Adding support in llvm-exegesis for Aarch64 for handling FPR64/128, PPR16 and ZPR128 reg class.

llvm/test/tools/llvm-exegesis/AArch64/setReg_init_check.s

@@ -0,0 +1,45 @@
+REQUIRES: aarch64-registered-target
+
+## PPR Register Class Initialization Testcase
+## Ideally, we should use PTRUE_{B/H/S/D} instead of FADDV_VPZ_D for an isolated test case; however, Exegesis does not yet support PTRUE_{B/H/S/D}.
+RUN: llvm-exegesis -mcpu=neoverse-v2 -mode=latency --dump-object-to-disk=%d --opcode-name=FADDV_VPZ_D 2>&1 | FileCheck %s --check-prefix=PPR
+RUN: llvm-objdump -d %d > %t.s
+RUN: FileCheck %s --check-prefix=PPR_ASM < %t.s
+PPR-NOT: setRegTo is not implemented, results will be unreliable
+PPR: assembled_snippet: {{.*}}C0035FD6
+PPR_ASM: {{<foo>:}}
+PPR_ASM: ptrue p{{[0-9]+}}.b
+PPR_ASM-NEXT: dupm z{{[0-9]+}}.s, #0x1
+PPR_ASM-NEXT: faddv d{{[0-9]+}}, p{{[0-9]+}}, z{{[0-9]+}}
+
+## ZPR Register Class Initialization Testcase
+## Ideally, we should use PTRUE_{B/H/S/D} instead of FADDV_VPZ_D for an isolated test case; however, Exegesis does not yet support PTRUE_{B/H/S/D}.
+RUN: llvm-exegesis -mcpu=neoverse-v2 -mode=latency --dump-object-to-disk=%d --opcode-name=FADDV_VPZ_D 2>&1 | FileCheck %s --check-prefix=ZPR
+RUN: llvm-objdump -d %d > %t.s
+RUN: FileCheck %s --check-prefix=ZPR_ASM < %t.s
+ZPR-NOT: setRegTo is not implemented, results will be unreliable
+ZPR: assembled_snippet: {{.*}}C0035FD6
+ZPR_ASM: {{<foo>:}}
+ZPR_ASM: ptrue p{{[0-9]+}}.b
+ZPR_ASM-NEXT: dupm z{{[0-9]+}}.s, #0x1
+ZPR_ASM-NEXT: faddv d{{[0-9]+}}, p{{[0-9]+}}, z{{[0-9]+}}
+
+## FPR128 Register Class Initialization Testcase
+RUN: llvm-exegesis -mcpu=neoverse-v2 -mode=latency --dump-object-to-disk=%d --opcode-name=ADDVv16i8v 2>&1 | FileCheck %s --check-prefix=FPR128
+RUN: llvm-objdump -d %d > %t.s
+RUN: FileCheck %s --check-prefix=FPR128-ASM < %t.s
+FPR128-NOT: setRegTo is not implemented, results will be unreliable
+FPR128: assembled_snippet: {{.*}}C0035FD6
+FPR128-ASM: {{<foo>:}}
+FPR128-ASM: movi v{{[0-9]+}}.2d, #0000000000000000
+FPR128-ASM-NEXT: addv b{{[0-9]+}}, v{{[0-9]+}}.16b
+
+## FPR64 Register Class Initialization Testcase
+RUN: llvm-exegesis -mcpu=neoverse-v2 -mode=latency --dump-object-to-disk=%d --opcode-name=ADDVv4i16v 2>&1 | FileCheck %s --check-prefix=FPR64
+RUN: llvm-objdump -d %d > %t.s
+RUN: FileCheck %s --check-prefix=FPR64-ASM < %t.s
+FPR64-NOT: setRegTo is not implemented, results will be unreliable
+FPR64: assembled_snippet: {{.*}}C0035FD6
+FPR64-ASM: {{<foo>:}}
+FPR64-ASM: fmov d{{[0-9]+}}, {{#2.0+|#2\.000000000000000000e\+00}}
+FPR64-ASM-NEXT: addv h{{[0-9]+}}, v{{[0-9]+}}.4h

llvm/tools/llvm-exegesis/lib/AArch64/Target.cpp

@@ -28,13 +28,66 @@ static unsigned getLoadImmediateOpcode(unsigned RegBitWidth) {
 // Generates instruction to load an immediate value into a register.
 static MCInst loadImmediate(MCRegister Reg, unsigned RegBitWidth,
                             const APInt &Value) {
-  if (Value.getBitWidth() > RegBitWidth)
-    llvm_unreachable("Value must fit in the Register");
+  // 0 <= Value.getZExtValue() < 2**16
+  assert(Value.getZExtValue() < (1 << 16) &&
+         "Value must be in the range of the immediate opcode");
   return MCInstBuilder(getLoadImmediateOpcode(RegBitWidth))
       .addReg(Reg)
       .addImm(Value.getZExtValue());
 }
 
+static MCInst loadZPRImmediate(MCRegister Reg, unsigned RegBitWidth,
+                               const APInt &Value) {
+  // 0 <= Value.getZExtValue() < 2**13
+  assert(Value.getZExtValue() < (1 << 13) &&
+         "Value must be in the range of the immediate opcode");
+  // For ZPR, we typically use DUPM instruction to load immediate values
+  return MCInstBuilder(AArch64::DUPM_ZI)
+      .addReg(Reg)
+      .addImm(Value.getZExtValue());
+}
+
+static MCInst loadPPRImmediate(MCRegister Reg, unsigned RegBitWidth,
+                               const APInt &Value) {
+  // For PPR, we typically use PTRUE instruction to set predicate registers
+  return MCInstBuilder(AArch64::PTRUE_B)
+      .addReg(Reg)
+      .addImm(31); // All lanes true for 16 bits
+}
+
+// Fetch base-instruction to load an FP immediate value into a register.
+static unsigned getLoadFPImmediateOpcode(unsigned RegBitWidth) {
+  switch (RegBitWidth) {
+  case 64:
+    return AArch64::FMOVDi; 
+  case 128:
+    return AArch64::MOVIv2d_ns;
+  }
+  llvm_unreachable("Invalid Value Width");
+}
+
+// Generates instruction to load an FP immediate value into a register.
+static MCInst loadFPImmediate(MCRegister Reg, unsigned RegBitWidth,
+                              const APInt &Value) {
+  // -31 <= Value.getZExtValue() <= 31
+  assert(Value.getZExtValue() <= 31 &&
+         "Value must be in the range of the immediate opcode");
+  return MCInstBuilder(getLoadFPImmediateOpcode(RegBitWidth))
+      .addReg(Reg)
+      .addImm(Value.getZExtValue());
+}
+
+// Generates instruction to load an FP128 immediate value into a register.
+static MCInst loadFP128Immediate(MCRegister Reg, unsigned RegBitWidth,
+                                 const APInt &Value) {
+  // 0 <= Value.getZExtValue() < 2**8
+  assert(Value.getZExtValue() < (1 << 8) &&
+         "Value must be in the range of the immediate opcode");
+  return MCInstBuilder(getLoadFPImmediateOpcode(RegBitWidth))
+      .addReg(Reg)
+      .addImm(Value.getZExtValue());
+}
+
 #include "AArch64GenExegesis.inc"
 
 namespace {
@@ -51,6 +104,15 @@ class ExegesisAArch64Target : public ExegesisTarget {
       return {loadImmediate(Reg, 32, Value)};
     if (AArch64::GPR64RegClass.contains(Reg))
       return {loadImmediate(Reg, 64, Value)};
+    if (AArch64::PPRRegClass.contains(Reg))
+      return {loadPPRImmediate(Reg, 16, Value)};
+    if (AArch64::FPR64RegClass.contains(Reg))
+      return {loadFPImmediate(Reg, 64, Value)};
+    if (AArch64::FPR128RegClass.contains(Reg))
+      return {loadFP128Immediate(Reg, 128, Value)};
+    if (AArch64::ZPRRegClass.contains(Reg))
+      return {loadZPRImmediate(Reg, 128, Value)};
+
     errs() << "setRegTo is not implemented, results will be unreliable\n";
     return {};
   }