[iovisor-dev] [PATCH RFC 3/4] New 32-bit register set

Jiong Wang jiong.wang at netronome.com
Tue Sep 19 23:10:40 UTC 2017


On 19/09/2017 07:44, Y Song wrote:
> Hi, Jiong,
>
> Thanks for the patch! It is a great start to support 32bit register in BPF.
> In the past, I have studied a little bit to see whether 32bit register
> support may reduce
> the number of unnecessary shifts on x86_64 and improve the
> performance. Looking through
> a few bpf programs and it looks like the opportunity is not great, but
> still nice to have if we
> have this capability. As you mentioned, this definitely helped 32bit
> architecture!
>
> I am not an expert in LLVM tablegen. I briefly looked through the code
> change and it looks like
> correct. Hopefully some llvm-dev tablegen experts can comment on the
> implementation.
>
> Below I only have a couple of minor comments.

Yong Hong,

   Thanks for the review.

   I have addressed your comments and attached the updated patch.

   Do you want me to put this patch set on to llvm review website? I 
guess it is the
   formal review channel?

Regards,
Jiong

-------------- next part --------------
Acked-by: Jakub Kicinski <jakub.kicinski at netronome.com>
Signed-off-by: Jiong Wang <jiong.wang at netronome.com>

diff --git a/lib/Target/BPF/BPFRegisterInfo.td b/lib/Target/BPF/BPFRegisterInfo.td
index c8e24f8..da1d6b5 100644
--- a/lib/Target/BPF/BPFRegisterInfo.td
+++ b/lib/Target/BPF/BPFRegisterInfo.td
@@ -11,31 +11,42 @@
 //  Declarations that describe the BPF register file
 //===----------------------------------------------------------------------===//
 
+let Namespace = "BPF" in {
+  def sub_32 : SubRegIndex<32>;
+}
+
+class Wi<bits<16> Enc, string n> : Register<n> {
+  let HWEncoding = Enc;
+  let Namespace = "BPF";
+}
+
 // Registers are identified with 4-bit ID numbers.
 // Ri - 64-bit integer registers
-class Ri<bits<16> Enc, string n> : Register<n> {
-  let Namespace = "BPF";
+class Ri<bits<16> Enc, string n, list<Register> subregs>
+  : RegisterWithSubRegs<n, subregs> {
   let HWEncoding = Enc;
+  let Namespace = "BPF";
+  let SubRegIndices = [sub_32];
 }
 
-// Integer registers
-def R0 : Ri< 0, "r0">, DwarfRegNum<[0]>;
-def R1 : Ri< 1, "r1">, DwarfRegNum<[1]>;
-def R2 : Ri< 2, "r2">, DwarfRegNum<[2]>;
-def R3 : Ri< 3, "r3">, DwarfRegNum<[3]>;
-def R4 : Ri< 4, "r4">, DwarfRegNum<[4]>;
-def R5 : Ri< 5, "r5">, DwarfRegNum<[5]>;
-def R6 : Ri< 6, "r6">, DwarfRegNum<[6]>;
-def R7 : Ri< 7, "r7">, DwarfRegNum<[7]>;
-def R8 : Ri< 8, "r8">, DwarfRegNum<[8]>;
-def R9 : Ri< 9, "r9">, DwarfRegNum<[9]>;
-def R10 : Ri<10, "r10">, DwarfRegNum<[10]>;
-def R11 : Ri<11, "r11">, DwarfRegNum<[11]>;
+foreach I = 0-11 in {
+  // 32-bit Integer (alias to low part of 64-bit register).
+  def W#I  : Wi<I,  "w"#I>,  DwarfRegNum<[I]>;
+  // 64-bit Integer registers
+  def R#I  : Ri<I,  "r"#I,  [!cast<Wi>("W"#I)]>,  DwarfRegNum<[I]>;
+}
 
 // Register classes.
-def GPR : RegisterClass<"BPF", [i64], 64, (add R1, R2, R3, R4, R5,
-                                           R6, R7, R8, R9, // callee saved
-                                           R0,  // return value
-                                           R11, // stack ptr
-                                           R10  // frame ptr
-                                          )>;
+def GPR32 : RegisterClass<"BPF", [i32], 32, (add
+  (sequence "W%u", 1, 9),
+  W0, // Return value
+  W11, // Stack Ptr
+  W10  // Frame Ptr
+)>;
+
+def GPR : RegisterClass<"BPF", [i64], 64, (add
+  (sequence "R%u", 1, 9),
+  R0, // Return value
+  R11, // Stack Ptr
+  R10  // Frame Ptr
+)>;


More information about the iovisor-dev mailing list