mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-10-30 07:22:55 +01:00
e5314eb2c2
XMM registers. There are many known deficiencies and fixmes, which will be addressed ASAP. The major benefit of this work is that it will allow the LLVM register allocator to allocate FP registers across basic blocks. The x86 backend will still default to x87 style FP. To enable this work, you must pass -enable-sse-scalar-fp and either -sse2 or -sse3 to llc. An example before and after would be for: double foo(double *P) { double Sum = 0; int i; for (i = 0; i < 1000; ++i) Sum += P[i]; return Sum; } The inner loop looks like the following: x87: .LBB_foo_1: # no_exit fldl (%esp) faddl (%eax,%ecx,8) fstpl (%esp) incl %ecx cmpl $1000, %ecx #FP_REG_KILL jne .LBB_foo_1 # no_exit SSE2: addsd (%eax,%ecx,8), %xmm0 incl %ecx cmpl $1000, %ecx #FP_REG_KILL jne .LBB_foo_1 # no_exit llvm-svn: 22340
123 lines
5.3 KiB
C++
123 lines
5.3 KiB
C++
//===- X86RegisterInfo.td - Describe the X86 Register File ------*- C++ -*-===//
|
|
//
|
|
// The LLVM Compiler Infrastructure
|
|
//
|
|
// This file was developed by the LLVM research group and is distributed under
|
|
// the University of Illinois Open Source License. See LICENSE.TXT for details.
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
//
|
|
// This file describes the X86 Register file, defining the registers themselves,
|
|
// aliases between the registers, and the register classes built out of the
|
|
// registers.
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// Register definitions...
|
|
//
|
|
let Namespace = "X86" in {
|
|
|
|
// In the register alias definitions below, we define which registers alias
|
|
// which others. We only specify which registers the small registers alias,
|
|
// because the register file generator is smart enough to figure out that
|
|
// AL aliases AX if we tell it that AX aliased AL (for example).
|
|
|
|
// 32-bit registers
|
|
def EAX : Register<"EAX">; def ECX : Register<"ECX">;
|
|
def EDX : Register<"EDX">; def EBX : Register<"EBX">;
|
|
def ESP : Register<"ESP">; def EBP : Register<"EBP">;
|
|
def ESI : Register<"ESI">; def EDI : Register<"EDI">;
|
|
|
|
// 16-bit registers
|
|
def AX : RegisterGroup<"AX", [EAX]>; def CX : RegisterGroup<"CX", [ECX]>;
|
|
def DX : RegisterGroup<"DX", [EDX]>; def BX : RegisterGroup<"BX", [EBX]>;
|
|
def SP : RegisterGroup<"SP", [ESP]>; def BP : RegisterGroup<"BP", [EBP]>;
|
|
def SI : RegisterGroup<"SI", [ESI]>; def DI : RegisterGroup<"DI", [EDI]>;
|
|
|
|
// 8-bit registers
|
|
def AL : RegisterGroup<"AL", [AX,EAX]>; def CL : RegisterGroup<"CL",[CX,ECX]>;
|
|
def DL : RegisterGroup<"DL", [DX,EDX]>; def BL : RegisterGroup<"BL",[BX,EBX]>;
|
|
def AH : RegisterGroup<"AH", [AX,EAX]>; def CH : RegisterGroup<"CH",[CX,ECX]>;
|
|
def DH : RegisterGroup<"DH", [DX,EDX]>; def BH : RegisterGroup<"BH",[BX,EBX]>;
|
|
|
|
// Pseudo Floating Point registers
|
|
def FP0 : Register<"FP0">; def FP1 : Register<"FP1">;
|
|
def FP2 : Register<"FP2">; def FP3 : Register<"FP3">;
|
|
def FP4 : Register<"FP4">; def FP5 : Register<"FP5">;
|
|
def FP6 : Register<"FP6">;
|
|
|
|
// XMM Registers, used by the various SSE instruction set extensions
|
|
def XMM0: Register<"XMM0">; def XMM1: Register<"XMM1">;
|
|
def XMM2: Register<"XMM2">; def XMM3: Register<"XMM3">;
|
|
def XMM4: Register<"XMM4">; def XMM5: Register<"XMM5">;
|
|
def XMM6: Register<"XMM6">; def XMM7: Register<"XMM7">;
|
|
|
|
// Floating point stack registers
|
|
def ST0 : Register<"ST(0)">; def ST1 : Register<"ST(1)">;
|
|
def ST2 : Register<"ST(2)">; def ST3 : Register<"ST(3)">;
|
|
def ST4 : Register<"ST(4)">; def ST5 : Register<"ST(5)">;
|
|
def ST6 : Register<"ST(6)">; def ST7 : Register<"ST(7)">;
|
|
|
|
// Flags, Segment registers, etc...
|
|
}
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// Register Class Definitions... now that we have all of the pieces, define the
|
|
// top-level register classes. The order specified in the register list is
|
|
// implicitly defined to be the register allocation order.
|
|
//
|
|
|
|
// List AL,CL,DL before AH,CH,DH, as X86 processors often suffer from false
|
|
// dependences between upper and lower parts of the register. BL and BH are
|
|
// last because they are call clobbered. Both Athlon and P4 chips suffer this
|
|
// issue.
|
|
def R8 : RegisterClass<i8, 8, [AL, CL, DL, AH, CH, DH, BL, BH]>;
|
|
|
|
def R16 : RegisterClass<i16, 16, [AX, CX, DX, SI, DI, BX, BP, SP]> {
|
|
let Methods = [{
|
|
iterator allocation_order_end(MachineFunction &MF) const {
|
|
if (hasFP(MF)) // Does the function dedicate EBP to being a frame ptr?
|
|
return end()-2; // If so, don't allocate SP or BP
|
|
else
|
|
return end()-1; // If not, just don't allocate SP
|
|
}
|
|
}];
|
|
}
|
|
|
|
def R32 : RegisterClass<i32, 32, [EAX, ECX, EDX, ESI, EDI, EBX, EBP, ESP]> {
|
|
let Methods = [{
|
|
iterator allocation_order_end(MachineFunction &MF) const {
|
|
if (hasFP(MF)) // Does the function dedicate EBP to being a frame ptr?
|
|
return end()-2; // If so, don't allocate ESP or EBP
|
|
else
|
|
return end()-1; // If not, just don't allocate ESP
|
|
}
|
|
}];
|
|
}
|
|
|
|
// FIXME: These registers can contain both integer and fp values. We should
|
|
// figure out the right way to deal with that. For now, since they'll be used
|
|
// for scalar FP, they are being declared f64
|
|
def RXMM : RegisterClass<f64, 32, [XMM0, XMM1, XMM2, XMM3,
|
|
XMM4, XMM5, XMM6, XMM7]>;
|
|
|
|
// FIXME: This sets up the floating point register files as though they are f64
|
|
// values, though they really are f80 values. This will cause us to spill
|
|
// values as 64-bit quantities instead of 80-bit quantities, which is much much
|
|
// faster on common hardware. In reality, this should be controlled by a
|
|
// command line option or something.
|
|
|
|
def RFP : RegisterClass<f64, 32, [FP0, FP1, FP2, FP3, FP4, FP5, FP6]>;
|
|
|
|
// Floating point stack registers (these are not allocatable by the
|
|
// register allocator - the floating point stackifier is responsible
|
|
// for transforming FPn allocations to STn registers)
|
|
def RST : RegisterClass<f64, 32, [ST0, ST1, ST2, ST3, ST4, ST5, ST6, ST7]> {
|
|
let Methods = [{
|
|
iterator allocation_order_end(MachineFunction &MF) const {
|
|
return begin();
|
|
}
|
|
}];
|
|
}
|