Add vmov between integers and floats in ARM assembler.

Adds vmovrs that implements moving from an integer (GP) register and a
float (S) register to the integrated ARM assembler.

The test also shows that moving from a float (S) register to an
integer (GP) register also works.

BUG= https://bugs.chromium.org/p/nativeclient/issues/detail?id=4334
R=eholk@chromium.org, stichnot@chromium.org

Review URL: https://codereview.chromium.org/1647683002 .

diff --git a/src/DartARM32/assembler_arm.cc b/src/DartARM32/assembler_arm.cc
index 8e25dde..c8bb8a3 100644 (file)
--- a/src/DartARM32/assembler_arm.cc
+++ b/src/DartARM32/assembler_arm.cc
@@ -620,8 +620,8 @@ void Assembler::vmovsr(SRegister sn, Register rt, Condition cond) {
                      ((static_cast<int32_t>(sn) & 1)*B7) | B4;
   Emit(encoding);
 }
-#endif
 
+// Moved to ARM32::AssemblerARM32::vmovrs().
 void Assembler::vmovrs(Register rt, SRegister sn, Condition cond) {
   ASSERT(TargetCPUFeatures::vfp_supported());
   ASSERT(sn != kNoSRegister);
@@ -636,6 +636,7 @@ void Assembler::vmovrs(Register rt, SRegister sn, Condition cond) {
                      ((static_cast<int32_t>(sn) & 1)*B7) | B4;
   Emit(encoding);
 }
+#endif
 
 
 void Assembler::vmovsrr(SRegister sm, Register rt, Register rt2,

diff --git a/src/DartARM32/assembler_arm.h b/src/DartARM32/assembler_arm.h
index ddb4987..344fe16 100644 (file)
--- a/src/DartARM32/assembler_arm.h
+++ b/src/DartARM32/assembler_arm.h
@@ -610,18 +610,19 @@ class Assembler : public ValueObject {
     return (AL << kConditionShift) | B24 | B21 |
            ((imm16 >> 4) << 8) | B6 | B5 | B4 | (imm16 & 0xf);
   }
-#endif
 
+  // Not ported. PNaCl doesn't allow breakpoint instructions.
   static uword GetBreakInstructionFiller() {
     return BkptEncoding(0);
   }
 
   // Floating point instructions (VFPv3-D16 and VFPv3-D32 profiles).
-#if 0
+
   // Moved to ARM32::AssemblerARM32::vmovsr().
   void vmovsr(SRegister sn, Register rt, Condition cond = AL);
-#endif
+  // Moved to ARM32::AssemblerARM32::vmovrs().
   void vmovrs(Register rt, SRegister sn, Condition cond = AL);
+#endif
   void vmovsrr(SRegister sm, Register rt, Register rt2, Condition cond = AL);
   void vmovrrs(Register rt, Register rt2, SRegister sm, Condition cond = AL);
   void vmovdrr(DRegister dm, Register rt, Register rt2, Condition cond = AL);

diff --git a/src/IceAssemblerARM32.cpp b/src/IceAssemblerARM32.cpp
index 29a9c25..f42a69a 100644 (file)
--- a/src/IceAssemblerARM32.cpp
+++ b/src/IceAssemblerARM32.cpp
@@ -2427,6 +2427,24 @@ void AssemblerARM32::vmovdd(const Operand *OpDd, const Operand *OpDm,
   emitVFPddd(Cond, VmovddOpcode, Dd, D0, Dm);
 }
 
+void AssemblerARM32::vmovrs(const Operand *OpRt, const Operand *OpSn,
+                            CondARM32::Cond Cond) {
+  // VMOV (between ARM core register and single-precision register)
+  //   ARM section A8.8.343, encoding A1.
+  //
+  //   vmov<c> <Rt>, <Sn>
+  //
+  // cccc11100001nnnntttt1010N0010000 where cccc=Cond, nnnnN = Sn, and tttt=Rt.
+  constexpr const char *Vmovrs = "vmovrs";
+  IValueT Rt = encodeGPRegister(OpRt, "Rt", Vmovrs);
+  IValueT Sn = encodeSRegister(OpSn, "Sn", Vmovrs);
+  assert(CondARM32::isDefined(Cond));
+  IValueT Encoding = (encodeCondition(Cond) << kConditionShift) | B27 | B26 |
+                     B25 | B20 | B11 | B9 | B4 | (getXXXXInRegXXXXY(Sn) << 16) |
+                     (Rt << kRdShift) | (getYInRegXXXXY(Sn) << 7);
+  emitInst(Encoding);
+}
+
 void AssemblerARM32::vmovs(const Operand *OpSd,
                            const OperandARM32FlexFpImm *OpFpImm,
                            CondARM32::Cond Cond) {
@@ -2465,7 +2483,7 @@ void AssemblerARM32::vmovsr(const Operand *OpSn, const Operand *OpRt,
   //
   //   vmov<c> <Sn>, <Rt>
   //
-  // cccc1110000onnnntttt1010N0010000 where cccc=Cond, nnnnN = Sn, and tttt=Rt.
+  // cccc11100000nnnntttt1010N0010000 where cccc=Cond, nnnnN = Sn, and tttt=Rt.
   constexpr const char *Vmovsr = "vmovsr";
   IValueT Sn = encodeSRegister(OpSn, "Sn", Vmovsr);
   IValueT Rt = encodeGPRegister(OpRt, "Rt", Vmovsr);

diff --git a/src/IceAssemblerARM32.h b/src/IceAssemblerARM32.h
index 2ff66b0..91ed038 100644 (file)
--- a/src/IceAssemblerARM32.h
+++ b/src/IceAssemblerARM32.h
@@ -387,6 +387,8 @@ public:
 
   void vmovdd(const Operand *OpDd, const Operand *OpDm, CondARM32::Cond Cond);
 
+  void vmovrs(const Operand *OpRt, const Operand *OpSn, CondARM32::Cond Cond);
+
   void vmovs(const Operand *OpSn, const OperandARM32FlexFpImm *OpFpImm,
              CondARM32::Cond Cond);
 

diff --git a/src/IceInstARM32.cpp b/src/IceInstARM32.cpp
index 9ca1085..0bfa540 100644 (file)
--- a/src/IceInstARM32.cpp
+++ b/src/IceInstARM32.cpp
@@ -1209,15 +1209,30 @@ void InstARM32Mov::emitIASCoreVFPMove(const Cfg *Func) const {
 
   // Move register to register.
   Variable *Dest = getDest();
+  // TODO(kschimpf) Consider merging methods emitIAS.. methods into
+  // a single case statement.
   switch (Dest->getType()) {
   default:
     // TODO(kschimpf): Fill this out more.
     return Asm->setNeedsTextFixup();
+  case IceType_i1:
+  case IceType_i8:
+  case IceType_i16:
+  case IceType_i32:
+    assert(Src0->getType() == IceType_f32 && "Expected int to float move");
+    Asm->vmovrs(Dest, Src0, getPredicate());
+    return;
+  case IceType_i64:
+    assert(false && "i64 to float moves not handled here!");
+    return;
   case IceType_f32:
     switch (Src0->getType()) {
     default:
-      // TODO(kschimpf): Fill this out more?
-      return Asm->setNeedsTextFixup();
+      assert(false && "Expected float to int move");
+      return;
+    case IceType_i1:
+    case IceType_i8:
+    case IceType_i16:
     case IceType_i32:
       return Asm->vmovsr(Dest, Src0, getPredicate());
     }

diff --git a/tests_lit/assembler/arm32/vmov-f2i.ll b/tests_lit/assembler/arm32/vmov-f2i.ll
new file mode 100644 (file)
index 0000000..c8b8504
--- /dev/null
+++ b/tests_lit/assembler/arm32/vmov-f2i.ll
@@ -0,0 +1,124 @@
+; Show that we can move between float (S) and integer (GPR) registers.
+
+; REQUIRES: allow_dump
+
+; Compile using standalone assembler.
+; RUN: %p2i --filetype=asm -i %s --target=arm32 --args -Om1 \
+; RUN:   --reg-use=s20,r5,r6 | FileCheck %s --check-prefix=ASM
+
+; Show bytes in assembled standalone code.
+; RUN: %p2i --filetype=asm -i %s --target=arm32 --assemble --disassemble \
+; RUN:   --args -Om1 --reg-use=s20,r5,r6  | FileCheck %s --check-prefix=DIS
+
+; Compile using integrated assembler.
+; RUN: %p2i --filetype=iasm -i %s --target=arm32 --args -Om1 \
+; RUN:   --reg-use=s20,r5,r6 \
+; RUN:   | FileCheck %s --check-prefix=IASM
+
+; Show bytes in assembled integrated code.
+; RUN: %p2i --filetype=iasm -i %s --target=arm32 --assemble --disassemble \
+; RUN:   --args -Om1 --reg-use=s20,r5,r6 | FileCheck %s --check-prefix=DIS
+
+define internal void @FloatToI1() {
+; ASM-LABEL: FloatToI1:
+; DIS-LABEL: {{.+}} <FloatToI1>:
+
+  %v = fptoui float 0.0 to i1
+
+; ASM:  vmov    r5, s20
+; DIS:   {{.+}}:   ee1a5a10
+; IASM-NOT: vmov
+
+  ret void
+}
+
+define internal void @FloatToI8() {
+; ASM-LABEL: FloatToI8:
+; DIS-LABEL: {{.+}} <FloatToI8>:
+
+  %v = fptoui float 0.0 to i8
+
+; ASM:  vmov    r5, s20
+; DIS:   {{.+}}:   ee1a5a10
+; IASM-NOT: vmov
+
+  ret void
+}
+
+define internal void @FloatToI16() {
+; ASM-LABEL: FloatToI16:
+; DIS-LABEL: {{.+}} <FloatToI16>:
+
+  %v = fptoui float 0.0 to i16
+
+; ASM:  vmov    r5, s20
+; DIS:   {{.+}}:   ee1a5a10
+; IASM-NOT: vmov
+
+  ret void
+}
+
+define internal void @FloatToI32() {
+; ASM-LABEL: FloatToI32:
+; DIS-LABEL: {{.+}} <FloatToI32>:
+
+  %v = fptoui float 0.0 to i32
+
+; ASM:  vmov    r5, s20
+; DIS:   {{.+}}:   ee1a5a10
+; IASM-NOT: vmov
+
+  ret void
+}
+
+define internal float @I1ToFloat() {
+; ASM-LABEL: I1ToFloat:
+; DIS-LABEL: {{.+}} <I1ToFloat>:
+
+  %v = uitofp i1 1 to float
+
+; ASM:  vmov    s20, r5
+; DIS:  {{.+}}:   ee0a5a10
+; IASM-NOT: vmov
+
+  ret float %v
+}
+
+define internal float @I8ToFloat() {
+; ASM-LABEL: I8ToFloat:
+; DIS-LABEL: {{.+}} <I8ToFloat>:
+
+  %v = uitofp i8 1 to float
+
+; ASM:  vmov    s20, r5
+; DIS:  {{.+}}:   ee0a5a10
+; IASM-NOT: vmov
+
+  ret float %v
+}
+
+define internal float @I16ToFloat() {
+; ASM-LABEL: I16ToFloat:
+; DIS-LABEL: {{.+}} <I16ToFloat>:
+
+  %v = uitofp i16 1 to float
+
+; ASM:  vmov    s20, r5
+; DIS:  {{.+}}:   ee0a5a10
+; IASM-NOT: vmov
+
+  ret float %v
+}
+
+define internal float @I32ToFloat() {
+; ASM-LABEL: I32ToFloat:
+; DIS-LABEL: {{.+}} <I32ToFloat>:
+
+  %v = uitofp i32 17 to float
+
+; ASM:  vmov    s20, r5
+; DIS:  {{.+}}:   ee0a5a10
+; IASM-NOT: vmov
+
+  ret float %v
+}