From de90aa00ac7f4e2ad5ead35cc0118a7666991fb7 Mon Sep 17 00:00:00 2001
From: Daniel Sanders <daniel.sanders@imgtec.com>
Date: Fri, 24 Oct 2014 13:09:19 +0000
Subject: [PATCH] [mips] For N32/N64, structs must be passed in the upper bits
 of a register.

Summary:
Most structs were fixed by r218451 but those of between >32-bits and
<64-bits remained broken since they were not marked with [ASZ]ExtUpper.
This patch fixes the remaining cases by using
CCPromoteToUpperBitsInType<i64> on i64's in addition to i32 and smaller.

Reviewers: vmedic

Reviewed By: vmedic

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D5963

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220556 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/Mips/MipsCallingConv.td       |  4 +-
 test/CodeGen/Mips/cconv/return-struct.ll | 54 ++++++++++++++++++++++++
 2 files changed, 56 insertions(+), 2 deletions(-)

diff --git a/lib/Target/Mips/MipsCallingConv.td b/lib/Target/Mips/MipsCallingConv.td
index bb45d4195ec..9e47338f09a 100644
--- a/lib/Target/Mips/MipsCallingConv.td
+++ b/lib/Target/Mips/MipsCallingConv.td
@@ -137,9 +137,9 @@ def RetCC_MipsN : CallingConv<[
   // both little and big-endian targets. When passing in registers, this
   // requires that big-endian targets shift the value into the upper bits.
   CCIfSubtarget<"isLittle()",
-      CCIfType<[i8, i16, i32], CCIfInReg<CCPromoteToType<i64>>>>,
+      CCIfType<[i8, i16, i32, i64], CCIfInReg<CCPromoteToType<i64>>>>,
   CCIfSubtargetNot<"isLittle()",
-      CCIfType<[i8, i16, i32], CCIfInReg<CCPromoteToUpperBitsInType<i64>>>>,
+      CCIfType<[i8, i16, i32, i64], CCIfInReg<CCPromoteToUpperBitsInType<i64>>>>,
 
   // i32 are returned in registers V0, V1
   CCIfType<[i32], CCAssignToReg<[V0, V1]>>,
diff --git a/test/CodeGen/Mips/cconv/return-struct.ll b/test/CodeGen/Mips/cconv/return-struct.ll
index 5252766121b..688aa3dbaa0 100644
--- a/test/CodeGen/Mips/cconv/return-struct.ll
+++ b/test/CodeGen/Mips/cconv/return-struct.ll
@@ -14,6 +14,7 @@
 
 @struct_byte = global {i8} zeroinitializer
 @struct_2byte = global {i8,i8} zeroinitializer
+@struct_3xi16 = global {[3 x i16]} zeroinitializer
 
 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture readonly, i64, i32, i1)
 
@@ -82,3 +83,56 @@ entry:
 ; N64-BE-DAG:        sh  [[R2]], 8([[SP:\$sp]])
 ; N64-BE-DAG:        lh  [[R3:\$[0-9]+]], 8([[SP:\$sp]])
 ; N64-BE-DAG:        dsll $2, [[R3]], 48
+
+; Ensure that structures bigger than 32-bits but smaller than 64-bits are
+; also returned in the upper bits on big endian targets. Previously, these were
+; missed by the CCPromoteToType and the shift didn't happen.
+define inreg {i48} @ret_struct_3xi16() nounwind {
+entry:
+        %0 = load volatile i48* bitcast ({[3 x i16]}* @struct_3xi16 to i48*), align 2
+        %1 = insertvalue {i48} undef, i48 %0, 0
+        ret {i48} %1
+}
+
+; ALL-LABEL: ret_struct_3xi16:
+
+; O32-BE-DAG:        lui [[PTR_HI:\$[0-9]+]], %hi(struct_3xi16)
+; O32-BE-DAG:        addiu [[PTR_LO:\$[0-9]+]], [[PTR_HI]], %lo(struct_3xi16)
+; O32-BE-DAG:        lhu [[R1:\$[0-9]+]], 4([[PTR_LO]])
+; O32-BE-DAG:        lw [[R2:\$[0-9]+]], %lo(struct_3xi16)([[PTR_HI]])
+; O32-BE-DAG:        sll [[R3:\$[0-9]+]], [[R2]], 16
+; O32-BE-DAG:        or  $3, [[R1]], [[R3]]
+; O32-BE-DAG:        srl $2, [[R2]], 16
+
+; O32-LE-DAG:        lui [[PTR_HI:\$[0-9]+]], %hi(struct_3xi16)
+; O32-LE-DAG:        addiu [[PTR_LO:\$[0-9]+]], [[PTR_HI]], %lo(struct_3xi16)
+; O32-LE-DAG:        lhu $3, 4([[PTR_LO]])
+; O32-LE-DAG:        lw $2, %lo(struct_3xi16)([[PTR_HI]])
+
+; N32-LE-DAG:        lui [[PTR_HI:\$[0-9]+]], %hi(struct_3xi16)
+; N32-LE-DAG:        addiu [[PTR_LO:\$[0-9]+]], [[PTR_HI]], %lo(struct_3xi16)
+; N32-LE-DAG:        lh [[R1:\$[0-9]+]], 4([[PTR_LO]])
+; N32-LE-DAG:        lwu [[R2:\$[0-9]+]], %lo(struct_3xi16)([[PTR_HI]])
+; N32-LE-DAG:        dsll [[R3:\$[0-9]+]], [[R1]], 32
+; N32-LE-DAG:        or $2, [[R2]], [[R3]]
+
+; N32-BE-DAG:        lui [[PTR_HI:\$[0-9]+]], %hi(struct_3xi16)
+; N32-BE-DAG:        addiu [[PTR_LO:\$[0-9]+]], [[PTR_HI]], %lo(struct_3xi16)
+; N32-BE-DAG:        lw [[R1:\$[0-9]+]], %lo(struct_3xi16)([[PTR_HI]])
+; N32-BE-DAG:        dsll [[R2:\$[0-9]+]], [[R1]], 16
+; N32-BE-DAG:        lhu [[R3:\$[0-9]+]], 4([[PTR_LO]])
+; N32-BE-DAG:        or [[R4:\$[0-9]+]], [[R3]], [[R2]]
+; N32-BE-DAG:        dsll $2, [[R4]], 16
+
+; N64-LE-DAG:        ld  [[PTR:\$[0-9]+]], %got_disp(struct_3xi16)($1)
+; N64-LE-DAG:        lh [[R1:\$[0-9]+]], 4([[PTR]])
+; N64-LE-DAG:        lwu [[R2:\$[0-9]+]], 0([[PTR]])
+; N64-LE-DAG:        dsll [[R3:\$[0-9]+]], [[R1]], 32
+; N64-LE-DAG:        or $2, [[R2]], [[R3]]
+
+; N64-BE-DAG:        ld  [[PTR:\$[0-9]+]], %got_disp(struct_3xi16)($1)
+; N64-BE-DAG:        lw [[R1:\$[0-9]+]], 0([[PTR]])
+; N64-BE-DAG:        dsll [[R2:\$[0-9]+]], [[R1]], 16
+; N64-BE-DAG:        lhu [[R3:\$[0-9]+]], 4([[PTR]])
+; N64-BE-DAG:        or [[R4:\$[0-9]+]], [[R3]], [[R2]]
+; N32-BE-DAG:        dsll $2, [[R4]], 16
-- 
2.34.1