Matt Arsenault
2014-Jul-09 19:51 UTC
[LLVMdev] Help!!!!Help!!!! " LLVM ERROR: Cannot select: 0x9fc9680: i32 = fp32_to_fp16 0x9fc0750 [ID=16] " problem!!!!!!!!!!!!!!!!!!
On 07/09/2014 12:41 PM, Matt Arsenault wrote:> On 07/09/2014 03:30 PM, yalong at multicorewareinc.com wrote: >> Thank you Kevin!!! >> If I use fptrunc and bitcast realise NEON vcvtt ( I can sure, >> "fptrunc double %tmp to float" is right, but "fptrunc float %tmp to >> half" is wrong). My target platform is MIPS. The command as following: >> >> NEON: >> vcvtt.f16.f32 s2, s0 >> >> llvm Code: >> >> %Vt_2 = load float* %VFP_s0, align 4 >> %Vt3_1 = fptrunc float %Vt_2 to half >> %Vt4_1 = bitcast half %Vt3_1 to i16 >> %Vt2_2 = bitcast float* %VFP_s2 to <2 x i16>* >> %Vrti_1 = load <2 x i16>* %Vt2_2, align 4 >> %Vrti_2 = insertelement <2 x i16> %Vrti_1, i16 %Vt4_1, i32 1 >> %Vt2_3 = bitcast float* %VFP_s2 to <2 x i16>* >> store <2 x i16> %Vrti_2, <2 x i16>* %Vt2_3, align 4 >> >> Error Log: >> LLVM ERROR: Cannot select: 0x9f554b0: ch = store 0x9d0f28c, >> 0x9f5d900, 0x9f54ba8, 0x9f54b20<ST2[FixedStack0](align=4), trunc to >> f16> [ID=52] >> 0x9f5d900: f32,ch = load 0x9f5e290, 0x9f5dd40, >> 0x9f54b20<LD4[%sunkaddr69]> [ORD=1810] [ID=51] >> 0x9f5dd40: i32 = add 0x9f55318, 0x9f5e0f8 [ORD=1808] [ID=31] >> 0x9f55318: i32,ch = CopyFromReg 0x9d0f28c, 0x9f6a3a0 [ORD=1796] [ID=26] >> 0x9f6a3a0: i32 = Register %vreg32 [ORD=1796] [ID=1] >> 0x9f5e0f8: i32 = Constant<64> [ORD=1808] [ID=17] >> 0x9f54b20: i32 = undef [ORD=1797] [ID=6] >> 0x9f54ba8: i32 = FrameIndex<0> [ID=24] >> 0x9f54b20: i32 = undef [ORD=1797] [ID=6] >> In function: testVCVTT32TO16Function >> >> > > I think that support for the half type is only implemented for ARM. > Last I tried to use it, I found that none of it works even on x86, and > the current handling of the half conversion SDNodes seem to rely on > ARM specific assumptionsHave you tried using the @llvm.convert.to/from.fp16 intrinsics instead? -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20140709/c963d972/attachment.html>
Andrea Di Biagio
2014-Jul-09 21:20 UTC
[LLVMdev] Help!!!!Help!!!! " LLVM ERROR: Cannot select: 0x9fc9680: i32 = fp32_to_fp16 0x9fc0750 [ID=16] " problem!!!!!!!!!!!!!!!!!!
Not sure if this can help, but if you really really want to have minimal half float support on Mips, then one thing you could try to do is to hack MipsISelLowering.cpp adding rules to expand float-half conversion SDNodes into library calls. + setOperationAction(ISD::FP16_TO_FP32, MVT::f32, Expand); + setOperationAction(ISD::FP32_TO_FP16, MVT::i32, Expand); (The MVT::i32 on the second rule is required because type i16 is promoted to i32). If you then convert every occurrence of 'fptrunc' from float to half with calls to @llvm.convert.to.fp16, then you should be able to compile (hopefully) with no errors. That means, in your original example you would convert the following IR statement: %Vt3_1 = fptrunc float %Vt_2 to half into %Vt3_1 = call i16 @llvm.convert.to.fp16(float %Vt_2) The downside is that you will have to add definitions for '__gnu_f2h_ieee' and '__gnu_h2f_ieee' in the compiler runtime. That is because the backend will expand all the float-half conversions into library calls... This workaround should work assuming that a) you can hack the backend, and b) it is acceptable (i.e. a reasonable solution in your case) to have a library call for every float-half conversion in your code. On Wed, Jul 9, 2014 at 8:51 PM, Matt Arsenault <Matthew.Arsenault at amd.com> wrote:> I think that support for the half type is only implemented for ARM. Last I > tried to use it, I found that none of it works even on x86, and the current > handling of the half conversion SDNodes seem to rely on ARM specific > assumptionsJust for the record, since revision 212293 (committed only five days ago), the x86 backend supports float half conversions. On x86, if the target has F16C, there are ISel patterns to map float-half conversions to specific instructions. If there is no F16C support, then the backend expands float-half conversions into runtime library calls. Cheers, Andrea
Wan, Xiaofei
2014-Jul-10 01:54 UTC
[LLVMdev] Help!!!!Help!!!! " LLVM ERROR: Cannot select: 0x9fc9680: i32 = fp32_to_fp16 0x9fc0750 [ID=16] " problem!!!!!!!!!!!!!!!!!!
This is fixed by Andrea's patch about one week ago From: llvmdev-bounces at cs.uiuc.edu [mailto:llvmdev-bounces at cs.uiuc.edu] On Behalf Of yalong at multicorewareinc.com Sent: Friday, July 11, 2014 12:45 AM To: Matt Arsenault; Kevin Qin Cc: llvmdev Subject: Re: [LLVMdev] Help!!!!Help!!!! " LLVM ERROR: Cannot select: 0x9fc9680: i32 = fp32_to_fp16 0x9fc0750 [ID=16] " problem!!!!!!!!!!!!!!!!!! Hi Matt, Thank you your replying. I try to use @llvm.convert.to/from.fp16 intrinsics before, I also meet "LLVM ERROR: Cannot select: fp32_to_fp16" problem, maybe these functions call fptrunc/fpext. Following is detail log: LLVM ERROR: Cannot select: 0x9fc9680: i32 = fp32_to_fp16 0x9fc0750 [ID=16] 0x9fc0750: f32,ch = load 0x3aafd68, 0x9fc2a20, 0x9feaab0<LD4[%sunkaddr85033]> [ORD=125117] [ID=15] 0x9fc2a20: i32 = add 0x9fed880, 0x9fd9ea0 [ORD=125115] [ID=13] 0x9fed880: i32,ch = CopyFromReg 0x3aafd68, 0x9fbea90 [ORD=125114] [ID=9] 0x9fbea90: i32 = Register %vreg13999 [ORD=125114] [ID=1] 0x9fd9ea0: i32 = Constant<80> [ORD=125115] [ID=2] 0x9feaab0: i32 = undef [ORD=125117] [ID=4] In function: internal_function_69 Command exited with non-zero status 1 ________________________________ yalong at multicorewareinc.com<mailto:yalong at multicorewareinc.com> From: Matt Arsenault<mailto:Matthew.Arsenault at amd.com> Date: 2014-07-09 12:51 To: yalong at multicorewareinc.com<mailto:yalong at multicorewareinc.com>; Kevin Qin<mailto:kevinqindev at gmail.com> CC: llvmdev<mailto:llvmdev at cs.uiuc.edu> Subject: Re: [LLVMdev] Help!!!!Help!!!! " LLVM ERROR: Cannot select: 0x9fc9680: i32 = fp32_to_fp16 0x9fc0750 [ID=16] " problem!!!!!!!!!!!!!!!!!! On 07/09/2014 12:41 PM, Matt Arsenault wrote: On 07/09/2014 03:30 PM, yalong at multicorewareinc.com<mailto:yalong at multicorewareinc.com> wrote: Thank you Kevin!!! If I use fptrunc and bitcast realise NEON vcvtt ( I can sure, "fptrunc double %tmp to float" is right, but "fptrunc float %tmp to half" is wrong). My target platform is MIPS. The command as following: NEON: vcvtt.f16.f32 s2, s0 llvm Code: %Vt_2 = load float* %VFP_s0, align 4 %Vt3_1 = fptrunc float %Vt_2 to half %Vt4_1 = bitcast half %Vt3_1 to i16 %Vt2_2 = bitcast float* %VFP_s2 to <2 x i16>* %Vrti_1 = load <2 x i16>* %Vt2_2, align 4 %Vrti_2 = insertelement <2 x i16> %Vrti_1, i16 %Vt4_1, i32 1 %Vt2_3 = bitcast float* %VFP_s2 to <2 x i16>* store <2 x i16> %Vrti_2, <2 x i16>* %Vt2_3, align 4 Error Log: LLVM ERROR: Cannot select: 0x9f554b0: ch = store 0x9d0f28c, 0x9f5d900, 0x9f54ba8, 0x9f54b20<ST2[FixedStack0](align=4), trunc to f16> [ID=52] 0x9f5d900: f32,ch = load 0x9f5e290, 0x9f5dd40, 0x9f54b20<LD4[%sunkaddr69]> [ORD=1810] [ID=51] 0x9f5dd40: i32 = add 0x9f55318, 0x9f5e0f8 [ORD=1808] [ID=31] 0x9f55318: i32,ch = CopyFromReg 0x9d0f28c, 0x9f6a3a0 [ORD=1796] [ID=26] 0x9f6a3a0: i32 = Register %vreg32 [ORD=1796] [ID=1] 0x9f5e0f8: i32 = Constant<64> [ORD=1808] [ID=17] 0x9f54b20: i32 = undef [ORD=1797] [ID=6] 0x9f54ba8: i32 = FrameIndex<0> [ID=24] 0x9f54b20: i32 = undef [ORD=1797] [ID=6] In function: testVCVTT32TO16Function I think that support for the half type is only implemented for ARM. Last I tried to use it, I found that none of it works even on x86, and the current handling of the half conversion SDNodes seem to rely on ARM specific assumptions Have you tried using the @llvm.convert.to/from.fp16 intrinsics instead? -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20140710/02bb371c/attachment.html>
yalong at multicorewareinc.com
2014-Jul-10 16:44 UTC
[LLVMdev] Help!!!!Help!!!! " LLVM ERROR: Cannot select: 0x9fc9680: i32 = fp32_to_fp16 0x9fc0750 [ID=16] " problem!!!!!!!!!!!!!!!!!!
Hi Matt, Thank you your replying. I try to use @llvm.convert.to/from.fp16 intrinsics before, I also meet "LLVM ERROR: Cannot select: fp32_to_fp16" problem, maybe these functions call fptrunc/fpext. Following is detail log: LLVM ERROR: Cannot select: 0x9fc9680: i32 = fp32_to_fp16 0x9fc0750 [ID=16] 0x9fc0750: f32,ch = load 0x3aafd68, 0x9fc2a20, 0x9feaab0<LD4[%sunkaddr85033]> [ORD=125117] [ID=15] 0x9fc2a20: i32 = add 0x9fed880, 0x9fd9ea0 [ORD=125115] [ID=13] 0x9fed880: i32,ch = CopyFromReg 0x3aafd68, 0x9fbea90 [ORD=125114] [ID=9] 0x9fbea90: i32 = Register %vreg13999 [ORD=125114] [ID=1] 0x9fd9ea0: i32 = Constant<80> [ORD=125115] [ID=2] 0x9feaab0: i32 = undef [ORD=125117] [ID=4] In function: internal_function_69 Command exited with non-zero status 1 yalong at multicorewareinc.com From: Matt ArsenaultDate: 2014-07-09 12:51To: yalong at multicorewareinc.com; Kevin QinCC: llvmdevSubject: Re: [LLVMdev] Help!!!!Help!!!! " LLVM ERROR: Cannot select: 0x9fc9680: i32 = fp32_to_fp16 0x9fc0750 [ID=16] " problem!!!!!!!!!!!!!!!!!! On 07/09/2014 12:41 PM, Matt Arsenault wrote: On 07/09/2014 03:30 PM, yalong at multicorewareinc.com wrote: Thank you Kevin!!! If I use fptrunc and bitcast realise NEON vcvtt ( I can sure, "fptrunc double %tmp to float" is right, but "fptrunc float %tmp to half" is wrong). My target platform is MIPS. The command as following: NEON: vcvtt.f16.f32 s2, s0 llvm Code: %Vt_2 = load float* %VFP_s0, align 4 %Vt3_1 = fptrunc float %Vt_2 to half %Vt4_1 = bitcast half %Vt3_1 to i16 %Vt2_2 = bitcast float* %VFP_s2 to <2 x i16>* %Vrti_1 = load <2 x i16>* %Vt2_2, align 4 %Vrti_2 = insertelement <2 x i16> %Vrti_1, i16 %Vt4_1, i32 1 %Vt2_3 = bitcast float* %VFP_s2 to <2 x i16>* store <2 x i16> %Vrti_2, <2 x i16>* %Vt2_3, align 4 Error Log: LLVM ERROR: Cannot select: 0x9f554b0: ch = store 0x9d0f28c, 0x9f5d900, 0x9f54ba8, 0x9f54b20<ST2[FixedStack0](align=4), trunc to f16> [ID=52] 0x9f5d900: f32,ch = load 0x9f5e290, 0x9f5dd40, 0x9f54b20<LD4[%sunkaddr69]> [ORD=1810] [ID=51] 0x9f5dd40: i32 = add 0x9f55318, 0x9f5e0f8 [ORD=1808] [ID=31] 0x9f55318: i32,ch = CopyFromReg 0x9d0f28c, 0x9f6a3a0 [ORD=1796] [ID=26] 0x9f6a3a0: i32 = Register %vreg32 [ORD=1796] [ID=1] 0x9f5e0f8: i32 = Constant<64> [ORD=1808] [ID=17] 0x9f54b20: i32 = undef [ORD=1797] [ID=6] 0x9f54ba8: i32 = FrameIndex<0> [ID=24] 0x9f54b20: i32 = undef [ORD=1797] [ID=6] In function: testVCVTT32TO16Function I think that support for the half type is only implemented for ARM. Last I tried to use it, I found that none of it works even on x86, and the current handling of the half conversion SDNodes seem to rely on ARM specific assumptions Have you tried using the @llvm.convert.to/from.fp16 intrinsics instead? -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20140710/dd105a5f/attachment.html>
yalong at multicorewareinc.com
2014-Jul-10 22:48 UTC
[LLVMdev] Help!!!!Help!!!! " LLVM ERROR: Cannot select: 0x9fc9680: i32 = fp32_to_fp16 0x9fc0750 [ID=16] " problem!!!!!!!!!!!!!!!!!!
Hi Andrea Thank you your replying. I do like your letter. Add following to line to MipsISelLowering.cpp. As your words, @llvm.convert.to.fp16 can compile successfully. However, the runtime is not right. + setOperationAction(ISD::FP16_TO_FP32, MVT::f32, Expand);+ setOperationAction(ISD::FP32_TO_FP16, MVT::i32, Expand); Robin yalong at multicorewareinc.com From: Andrea Di BiagioDate: 2014-07-09 14:20To: Matt ArsenaultCC: yalong at multicorewareinc.com; Kevin Qin; llvmdevSubject: Re: [LLVMdev] Help!!!!Help!!!! " LLVM ERROR: Cannot select: 0x9fc9680: i32 = fp32_to_fp16 0x9fc0750 [ID=16] " problem!!!!!!!!!!!!!!!!!!Not sure if this can help, but if you really really want to have minimal half float support on Mips, then one thing you could try to do is to hack MipsISelLowering.cpp adding rules to expand float-half conversion SDNodes into library calls. + setOperationAction(ISD::FP16_TO_FP32, MVT::f32, Expand); + setOperationAction(ISD::FP32_TO_FP16, MVT::i32, Expand); (The MVT::i32 on the second rule is required because type i16 is promoted to i32). If you then convert every occurrence of 'fptrunc' from float to half with calls to @llvm.convert.to.fp16, then you should be able to compile (hopefully) with no errors. That means, in your original example you would convert the following IR statement: %Vt3_1 = fptrunc float %Vt_2 to half into %Vt3_1 = call i16 @llvm.convert.to.fp16(float %Vt_2) The downside is that you will have to add definitions for '__gnu_f2h_ieee' and '__gnu_h2f_ieee' in the compiler runtime. That is because the backend will expand all the float-half conversions into library calls... This workaround should work assuming that a) you can hack the backend, and b) it is acceptable (i.e. a reasonable solution in your case) to have a library call for every float-half conversion in your code. On Wed, Jul 9, 2014 at 8:51 PM, Matt Arsenault <Matthew.Arsenault at amd.com> wrote:> I think that support for the half type is only implemented for ARM. Last I > tried to use it, I found that none of it works even on x86, and the current > handling of the half conversion SDNodes seem to rely on ARM specific > assumptionsJust for the record, since revision 212293 (committed only five days ago), the x86 backend supports float half conversions. On x86, if the target has F16C, there are ISel patterns to map float-half conversions to specific instructions. If there is no F16C support, then the backend expands float-half conversions into runtime library calls. Cheers, Andrea -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20140710/c0df4b12/attachment.html>
Maybe Matching Threads
- [LLVMdev] Help!!!!Help!!!! " LLVM ERROR: Cannot select: 0x9fc9680: i32 = fp32_to_fp16 0x9fc0750 [ID=16] " problem!!!!!!!!!!!!!!!!!!
- [LLVMdev] Help!!!!Help!!!! " LLVM ERROR: Cannot select: 0x9fc9680: i32 = fp32_to_fp16 0x9fc0750 [ID=16] " problem!!!!!!!!!!!!!!!!!!
- [LLVMdev] Help!!!!Help!!!! " LLVM ERROR: Cannot select: 0x9fc9680: i32 = fp32_to_fp16 0x9fc0750 [ID=16] " problem!!!!!!!!!!!!!!!!!!
- [LLVMdev] Help!!!!Help!!!! " LLVM ERROR: Cannot select: 0x9fc9680: i32 = fp32_to_fp16 0x9fc0750 [ID=16] " problem!!!!!!!!!!!!!!!!!!
- [LLVMdev] __fp16 suport in llvm back-end