64 -mattr=+m,+fast-unaligned-access \
+; RUN: llc < %s -mtriple=riscv64 -mattr=+m,+unaligned-scalar-mem \
 ; RUN:   | FileCheck %s --check-prefixes=RV64-BOTH,RV64-FAST
 %struct.x = type { i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8 }
 

llvm/test/CodeGen/RISCV/pr56110.ll

@@ -1,6 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=riscv32 | FileCheck %s
-; RUN: llc < %s -mtriple=riscv32 -mattr=+fast-unaligned-access | FileCheck %s
+; RUN: llc < %s -mtriple=riscv32 -mattr=+unaligned-scalar-mem | FileCheck %s
 
 define void @foo_set(ptr nocapture noundef %a, i32 noundef %v) {
 ; CHECK-LABEL: foo_set:

llvm/test/CodeGen/RISCV/riscv-func-target-feature.ll

@@ -36,7 +36,7 @@ entry:
 }
 
 ; CHECK-NOT: .option push
-define void @test5() "target-features"="+fast-unaligned-access" {
+define void @test5() "target-features"="+unaligned-scalar-mem" {
 ; CHECK-LABEL: test5
 ; CHECK-NOT: .option pop
 entry:

llvm/test/CodeGen/RISCV/rvv/concat-vectors-constant-stride.ll

@@ -1,7 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=riscv32 -mattr=+v,+fast-unaligned-access -target-abi=ilp32 \
+; RUN: llc -mtriple=riscv32 -mattr=+v,+unaligned-vector-mem -target-abi=ilp32 \
 ; RUN:     -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV32
-; RUN: llc -mtriple=riscv64 -mattr=+v,+fast-unaligned-access -target-abi=lp64 \
+; RUN: llc -mtriple=riscv64 -mattr=+v,+unaligned-vector-mem -target-abi=lp64 \
 ; RUN:     -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV64
 
 define void @constant_forward_stride(ptr %s, ptr %d) {

llvm/test/CodeGen/RISCV/rvv/fixed-vectors-strided-load-combine.ll

@@ -1,7 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
 ; RUN: llc -mtriple=riscv32 -mattr=+v,+zfh,+zvfh -verify-machineinstrs < %s | FileCheck %s -check-prefixes=CHECK,CHECK-NO-MISALIGN,RV32
 ; RUN: llc -mtriple=riscv64 -mattr=+v,+zfh,+zvfh -verify-machineinstrs < %s | FileCheck %s -check-prefixes=CHECK,CHECK-NO-MISALIGN,RV64
-; RUN: llc -mtriple=riscv64 -mattr=+v,+zfh,+zvfh,+fast-unaligned-access -verify-machineinstrs < %s | FileCheck %s -check-prefixes=CHECK,RV64,RV64-MISALIGN
+; RUN: llc -mtriple=riscv64 -mattr=+v,+zfh,+zvfh,+unaligned-vector-mem -verify-machineinstrs < %s | FileCheck %s -check-prefixes=CHECK,RV64,RV64-MISALIGN
 
 ; RUN: llc -mtriple=riscv64 -mattr=+f,+zfh,+zve64f,+zvl128b,+zvfh -verify-machineinstrs < %s | FileCheck %s -check-prefixes=CHECK,CHECK-NO-MISALIGN,ZVE64F
 

llvm/test/CodeGen/RISCV/rvv/fixed-vectors-unaligned.ll

@@ -3,9 +3,9 @@
 ; RUN:   | FileCheck %s --check-prefixes=SLOW,RV32-SLOW
 ; RUN: llc -mtriple=riscv64 -mattr=+m,+v -verify-machineinstrs < %s \
 ; RUN:   | FileCheck %s --check-prefixes=SLOW,RV64-SLOW
-; RUN: llc -mtriple=riscv32 -mattr=+m,+v,+fast-unaligned-access -verify-machineinstrs < %s \
+; RUN: llc -mtriple=riscv32 -mattr=+m,+v,+unaligned-vector-mem -verify-machineinstrs < %s \
 ; RUN:   | FileCheck %s --check-prefixes=FAST,RV32-FAST
-; RUN: llc -mtriple=riscv64 -mattr=+m,+v,+fast-unaligned-access -verify-machineinstrs < %s \
+; RUN: llc -mtriple=riscv64 -mattr=+m,+v,+unaligned-vector-mem -verify-machineinstrs < %s \
 ; RUN:   | FileCheck %s --check-prefixes=FAST,RV64-FAST
 
 define <4 x i32> @load_v4i32_align1(ptr %ptr) {

llvm/test/CodeGen/RISCV/rvv/memcpy-inline.ll

@@ -3,9 +3,9 @@
 ; RUN:   | FileCheck %s --check-prefixes=RV32-BOTH,RV32
 ; RUN: llc < %s -mtriple=riscv64 -mattr=+v \
 ; RUN:   | FileCheck %s --check-prefixes=RV64-BOTH,RV64
-; RUN: llc < %s -mtriple=riscv32 -mattr=+v,+fast-unaligned-access \
+; RUN: llc < %s -mtriple=riscv32 -mattr=+v,+unaligned-scalar-mem,+unaligned-vector-mem \
 ; RUN:   | FileCheck %s --check-prefixes=RV32-BOTH,RV32-FAST
-; RUN: llc < %s -mtriple=riscv64 -mattr=+v,+fast-unaligned-access \
+; RUN: llc < %s -mtriple=riscv64 -mattr=+v,+unaligned-scalar-mem,+unaligned-vector-mem \
 ; RUN:   | FileCheck %s --check-prefixes=RV64-BOTH,RV64-FAST
 
 ; ----------------------------------------------------------------------

llvm/test/CodeGen/RISCV/rvv/memset-inline.ll

@@ -3,9 +3,9 @@
 ; RUN:   | FileCheck %s --check-prefixes=RV32-BOTH,RV32
 ; RUN: llc < %s -mtriple=riscv64 -mattr=+m,+v \
 ; RUN:   | FileCheck %s --check-prefixes=RV64-BOTH,RV64
-; RUN: llc < %s -mtriple=riscv32 -mattr=+m,+v,+fast-unaligned-access \
+; RUN: llc < %s -mtriple=riscv32 -mattr=+m,+v,+unaligned-scalar-mem,,+unaligned-vector-mem \
 ; RUN:   | FileCheck %s --check-prefixes=RV32-BOTH,RV32-FAST
-; RUN: llc < %s -mtriple=riscv64 -mattr=+m,+v,+fast-unaligned-access \
+; RUN: llc < %s -mtriple=riscv64 -mattr=+m,+v,+unaligned-scalar-mem,+unaligned-vector-mem \
 ; RUN:   | FileCheck %s --check-prefixes=RV64-BOTH,RV64-FAST
 %struct.x = type { i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8 }
 

llvm/test/CodeGen/RISCV/rvv/unaligned-loads-stores.ll

@@ -3,9 +3,9 @@
 ; RUN:    -verify-machineinstrs | FileCheck %s
 ; RUN: llc -mtriple riscv64 -mattr=+d,+zfh,+zvfh,+v < %s \
 ; RUN:    -verify-machineinstrs | FileCheck %s
-; RUN: llc -mtriple riscv32 -mattr=+d,+zfh,+zvfh,+v,+fast-unaligned-access < %s \
+; RUN: llc -mtriple riscv32 -mattr=+d,+zfh,+zvfh,+v,+unaligned-vector-mem < %s \
 ; RUN:    -verify-machineinstrs | FileCheck --check-prefix=FAST %s
-; RUN: llc -mtriple riscv64 -mattr=+d,+zfh,+zvfh,+v,+fast-unaligned-access < %s \
+; RUN: llc -mtriple riscv64 -mattr=+d,+zfh,+zvfh,+v,+unaligned-vector-mem < %s \
 ; RUN:    -verify-machineinstrs | FileCheck --check-prefix=FAST %s
 
 

llvm/test/CodeGen/RISCV/unaligned-load-store.ll

@@ -3,9 +3,9 @@
 ; RUN:   | FileCheck -check-prefixes=ALL,SLOW,RV32I %s
 ; RUN: llc -mtriple=riscv64 -verify-machineinstrs < %s \
 ; RUN:   | FileCheck -check-prefixes=ALL,SLOW,RV64I %s
-; RUN: llc -mtriple=riscv32 -mattr=+fast-unaligned-access -verify-machineinstrs < %s \
+; RUN: llc -mtriple=riscv32 -mattr=+unaligned-scalar-mem -verify-machineinstrs < %s \
 ; RUN:   | FileCheck -check-prefixes=ALL,FAST,RV32I-FAST %s
-; RUN: llc -mtriple=riscv64 -mattr=+fast-unaligned-access -verify-machineinstrs < %s \
+; RUN: llc -mtriple=riscv64 -mattr=+unaligned-scalar-mem -verify-machineinstrs < %s \
 ; RUN:   | FileCheck -check-prefixes=ALL,FAST,RV64I-FAST %s
 
 ; A collection of cases showing codegen for unaligned loads and stores

llvm/utils/TableGen/RISCVTargetDefEmitter.cpp

@@ -60,11 +60,19 @@ static void EmitRISCVTargetDef(RecordKeeper &RK, raw_ostream &OS) {
     if (MArch.empty())
       MArch = getMArch(*Rec);
 
-    const bool FastUnalignedAccess =
+    bool FastScalarUnalignedAccess =
         any_of(Rec->getValueAsListOfDefs("Features"), [&](auto &Feature) {
-          return Feature->getValueAsString("Name") == "fast-unaligned-access";
+          return Feature->getValueAsString("Name") == "unaligned-scalar-mem";
         });
 
+    bool FastVectorUnalignedAccess =
+        any_of(Rec->getValueAsListOfDefs("Features"), [&](auto &Feature) {
+          return Feature->getValueAsString("Name") == "unaligned-vector-mem";
+        });
+
+    bool FastUnalignedAccess =
+        FastScalarUnalignedAccess && FastVectorUnalignedAccess;
+
     OS << "PROC(" << Rec->getName() << ", "
        << "{\"" << Rec->getValueAsString("Name") << "\"}, "
        << "{\"" << MArch << "\"}, " << FastUnalignedAccess << ")\n";