Should tablegen detect this as an error, or is it documented as a limitation
somewhere that we've missed?
In the tablegen-generated file AMDILGenIntrinsics.inc, we have a bunch of if
statements comparing strings, many of which are dead, preventing correct
recognition of some intrinsics in the their text form. I'm not quite sure
what GET_FUNCTION_RECOGNIZER is used for, but if it's used, it's broken…
;-)
Here's a small segment:
// Function name -> enum value recognizer code.
#ifdef GET_FUNCTION_RECOGNIZER
StringRef NameR(Name+6, Len-6); // Skip over 'llvm.'
switch (Name[5]) { // Dispatch on first letter.
default: break;
case 'A':
…
if (NameR.startswith("MDIL.barrier.")) return
AMDILIntrinsic::AMDIL_barrier;
if (NameR.startswith("MDIL.barrier.global.")) return
AMDILIntrinsic::AMDIL_barrier_global;
if (NameR.startswith("MDIL.barrier.local.")) return
AMDILIntrinsic::AMDIL_barrier_local;
if (NameR.startswith("MDIL.barrier.region.")) return
AMDILIntrinsic::AMDIL_barrier_region;
…
if (NameR.startswith("MDIL.fma.")) return
AMDILIntrinsic::AMDIL_fma;
if (NameR.startswith("MDIL.fma.rte.")) return
AMDILIntrinsic::AMDIL_fma_rte;
if (NameR.startswith("MDIL.fma.rtn.")) return
AMDILIntrinsic::AMDIL_fma_rtn;
if (NameR.startswith("MDIL.fma.rtp.")) return
AMDILIntrinsic::AMDIL_fma_rtp;
if (NameR.startswith("MDIL.fma.rtz.")) return
AMDILIntrinsic::AMDIL_fma_rtz;
…
and several other similar instances.
Out of curiosity, what is wrong about that? It looks ok to me. -Chris On Nov 29, 2012, at 6:52 PM, "Relph, Richard" <Richard.Relph at amd.com> wrote:> Should tablegen detect this as an error, or is it documented as a limitation somewhere that we've missed? > > In the tablegen-generated file AMDILGenIntrinsics.inc, we have a bunch of if statements comparing strings, many of which are dead, preventing correct recognition of some intrinsics in the their text form. I'm not quite sure what GET_FUNCTION_RECOGNIZER is used for, but if it's used, it's broken… ;-) > > Here's a small segment: > > // Function name -> enum value recognizer code. > #ifdef GET_FUNCTION_RECOGNIZER > StringRef NameR(Name+6, Len-6); // Skip over 'llvm.' > switch (Name[5]) { // Dispatch on first letter. > default: break; > case 'A': > … > if (NameR.startswith("MDIL.barrier.")) return AMDILIntrinsic::AMDIL_barrier; > if (NameR.startswith("MDIL.barrier.global.")) return AMDILIntrinsic::AMDIL_barrier_global; > if (NameR.startswith("MDIL.barrier.local.")) return AMDILIntrinsic::AMDIL_barrier_local; > if (NameR.startswith("MDIL.barrier.region.")) return AMDILIntrinsic::AMDIL_barrier_region; > … > if (NameR.startswith("MDIL.fma.")) return AMDILIntrinsic::AMDIL_fma; > if (NameR.startswith("MDIL.fma.rte.")) return AMDILIntrinsic::AMDIL_fma_rte; > if (NameR.startswith("MDIL.fma.rtn.")) return AMDILIntrinsic::AMDIL_fma_rtn; > if (NameR.startswith("MDIL.fma.rtp.")) return AMDILIntrinsic::AMDIL_fma_rtp; > if (NameR.startswith("MDIL.fma.rtz.")) return AMDILIntrinsic::AMDIL_fma_rtz; > … > and several other similar instances. > > _______________________________________________ > LLVM Developers mailing list > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
If the source being scanned has "llvm.AMDIL.barrier.global, it will match the first barrier test and return AMDIL_barrier, not AMDIL_barrier_global. On Nov 29, 2012, at 7:19 PM, Chris Lattner <clattner at apple.com> wrote:> Out of curiosity, what is wrong about that? It looks ok to me. > > -Chris > > On Nov 29, 2012, at 6:52 PM, "Relph, Richard" <Richard.Relph at amd.com> wrote: > >> Should tablegen detect this as an error, or is it documented as a limitation somewhere that we've missed? >> >> In the tablegen-generated file AMDILGenIntrinsics.inc, we have a bunch of if statements comparing strings, many of which are dead, preventing correct recognition of some intrinsics in the their text form. I'm not quite sure what GET_FUNCTION_RECOGNIZER is used for, but if it's used, it's broken… ;-) >> >> Here's a small segment: >> >> // Function name -> enum value recognizer code. >> #ifdef GET_FUNCTION_RECOGNIZER >> StringRef NameR(Name+6, Len-6); // Skip over 'llvm.' >> switch (Name[5]) { // Dispatch on first letter. >> default: break; >> case 'A': >> … >> if (NameR.startswith("MDIL.barrier.")) return AMDILIntrinsic::AMDIL_barrier; >> if (NameR.startswith("MDIL.barrier.global.")) return AMDILIntrinsic::AMDIL_barrier_global; >> if (NameR.startswith("MDIL.barrier.local.")) return AMDILIntrinsic::AMDIL_barrier_local; >> if (NameR.startswith("MDIL.barrier.region.")) return AMDILIntrinsic::AMDIL_barrier_region; >> … >> if (NameR.startswith("MDIL.fma.")) return AMDILIntrinsic::AMDIL_fma; >> if (NameR.startswith("MDIL.fma.rte.")) return AMDILIntrinsic::AMDIL_fma_rte; >> if (NameR.startswith("MDIL.fma.rtn.")) return AMDILIntrinsic::AMDIL_fma_rtn; >> if (NameR.startswith("MDIL.fma.rtp.")) return AMDILIntrinsic::AMDIL_fma_rtp; >> if (NameR.startswith("MDIL.fma.rtz.")) return AMDILIntrinsic::AMDIL_fma_rtz; >> … >> and several other similar instances. >> >> _______________________________________________ >> LLVM Developers mailing list >> LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu >> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev >
> From: llvmdev-bounces at cs.uiuc.edu [mailto:llvmdev-bounces at cs.uiuc.edu] On Behalf Of Chris Lattner > Subject: Re: [LLVMdev] Tablegen bug???> Out of curiosity, what is wrong about that? It looks ok to me.This:> > if (NameR.startswith("MDIL.barrier.")) return AMDILIntrinsic::AMDIL_barrier;prevents the three following tests:> > if (NameR.startswith("MDIL.barrier.global.")) return AMDILIntrinsic::AMDIL_barrier_global; > > if (NameR.startswith("MDIL.barrier.local.")) return AMDILIntrinsic::AMDIL_barrier_local; > > if (NameR.startswith("MDIL.barrier.region.")) return AMDILIntrinsic::AMDIL_barrier_region;from ever matching. The order needs to be changed so the common substring is checked last. - Chuck THIS COMMUNICATION MAY CONTAIN CONFIDENTIAL AND/OR OTHERWISE PROPRIETARY MATERIAL and is thus for use only by the intended recipient. If you received this in error, please contact the sender and delete the e-mail and its attachments from all computers.