Message ID | 20190301214945.4353-1-palmer@sifive.com |
---|---|
State | New |
Headers | show |
Series | [PULL] target/riscv: Convert to decodetree | expand |
On Fri, 1 Mar 2019 at 21:49, Palmer Dabbelt <palmer@sifive.com> wrote: > > merged tag 'i2c-for-release-20190228' > Primary key fingerprint: FD0D 5CE6 7CE0 F59A 6688 2686 61F3 8C90 919B FF81 > The following changes since commit 20b084c4b1401b7f8fbc385649d48c67b6f43d44: > > Merge remote-tracking branch 'remotes/cminyard/tags/i2c-for-release-20190228' into staging (2019-03-01 11:20:49 +0000) > > are available in the Git repository at: > > git://github.com/palmer-dabbelt/qemu.git tags/riscv-for-master-4.0-sf2 > > for you to fetch changes up to 0bcba29464ea9969fc69cd729e4c8bddfb2e18e3: > > target/riscv: Remaining rvc insn reuse 32 bit translators (2019-03-01 13:16:18 -0800) > > ---------------------------------------------------------------- > target/riscv: Convert to decodetree > > Bastian: this patchset converts the RISC-V decoder to decodetree in four major steps: > > 1) Convert 32-bit instructions to decodetree [Patch 1-15]: > Many of the gen_* functions are called by the decode functions for 16-bit > and 32-bit functions. If we move translation code from the gen_* > functions to the generated trans_* functions of decode-tree, we get a lot of > duplication. Therefore, we mostly generate calls to the old gen_* function > which are properly replaced after step 2). > > Each of the trans_ functions are grouped into files corresponding to their > ISA extension, e.g. addi which is in RV32I is translated in the file > 'trans_rvi.inc.c'. > > 2) Convert 16-bit instructions to decodetree [Patch 16-18]: > All 16 bit instructions have a direct mapping to a 32 bit instruction. Thus, > we convert the arguments in the 16 bit trans_ function to the arguments of > the corresponding 32 bit instruction and call the 32 bit trans_ function. > > 3) Remove old manual decoding in gen_* function [Patch 19-29]: > this move all manual translation code into the trans_* instructions of > decode tree, such that we can remove the old decode_* functions. > > 4) Simplify RVC by reusing as much as possible from the RVG decoder as suggested > by Richard. [Patch 30-34] > > Palmer: This passed Alistar's testing on rv32 and rv64 as well as my > testing on rv64, so I think it's good to go. Thanks for the cleanup! > Hi; I'm afraid this has compile errors on the OSX build: In file included from /Users/pm215/src/qemu-for-merges/target/riscv/translate.c:377: target/riscv/decode_insn16.inc.c:37:15: error: redefinition of typedef 'arg_fld' is a C11 feature [-Werror,-Wtypedef-redefinition] typedef arg_i arg_fld; ^ target/riscv/decode_insn32.inc.c:293:15: note: previous definition is here typedef arg_i arg_fld; ^ In file included from /Users/pm215/src/qemu-for-merges/target/riscv/translate.c:377: target/riscv/decode_insn16.inc.c:39:15: error: redefinition of typedef 'arg_lw' is a C11 feature [-Werror,-Wtypedef-redefinition] typedef arg_i arg_lw; ^ target/riscv/decode_insn32.inc.c:137:15: note: previous definition is here typedef arg_i arg_lw; ^ In file included from /Users/pm215/src/qemu-for-merges/target/riscv/translate.c:377: target/riscv/decode_insn16.inc.c:41:15: error: redefinition of typedef 'arg_fsd' is a C11 feature [-Werror,-Wtypedef-redefinition] typedef arg_s arg_fsd; ^ target/riscv/decode_insn32.inc.c:295:15: note: previous definition is here typedef arg_s arg_fsd; ^ In file included from /Users/pm215/src/qemu-for-merges/target/riscv/translate.c:377: target/riscv/decode_insn16.inc.c:43:15: error: redefinition of typedef 'arg_sw' is a C11 feature [-Werror,-Wtypedef-redefinition] typedef arg_s arg_sw; ^ target/riscv/decode_insn32.inc.c:147:15: note: previous definition is here typedef arg_s arg_sw; ^ In file included from /Users/pm215/src/qemu-for-merges/target/riscv/translate.c:377: target/riscv/decode_insn16.inc.c:49:28: error: redefinition of typedef 'arg_c_addi16sp_lui' is a C11 feature [-Werror,-Wtypedef-redefinition] typedef arg_c_addi16sp_lui arg_c_addi16sp_lui; ^ target/riscv/decode_insn16.inc.c:7:3: note: previous definition is here } arg_c_addi16sp_lui; ^ (and a lot of other similar errors) thanks -- PMM
On 3/4/19 12:02 PM, Peter Maydell wrote: > On Fri, 1 Mar 2019 at 21:49, Palmer Dabbelt <palmer@sifive.com> wrote: >> merged tag 'i2c-for-release-20190228' >> Primary key fingerprint: FD0D 5CE6 7CE0 F59A 6688 2686 61F3 8C90 919B FF81 >> The following changes since commit 20b084c4b1401b7f8fbc385649d48c67b6f43d44: >> >> Merge remote-tracking branch 'remotes/cminyard/tags/i2c-for-release-20190228' into staging (2019-03-01 11:20:49 +0000) >> >> are available in the Git repository at: >> >> git://github.com/palmer-dabbelt/qemu.git tags/riscv-for-master-4.0-sf2 >> >> for you to fetch changes up to 0bcba29464ea9969fc69cd729e4c8bddfb2e18e3: >> >> target/riscv: Remaining rvc insn reuse 32 bit translators (2019-03-01 13:16:18 -0800) >> >> ---------------------------------------------------------------- >> target/riscv: Convert to decodetree >> >> Bastian: this patchset converts the RISC-V decoder to decodetree in four major steps: >> >> 1) Convert 32-bit instructions to decodetree [Patch 1-15]: >> Many of the gen_* functions are called by the decode functions for 16-bit >> and 32-bit functions. If we move translation code from the gen_* >> functions to the generated trans_* functions of decode-tree, we get a lot of >> duplication. Therefore, we mostly generate calls to the old gen_* function >> which are properly replaced after step 2). >> >> Each of the trans_ functions are grouped into files corresponding to their >> ISA extension, e.g. addi which is in RV32I is translated in the file >> 'trans_rvi.inc.c'. >> >> 2) Convert 16-bit instructions to decodetree [Patch 16-18]: >> All 16 bit instructions have a direct mapping to a 32 bit instruction. Thus, >> we convert the arguments in the 16 bit trans_ function to the arguments of >> the corresponding 32 bit instruction and call the 32 bit trans_ function. >> >> 3) Remove old manual decoding in gen_* function [Patch 19-29]: >> this move all manual translation code into the trans_* instructions of >> decode tree, such that we can remove the old decode_* functions. >> >> 4) Simplify RVC by reusing as much as possible from the RVG decoder as suggested >> by Richard. [Patch 30-34] >> >> Palmer: This passed Alistar's testing on rv32 and rv64 as well as my >> testing on rv64, so I think it's good to go. Thanks for the cleanup! >> > Hi; I'm afraid this has compile errors on the OSX build: > > > In file included from > /Users/pm215/src/qemu-for-merges/target/riscv/translate.c:377: > target/riscv/decode_insn16.inc.c:37:15: error: redefinition of typedef > 'arg_fld' is a C11 feature [-Werror,-Wtypedef-redefinition] > typedef arg_i arg_fld; > ^ > target/riscv/decode_insn32.inc.c:293:15: note: previous definition is here > typedef arg_i arg_fld; This looks like an unforeseen decodetree problem (CC' Richard). As these 16/32 instructions share the same trans_* function, we emit the same typedef once for 16 bit and once for 32 bit. I don't see an easy fix in decodetree other than annotating one of the instructions with something like !shared , since both insn are parsed in two runs of decodetree and later included together into one file. I'll try to prepare a patch. Cheers, Bastian
On 3/4/19 1:52 PM, Bastian Koppelmann wrote: > > On 3/4/19 12:02 PM, Peter Maydell wrote: >> On Fri, 1 Mar 2019 at 21:49, Palmer Dabbelt <palmer@sifive.com> wrote: >>> merged tag 'i2c-for-release-20190228' >>> Primary key fingerprint: FD0D 5CE6 7CE0 F59A 6688 2686 61F3 8C90 >>> 919B FF81 >>> The following changes since commit >>> 20b084c4b1401b7f8fbc385649d48c67b6f43d44: >>> >>> Merge remote-tracking branch >>> 'remotes/cminyard/tags/i2c-for-release-20190228' into staging >>> (2019-03-01 11:20:49 +0000) >>> >>> are available in the Git repository at: >>> >>> git://github.com/palmer-dabbelt/qemu.git >>> tags/riscv-for-master-4.0-sf2 >>> >>> for you to fetch changes up to >>> 0bcba29464ea9969fc69cd729e4c8bddfb2e18e3: >>> >>> target/riscv: Remaining rvc insn reuse 32 bit translators >>> (2019-03-01 13:16:18 -0800) >>> >>> ---------------------------------------------------------------- >>> target/riscv: Convert to decodetree >>> >>> Bastian: this patchset converts the RISC-V decoder to decodetree in >>> four major steps: >>> >>> 1) Convert 32-bit instructions to decodetree [Patch 1-15]: >>> Many of the gen_* functions are called by the decode functions >>> for 16-bit >>> and 32-bit functions. If we move translation code from the gen_* >>> functions to the generated trans_* functions of decode-tree, we >>> get a lot of >>> duplication. Therefore, we mostly generate calls to the old >>> gen_* function >>> which are properly replaced after step 2). >>> >>> Each of the trans_ functions are grouped into files >>> corresponding to their >>> ISA extension, e.g. addi which is in RV32I is translated in the >>> file >>> 'trans_rvi.inc.c'. >>> >>> 2) Convert 16-bit instructions to decodetree [Patch 16-18]: >>> All 16 bit instructions have a direct mapping to a 32 bit >>> instruction. Thus, >>> we convert the arguments in the 16 bit trans_ function to the >>> arguments of >>> the corresponding 32 bit instruction and call the 32 bit trans_ >>> function. >>> >>> 3) Remove old manual decoding in gen_* function [Patch 19-29]: >>> this move all manual translation code into the trans_* >>> instructions of >>> decode tree, such that we can remove the old decode_* functions. >>> >>> 4) Simplify RVC by reusing as much as possible from the RVG decoder >>> as suggested >>> by Richard. [Patch 30-34] >>> >>> Palmer: This passed Alistar's testing on rv32 and rv64 as well as my >>> testing on rv64, so I think it's good to go. Thanks for the cleanup! >>> >> Hi; I'm afraid this has compile errors on the OSX build: >> >> >> In file included from >> /Users/pm215/src/qemu-for-merges/target/riscv/translate.c:377: >> target/riscv/decode_insn16.inc.c:37:15: error: redefinition of typedef >> 'arg_fld' is a C11 feature [-Werror,-Wtypedef-redefinition] >> typedef arg_i arg_fld; >> ^ >> target/riscv/decode_insn32.inc.c:293:15: note: previous definition is >> here >> typedef arg_i arg_fld; > > > This looks like an unforeseen decodetree problem (CC' Richard). As > these 16/32 instructions share the same trans_* function, we emit the > same typedef once for 16 bit and once for 32 bit. I don't see an easy > fix in decodetree other than annotating one of the instructions with > something like !shared , since both insn are parsed in two runs of > decodetree and later included together into one file. I'll try to > prepare a patch. This is something we can figure out after this PR. @Palmer I think you can drop patches 30-34 for now as these are the offenders and the rest of the PR is fine. Cheers, Bastian
On 3/4/19 4:52 AM, Bastian Koppelmann wrote: > This looks like an unforeseen decodetree problem (CC' Richard). As these 16/32 > instructions share the same trans_* function, we emit the same typedef once for > 16 bit and once for 32 bit. I don't see an easy fix in decodetree Use an argument set, and add !extern to the argument set definition for the "second" decoder. (Leastwise I don't think that feature is sitting on the branch...) r~
On Mon, 04 Mar 2019 11:30:21 PST (-0800), richard.henderson@linaro.org wrote: > On 3/4/19 4:52 AM, Bastian Koppelmann wrote: >> This looks like an unforeseen decodetree problem (CC' Richard). As these 16/32 >> instructions share the same trans_* function, we emit the same typedef once for >> 16 bit and once for 32 bit. I don't see an easy fix in decodetree > > Use an argument set, and add !extern to the argument set definition for the > "second" decoder. > > (Leastwise I don't think that feature is sitting on the branch...) > Thanks, I'll do another PR and keep the last 5 patches for later.