[PATCH 1/2] Triple: Add gcn triple

Tom Stellard tom at stellard.net
Mon Dec 29 07:35:14 PST 2014


On Thu, Dec 25, 2014 at 05:47:35AM +0000, Sahasrabuddhe, Sameer wrote:
> Matthew Arsenault wrote:
> 
> > I was thinking keep r600 for r600, and amdgpu for SI+, mostly because I
> > hate the gcn name. What do you think about 32/64 in case somebody
> > decides to add 32-bit support? gcn and then gcn32 for that, or gcn32/gcn64?
> 
> "gcn" indicates a particular architecture while "amdgpu" indicates a generic manufacturer. What would we choose for the next big architecture, if the current is called "amdgpu"? It's easier to just let the manufacturer come up with a suitable indicator and use that, like "gcn". The next already has a name: "gcn2", if it ever needed its own triple. In that sense, suffixes like "gcn32" and "gcn64" might cause more confusion somewhere down the path. For all we know, some future architecture might get called "amd-one" or "gcn-360", or whatever phrase catches the fancy of the market at that point in time.
> 

We discussed this on IRC and decided to go with amdgcn as the triple name.

-Tom

> Sameer.
> 
> ________________________________________
> From: llvm-commits-bounces at cs.uiuc.edu [llvm-commits-bounces at cs.uiuc.edu] on behalf of Arsenault, Matthew
> Sent: Friday, December 19, 2014 10:06 PM
> To: Tom Stellard
> Cc: llvm-commits at cs.uiuc.edu; Stellard, Thomas
> Subject: Re: [PATCH 1/2] Triple: Add gcn triple
> 
> On 12/19/2014 11:28 AM, Tom Stellard wrote:
> > On Fri, Dec 19, 2014 at 11:10:05AM -0500, Matt Arsenault wrote:
> >> On 12/19/2014 10:43 AM, Tom Stellard wrote:
> >>> On Thu, Dec 04, 2014 at 10:40:23AM -0500, Tom Stellard wrote:
> >>>> This will be used for AMD GPUs with the Graphics Core Next architecture.
> >>> Ping.
> >> I thought you were going to change this to amdgpu?
> >>
> > I think it will make it easier to separate the functionality if there are different
> > triples for each architecture.  I still plan to rename the directory to AMDGPU.
> >
> > -Tom
> 
> I was thinking keep r600 for r600, and amdgpu for SI+, mostly because I
> hate the gcn name. What do you think about 32/64 in case somebody
> decides to add 32-bit support? gcn and then gcn32 for that, or gcn32/gcn64?
> >
> >>>> ---
> >>>>   include/llvm/ADT/Triple.h | 1 +
> >>>>   lib/Support/Triple.cpp    | 7 +++++++
> >>>>   2 files changed, 8 insertions(+)
> >>>>
> >>>> diff --git a/include/llvm/ADT/Triple.h b/include/llvm/ADT/Triple.h
> >>>> index 0932040..643f29e 100644
> >>>> --- a/include/llvm/ADT/Triple.h
> >>>> +++ b/include/llvm/ADT/Triple.h
> >>>> @@ -50,6 +50,7 @@ public:
> >>>>       armeb,      // ARM (big endian): armeb
> >>>>       aarch64,    // AArch64 (little endian): aarch64
> >>>>       aarch64_be, // AArch64 (big endian): aarch64_be
> >>>> +    gcn,        // GCN: AMD GCN GPUs
> >> This should probably be next to r600 in the enum order and in all the cases
> >>>>       hexagon,    // Hexagon: hexagon
> >>>>       mips,       // MIPS: mips, mipsallegrex
> >>>>       mipsel,     // MIPSEL: mipsel, mipsallegrexel
> >>>> diff --git a/lib/Support/Triple.cpp b/lib/Support/Triple.cpp
> >>>> index 64cd6e7..6570fe6 100644
> >>>> --- a/lib/Support/Triple.cpp
> >>>> +++ b/lib/Support/Triple.cpp
> >>>> @@ -24,6 +24,7 @@ const char *Triple::getArchTypeName(ArchType Kind) {
> >>>>     case arm:         return "arm";
> >>>>     case armeb:       return "armeb";
> >>>>     case hexagon:     return "hexagon";
> >>>> +  case gcn:         return "gcn";
> >>>>     case mips:        return "mips";
> >>>>     case mipsel:      return "mipsel";
> >>>>     case mips64:      return "mips64";
> >>>> @@ -82,6 +83,7 @@ const char *Triple::getArchTypePrefix(ArchType Kind) {
> >>>>     case hexagon:     return "hexagon";
> >>>> +  case gcn:
> >>>>     case r600:        return "r600";
> >>>>     case sparcv9:
> >>>> @@ -199,6 +201,7 @@ Triple::ArchType Triple::getArchTypeForLLVMName(StringRef Name) {
> >>>>       .Case("ppc", ppc)
> >>>>       .Case("ppc64le", ppc64le)
> >>>>       .Case("r600", r600)
> >>>> +    .Case("gcn", gcn)
> >>>>       .Case("hexagon", hexagon)
> >>>>       .Case("sparc", sparc)
> >>>>       .Case("sparcv9", sparcv9)
> >>>> @@ -286,6 +289,7 @@ static Triple::ArchType parseArch(StringRef ArchName) {
> >>>>       .Cases("mips64", "mips64eb", Triple::mips64)
> >>>>       .Case("mips64el", Triple::mips64el)
> >>>>       .Case("r600", Triple::r600)
> >>>> +    .Case("gcn", Triple::gcn)
> >>>>       .Case("hexagon", Triple::hexagon)
> >>>>       .Case("s390x", Triple::systemz)
> >>>>       .Case("sparc", Triple::sparc)
> >>>> @@ -864,6 +868,7 @@ static unsigned getArchPointerBitWidth(llvm::Triple::ArchType Arch) {
> >>>>     case llvm::Triple::aarch64:
> >>>>     case llvm::Triple::aarch64_be:
> >>>> +  case llvm::Triple::gcn:
> >>>>     case llvm::Triple::le64:
> >>>>     case llvm::Triple::mips64:
> >>>>     case llvm::Triple::mips64el:
> >>>> @@ -899,6 +904,7 @@ Triple Triple::get32BitArchVariant() const {
> >>>>     case Triple::UnknownArch:
> >>>>     case Triple::aarch64:
> >>>>     case Triple::aarch64_be:
> >>>> +  case Triple::gcn:
> >>>>     case Triple::msp430:
> >>>>     case Triple::systemz:
> >>>>     case Triple::ppc64le:
> >>>> @@ -962,6 +968,7 @@ Triple Triple::get64BitArchVariant() const {
> >>>>     case Triple::aarch64_be:
> >>>>     case Triple::le64:
> >>>>     case Triple::amdil64:
> >>>> +  case Triple::gcn:
> >>>>     case Triple::hsail64:
> >>>>     case Triple::spir64:
> >>>>     case Triple::mips64:
> >>>> --
> >>>> 1.8.5.5
> >>>>
> >>>> _______________________________________________
> >>>> llvm-commits mailing list
> >>>> llvm-commits at cs.uiuc.edu
> >>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
> >>> _______________________________________________
> >>> llvm-commits mailing list
> >>> llvm-commits at cs.uiuc.edu
> >>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
> 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits



More information about the llvm-commits mailing list