Register Calling Convention - CodeGen Basic Additions

[oren_ben_simhon]

Summary

The Register Calling Convention (RegCall) was introduced by Intel to optimize parameter transfer on function call.
This calling convention ensures that as many values as possible are passed or returned in registers.

The following review presents the basic additions to LLVM CodeGen in order to support RegCall in X86.
The review also includes two lit files that exhaustively test regcall in various configurations.

[rnk]

lib/Target/X86/X86CallingConv.td

59

Maybe say "SysV" instead of "Lin". You also want to use this on the BSDs and all non-Windows platforms, right?

67

This would be CCIfSubtargetSysV, and the logic would probably be "!isTargetWindows()" or something like that.

1120

Do you anticipate that users will use this convention in a context where SSE is disabled? If so, you should look at r282819 and do something similar.

Honestly, we should refactor this code. We should probably compute these lists by filtering out register classes in X86RegisterInfo rather than table generating N different lists for every calling convention.

[asl]

Can't you yield the same outcome via 'inreg' attribute?

[rnk]

In D25022#557751, @asl wrote:

Can't you yield the same outcome via 'inreg' attribute?

It will have similar effects, but it is not ABI compatible with Intel's __regcall.

[mkuper]

lib/Target/X86/X86CallingConv.td

37

This is really odd. Maybe fix AssignToReg?

53

GRP_8 and GPR_16 are only used for return values, right?
When will you use anything but AL/AX to return?

95

If I'm reading this correctly, with this definition, you'll pass f80 in an incompatible way - which sounds fairly bad. The right thing to do would be to either implement it in one go, or leave a TODO, but refuse to compile cases which use f80.
Otherwise, once you want to fix this, you'll have an unsolvable compatibility problem, since you can't be compatible with both old versions of clang, and with ICC.

125

Why <0, 0> vs. <0, 4>? Should the 4 have been 0 as well?

128

How can we even get here? Is regcall supported for FP on non-SSE platforms?

141

Extra space.

144

Extra space.

156

Same as for the fp80 case - leaving TODOs in calling conventions is not generally a good idea, unless the temporary behavior is "fail".

test/CodeGen/X86/avx512-regcall-NoMask.ll

347

Have you checked what happens when you try to pass a 256-bit vector without AVX, 512-bit without AVX512, etc?
I assume there's no compatibility requirement for these cases, but it'd be good to check it does something that makes sense.

test/CodeGen/X86/sse-regcall.ll

208

Please add a newline.

[oren_ben_simhon]

In D25022#559221, @rnk wrote:

In D25022#557751, @asl wrote:

Can't you yield the same outcome via 'inreg' attribute?

It will have similar effects, but it is not ABI compatible with Intel's __regcall.

Also, InReg attribute tells backend to save an argument into a register,
but doesn’t tell you: how, where, in which order?
To answer all these questions __regcall spec was defined.

lib/Target/X86/X86CallingConv.td

53

You are right, GPR8/GPR16 are used just for return values.
In regcall calling convention there are cases in which you return a value in multiple registers.
Some examples:

• v64i1 (mask) in IA32 is splitted into two registers
• a structure that is returned, will be splitted to multiple registers according to its fields.

95

As you said, interoperability is not possible. At this moment, this is true due to many reasons.

I had to split the feature to multiple small reviews, because there are 3 additional features that required a lot of changes:

• v64i1 - split values to ultiple registers
• v*i1 types - mask registers support
• fp80 - update floating point support to enable register based arguments

So, I believe that it is better to handle this gracefully for the time being.

125

In the case of f80 on 32 bit architecture, we know that the stack alignment is 4 bytes.
So there is no need to pass 0 as the second parameter (which is translated to another function query of the type)
See also CC_X86_32_Common.

128

Yes, regcall supports non-sse platforms. In that case, FP will be saved on the stack.

1120

Yes, SSE disabled is possible. I will make similar changes to what you did in r282819.
I wanted to ask you, why did you update the returned SaveList in x86RegisterInfo.cpp but not the corresponding RegMask?

test/CodeGen/X86/avx512-regcall-NoMask.ll

347

Indeed this is a nice test. Such test is part of sse_regcall.ll

[oren_ben_simhon]

lib/Target/X86/X86CallingConv.td

37

AssignToReg, takes the list of registers and creates a static constant list of registers. For example:

static const MCPhysReg RegList1[] = {
  X86::ECX, X86::EDX, X86::R8D, X86::R9D
};

Empty static constant list is forbidden in C.

Changing the expansion of AssignToReg (to check if the list is empty) will require LLVM infrastructure change.
I would prefer to leave it out of the scope of this review.

95

Anyway, i added an assertion for the unhandled types.

[oren_ben_simhon]

Uploaded updates according to the comments posted until 10/04

[rnk]

lib/Target/X86/X86CallingConv.h

47 ↗

(On Diff #73629)

format

49–50 ↗

(On Diff #73629)

Use report_fatal_error so that it fails in release builds

lib/Target/X86/X86CallingConv.td

1120

Hm, that's an oversight, but I guess it isn't observable because a function with SSE disabled won't have any XMM* registers live across the call.

[oren_ben_simhon]

Added updates according to comments posted until 10/06

[mkuper]

lib/Target/X86/X86CallingConv.h

51 ↗

(On Diff #73742)

The whole point is that we don't gracefully continue. :-)
Can you make this llvm_unreachable?

lib/Target/X86/X86CallingConv.td

37

I agree this is not in scope of this review. The way we tend to do this is by first fixing the problem (in a separate review), and then using the fix in a the functional change.

In any case, I'm not sure I see what the problem is - an empty static constant array compiles just fine, e.g.
static const int foo[] = {};
What am I missing?

53

Ah, ok, I see, thanks.

117

Can you remove this until you reintroduce f80 support?
I think leaving it in is just confusing at this point.

128

Is this behavior documented somewhere?
https://software.intel.com/en-us/node/693069 suggests float/double values are always classified as XMM.

test/CodeGen/X86/avx512-regcall-NoMask.ll

145

Perhaps a test that exercises this case more fully (modifies both registers)?

181

In all the float/double/vector tests - shouldn't you be looking for a specific register in a lot of places?

542

I don't see any struct tests.
Do you expect the FE to completely break down structs for you?

If you do:

1. What about returning structs?
2. Would this actually do the right thing for regcall? That is, can you pass "half a struct" in registers, and half on the stack? I assume you don't expect to use inreg.

test/CodeGen/X86/sse-regcall.ll

86

This doesn't actually check how the "oversize" arguments are passed.
What I'm driving at here is that if this is supported (and documented) by ICC, we need to do it in a compatible way.

[oren_ben_simhon]

lib/Target/X86/X86CallingConv.td

37

I am not sure what is the difference between our environments (are you compiling for C instead for C++?), here is the error message that i get:
C:\work\obensimh\ICL\llorg_ws\builds\lib\Target\X86\X86GenCallingConv.inc(817): error C2466: cannot allocate an array of constant size 0

You can find more details in the following link:
http://stackoverflow.com/questions/9881777/why-do-i-get-cannot-allocate-an-array-of-constant-size-0

128

I will make sure it is added to the documentation and add a comment in the code as well

test/CodeGen/X86/avx512-regcall-NoMask.ll

542

Yes, I do expect the structures to be broken down by FE.
However, the FE should break the structure down only if it can break down all of it. No partial break down allowed.
In other words, you can't pass half structure on stack and half on registers. In that case, the structure will be passed through a pointer.

You are right, we are not depending on inreg attribute.

In the case of returned structure, the behaviour is similar to passing a structure (that was broken down) to a function.
So actually, all tests that call a regcall function and pass to it parameters, have similar behaviour to tests that return a structure.
In order to make it clearer, I added a test that pass/return a structure by value.

test/CodeGen/X86/sse-regcall.ll

86

If i undersand you correctly, test_f32stack checks for "oversize" arguments which are saved on the stack.
All tests are compatible with the ABI.

[oren_ben_simhon]

Fixed all comments submitted until 10/09

[mkuper]

It would be great if another person at Intel went through the test-cases, and verified that they really confirm to the ABI - I'm not familiar with the calling convention beyond what's documented, so...

lib/Target/X86/X86CallingConv.h

51 ↗

(On Diff #73742)

I'm sorry, what I meant was - keep report_fatal_error, but make the *return* llvm_unreachable, if possible (I don't remember if this elicits warnings or not.)

lib/Target/X86/X86CallingConv.td

37

Ok, I see. Both GCC and Clang accept this by default as an extension, but I guess MSVC doesn't (and maybe GCC/Clang don't either with the flags we use for LLVM).

Anyway, I'm ok with deferring fixing this to a separate patch. Please make the comment here into a TODO.

103

AVX? Or should this get passed as 2 * XMM if we have SSE but not AVX?

108

AVX512?

128

Ok, thanks.

170

Same as above.

174

Same as above.

[MatzeB]

lib/Target/X86/X86CallingConv.h

51 ↗

(On Diff #73742)

report_fatal_error() should be enough (and is declared with the noreturn attribute). Having both is unnecessary (if the unreachable comes 2nd we won't reach it) or wrong (if llvm_unreachable comes first the compiler can optimize out everything including the following report_fatal_error).

[mkuper]

lib/Target/X86/X86CallingConv.h

51 ↗

(On Diff #73742)

Ah, ok, didn't realize report_fatal_error is declared noreturn.
So we can just remove the "return false".

[oren_ben_simhon]

lib/Target/X86/X86CallingConv.td

103

You are right, AVX is the correct subtarget. If there is no AVX, it should be passed through the stack. GMAR HATIMA TOVA :)

[oren_ben_simhon]

Fixed comments submitted until 10/10

[mkuper]

LGTM, with a caveat - it would be really good to have an additional person at Intel (very preferably someone familiar with regcall) go over the tests and make sure that this does what the calling convention says it should.

[mkuper]

Oren, are you planning to have someone else at Intel review the test post-commit, or has that already happened offline?

[oren_ben_simhon]

In D25022#569388, @mkuper wrote:

Oren, are you planning to have someone else at Intel review the test post-commit, or has that already happened offline?

Thanks Michael for the review.
Your inputs were valuable and helped me a lot (as I am a new member in the LLVM/compilers community).

Elena Demikhovsky (also CCed on this review) did an internal review on the tests.
Elena, whom I think you are familiar with, is part of Intel's compilers group.

include/llvm/IR/CallingConv.h

enum {
     AMDGPU_PS = 89,
 
     /// Calling convention used for Mesa compute shaders.
     AMDGPU_CS = 90,
 
     /// Calling convention for AMDGPU code object kernels.
     AMDGPU_KERNEL = 91,
 
+    /// Register calling convention used for parameters transfer optimization
+    X86_RegCall = 92,
+
     /// The highest possible calling convention ID. Must be some 2^k - 1.
     MaxID = 1023
   };
 } // End CallingConv namespace
 
 } // End llvm namespace
 
 #endif

lib/AsmParser/LLLexer.cpp

#define KEYWORD(STR)                                                           \
   KEYWORD(avr_signalcc);
   KEYWORD(ptx_kernel);
   KEYWORD(ptx_device);
   KEYWORD(spir_kernel);
   KEYWORD(spir_func);
   KEYWORD(intel_ocl_bicc);
   KEYWORD(x86_64_sysvcc);
   KEYWORD(x86_64_win64cc);
+  KEYWORD(x86_regcallcc);
   KEYWORD(webkit_jscc);
   KEYWORD(swiftcc);
   KEYWORD(anyregcc);
   KEYWORD(preserve_mostcc);
   KEYWORD(preserve_allcc);
   KEYWORD(ghccc);
   KEYWORD(x86_intrcc);
   KEYWORD(hhvmcc);

lib/AsmParser/LLParser.cpp

 bool LLParser::ParseOptionalCallingConv(unsigned &CC) {
   switch (Lex.getKind()) {
   default:                       CC = CallingConv::C; return false;
   case lltok::kw_ccc:            CC = CallingConv::C; break;
   case lltok::kw_fastcc:         CC = CallingConv::Fast; break;
   case lltok::kw_coldcc:         CC = CallingConv::Cold; break;
   case lltok::kw_x86_stdcallcc:  CC = CallingConv::X86_StdCall; break;
   case lltok::kw_x86_fastcallcc: CC = CallingConv::X86_FastCall; break;
+  case lltok::kw_x86_regcallcc:  CC = CallingConv::X86_RegCall; break;
   case lltok::kw_x86_thiscallcc: CC = CallingConv::X86_ThisCall; break;
   case lltok::kw_x86_vectorcallcc:CC = CallingConv::X86_VectorCall; break;
   case lltok::kw_arm_apcscc:     CC = CallingConv::ARM_APCS; break;
   case lltok::kw_arm_aapcscc:    CC = CallingConv::ARM_AAPCS; break;
   case lltok::kw_arm_aapcs_vfpcc:CC = CallingConv::ARM_AAPCS_VFP; break;
   case lltok::kw_msp430_intrcc:  CC = CallingConv::MSP430_INTR; break;
   case lltok::kw_avr_intrcc:     CC = CallingConv::AVR_INTR; break;
   case lltok::kw_avr_signalcc:   CC = CallingConv::AVR_SIGNAL; break;

lib/AsmParser/LLToken.h

enum Kind {
   kw_ccc,
   kw_fastcc,
   kw_coldcc,
   kw_intel_ocl_bicc,
   kw_x86_stdcallcc,
   kw_x86_fastcallcc,
   kw_x86_thiscallcc,
   kw_x86_vectorcallcc,
+  kw_x86_regcallcc,
   kw_arm_apcscc,
   kw_arm_aapcscc,
   kw_arm_aapcs_vfpcc,
   kw_msp430_intrcc,
   kw_avr_intrcc,
   kw_avr_signalcc,
   kw_ptx_kernel,
   kw_ptx_device,

lib/IR/AsmWriter.cpp

static void PrintCallingConv(unsigned cc, raw_ostream &Out) {
   case CallingConv::AnyReg:        Out << "anyregcc"; break;
   case CallingConv::PreserveMost:  Out << "preserve_mostcc"; break;
   case CallingConv::PreserveAll:   Out << "preserve_allcc"; break;
   case CallingConv::CXX_FAST_TLS:  Out << "cxx_fast_tlscc"; break;
   case CallingConv::GHC:           Out << "ghccc"; break;
   case CallingConv::X86_StdCall:   Out << "x86_stdcallcc"; break;
   case CallingConv::X86_FastCall:  Out << "x86_fastcallcc"; break;
   case CallingConv::X86_ThisCall:  Out << "x86_thiscallcc"; break;
+  case CallingConv::X86_RegCall:   Out << "x86_regcallcc"; break;
   case CallingConv::X86_VectorCall:Out << "x86_vectorcallcc"; break;
   case CallingConv::Intel_OCL_BI:  Out << "intel_ocl_bicc"; break;
   case CallingConv::ARM_APCS:      Out << "arm_apcscc"; break;
   case CallingConv::ARM_AAPCS:     Out << "arm_aapcscc"; break;
   case CallingConv::ARM_AAPCS_VFP: Out << "arm_aapcs_vfpcc"; break;
   case CallingConv::MSP430_INTR:   Out << "msp430_intrcc"; break;
   case CallingConv::AVR_INTR:      Out << "avr_intrcc "; break;
   case CallingConv::AVR_SIGNAL:    Out << "avr_signalcc "; break;

lib/Target/X86/X86CallingConv.td

 //===----------------------------------------------------------------------===//
 
 /// CCIfSubtarget - Match if the current subtarget has a feature F.
 class CCIfSubtarget<string F, CCAction A>
     : CCIf<!strconcat("static_cast<const X86Subtarget&>"
                        "(State.getMachineFunction().getSubtarget()).", F),
            A>;
 
+// Register classes for RegCall
+class RC_X86_RegCall {
+  list<Register> GPR_8 = [];
+  list<Register> GPR_16 = [];
+  list<Register> GPR_32 = [];
+  list<Register> GPR_64 = [];
+  list<Register> XMM = [];
+  list<Register> YMM = [];
+  list<Register> ZMM = [];
+}
+
+// RegCall register classes for 32 bits
+def RC_X86_32_RegCall : RC_X86_RegCall {
+  let GPR_8 = [AL, CL, DL, DIL, SIL];
+  let GPR_16 = [AX, CX, DX, DI, SI];
+  let GPR_32 = [EAX, ECX, EDX, EDI, ESI];
+  let GPR_64 = [RAX]; /* not actually used, but AssignToReg can't handle [] */
+  let XMM = [XMM0, XMM1, XMM2, XMM3, XMM4, XMM5, XMM6, XMM7];
+  let YMM = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7];
+  let ZMM = [ZMM0, ZMM1, ZMM2, ZMM3, ZMM4, ZMM5, ZMM6, ZMM7];
+}
+
+class RC_X86_64_RegCall : RC_X86_RegCall {
+  let XMM = [XMM0, XMM1, XMM2, XMM3, XMM4, XMM5, XMM6, XMM7,
+             XMM8, XMM9, XMM10, XMM11, XMM12, XMM13, XMM14, XMM15];
+  let YMM = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
+             YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15];
+  let ZMM = [ZMM0, ZMM1, ZMM2, ZMM3, ZMM4, ZMM5, ZMM6, ZMM7,
+             ZMM8, ZMM9, ZMM10, ZMM11, ZMM12, ZMM13, ZMM14, ZMM15];
+}
+
+def RC_X86_64_RegCall_Win : RC_X86_64_RegCall {
+  let GPR_8 = [AL, CL, DL, DIL, SIL, R8B, R9B, R10B, R11B, R12B, R14B, R15B];
+  let GPR_16 = [AX, CX, DX, DI, SI, R8W, R9W, R10W, R11W, R12W, R14W, R15W];
+  let GPR_32 = [EAX, ECX, EDX, EDI, ESI, R8D, R9D, R10D, R11D, R12D, R14D, R15D];
+  let GPR_64 = [RAX, RCX, RDX, RDI, RSI, R8, R9, R10, R11, R12, R14, R15];
+}
+
+def RC_X86_64_RegCall_Lin : RC_X86_64_RegCall {
+  let GPR_8 = [AL, CL, DL, DIL, SIL, R8B, R9B, R12B, R13B, R14B, R15B];
+  let GPR_16 = [AX, CX, DX, DI, SI, R8W, R9W, R12W, R13W, R14W, R15W];
+  let GPR_32 = [EAX, ECX, EDX, EDI, ESI, R8D, R9D, R12D, R13D, R14D, R15D];
+  let GPR_64 = [RAX, RCX, RDX, RDI, RSI, R8, R9, R12, R13, R14, R15];
+}
+
+// Match if the current subtarget is either Linux or OSX
+class CCIfSubtargetLinuxOSX<CCAction A> : CCIf<
+  "static_cast<const X86Subtarget&>(State.getMachineFunction().getSubtarget())."
+  "isTargetLinux() ||"
+  "static_cast<const X86Subtarget&>(State.getMachineFunction().getSubtarget())."
+  "getTargetTriple().isMacOSX()",
+  A
+>;
+
+// X86-64 Intel regcall calling convention.
+multiclass X86_RegCall_base<RC_X86_RegCall RC> {
+def CC_#NAME : CallingConv<[
+  // Handles byval parameters.
+    CCIfSubtarget<"is64Bit()", CCIfByVal<CCPassByVal<8, 8>>>,
+    CCIfByVal<CCPassByVal<4, 4>>,
+
+    // Promote i1/i8/i16 arguments to i32.
+    CCIfType<[i1, i8, i16], CCPromoteToType<i32>>,
+
+    // bool, char, int, enum, long, pointer --> GPR
+    CCIfType<[i32], CCAssignToReg<RC.GPR_32>>,
+
+    // TODO: Handle the case of mask types (v*i1)
+    // TODO: Handle the case of 32 bit machine with v64i1 argument 
+    //       (split to 2 registers)
+
+    // long long, __int64 --> GPR
+    CCIfType<[i64], CCAssignToReg<RC.GPR_64>>,
+
+    // TODO: Handle the case of long double (f80)
+
+    // float, double, float128 --> XMM
+    CCIfType<[f32, f64, f128], CCAssignToReg<RC.XMM>>,
+
+    // __m128, __m128i, __m128d --> XMM
+    CCIfType<[v16i8, v8i16, v4i32, v2i64, v4f32, v2f64], 
+      CCAssignToReg<RC.XMM>>,
+
+    // __m256, __m256i, __m256d --> YMM
+    CCIfType<[v32i8, v16i16, v8i32, v4i64, v8f32, v4f64], 
+      CCAssignToReg<RC.YMM>>,
+
+    // __m512, __m512i, __m512d --> ZMM
+    CCIfType<[v64i8, v32i16, v16i32, v8i64, v16f32, v8f64], 
+      CCAssignToReg<RC.ZMM>>,
+
+    // If no register was found -> assign to stack
+
+    // In 64 bit, assign 64/32 bit values to 8 byte stack
+    CCIfSubtarget<"is64Bit()", CCIfType<[i32, i64, f32, f64], 
+      CCAssignToStack<8, 8>>>,
+
+    // In 32 bit, assign 64/32 bit values to 8/4 byte stack
+    CCIfType<[i32, f32], CCAssignToStack<4, 4>>,
+    CCIfType<[f64], CCAssignToStack<8, 4>>,
+
+    // Long doubles get stack slots whose size and alignment depends on the
+    // subtarget.
+    CCIfSubtarget<"is64Bit()", CCIfType<[f80], CCAssignToStack<0, 0>>>,
+    CCIfType<[f80], CCAssignToStack<0, 4>>,
+
+    // MMX type gets 8 byte slot in stack , while alignment depends on target
+    CCIfSubtarget<"is64Bit()", CCIfType<[x86mmx], CCAssignToStack<8, 8>>>,
+    CCIfType<[x86mmx], CCAssignToStack<8, 4>>,
+
+    // float 128 get stack slots whose size and alignment depends 
+    // on the subtarget.
+    CCIfType<[f128], CCAssignToStack<0, 0>>,
+
+    // Vectors get 16-byte stack slots that are 16-byte aligned.
+    CCIfType<[v16i8, v8i16, v4i32, v2i64, v4f32, v2f64], 
+      CCAssignToStack<16, 16>>,
+
+    // 256-bit vectors get 32-byte stack slots that are 32-byte aligned.
+    CCIfType<[v32i8, v16i16, v8i32, v4i64, v8f32, v4f64], 
+      CCAssignToStack<32, 32 >>,
+
+    // 512-bit vectors get 64-byte stack slots that are 64-byte aligned.
+    CCIfType<[v16i32, v8i64, v16f32, v8f64], CCAssignToStack<64, 64 >>
+]>;
+
+def RetCC_#NAME : CallingConv<[
+    // Promote i1 arguments to i8.
+    CCIfType<[i1], CCPromoteToType<i8>>,
+
+    // bool, char, int, enum, long, pointer --> GPR
+    CCIfType<[i8], CCAssignToReg<RC.GPR_8>>,
+    CCIfType<[i16], CCAssignToReg<RC.GPR_16>>,
+    CCIfType<[i32], CCAssignToReg<RC.GPR_32>>,
+
+    // TODO: Handle the case of mask types (v*i1)
+    // TODO: Handle the case of 32 bit machine with v64i1 argument 
+      (split to 2 registers)
+
+    // long long, __int64 --> GPR
+    CCIfType<[i64], CCAssignToReg<RC.GPR_64>>,
+
+    // long double --> FP
+    CCIfType<[f80], CCAssignToReg<[FP0]>>,
+
+    // float, double, float128 --> XMM
+    CCIfType<[f32, f64, f128], CCAssignToReg<RC.XMM>>,
+
+    // __m128, __m128i, __m128d --> XMM
+    CCIfType<[v16i8, v8i16, v4i32, v2i64, v4f32, v2f64], 
+      CCAssignToReg<RC.XMM>>,
+
+    // __m256, __m256i, __m256d --> YMM
+    CCIfType<[v32i8, v16i16, v8i32, v4i64, v8f32, v4f64], 
+      CCAssignToReg<RC.YMM>>,
+
+    // __m512, __m512i, __m512d --> ZMM
+    CCIfType<[v64i8, v32i16, v16i32, v8i64, v16f32, v8f64], 
+      CCAssignToReg<RC.ZMM>>
+]>;
+}
+
 //===----------------------------------------------------------------------===//
 // Return Value Calling Conventions
 //===----------------------------------------------------------------------===//
 
 // Return-value conventions common to all X86 CC's.
 def RetCC_X86Common : CallingConv<[
   // Scalar values are returned in AX first, then DX.  For i8, the ABI
   // requires the values to be in AL and AH, however this code uses AL and DL
def RetCC_X86_64_HHVM: CallingConv<[
   // Promote all types to i64
   CCIfType<[i8, i16, i32], CCPromoteToType<i64>>,
 
   // Return: could return in any GP register save RSP and R12.
   CCIfType<[i64], CCAssignToReg<[RBX, RBP, RDI, RSI, RDX, RCX, R8, R9,
                                  RAX, R10, R11, R13, R14, R15]>>
 ]>;
 
+
+defm X86_32_RegCall :
+   X86_RegCall_base<RC_X86_32_RegCall>;
+defm X86_Win64_RegCall :
+     X86_RegCall_base<RC_X86_64_RegCall_Win>;
+defm X86_Lin64_RegCall :
+     X86_RegCall_base<RC_X86_64_RegCall_Lin>;
+
 // This is the root return-value convention for the X86-32 backend.
 def RetCC_X86_32 : CallingConv<[
   // If FastCC, use RetCC_X86_32_Fast.
   CCIfCC<"CallingConv::Fast", CCDelegateTo<RetCC_X86_32_Fast>>,
   // If HiPE, use RetCC_X86_32_HiPE.
   CCIfCC<"CallingConv::HiPE", CCDelegateTo<RetCC_X86_32_HiPE>>,
   CCIfCC<"CallingConv::X86_VectorCall", CCDelegateTo<RetCC_X86_32_VectorCall>>,
+  CCIfCC<"CallingConv::X86_RegCall", CCDelegateTo<RetCC_X86_32_RegCall>>,
 
   // Otherwise, use RetCC_X86_32_C.
   CCDelegateTo<RetCC_X86_32_C>
 ]>;
 
 // This is the root return-value convention for the X86-64 backend.
 def RetCC_X86_64 : CallingConv<[
   // HiPE uses RetCC_X86_64_HiPE
   CCIfCC<"CallingConv::HiPE", CCDelegateTo<RetCC_X86_64_HiPE>>,
 
   // Handle JavaScript calls.
   CCIfCC<"CallingConv::WebKit_JS", CCDelegateTo<RetCC_X86_64_WebKit_JS>>,
   CCIfCC<"CallingConv::AnyReg", CCDelegateTo<RetCC_X86_64_AnyReg>>,
 
   // Handle Swift calls.
   CCIfCC<"CallingConv::Swift", CCDelegateTo<RetCC_X86_64_Swift>>,
 
   // Handle explicit CC selection
   CCIfCC<"CallingConv::X86_64_Win64", CCDelegateTo<RetCC_X86_Win64_C>>,
   CCIfCC<"CallingConv::X86_64_SysV", CCDelegateTo<RetCC_X86_64_C>>,
 
   // Handle HHVM calls.
   CCIfCC<"CallingConv::HHVM", CCDelegateTo<RetCC_X86_64_HHVM>>,
 
+   CCIfCC<"CallingConv::X86_RegCall",
+          CCIfSubtarget<"isTargetWin64()",
+                        CCDelegateTo<RetCC_X86_Win64_RegCall>>>,
+   CCIfCC<"CallingConv::X86_RegCall",
+          CCIfSubtargetLinuxOSX<CCDelegateTo<RetCC_X86_Lin64_RegCall>>>,
+          
   // Mingw64 and native Win64 use Win64 CC
   CCIfSubtarget<"isTargetWin64()", CCDelegateTo<RetCC_X86_Win64_C>>,
 
   // Otherwise, drop to normal X86-64 CC
   CCDelegateTo<RetCC_X86_64_C>
 ]>;
 
 // This is the return-value convention used for the entire X86 backend.
def CC_X86_32 : CallingConv<[
   CCIfCC<"CallingConv::X86_INTR", CCDelegateTo<CC_X86_32_Intr>>,
   CCIfSubtarget<"isTargetMCU()", CCDelegateTo<CC_X86_32_MCU>>,
   CCIfCC<"CallingConv::X86_FastCall", CCDelegateTo<CC_X86_32_FastCall>>,
   CCIfCC<"CallingConv::X86_VectorCall", CCDelegateTo<CC_X86_32_VectorCall>>,
   CCIfCC<"CallingConv::X86_ThisCall", CCDelegateTo<CC_X86_32_ThisCall>>,
   CCIfCC<"CallingConv::Fast", CCDelegateTo<CC_X86_32_FastCC>>,
   CCIfCC<"CallingConv::GHC", CCDelegateTo<CC_X86_32_GHC>>,
   CCIfCC<"CallingConv::HiPE", CCDelegateTo<CC_X86_32_HiPE>>,
+  CCIfCC<"CallingConv::X86_RegCall", CCDelegateTo<CC_X86_32_RegCall>>,
 
   // Otherwise, drop to normal X86-32 CC
   CCDelegateTo<CC_X86_32_C>
 ]>;
 
 // This is the root argument convention for the X86-64 backend.
 def CC_X86_64 : CallingConv<[
   CCIfCC<"CallingConv::GHC", CCDelegateTo<CC_X86_64_GHC>>,
   CCIfCC<"CallingConv::HiPE", CCDelegateTo<CC_X86_64_HiPE>>,
   CCIfCC<"CallingConv::WebKit_JS", CCDelegateTo<CC_X86_64_WebKit_JS>>,
   CCIfCC<"CallingConv::AnyReg", CCDelegateTo<CC_X86_64_AnyReg>>,
   CCIfCC<"CallingConv::X86_64_Win64", CCDelegateTo<CC_X86_Win64_C>>,
   CCIfCC<"CallingConv::X86_64_SysV", CCDelegateTo<CC_X86_64_C>>,
   CCIfCC<"CallingConv::X86_VectorCall", CCDelegateTo<CC_X86_Win64_VectorCall>>,
   CCIfCC<"CallingConv::HHVM", CCDelegateTo<CC_X86_64_HHVM>>,
   CCIfCC<"CallingConv::HHVM_C", CCDelegateTo<CC_X86_64_HHVM_C>>,
+  CCIfCC<"CallingConv::X86_RegCall",
+    CCIfSubtarget<"isTargetWin64()", CCDelegateTo<CC_X86_Win64_RegCall>>>,
+  CCIfCC<"CallingConv::X86_RegCall",
+    CCIfSubtargetLinuxOSX<CCDelegateTo<CC_X86_Lin64_RegCall>>>,
   CCIfCC<"CallingConv::X86_INTR", CCDelegateTo<CC_X86_64_Intr>>,
 
   // Mingw64 and native Win64 use Win64 CC
   CCIfSubtarget<"isTargetWin64()", CCDelegateTo<CC_X86_Win64_C>>,
 
   // Otherwise, drop to normal X86-64 CC
   CCDelegateTo<CC_X86_64_C>
 ]>;
def CSR_64_Intel_OCL_BI_AVX    : CalleeSavedRegs<(add CSR_64,
                                                   (sequence "YMM%u", 8, 15))>;
 
 def CSR_64_Intel_OCL_BI_AVX512 : CalleeSavedRegs<(add RBX, RDI, RSI, R14, R15,
                                                   (sequence "ZMM%u", 16, 31),
                                                   K4, K5, K6, K7)>;
 
 // Only R12 is preserved for PHP calls in HHVM.
 def CSR_64_HHVM : CalleeSavedRegs<(add R12)>;
+
+// Register calling convention preserves few GPR and XMM8-15
+// TODO: Register that passed as arguments doesn't need to be preserved
+def CSR_32_RegCall : CalleeSavedRegs<(add ESI, EDI, EBX, EBP, ESP,
+                                      (sequence "XMM%u", 4, 7))>;
+def CSR_Win64_RegCall : CalleeSavedRegs<(add RBX, RBP, RSP,
+                                         (sequence "R%u", 10, 15),
+                                         (sequence "XMM%u", 8, 15))>;
+def CSR_Lin64_RegCall : CalleeSavedRegs<(add RBX, RBP, RSP,
+                                         (sequence "R%u", 12, 15),
+                                         (sequence "XMM%u", 8, 15))>;

lib/Target/X86/X86RegisterInfo.cpp

case CallingConv::Intel_OCL_BI: {
     if (HasAVX && Is64Bit)
       return CSR_64_Intel_OCL_BI_AVX_SaveList;
     if (!HasAVX && !IsWin64 && Is64Bit)
       return CSR_64_Intel_OCL_BI_SaveList;
     break;
   }
   case CallingConv::HHVM:
     return CSR_64_HHVM_SaveList;
+  case CallingConv::X86_RegCall:
+    return Is64Bit ? (IsWin64 ? CSR_Win64_RegCall_SaveList
+                              : CSR_Lin64_RegCall_SaveList)
+                   : CSR_32_RegCall_SaveList;
   case CallingConv::Cold:
     if (Is64Bit)
       return CSR_64_MostRegs_SaveList;
     break;
   case CallingConv::X86_64_Win64:
     return CSR_Win64_SaveList;
   case CallingConv::X86_64_SysV:
     if (CallsEHReturn)
case CallingConv::Intel_OCL_BI: {
     if (HasAVX && Is64Bit)
       return CSR_64_Intel_OCL_BI_AVX_RegMask;
     if (!HasAVX && !IsWin64 && Is64Bit)
       return CSR_64_Intel_OCL_BI_RegMask;
     break;
   }
   case CallingConv::HHVM:
     return CSR_64_HHVM_RegMask;
+  case CallingConv::X86_RegCall:
+    return Is64Bit ? (IsWin64 ? CSR_Win64_RegCall_RegMask
+                              : CSR_Lin64_RegCall_RegMask)
+                   : CSR_32_RegCall_RegMask;
   case CallingConv::Cold:
     if (Is64Bit)
       return CSR_64_MostRegs_RegMask;
     break;
   case CallingConv::X86_64_Win64:
     return CSR_Win64_RegMask;
   case CallingConv::X86_64_SysV:
     return CSR_64_RegMask;

test/CodeGen/X86/avx512-regcall-NoMask.ll

+; RUN: llc < %s -mtriple=i386-pc-win32       -mattr=+avx512f -mattr=+avx512vl -mattr=+avx512bw -mattr=+avx512dq  | FileCheck --check-prefix=X32 %s
+; RUN: llc < %s -mtriple=x86_64-win32        -mattr=+avx512f -mattr=+avx512vl -mattr=+avx512bw -mattr=+avx512dq  | FileCheck --check-prefix=WIN64 %s
+; RUN: llc < %s -mtriple=x86_64-linux-gnu    -mattr=+avx512f -mattr=+avx512vl -mattr=+avx512bw -mattr=+avx512dq  | FileCheck --check-prefix=LINUXOSX64 %s 
+
+; X32-LABEL:  test_argReti1:
+; X32:        kmov{{.*}}  %eax, %k{{[0-7]}}
+; X32:        kmov{{.*}}  %k{{[0-7]}}, %eax
+; X32:        ret{{.*}}
+
+; WIN64-LABEL:  test_argReti1:
+; WIN64:        kmov{{.*}}  %eax, %k{{[0-7]}}
+; WIN64:        kmov{{.*}}  %k{{[0-7]}}, %eax
+; WIN64:        ret{{.*}}
+
+; Test regcall when receiving/returning i1
+define x86_regcallcc i1 @test_argReti1(i1 %a)  {
+  %add = add i1 %a, 1
+  ret i1 %add
+}
+
+; X32-LABEL:  test_CallargReti1:
+; X32:        kmov{{.*}}  %k{{[0-7]}}, %eax
+; X32:        call{{.*}}   {{.*}}test_argReti1
+; X32:        kmov{{.*}}  %eax, %k{{[0-7]}}
+; X32:        ret{{.*}}
+
+; WIN64-LABEL:  test_CallargReti1:
+; WIN64:        kmov{{.*}}  %k{{[0-7]}}, %eax
+; WIN64:        call{{.*}}   {{.*}}test_argReti1
+; WIN64:        kmov{{.*}}  %eax, %k{{[0-7]}}
+; WIN64:        ret{{.*}}
+
+; Test regcall when passing/retrieving i1
+define x86_regcallcc i1 @test_CallargReti1(i1 %a)  {
+  %b = add i1 %a, 1
+  %c = call x86_regcallcc i1 @test_argReti1(i1 %b)
+  %d = add i1 %c, 1
+  ret i1 %d
+}
+
+; X32-LABEL:  test_argReti8:
+; X32:        incb  %al
+; X32:        ret{{.*}}
+
+; WIN64-LABEL:  test_argReti8:
+; WIN64:        incb %al
+; WIN64:        ret{{.*}}
+
+; Test regcall when receiving/returning i8
+define x86_regcallcc i8 @test_argReti8(i8 %a)  {
+  %add = add i8 %a, 1
+  ret i8 %add
+}
+
+; X32-LABEL:  test_CallargReti8:
+; X32:        incb %al
+; X32:        call{{.*}}   {{.*}}test_argReti8
+; X32:        incb %al
+; X32:        ret{{.*}}
+
+; WIN64-LABEL:  test_CallargReti8:
+; WIN64:        incb %al
+; WIN64:        call{{.*}}   {{.*}}test_argReti8
+; WIN64:        incb %al
+; WIN64:        ret{{.*}}
+
+; Test regcall when passing/retrieving i8
+define x86_regcallcc i8 @test_CallargReti8(i8 %a)  {
+  %b = add i8 %a, 1
+  %c = call x86_regcallcc i8 @test_argReti8(i8 %b)
+  %d = add i8 %c, 1
+  ret i8 %d
+}
+
+; X32-LABEL:  test_argReti16:
+; X32:        incl %eax
+; X32:        ret{{.*}}
+
+; WIN64-LABEL:  test_argReti16:
+; WIN64:        incl %eax
+; WIN64:        ret{{.*}}
+
+; Test regcall when receiving/returning i16
+define x86_regcallcc i16 @test_argReti16(i16 %a)  {
+  %add = add i16 %a, 1
+  ret i16 %add
+}
+
+; X32-LABEL:  test_CallargReti16:
+; X32:        incl %eax
+; X32:        call{{.*}}   {{.*}}test_argReti16
+; X32:        incl %eax
+; X32:        ret{{.*}}
+
+; WIN64-LABEL:  test_CallargReti16:
+; WIN64:        incl %eax
+; WIN64:        call{{.*}}   {{.*}}test_argReti16
+; WIN64:        incl %eax
+; WIN64:        ret{{.*}}
+
+; Test regcall when passing/retrieving i16
+define x86_regcallcc i16 @test_CallargReti16(i16 %a)  {
+  %b = add i16 %a, 1
+  %c = call x86_regcallcc i16 @test_argReti16(i16 %b)
+  %d = add i16 %c, 1
+  ret i16 %d
+}
+
+; X32-LABEL:  test_argReti32:
+; X32:        incl %eax
+; X32:        ret{{.*}}
+
+; WIN64-LABEL:  test_argReti32:
+; WIN64:        incl %eax
+; WIN64:        ret{{.*}}
+
+; Test regcall when receiving/returning i32
+define x86_regcallcc i32 @test_argReti32(i32 %a)  {
+  %add = add i32 %a, 1
+  ret i32 %add
+}
+
+; X32-LABEL:  test_CallargReti32:
+; X32:        incl %eax
+; X32:        call{{.*}}   {{.*}}test_argReti32
+; X32:        incl %eax
+; X32:        ret{{.*}}
+
+; WIN64-LABEL:  test_CallargReti32:
+; WIN64:        incl %eax
+; WIN64:        call{{.*}}   {{.*}}test_argReti32
+; WIN64:        incl %eax
+; WIN64:        ret{{.*}}
+
+; Test regcall when passing/retrieving i32
+define x86_regcallcc i32 @test_CallargReti32(i32 %a)  {
+  %b = add i32 %a, 1
+  %c = call x86_regcallcc i32 @test_argReti32(i32 %b)
+  %d = add i32 %c, 1
+  ret i32 %d
+}
+
+; X32-LABEL:  test_argReti64:
+; X32:        addl $1, %eax
+; X32:        ret{{.*}}
+
+; WIN64-LABEL:  test_argReti64:
+; WIN64:        incq %rax
+; WIN64:        ret{{.*}}
+
+; Test regcall when receiving/returning i64
+define x86_regcallcc i64 @test_argReti64(i64 %a)  {
+  %add = add i64 %a, 1
+  ret i64 %add
+}
+
+; X32-LABEL:  test_CallargReti64:
+; X32:        add{{.*}}  $1, %eax
+; X32:        adcl   $0, {{%e(cx|dx|si|di|bx|bp)}}
+; X32:        call{{.*}}   {{.*}}test_argReti64
+; X32:        add{{.*}}  $1, %eax
+; X32:        adcl   $0, {{%e(cx|dx|si|di|bx|bp)}}
+; X32:        ret{{.*}}
+
+; WIN64-LABEL:  test_CallargReti64:
+; WIN64:        incq %rax
+; WIN64:        call{{.*}}   {{.*}}test_argReti64
+; WIN64:        incq %rax
+; WIN64:        ret{{.*}}
+
+; Test regcall when passing/retrieving i64
+define x86_regcallcc i64 @test_CallargReti64(i64 %a)  {
+  %b = add i64 %a, 1
+  %c = call x86_regcallcc i64 @test_argReti64(i64 %b)
+  %d = add i64 %c, 1
+  ret i64 %d
+}
+
+; X32-LABEL:  test_argRetFloat:
+; X32:        vadd{{.*}}  {{.*}}, {{%xmm([0-7])}}
+; X32:        ret{{.*}}
+
+; WIN64-LABEL:  test_argRetFloat:
+; WIN64:        vadd{{.*}}  {{.*}}, {{%xmm([0-7])}}
+; WIN64:        ret{{.*}}
+
+; Test regcall when receiving/returning float
+define x86_regcallcc float @test_argRetFloat(float %a)  {
+  %add = fadd float 1.0, %a
+  ret float %add
+}
+
+; X32-LABEL:  test_CallargRetFloat:
+; X32:        vadd{{.*}}  {{%xmm([0-7])}}, {{%xmm([0-7])}}, {{%xmm([0-7])}}
+; X32:        call{{.*}}   {{.*}}test_argRetFloat
+; X32:        vadd{{.*}}  {{%xmm([0-7])}}, {{%xmm([0-7])}}, {{%xmm([0-7])}}
+; X32:        ret{{.*}}
+
+; WIN64-LABEL:  test_CallargRetFloat:
+; WIN64:        vadd{{.*}}  {{%xmm([0-9]+)}}, {{%xmm([0-9]+)}}, {{%xmm([0-9]+)}}
+; WIN64:        call{{.*}}   {{.*}}test_argRetFloat
+; WIN64:        vadd{{.*}}  {{%xmm([0-9]+)}}, {{%xmm([0-9]+)}}, {{%xmm([0-9]+)}}
+; WIN64:        ret{{.*}}
+
+; Test regcall when passing/retrieving float
+define x86_regcallcc float @test_CallargRetFloat(float %a)  {
+  %b = fadd float 1.0, %a
+  %c = call x86_regcallcc float @test_argRetFloat(float %b)
+  %d = fadd float 1.0, %c
+  ret float %d
+}
+
+; X32-LABEL:  test_argRetDouble:
+; X32:        vadd{{.*}}  {{.*}}, {{%xmm([0-7])}}
+; X32:        ret{{.*}}
+
+; WIN64-LABEL:  test_argRetDouble:
+; WIN64:        vadd{{.*}}  {{.*}}, {{%xmm([0-7])}}
+; WIN64:        ret{{.*}}
+
+; Test regcall when receiving/returning double
+define x86_regcallcc double @test_argRetDouble(double %a)  {
+  %add = fadd double %a, 1.0
+  ret double %add
+}
+
+; X32-LABEL:  test_CallargRetDouble:
+; X32:        vadd{{.*}}  {{%xmm([0-7])}}, {{%xmm([0-7])}}, {{%xmm([0-7])}}
+; X32:        call{{.*}}   {{.*}}test_argRetDouble
+; X32:        vadd{{.*}}  {{%xmm([0-7])}}, {{%xmm([0-7])}}, {{%xmm([0-7])}}
+; X32:        ret{{.*}}
+
+; WIN64-LABEL:  test_CallargRetDouble:
+; WIN64:        vadd{{.*}}  {{%xmm([0-9]+)}}, {{%xmm([0-9]+)}}, {{%xmm([0-9]+)}}
+; WIN64:        call{{.*}}   {{.*}}test_argRetDouble
+; WIN64:        vadd{{.*}}  {{%xmm([0-9]+)}}, {{%xmm([0-9]+)}}, {{%xmm([0-9]+)}}
+; WIN64:        ret{{.*}}
+
+; Test regcall when passing/retrieving double
+define x86_regcallcc double @test_CallargRetDouble(double %a)  {
+  %b = fadd double 1.0, %a
+  %c = call x86_regcallcc double @test_argRetDouble(double %b)
+  %d = fadd double 1.0, %c
+  ret double %d
+}
+
+; X32-LABEL:  test_argRetPointer:
+; X32:        incl %eax
+; X32:        ret{{.*}}
+
+; WIN64-LABEL:  test_argRetPointer:
+; WIN64:        incl %eax
+; WIN64:        ret{{.*}}
+
+; Test regcall when receiving/returning pointer
+define x86_regcallcc [4 x i32]* @test_argRetPointer([4 x i32]* %a)  {
+  %b = ptrtoint [4 x i32]* %a to i32
+  %c = add i32 %b, 1
+  %d = inttoptr i32 %c to [4 x i32]*
+  ret [4 x i32]* %d
+}
+
+; X32-LABEL:  test_CallargRetPointer:
+; X32:        incl %eax
+; X32:        call{{.*}}   {{.*}}test_argRetPointer
+; X32:        incl %eax
+; X32:        ret{{.*}}
+
+; WIN64-LABEL:  test_CallargRetPointer:
+; WIN64:        incl %eax
+; WIN64:        call{{.*}}   {{.*}}test_argRetPointer
+; WIN64:        incl %eax
+; WIN64:        ret{{.*}}
+
+; Test regcall when passing/retrieving pointer
+define x86_regcallcc [4 x i32]* @test_CallargRetPointer([4 x i32]* %a)  {
+  %b = ptrtoint [4 x i32]* %a to i32
+  %c = add i32 %b, 1
+  %d = inttoptr i32 %c to [4 x i32]*
+  %e = call x86_regcallcc [4 x i32]* @test_argRetPointer([4 x i32]* %d)
+  %f = ptrtoint [4 x i32]* %e to i32
+  %g = add i32 %f, 1
+  %h = inttoptr i32 %g to [4 x i32]*
+  ret [4 x i32]* %h
+}
+
+; X32-LABEL:  test_argRet128Vector:
+; X32:        vpblend{{.*}}  {{%xmm([0-7])}}, {{%xmm([0-7])}}, {{%xmm([0-7])}}
+; X32:        ret{{.*}}
+
+; WIN64-LABEL:  test_argRet128Vector:
+; WIN64:        vpblend{{.*}}  {{%xmm([0-9]+)}}, {{%xmm([0-9]+)}}, {{%xmm([0-9]+)}}
+; WIN64:        ret{{.*}}
+
+; Test regcall when receiving/returning 128 bit vector
+define x86_regcallcc <4 x i32> @test_argRet128Vector(<4 x i32> %a, <4 x i32> %b)  {
+  %d = select <4 x i1> undef , <4 x i32> %a, <4 x i32> %b
+  ret <4 x i32> %d
+}
+
+; X32-LABEL:  test_CallargRet128Vector:
+; X32:        vmov{{.*}}  {{%xmm([0-7])}}, {{%xmm([0-7])}}
+; X32:        call{{.*}}   {{.*}}test_argRet128Vector
+; X32:        vpblend{{.*}}  {{%xmm([0-7])}}, {{%xmm([0-7])}}, {{%xmm([0-7])}}
+; X32:        ret{{.*}}
+
+; WIN64-LABEL:  test_CallargRet128Vector:
+; WIN64:        vmov{{.*}}  {{%xmm([0-9]+)}}, {{%xmm([0-9]+)}}
+; WIN64:        call{{.*}}   {{.*}}test_argRet128Vector
+; WIN64:        vpblend{{.*}}  {{%xmm([0-9]+)}}, {{%xmm([0-9]+)}}, {{%xmm([0-9]+)}}
+; WIN64:        ret{{.*}}
+
+; Test regcall when passing/retrieving 128 bit vector
+define x86_regcallcc <4 x i32> @test_CallargRet128Vector(<4 x i32> %a)  {
+  %b = call x86_regcallcc <4 x i32> @test_argRet128Vector(<4 x i32> %a, <4 x i32> %a)
+  %c = select <4 x i1> undef , <4 x i32> %a, <4 x i32> %b
+  ret <4 x i32> %c
+}
+
+; X32-LABEL:  test_argRet256Vector:
+; X32:        vpblend{{.*}}  {{%ymm([0-7])}}, {{%ymm([0-7])}}, {{%ymm([0-7])}}
+; X32:        ret{{.*}}
+
+; WIN64-LABEL:  test_argRet256Vector:
+; WIN64:        vpblend{{.*}}  {{%ymm([0-9]+)}}, {{%ymm([0-9]+)}}, {{%ymm([0-9]+)}}
+; WIN64:        ret{{.*}}
+
+; Test regcall when receiving/returning 256 bit vector
+define x86_regcallcc <8 x i32> @test_argRet256Vector(<8 x i32> %a, <8 x i32> %b)  {
+  %d = select <8 x i1> undef , <8 x i32> %a, <8 x i32> %b
+  ret <8 x i32> %d
+}
+
+; X32-LABEL:  test_CallargRet256Vector:
+; X32:        vmov{{.*}}  {{%ymm([0-7])}}, {{%ymm([0-7])}}
+; X32:        call{{.*}}   {{.*}}test_argRet256Vector
+; X32:        vpblend{{.*}}  {{%ymm([0-7])}}, {{%ymm([0-7])}}, {{%ymm([0-7])}}
+; X32:        ret{{.*}}
+
+; WIN64-LABEL:  test_CallargRet256Vector:
+; WIN64:        vmov{{.*}}  {{%ymm([0-9]+)}}, {{%ymm([0-9]+)}}
+; WIN64:        call{{.*}}   {{.*}}test_argRet256Vector
+; WIN64:        vpblend{{.*}}  {{%ymm([0-9]+)}}, {{%ymm([0-9]+)}}, {{%ymm([0-9]+)}}
+; WIN64:        ret{{.*}}
+
+; Test regcall when passing/retrieving 256 bit vector
+define x86_regcallcc <8 x i32> @test_CallargRet256Vector(<8 x i32> %a)  {
+  %b = call x86_regcallcc <8 x i32> @test_argRet256Vector(<8 x i32> %a, <8 x i32> %a)
+  %c = select <8 x i1> undef , <8 x i32> %a, <8 x i32> %b
+  ret <8 x i32> %c
+}
+
+; X32-LABEL:  test_argRet512Vector:
+; X32:        vpblend{{.*}}  {{%zmm([0-7])}}, {{%zmm([0-7])}}, {{%zmm([0-7])}}
+; X32:        ret{{.*}}
+
+; WIN64-LABEL:  test_argRet512Vector:
+; WIN64:        vpblend{{.*}}  {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}
+; WIN64:        ret{{.*}}
+
+; Test regcall when receiving/returning 512 bit vector
+define x86_regcallcc <16 x i32> @test_argRet512Vector(<16 x i32> %a, <16 x i32> %b)  {
+  %d = select <16 x i1> undef , <16 x i32> %a, <16 x i32> %b
+  ret <16 x i32> %d
+}
+
+; X32-LABEL:  test_CallargRet512Vector:
+; X32:        vmov{{.*}}  {{%zmm([0-7])}}, {{%zmm([0-7])}}
+; X32:        call{{.*}}   {{.*}}test_argRet512Vector
+; X32:        vpblend{{.*}}  {{%zmm([0-7])}}, {{%zmm([0-7])}}, {{%zmm([0-7])}}
+; X32:        ret{{.*}}
+
+; WIN64-LABEL:  test_CallargRet512Vector:
+; WIN64:        vmov{{.*}}  {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}
+; WIN64:        call{{.*}}   {{.*}}test_argRet512Vector
+; WIN64:        vpblend{{.*}}  {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}
+; WIN64:        ret{{.*}}
+
+; Test regcall when passing/retrieving 512 bit vector
+define x86_regcallcc <16 x i32> @test_CallargRet512Vector(<16 x i32> %a)  {
+  %b = call x86_regcallcc <16 x i32> @test_argRet512Vector(<16 x i32> %a, <16 x i32> %a)
+  %c = select <16 x i1> undef , <16 x i32> %a, <16 x i32> %b
+  ret <16 x i32> %c
+}
+
+; WIN64: testf32_inp
+; WIN64: {{.*}} {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}
+; WIN64: {{.*}} {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}
+; WIN64: {{.*}} {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}
+; WIN64: {{.*}} {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}
+; WIN64: {{.*}} {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}
+; WIN64: {{.*}} {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}
+; WIN64: {{.*}} {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}
+; WIN64: {{.*}} {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}
+; WIN64: retq
+
+; X32: testf32_inp
+; X32: vmovups {{%xmm([0-7])}}, {{.*(%esp).*}}  {{#+}} 16-byte Spill
+; X32: vmovups {{%xmm([0-7])}}, {{.*(%esp).*}}  {{#+}} 16-byte Spill
+; X32: {{.*}} {{%zmm[0-7]}}, {{%zmm[0-7]}}, {{%zmm[0-7]}}
+; X32: {{.*}} {{%zmm[0-7]}}, {{%zmm[0-7]}}, {{%zmm[0-7]}}
+; X32: {{.*}} {{%zmm[0-7]}}, {{%zmm[0-7]}}, {{%zmm[0-7]}}
+; X32: {{.*}} {{%zmm[0-7]}}, {{%zmm[0-7]}}, {{%zmm[0-7]}}
+; X32: {{.*}} {{%zmm[0-7]}}, {{%zmm[0-7]}}, {{%zmm[0-7]}}
+; X32: {{.*}} {{%zmm[0-7]}}, {{%zmm[0-7]}}, {{%zmm[0-7]}}
+; X32: {{.*}} {{%zmm[0-7]}}, {{%zmm[0-7]}}, {{%zmm[0-7]}}
+; X32: {{.*}} {{%zmm[0-7]}}, {{%zmm[0-7]}}, {{%zmm[0-7]}}
+; X32: vmovups {{.*(%esp).*}}, {{%xmm([0-7])}}  {{#+}} 16-byte Reload
+; X32: vmovups {{.*(%esp).*}}, {{%xmm([0-7])}}  {{#+}} 16-byte Reload
+; X32: retl
+
+; LINUXOSX64: testf32_inp
+; LINUXOSX64: {{.*}} {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}
+; LINUXOSX64: {{.*}} {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}
+; LINUXOSX64: {{.*}} {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}
+; LINUXOSX64: {{.*}} {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}
+; LINUXOSX64: {{.*}} {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}
+; LINUXOSX64: {{.*}} {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}
+; LINUXOSX64: {{.*}} {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}
+; LINUXOSX64: {{.*}} {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}, {{%zmm([0-9]|1[0-1])}}
+; LINUXOSX64: retq
+
+; Test regcall when running multiple input parameters - callee saved XMMs
+define x86_regcallcc <32 x float> @testf32_inp(<32 x float> %a, <32 x float> %b, <32 x float> %c) nounwind {
+  %x1 = fadd <32 x float> %a, %b
+  %x2 = fmul <32 x float> %a, %b
+  %x3 = fsub <32 x float> %x1, %x2
+  %x4 = fadd <32 x float> %x3, %c
+  ret <32 x float> %x4
+}
+
+; X32: pushl {{%e(si|di|bx|bp)}}
+; X32: pushl {{%e(si|di|bx|bp)}}
+; X32: pushl {{%e(si|di|bx|bp)}}
+; X32: pushl {{%e(si|di|bx|bp)}}
+; X32: popl {{%e(si|di|bx|bp)}}
+; X32: popl {{%e(si|di|bx|bp)}}
+; X32: popl {{%e(si|di|bx|bp)}}
+; X32: popl {{%e(si|di|bx|bp)}}
+; X32: retl
+
+; WIN64: pushq  {{%r(bp|bx|1[0-5])}}
+; WIN64: pushq  {{%r(bp|bx|1[0-5])}}
+; WIN64: pushq  {{%r(bp|bx|1[0-5])}}
+; WIN64: pushq  {{%r(bp|bx|1[0-5])}}
+; WIN64: popq {{%r(bp|bx|1[0-5])}}
+; WIN64: popq {{%r(bp|bx|1[0-5])}}
+; WIN64: popq {{%r(bp|bx|1[0-5])}}
+; WIN64: popq {{%r(bp|bx|1[0-5])}}
+; WIN64: retq
+
+; LINUXOSX64: pushq {{%r(bp|bx|1[2-5])}}
+; LINUXOSX64: pushq {{%r(bp|bx|1[2-5])}}
+; LINUXOSX64: pushq {{%r(bp|bx|1[2-5])}}
+; LINUXOSX64: popq  {{%r(bp|bx|1[2-5])}}
+; LINUXOSX64: popq  {{%r(bp|bx|1[2-5])}}
+; LINUXOSX64: popq  {{%r(bp|bx|1[2-5])}}
+; LINUXOSX64: retq
+
+; Test regcall when running multiple input parameters - callee saved GPRs
+define x86_regcallcc i32 @testi32_inp(i32 %a1, i32 %a2, i32 %a3, i32 %a4, i32 %a5, i32 %a6,
+                                      i32 %b1, i32 %b2, i32 %b3, i32 %b4, i32 %b5, i32 %b6) nounwind {
+  %x1 = sub i32 %a1, %a2
+  %x2 = sub i32 %a3, %a4
+  %x3 = sub i32 %a5, %a6
+  %y1 = sub i32 %b1, %b2
+  %y2 = sub i32 %b3, %b4
+  %y3 = sub i32 %b5, %b6
+  %v1 = add i32 %a1, %a2
+  %v2 = add i32 %a3, %a4
+  %v3 = add i32 %a5, %a6
+  %w1 = add i32 %b1, %b2
+  %w2 = add i32 %b3, %b4
+  %w3 = add i32 %b5, %b6
+  %s1 = mul i32 %x1, %y1
+  %s2 = mul i32 %x2, %y2
+  %s3 = mul i32 %x3, %y3
+  %t1 = mul i32 %v1, %w1
+  %t2 = mul i32 %v2, %w2
+  %t3 = mul i32 %v3, %w3
+  %m1 = add i32 %s1, %s2
+  %m2 = add i32 %m1, %s3
+  %n1 = add i32 %t1, %t2
+  %n2 = add i32 %n1, %t3
+  %r1 = add i32 %m2, %n2
+  ret i32 %r1
+}
+
+; X32: testf32_stack
+; X32: vaddps {{%zmm([0-7])}}, {{%zmm([0-7])}}, {{%zmm([0-7])}}
+; X32: vaddps {{%zmm([0-7])}}, {{%zmm([0-7])}}, {{%zmm([0-7])}}
+; X32: vaddps {{%zmm([0-7])}}, {{%zmm([0-7])}}, {{%zmm([0-7])}}
+; X32: vaddps {{%zmm([0-7])}}, {{%zmm([0-7])}}, {{%zmm([0-7])}}
+; X32: vaddps {{%zmm([0-7])}}, {{%zmm([0-7])}}, {{%zmm([0-7])}}
+; X32: vaddps {{%zmm([0-7])}}, {{%zmm([0-7])}}, {{%zmm([0-7])}}
+; X32: vaddps {{([0-9])+}}(%ebp), {{%zmm([0-7])}}, {{%zmm([0-7])}}
+; X32: vaddps {{([0-9])+}}(%ebp), {{%zmm([0-7])}}, {{%zmm([0-7])}}
+; X32: vaddps {{([0-9])+}}(%ebp), {{%zmm([0-7])}}, {{%zmm([0-7])}}
+; X32: vaddps {{([0-9])+}}(%ebp), {{%zmm([0-7])}}, {{%zmm([0-7])}}
+; X32: vaddps {{([0-9])+}}(%ebp), {{%zmm([0-7])}}, {{%zmm([0-7])}}
+; X32: vaddps {{([0-9])+}}(%ebp), {{%zmm([0-7])}}, {{%zmm([0-7])}}
+; X32: vaddps {{([0-9])+}}(%ebp), {{%zmm([0-7])}}, {{%zmm([0-7])}}
+; X32: vaddps {{([0-9])+}}(%ebp), {{%zmm([0-7])}}, {{%zmm([0-7])}}
+; X32: vaddps {{([0-9])+}}(%ebp), {{%zmm([0-7])}}, {{%zmm([0-1])}}
+; X32: vaddps {{([0-9])+}}(%ebp), {{%zmm([0-7])}}, {{%zmm([0-1])}}
+; X32: retl
+
+; LINUXOSX64: testf32_stack
+; LINUXOSX64: vaddps {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}
+; LINUXOSX64: vaddps {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}
+; LINUXOSX64: vaddps {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}
+; LINUXOSX64: vaddps {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}
+; LINUXOSX64: vaddps {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}
+; LINUXOSX64: vaddps {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}
+; LINUXOSX64: vaddps {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}
+; LINUXOSX64: vaddps {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}
+; LINUXOSX64: vaddps {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}
+; LINUXOSX64: vaddps {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}
+; LINUXOSX64: vaddps {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}
+; LINUXOSX64: vaddps {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}
+; LINUXOSX64: vaddps {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}
+; LINUXOSX64: vaddps {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}, {{%zmm([0-9]+)}}
+; LINUXOSX64: vaddps {{([0-9])+}}(%rbp), {{%zmm([0-9]+)}}, {{%zmm([0-1])}}
+; LINUXOSX64: vaddps {{([0-9])+}}(%rbp), {{%zmm([0-9]+)}}, {{%zmm([0-1])}}
+; LINUXOSX64: retq
+
+; Test that parameters, overflowing register capacity, are passed through the stack
+define x86_regcallcc <32 x float> @testf32_stack(<32 x float> %a0, <32 x float> %b0, <32 x float> %c0, 
+                                               <32 x float> %a1, <32 x float> %b1, <32 x float> %c1,
+                                               <32 x float> %a2, <32 x float> %b2, <32 x float> %c2) nounwind {
+  %x1 = fadd <32 x float> %a0, %b0
+  %x2 = fadd <32 x float> %c0, %x1
+  %x3 = fadd <32 x float> %a1, %x2
+  %x4 = fadd <32 x float> %b1, %x3
+  %x5 = fadd <32 x float> %c1, %x4
+  %x6 = fadd <32 x float> %a2, %x5
+  %x7 = fadd <32 x float> %b2, %x6
+  %x8 = fadd <32 x float> %c2, %x7
+  ret <32 x float> %x8
+}

test/CodeGen/X86/sse-regcall.ll

+; RUN: llc < %s -mtriple=i386-pc-win32 -mattr=+sse | FileCheck --check-prefix=WIN32 %s
+; RUN: llc < %s -mtriple=x86_64-win32 -mattr=+sse | FileCheck --check-prefix=WIN64 %s
+; RUN: llc < %s -mtriple=x86_64-linux-gnu -mattr=+sse | FileCheck --check-prefix=LINUXOSX %s
+
+; WIN32-LABEL:  test_argReti1:
+; WIN32:        incb  %al
+; WIN32:        ret{{.*}}
+
+; WIN64-LABEL:  test_argReti1:
+; WIN64:        incb  %al
+; WIN64:        ret{{.*}}
+
+; Test regcall when receiving/returning i1
+define x86_regcallcc i1 @test_argReti1(i1 %a)  {
+  %add = add i1 %a, 1
+  ret i1 %add
+}
+
+; WIN32-LABEL:  test_CallargReti1:
+; WIN32:        movzbl  %al, %eax
+; WIN32:        call{{.*}}   {{.*}}test_argReti1
+; WIN32:        incb  %al
+; WIN32:        ret{{.*}}
+
+; WIN64-LABEL:  test_CallargReti1:
+; WIN64:        movzbl  %al, %eax
+; WIN64:        call{{.*}}   {{.*}}test_argReti1
+; WIN64:        incb  %al
+; WIN64:        ret{{.*}}
+
+; Test regcall when passing/retrieving i1
+define x86_regcallcc i1 @test_CallargReti1(i1 %a)  {
+  %b = add i1 %a, 1
+  %c = call x86_regcallcc i1 @test_argReti1(i1 %b)
+  %d = add i1 %c, 1
+  ret i1 %d
+}
+
+; WIN64-LABEL: testf32_inp
+; WIN64: movaps {{%xmm(1[2-5])}}, {{.*(%rsp).*}}  {{#+}} 16-byte Spill
+; WIN64: movaps {{%xmm(1[2-5])}}, {{.*(%rsp).*}}  {{#+}} 16-byte Spill
+; WIN64: movaps {{%xmm(1[2-5])}}, {{.*(%rsp).*}}  {{#+}} 16-byte Spill
+; WIN64: movaps {{%xmm(1[2-5])}}, {{.*(%rsp).*}}  {{#+}} 16-byte Spill
+; WIN64: {{.*}} {{%xmm([0-9]|1[0-1])}}, {{%xmm(1[2-5])}}
+; WIN64: {{.*}} {{%xmm([0-9]|1[0-1])}}, {{%xmm(1[2-5])}}
+; WIN64: {{.*}} {{%xmm([0-9]|1[0-1])}}, {{%xmm(1[2-5])}}
+; WIN64: {{.*}} {{%xmm([0-9]|1[0-1])}}, {{%xmm(1[2-5])}}
+; WIN64: movaps {{.*(%rsp).*}}, {{%xmm(1[2-5])}}  {{#+}} 16-byte Reload
+; WIN64: movaps {{.*(%rsp).*}}, {{%xmm(1[2-5])}}  {{#+}} 16-byte Reload
+; WIN64: movaps {{.*(%rsp).*}}, {{%xmm(1[2-5])}}  {{#+}} 16-byte Reload
+; WIN64: movaps {{.*(%rsp).*}}, {{%xmm(1[2-5])}}  {{#+}} 16-byte Reload
+; WIN64: retq
+
+; WIN32-LABEL: testf32_inp
+; WIN32: movaps {{%xmm([4-7])}}, {{.*(%ebp).*}}  {{#+}} 16-byte Spill
+; WIN32: movaps {{%xmm([4-7])}}, {{.*(%ebp).*}}  {{#+}} 16-byte Spill
+; WIN32: movaps {{%xmm([4-7])}}, {{.*(%ebp).*}}  {{#+}} 16-byte Spill
+; WIN32: movaps {{%xmm([4-7])}}, {{.*(%ebp).*}}  {{#+}} 16-byte Spill
+; WIN32: {{.*}} {{%xmm[0-7]}}, {{%xmm[4-7]}}
+; WIN32: {{.*}} {{%xmm[0-7]}}, {{%xmm[4-7]}}
+; WIN32: {{.*}} {{%xmm[0-7]}}, {{%xmm[4-7]}}
+; WIN32: {{.*}} {{%xmm[0-7]}}, {{%xmm[4-7]}}
+; WIN32: movaps {{.*(%ebp).*}}, {{%xmm([4-7])}}  {{#+}} 16-byte Reload
+; WIN32: movaps {{.*(%ebp).*}}, {{%xmm([4-7])}}  {{#+}} 16-byte Reload
+; WIN32: movaps {{.*(%ebp).*}}, {{%xmm([4-7])}}  {{#+}} 16-byte Reload
+; WIN32: movaps {{.*(%ebp).*}}, {{%xmm([4-7])}}  {{#+}} 16-byte Reload
+; WIN32: retl
+
+; LINUXOSX-LABEL: testf32_inp
+; LINUXOSX: movaps {{%xmm(1[2-5])}}, {{.*(%rsp).*}}  {{#+}} 16-byte Spill
+; LINUXOSX: movaps {{%xmm(1[2-5])}}, {{.*(%rsp).*}}  {{#+}} 16-byte Spill
+; LINUXOSX: movaps {{%xmm(1[2-5])}}, {{.*(%rsp).*}}  {{#+}} 16-byte Spill
+; LINUXOSX: movaps {{%xmm(1[2-5])}}, {{.*(%rsp).*}}  {{#+}} 16-byte Spill
+; LINUXOSX: {{.*}} {{%xmm([0-9]|1[0-1])}}, {{%xmm(1[2-5])}}
+; LINUXOSX: {{.*}} {{%xmm([0-9]|1[0-1])}}, {{%xmm(1[2-5])}}
+; LINUXOSX: {{.*}} {{%xmm([0-9]|1[0-1])}}, {{%xmm(1[2-5])}}
+; LINUXOSX: {{.*}} {{%xmm([0-9]|1[0-1])}}, {{%xmm(1[2-5])}}
+; LINUXOSX: movaps {{.*(%rsp).*}}, {{%xmm(1[2-5])}}  {{#+}} 16-byte Reload
+; LINUXOSX: movaps {{.*(%rsp).*}}, {{%xmm(1[2-5])}}  {{#+}} 16-byte Reload
+; LINUXOSX: movaps {{.*(%rsp).*}}, {{%xmm(1[2-5])}}  {{#+}} 16-byte Reload
+; LINUXOSX: movaps {{.*(%rsp).*}}, {{%xmm(1[2-5])}}  {{#+}} 16-byte Reload
+; LINUXOSX: retq
+
+;test calling conventions - input parameters, callee saved XMMs
+define x86_regcallcc <16 x float> @testf32_inp(<16 x float> %a, <16 x float> %b, <16 x float> %c) nounwind {
+  %x1 = fadd <16 x float> %a, %b
+  %x2 = fmul <16 x float> %a, %b
+  %x3 = fsub <16 x float> %x1, %x2
+  %x4 = fadd <16 x float> %x3, %c
+  ret <16 x float> %x4
+}
+
+; WIN32-LABEL: testi32_inp
+; WIN32: pushl {{%e(si|di|bx|bp)}}
+; WIN32: pushl {{%e(si|di|bx|bp)}}
+; WIN32: pushl {{%e(si|di|bx|bp)}}
+; WIN32: pushl {{%e(si|di|bx|bp)}}
+; WIN32: popl {{%e(si|di|bx|bp)}}
+; WIN32: popl {{%e(si|di|bx|bp)}}
+; WIN32: popl {{%e(si|di|bx|bp)}}
+; WIN32: popl {{%e(si|di|bx|bp)}}
+; WIN32: retl
+
+; WIN64-LABEL: testi32_inp
+; WIN64: pushq  {{%r(bp|bx|1[0-5])}}
+; WIN64: pushq  {{%r(bp|bx|1[0-5])}}
+; WIN64: pushq  {{%r(bp|bx|1[0-5])}}
+; WIN64: pushq  {{%r(bp|bx|1[0-5])}}
+; WIN64: pushq  {{%r(bp|bx|1[0-5])}}
+; WIN64: popq {{%r(bp|bx|1[0-5])}}
+; WIN64: popq {{%r(bp|bx|1[0-5])}}
+; WIN64: popq {{%r(bp|bx|1[0-5])}}
+; WIN64: popq {{%r(bp|bx|1[0-5])}}
+; WIN64: popq {{%r(bp|bx|1[0-5])}}
+; WIN64: retq
+
+; LINUXOSX-LABEL: testi32_inp
+; LINUXOSX: pushq {{%r(bp|bx|1[2-5])}}
+; LINUXOSX: pushq {{%r(bp|bx|1[2-5])}}
+; LINUXOSX: pushq {{%r(bp|bx|1[2-5])}}
+; LINUXOSX: pushq {{%r(bp|bx|1[2-5])}}
+; LINUXOSX: popq  {{%r(bp|bx|1[2-5])}}
+; LINUXOSX: popq  {{%r(bp|bx|1[2-5])}}
+; LINUXOSX: popq  {{%r(bp|bx|1[2-5])}}
+; LINUXOSX: popq  {{%r(bp|bx|1[2-5])}}
+; LINUXOSX: retq
+
+;test calling conventions - input parameters, callee saved GPRs
+define x86_regcallcc i32 @testi32_inp(i32 %a1, i32 %a2, i32 %a3, i32 %a4, i32 %a5, i32 %a6,
+                                      i32 %b1, i32 %b2, i32 %b3, i32 %b4, i32 %b5, i32 %b6) nounwind {
+  %x1 = sub i32 %a1, %a2
+  %x2 = sub i32 %a3, %a4
+  %x3 = sub i32 %a5, %a6
+  %y1 = sub i32 %b1, %b2
+  %y2 = sub i32 %b3, %b4
+  %y3 = sub i32 %b5, %b6
+  %v1 = add i32 %a1, %a2
+  %v2 = add i32 %a3, %a4
+  %v3 = add i32 %a5, %a6
+  %w1 = add i32 %b1, %b2
+  %w2 = add i32 %b3, %b4
+  %w3 = add i32 %b5, %b6
+  %s1 = mul i32 %x1, %y1
+  %s2 = mul i32 %x2, %y2
+  %s3 = mul i32 %x3, %y3
+  %t1 = mul i32 %v1, %w1
+  %t2 = mul i32 %v2, %w2
+  %t3 = mul i32 %v3, %w3
+  %m1 = add i32 %s1, %s2
+  %m2 = add i32 %m1, %s3
+  %n1 = add i32 %t1, %t2
+  %n2 = add i32 %n1, %t3
+  %r1 = add i32 %m2, %n2
+  ret i32 %r1
+}
+
+; X32: testf32_stack
+; X32: movaps {{%xmm([0-7])}}, {{(-*[0-9])+}}(%ebp)
+; X32: movaps {{%xmm([0-7])}}, {{(-*[0-9])+}}(%ebp)
+; X32: movaps {{%xmm([0-7])}}, {{(-*[0-9])+}}(%ebp)
+; X32: movaps {{%xmm([0-7])}}, {{(-*[0-9])+}}(%ebp)
+; X32: addps {{([0-9])+}}(%ebp), {{%xmm([0-7])}}
+; X32: addps {{([0-9])+}}(%ebp), {{%xmm([0-7])}}
+; X32: addps {{([0-9])+}}(%ebp), {{%xmm([0-7])}}
+; X32: addps {{([0-9])+}}(%ebp), {{%xmm([0-7])}}
+; X32: addps {{([0-9])+}}(%ebp), {{%xmm([0-7])}}
+; X32: addps {{([0-9])+}}(%ebp), {{%xmm([0-7])}}
+; X32: addps {{([0-9])+}}(%ebp), {{%xmm([0-7])}}
+; X32: addps {{([0-9])+}}(%ebp), {{%xmm([0-7])}}
+; X32: addps {{([0-9])+}}(%ebp), {{%xmm([0-7])}}
+; X32: addps {{([0-9])+}}(%ebp), {{%xmm([0-7])}}
+; X32: addps {{([0-9])+}}(%ebp), {{%xmm([0-7])}}
+; X32: addps {{([0-9])+}}(%ebp), {{%xmm([0-7])}}
+; X32: addps {{([0-9])+}}(%ebp), {{%xmm([0-7])}}
+; X32: addps {{([0-9])+}}(%ebp), {{%xmm([0-7])}}
+; X32: addps {{([0-9])+}}(%ebp), {{%xmm([0-7])}}
+; X32: addps {{([0-9])+}}(%ebp), {{%xmm([0-7])}}
+; X32: movaps {{(-*[0-9])+}}(%ebp), {{%xmm([0-7])}}
+; X32: movaps {{(-*[0-9])+}}(%ebp), {{%xmm([0-7])}}
+; X32: movaps {{(-*[0-9])+}}(%ebp), {{%xmm([0-7])}}
+; X32: movaps {{(-*[0-9])+}}(%ebp), {{%xmm([0-7])}}
+
+; LINUXOSX: testf32_stack
+; LINUXOSX: addps {{%xmm([0-9]+)}}, {{%xmm([0-9]+)}}
+; LINUXOSX: addps {{%xmm([0-9]+)}}, {{%xmm([0-9]+)}}
+; LINUXOSX: addps {{%xmm([0-9]+)}}, {{%xmm([0-9]+)}}
+; LINUXOSX: addps {{%xmm([0-9]+)}}, {{%xmm([0-9]+)}}
+; LINUXOSX: addps {{%xmm([0-9]+)}}, {{%xmm([0-9]+)}}
+; LINUXOSX: addps {{%xmm([0-9]+)}}, {{%xmm([0-9]+)}}
+; LINUXOSX: addps {{%xmm([0-9]+)}}, {{%xmm([0-9]+)}}
+; LINUXOSX: addps {{%xmm([0-9]+)}}, {{%xmm([0-9]+)}}
+; LINUXOSX: addps {{([0-9])+}}(%rsp), {{%xmm([0-7])}}
+; LINUXOSX: addps {{([0-9])+}}(%rsp), {{%xmm([0-7])}}
+; LINUXOSX: addps {{([0-9])+}}(%rsp), {{%xmm([0-7])}}
+; LINUXOSX: addps {{([0-9])+}}(%rsp), {{%xmm([0-7])}}
+; LINUXOSX: addps {{([0-9])+}}(%rsp), {{%xmm([0-7])}}
+; LINUXOSX: addps {{([0-9])+}}(%rsp), {{%xmm([0-7])}}
+; LINUXOSX: addps {{([0-9])+}}(%rsp), {{%xmm([0-7])}}
+; LINUXOSX: addps {{([0-9])+}}(%rsp), {{%xmm([0-7])}}
+; LINUXOSX: retq
+
+; Test that parameters, overflowing register capacity, are passed through the stack
+define x86_regcallcc <32 x float> @testf32_stack(<32 x float> %a, <32 x float> %b, <32 x float> %c) nounwind {
+  %x1 = fadd <32 x float> %a, %b
+  %x2 = fadd <32 x float> %x1, %c
+  ret <32 x float> %x2
+}
 No newline at end of file