[PATCH] Allow __fp16 as a function arg or return type for AArch64

[Oliver Stannard]

ACLE 2.0 [1] allows __fp16 to be used as a function argument or return type. This enables this for AArch64.

I have not enabled this for 32-bit ARM targets yet, as we are expecting to release an updated version of the 32-bit AAPCS soon, which changes the handling of __fp16 in a non backwards-compatible way.

[1] http://infocenter.arm.com/help/topic/com.arm.doc.ihi0053c/IHI0053C_acle_2_0.pdf

http://reviews.llvm.org/D4455

Files:
  lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
  lib/Target/AArch64/AArch64CallingConvention.td
  lib/Target/AArch64/AArch64ISelLowering.cpp
  test/CodeGen/AArch64/arm64-aapcs.ll

Index: lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
===================================================================
--- lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
+++ lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
@@ -270,7 +270,7 @@
 
   EVT OrigVT = VT;
   EVT SVT = VT;
-  while (SVT != MVT::f32) {
+  while (SVT != MVT::f32 && SVT != MVT::f16) {
     SVT = (MVT::SimpleValueType)(SVT.getSimpleVT().SimpleTy - 1);
     if (ConstantFPSDNode::isValueValidForType(SVT, CFP->getValueAPF()) &&
         // Only do this if the target has a native EXTLOAD instruction from
Index: lib/Target/AArch64/AArch64CallingConvention.td
===================================================================
--- lib/Target/AArch64/AArch64CallingConvention.td
+++ lib/Target/AArch64/AArch64CallingConvention.td
@@ -54,6 +54,8 @@
 
   CCIfType<[i64], CCAssignToRegWithShadow<[X0, X1, X2, X3, X4, X5, X6, X7],
                                           [W0, W1, W2, W3, W4, W5, W6, W7]>>,
+  CCIfType<[f16], CCAssignToRegWithShadow<[H0, H1, H2, H3, H4, H5, H6, H7],
+                                          [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7]>>,
   CCIfType<[f32], CCAssignToRegWithShadow<[S0, S1, S2, S3, S4, S5, S6, S7],
                                           [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7]>>,
   CCIfType<[f64], CCAssignToRegWithShadow<[D0, D1, D2, D3, D4, D5, D6, D7],
@@ -65,7 +67,7 @@
            CCAssignToReg<[Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7]>>,
 
   // If more than will fit in registers, pass them on the stack instead.
-  CCIfType<[i1, i8, i16], CCAssignToStack<8, 8>>,
+  CCIfType<[i1, i8, i16, f16], CCAssignToStack<8, 8>>,
   CCIfType<[i32, f32], CCAssignToStack<8, 8>>,
   CCIfType<[i64, f64, v1f64, v2f32, v1i64, v2i32, v4i16, v8i8],
            CCAssignToStack<8, 8>>,
@@ -88,6 +90,8 @@
                                           [X0, X1, X2, X3, X4, X5, X6, X7]>>,
   CCIfType<[i64], CCAssignToRegWithShadow<[X0, X1, X2, X3, X4, X5, X6, X7],
                                           [W0, W1, W2, W3, W4, W5, W6, W7]>>,
+  CCIfType<[f16], CCAssignToRegWithShadow<[H0, H1, H2, H3, H4, H5, H6, H7],
+                                          [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7]>>,
   CCIfType<[f32], CCAssignToRegWithShadow<[S0, S1, S2, S3, S4, S5, S6, S7],
                                           [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7]>>,
   CCIfType<[f64], CCAssignToRegWithShadow<[D0, D1, D2, D3, D4, D5, D6, D7],
Index: lib/Target/AArch64/AArch64ISelLowering.cpp
===================================================================
--- lib/Target/AArch64/AArch64ISelLowering.cpp
+++ lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -1711,6 +1711,8 @@
         RC = &AArch64::GPR32RegClass;
       else if (RegVT == MVT::i64)
         RC = &AArch64::GPR64RegClass;
+      else if (RegVT == MVT::f16)
+        RC = &AArch64::FPR16RegClass;
       else if (RegVT == MVT::f32)
         RC = &AArch64::FPR32RegClass;
       else if (RegVT == MVT::f64 || RegVT.is64BitVector())
Index: test/CodeGen/AArch64/arm64-aapcs.ll
===================================================================
--- test/CodeGen/AArch64/arm64-aapcs.ll
+++ test/CodeGen/AArch64/arm64-aapcs.ll
@@ -109,3 +109,17 @@
 ; CHECK: ldr {{q[0-9]+}}, [sp]
   ret <2 x double> %varg_stack;
 }
+
+; Check that f16 can be passed and returned (ACLE 2.0 extension)
+define half @test_half(float, half %arg) {
+; CHECK-LABEL: test_half:
+; CHECK: mov v0.16b, v{{[0-9]+}}.16b
+  ret half %arg;
+}
+
+; Check that f16 constants are materialized correctly
+define half @test_half_const() {
+; CHECK-LABEL: test_half_const:
+; CHECK: ldr h0, [x{{[0-9]+}}, :lo12:{{.*}}]
+  ret half 0xH4248
+}
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D4455.11254.patch
Type: text/x-patch
Size: 3630 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140710/a64e251a/attachment.bin>