Index: lib/Target/X86/X86ISelLowering.cpp
===================================================================
--- lib/Target/X86/X86ISelLowering.cpp
+++ lib/Target/X86/X86ISelLowering.cpp
@@ -15382,10 +15382,11 @@
   MemOps.push_back(Store);
 
   // Store ptr to reg_save_area.
-  FIN = DAG.getNode(ISD::ADD, DL, PtrVT, FIN, DAG.getIntPtrConstant(8, DL));
+  FIN = DAG.getNode(ISD::ADD, DL, PtrVT, FIN, DAG.getIntPtrConstant(
+      Subtarget->isTarget64BitLP64() ? 8 : 4, DL));
   SDValue RSFIN = DAG.getFrameIndex(FuncInfo->getRegSaveFrameIndex(), PtrVT);
-  Store = DAG.getStore(Op.getOperand(0), DL, RSFIN, FIN,
-                       MachinePointerInfo(SV, 16), false, false, 0);
+  Store = DAG.getStore(Op.getOperand(0), DL, RSFIN, FIN, MachinePointerInfo(
+      SV, Subtarget->isTarget64BitLP64() ? 16 : 12), false, false, 0);
   MemOps.push_back(Store);
   return DAG.getNode(ISD::TokenFactor, DL, MVT::Other, MemOps);
 }
Index: test/CodeGen/X86/musttail-varargs.ll
===================================================================
--- test/CodeGen/X86/musttail-varargs.ll
+++ test/CodeGen/X86/musttail-varargs.ll
@@ -1,4 +1,5 @@
 ; RUN: llc < %s -enable-tail-merge=0 -mtriple=x86_64-linux | FileCheck %s --check-prefix=LINUX
+; RUN: llc < %s -enable-tail-merge=0 -mtriple=x86_64-linux-gnux32 | FileCheck %s --check-prefix=LINUX-X32
 ; RUN: llc < %s -enable-tail-merge=0 -mtriple=x86_64-windows | FileCheck %s --check-prefix=WINDOWS
 ; RUN: llc < %s -enable-tail-merge=0 -mtriple=i686-windows | FileCheck %s --check-prefix=X86
 
@@ -57,6 +58,40 @@
 ; LINUX-DAG: movb {{.*}}, %al
 ; LINUX: jmpq *{{.*}}  # TAILCALL
 
+; LINUX-X32-LABEL: f_thunk:
+; LINUX-X32-DAG: movl %edi, {{.*}}
+; LINUX-X32-DAG: movq %rsi, {{.*}}
+; LINUX-X32-DAG: movq %rdx, {{.*}}
+; LINUX-X32-DAG: movq %rcx, {{.*}}
+; LINUX-X32-DAG: movq %r8, {{.*}}
+; LINUX-X32-DAG: movq %r9, {{.*}}
+; LINUX-X32-DAG: movb %al, {{.*}}
+; LINUX-X32-DAG: movaps %xmm0, {{[0-9]*}}(%esp)
+; LINUX-X32-DAG: movaps %xmm1, {{[0-9]*}}(%esp)
+; LINUX-X32-DAG: movaps %xmm2, {{[0-9]*}}(%esp)
+; LINUX-X32-DAG: movaps %xmm3, {{[0-9]*}}(%esp)
+; LINUX-X32-DAG: movaps %xmm4, {{[0-9]*}}(%esp)
+; LINUX-X32-DAG: movaps %xmm5, {{[0-9]*}}(%esp)
+; LINUX-X32-DAG: movaps %xmm6, {{[0-9]*}}(%esp)
+; LINUX-X32-DAG: movaps %xmm7, {{[0-9]*}}(%esp)
+; LINUX-X32: callq get_f
+; LINUX-X32-DAG: movaps {{[0-9]*}}(%esp), %xmm0
+; LINUX-X32-DAG: movaps {{[0-9]*}}(%esp), %xmm1
+; LINUX-X32-DAG: movaps {{[0-9]*}}(%esp), %xmm2
+; LINUX-X32-DAG: movaps {{[0-9]*}}(%esp), %xmm3
+; LINUX-X32-DAG: movaps {{[0-9]*}}(%esp), %xmm4
+; LINUX-X32-DAG: movaps {{[0-9]*}}(%esp), %xmm5
+; LINUX-X32-DAG: movaps {{[0-9]*}}(%esp), %xmm6
+; LINUX-X32-DAG: movaps {{[0-9]*}}(%esp), %xmm7
+; LINUX-X32-DAG: movl {{.*}}, %edi
+; LINUX-X32-DAG: movq {{.*}}, %rsi
+; LINUX-X32-DAG: movq {{.*}}, %rdx
+; LINUX-X32-DAG: movq {{.*}}, %rcx
+; LINUX-X32-DAG: movq {{.*}}, %r8
+; LINUX-X32-DAG: movq {{.*}}, %r9
+; LINUX-X32-DAG: movb {{.*}}, %al
+; LINUX-X32: jmpq *{{.*}}  # TAILCALL
+
 ; WINDOWS-LABEL: f_thunk:
 ; WINDOWS-NOT: mov{{.}}ps
 ; WINDOWS-DAG: movq %rdx, {{.*}}
@@ -92,6 +127,10 @@
 ; LINUX-NOT: movq
 ; LINUX: jmpq *%rdi  # TAILCALL
 
+; LINUX-X32-LABEL: g_thunk:
+; LINUX-X32-DAG: movl %edi, %[[REG:e[abcd]x|ebp|esi|edi|r8|r9|r1[0-5]]]
+; LINUX-X32-DAG: jmpq *%[[REG]]  # TAILCALL
+
 ; WINDOWS-LABEL: g_thunk:
 ; WINDOWS-NOT: movq
 ; WINDOWS: jmpq *%rcx # TAILCALL
@@ -130,6 +169,10 @@
 ; LINUX: jne
 ; LINUX: jmpq *{{.*}} # TAILCALL
 ; LINUX: jmpq *{{.*}} # TAILCALL
+; LINUX-X32-LABEL: h_thunk:
+; LINUX-X32: jne
+; LINUX-X32: jmpq *{{.*}} # TAILCALL
+; LINUX-X32: jmpq *{{.*}} # TAILCALL
 ; WINDOWS-LABEL: h_thunk:
 ; WINDOWS: jne
 ; WINDOWS: jmpq *{{.*}} # TAILCALL
Index: test/CodeGen/X86/soft-fp.ll
===================================================================
--- test/CodeGen/X86/soft-fp.ll
+++ test/CodeGen/X86/soft-fp.ll
@@ -1,5 +1,6 @@
 ; RUN: llc < %s -march=x86    -mattr=+sse2,+soft-float | FileCheck %s
 ; RUN: llc < %s -march=x86-64 -mattr=+sse2,+soft-float | FileCheck %s
+; RUN: llc < %s -mtriple=x86_64-gnux32 -mattr=+sse2,+soft-float | FileCheck %s
 
 ; CHECK-NOT: xmm{[0-9]+}
 
Index: test/CodeGen/X86/stdarg.ll
===================================================================
--- test/CodeGen/X86/stdarg.ll
+++ test/CodeGen/X86/stdarg.ll
@@ -1,5 +1,4 @@
-; RUN: llc < %s -mtriple=x86_64-linux | FileCheck %s
-; CHECK: testb %al, %al
+; RUN: llc < %s -mtriple=x86_64-linux | FileCheck %s -check-prefix=CHECK
 
 %struct.__va_list_tag = type { i32, i32, i8*, i8* }
 
@@ -8,6 +7,15 @@
   %ap = alloca [1 x %struct.__va_list_tag], align 8; <[1 x %struct.__va_list_tag]*> [#uses=2]
   %ap12 = bitcast [1 x %struct.__va_list_tag]* %ap to i8*; <i8*> [#uses=2]
   call void @llvm.va_start(i8* %ap12)
+; CHECK: testb %al, %al
+
+; These test for specific offsets, which is very fragile. Still, the test needs
+; to ensure that va_list is correctly handled by both x86-64 and x32.
+;
+; CHECK-DAG: movq {{.*}}, 192(%rsp)
+; CHECK-DAG: movq {{.*}}, 184(%rsp)
+; CHECK-DAG: movl {{.*}}, 180(%rsp)
+; CHECK-DAG: movl {{.*}}, 176(%rsp)
   %ap3 = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i64 0, i64 0; <%struct.__va_list_tag*> [#uses=1]
   call void @bar(%struct.__va_list_tag* %ap3) nounwind
   call void @llvm.va_end(i8* %ap12)
Index: test/CodeGen/X86/x32-va_start.ll
===================================================================
--- /dev/null
+++ test/CodeGen/X86/x32-va_start.ll
@@ -0,0 +1,99 @@
+; RUN: llc < %s -mtriple=x86_64-linux-gnux32 | FileCheck %s -check-prefix=CHECK -check-prefix=CHECK-SSE
+; RUN: llc < %s -mtriple=x86_64-linux-gnux32 -mattr=-sse | FileCheck %s -check-prefix=CHECK -check-prefix=CHECK-NOSSE
+;
+; Verifies that x32 va_start lowering is sane. To regenerate this test, use
+; cat <<EOF |
+; #include <stdarg.h>
+;
+; int foo(float a, const char* fmt, ...) {
+;   va_list ap;
+;   va_start(ap, fmt);
+;   int value = va_arg(ap, int);
+;   va_end(ap);
+;   return value;
+; }
+; EOF
+; build/bin/clang -mx32 -O3 -o- -S -emit-llvm -xc -
+;
+target datalayout = "e-m:e-p:32:32-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnux32"
+
+%struct.__va_list_tag = type { i32, i32, i8*, i8* }
+
+define i32 @foo(float %a, i8* nocapture readnone %fmt, ...) nounwind {
+entry:
+  %ap = alloca [1 x %struct.__va_list_tag], align 16
+  %0 = bitcast [1 x %struct.__va_list_tag]* %ap to i8*
+  call void @llvm.lifetime.start(i64 16, i8* %0) #2
+  call void @llvm.va_start(i8* %0)
+; CHECK-SSE: subl $72, %esp
+; CHECK-SSE: testb %al, %al
+; CHECK-SSE: je .[[NOFP:.*]]
+; CHECK-SSE-DAG: movaps %xmm1
+; CHECK-SSE-DAG: movaps %xmm2
+; CHECK-SSE-DAG: movaps %xmm3
+; CHECK-SSE-DAG: movaps %xmm4
+; CHECK-SSE-DAG: movaps %xmm5
+; CHECK-SSE-DAG: movaps %xmm6
+; CHECK-SSE-DAG: movaps %xmm7
+; CHECK-NOSSE-NOT: xmm
+; CHECK-SSE: .[[NOFP]]:
+; CHECK-DAG: movq %r9
+; CHECK-DAG: movq %r8
+; CHECK-DAG: movq %rcx
+; CHECK-DAG: movq %rdx
+; CHECK-DAG: movq %rsi
+  %gp_offset_p = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0, i32 0
+  %gp_offset = load i32, i32* %gp_offset_p, align 16
+  %fits_in_gp = icmp ult i32 %gp_offset, 41
+  br i1 %fits_in_gp, label %vaarg.in_reg, label %vaarg.in_mem
+; CHECK: cmpl $40, [[COUNT:.*]]
+; CHECK: ja .[[IN_MEM:.*]]
+
+vaarg.in_reg:                                     ; preds = %entry
+  %1 = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0, i32 3
+  %reg_save_area = load i8*, i8** %1, align 4
+  %2 = getelementptr i8, i8* %reg_save_area, i32 %gp_offset
+  %3 = add i32 %gp_offset, 8
+  store i32 %3, i32* %gp_offset_p, align 16
+  br label %vaarg.end
+; CHECK: movl {{[^,]*}}, [[ADDR:.*]]
+; CHECK: addl [[COUNT]], [[ADDR]]
+; CHECK-SSE: jmp .[[END:.*]]
+; CHECK-NOSSE: movl ([[ADDR]]), %eax
+; CHECK-NOSSE: retq
+; CHECK: .[[IN_MEM]]:
+vaarg.in_mem:                                     ; preds = %entry
+  %overflow_arg_area_p = getelementptr inbounds [1 x %struct.__va_list_tag], [1 x %struct.__va_list_tag]* %ap, i32 0, i32 0, i32 2
+  %overflow_arg_area = load i8*, i8** %overflow_arg_area_p, align 8
+  %overflow_arg_area.next = getelementptr i8, i8* %overflow_arg_area, i32 8
+  store i8* %overflow_arg_area.next, i8** %overflow_arg_area_p, align 8
+  br label %vaarg.end
+; CHECK: movl {{[^,]*}}, [[ADDR]]
+; CHECK-NOSSE: movl ([[ADDR]]), %eax
+; CHECK-NOSSE: retq
+; CHECK-SSE: .[[END]]:
+
+vaarg.end:                                        ; preds = %vaarg.in_mem, %vaarg.in_reg
+  %vaarg.addr.in = phi i8* [ %2, %vaarg.in_reg ], [ %overflow_arg_area, %vaarg.in_mem ]
+  %vaarg.addr = bitcast i8* %vaarg.addr.in to i32*
+  %4 = load i32, i32* %vaarg.addr, align 4
+  call void @llvm.va_end(i8* %0)
+  call void @llvm.lifetime.end(i64 16, i8* %0) #2
+  ret i32 %4
+; CHECK-SSE: movl ([[ADDR]]), %eax
+; CHECK-SSE: retq
+}
+
+; Function Attrs: nounwind argmemonly
+declare void @llvm.lifetime.start(i64, i8* nocapture) nounwind
+
+; Function Attrs: nounwind
+declare void @llvm.va_start(i8*) nounwind
+
+; Function Attrs: nounwind
+declare void @llvm.va_end(i8*) nounwind
+
+; Function Attrs: nounwind argmemonly
+declare void @llvm.lifetime.end(i64, i8* nocapture) nounwind
+