https://gcc.gnu.org/g:34bfaf7aedcdf6ef482fdface3a10ce30d3c8a42
commit 34bfaf7aedcdf6ef482fdface3a10ce30d3c8a42 Author: Michael Meissner <[email protected]> Date: Tue Jun 23 23:09:57 2026 -0400 Update ChangeLog.* Diff: --- gcc/ChangeLog.sha | 640 ++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 640 insertions(+) diff --git a/gcc/ChangeLog.sha b/gcc/ChangeLog.sha index 4ee43fdc2ada..b95d523d7228 100644 --- a/gcc/ChangeLog.sha +++ b/gcc/ChangeLog.sha @@ -1,3 +1,643 @@ +==================== Branch work248-sha, patch #410 ==================== + +Add xvrlw support. + +This patch adds support for a possible new variant of the vector rotate left +instruction that might be added to a future PowerPC. This variant (xvrlw) can +use any VSX register instead of requiring only Altivec registers. + +As a potential test, add other variants of vxvrl<x> enabled with the -mxvrld option. + +2026-06-23 Michael Meissner <[email protected]> + +gcc/ + + * config/rs6000/altivec.md (xvrlw): New insn. + (test_xvrl<VI_char>): Add test insns. + * config/rs6000/rs6000.h (TARGET_XVRLD): New macro. + +gcc/testsuite/ + + * gcc.target/powerpc/vector-rotate-left.c: New test. + +==================== Branch work248-sha, patch #408 ==================== + +PR target/117251: Add tests + +This is patch #45 of 45 to generate the 'XXEVAL' instruction on power10 +and power11 instead of using the Altivec 'VAND' instruction feeding +into 'VNAND'. The 'XXEVAL' instruction can use all 64 vector +registers, instead of the 32 registers that traditional Altivec vector +instructions use. By allowing all of the vector registers to be used, +it reduces the amount of spilling that a large benchmark generated. + +This patch adds the tests for generating 'XXEVAL' to the testsuite. + +I have tested these patches on both big endian and little endian +PowerPC servers, with no regressions. Can I check these patchs into +the trunk? + +2026-06-23 Michael Meissner <[email protected]> + +gcc/testsuite/ + + PR target/117251 + * gcc.target/powerpc/p10-vector-fused-1.c: New test. + * gcc.target/powerpc/p10-vector-fused-2.c: Likewise. + +==================== Branch work248-sha, patch #407 ==================== + +PR target/117251: Improve vector fusion #7 + +See the following post for a complete explanation of what the patches +for PR target/117251: + + * https://gcc.gnu.org/pipermail/gcc-patches/2025-June/686474.html + +This is patch #2 of 45 to generate the 'XXEVAL' instruction on power10 +and power11 instead of using the Altivec 'VANDC' instruction feeding +into 'VAND'. The 'XXEVAL' instruction can use all 64 vector registers, +instead of the 32 registers that traditional Altivec vector +instructions use. By allowing all of the vector registers to be used, +it reduces the amount of spilling that a large benchmark generated. + +For vectors such as: + + vector int a, b, c, d; + +This patch will generate a call to xxeval if any of the registers are allocated +to traditional floating point registers: + + Code: Old: New: + ===== ==== ==== + a = ~ ((~ (c | d)) & b); vnor t,c,d xxeval a,b,c,d,247 + vnand a,t,b + + a = ~ ((c | d) & b); vor t,c,d xxeval a,b,c,d,248 + vnand a,t,b + + a = ~ ((c ^ d) & b); vxor t,c,d xxeval a,b,c,d,249 + vnand a,t,b + + a = ~ ((c & ~ d) & b); vandc t,c,d xxeval a,b,c,d,253 + vnand a,t,b + + a = ~ ((c & d) & b); vand t,c,d xxeval a,b,c,d,254 + vnand a,t,b + +Since fusion using 2 Altivec instructions is slightly faster than using +the 'XXEVAL' instruction we prefer to generate the Altivec instructions +if we can. In addition, because 'XXEVAL' is a prefixed instruction, it +possibly might generate an extra NOP instruction to align the 'XXEVAL' +instruction. + +I have tested these patches on both big endian and little endian +PowerPC servers, with no regressions. Can I check these patchs into +the trunk? + +2026-06-23 Michael Meissner <[email protected]> + +gcc/ + + PR target/117251 + * config/rs6000/fusion.md: Regenerate. + * config/rs6000/genfusion.pl (gen_logical_addsubf): Add support + to generate vector/vector fusion if XXEVAL is supported. + +==================== Branch work248-sha, patch #406 ==================== + +PR target/117251: Improve vector fusion #6 + +See the following post for a complete explanation of what the patches +for PR target/117251: + + * https://gcc.gnu.org/pipermail/gcc-patches/2025-June/686474.html + +This is patch #2 of 45 to generate the 'XXEVAL' instruction on power10 +and power11 instead of using the Altivec 'VANDC' instruction feeding +into 'VAND'. The 'XXEVAL' instruction can use all 64 vector registers, +instead of the 32 registers that traditional Altivec vector +instructions use. By allowing all of the vector registers to be used, +it reduces the amount of spilling that a large benchmark generated. + +For vectors such as: + + vector int a, b, c, d; + +This patch will generate a call to xxeval if any of the registers are allocated +to traditional floating point registers: + + Code: Old: New: + ===== ==== ==== + a = ~ ((c & ~ d) | b); vandc t,c,d xxeval a,b,c,d,208 + vnor a,t,b + + a = ~ ((c & ~ d) ^ b); vandc t,c,d xxeval a,b,c,d,210 + veqv a,t,b + + a = ~ ((c & d) | b); vand t,c,d xxeval a,b,c,d,224 + vnor a,t,b + + a = (~ (c & d)) ^ b; vnand t,c,d xxeval a,b,c,d,225 + vxor a,t,b + + a = (~ (c & d)) | b; vnand t,c,d xxeval a,b,c,d,239 + vor a,t,b + + a = ~ ((~ (c & d)) & b); vnand t,c,d xxeval a,b,c,d,241 + vnand a,t,b + + a = ~ ((c | ~ d) & b); vorc t,c,d xxeval a,b,c,d,244 + vnand a,t,b + + a = ~ ((~ (c ^ d)) & b); veqv t,c,d xxeval a,b,c,d,246 + vnand a,t,b + +Since fusion using 2 Altivec instructions is slightly faster than using +the 'XXEVAL' instruction we prefer to generate the Altivec instructions +if we can. In addition, because 'XXEVAL' is a prefixed instruction, it +possibly might generate an extra NOP instruction to align the 'XXEVAL' +instruction. + +I have tested these patches on both big endian and little endian +PowerPC servers, with no regressions. Can I check these patchs into +the trunk? + +2026-06-23 Michael Meissner <[email protected]> + +gcc/ + + PR target/117251 + * config/rs6000/fusion.md: Regenerate. + * config/rs6000/genfusion.pl (gen_logical_addsubf): Add support + to generate vector/vector fusion if XXEVAL is supported. + +==================== Branch work248-sha, patch #405 ==================== + +PR target/117251: Improve vector fusion #5 + +See the following post for a complete explanation of what the patches +for PR target/117251: + + * https://gcc.gnu.org/pipermail/gcc-patches/2025-June/686474.html + +This is patch #2 of 45 to generate the 'XXEVAL' instruction on power10 +and power11 instead of using the Altivec 'VANDC' instruction feeding +into 'VAND'. The 'XXEVAL' instruction can use all 64 vector registers, +instead of the 32 registers that traditional Altivec vector +instructions use. By allowing all of the vector registers to be used, +it reduces the amount of spilling that a large benchmark generated. + +For vectors such as: + + vector int a, b, c, d; + +This patch will generate a call to xxeval if any of the registers are allocated +to traditional floating point registers: + + Code: Old: New: + ===== ==== ==== + a = ~ ((c | d) | b); vor t,c,d xxeval a,b,c,d,128 + vnor a,t,b + + a = (~ (c | d)) ^ b; vnor t,c,d xxeval a,b,c,d,135 + vxor a,t,b + + a = (~ (c | d)) | b; vnor t,c,d xxeval a,b,c,d,143 + vor a,t,b + + a = ~ ((c ^ d) | b); vxor t,c,d xxeval a,b,c,d,144 + vnor a,t,b + + a = (~ (c ^ d)) ^ b; veqv t,c,d xxeval a,b,c,d,150 + vxor a,t,b + + a = (~ (c ^ d)) | b; veqv t,c,d xxeval a,b,c,d,159 + vor a,t,b + + a = (c | ~ d) ^ b; vorc t,c,d xxeval a,b,c,d,180 + vxor a,t,b + + a = (c | ~ d) | b; vorc t,c,d xxeval a,b,c,d,191 + vor a,t,b + +Since fusion using 2 Altivec instructions is slightly faster than using +the 'XXEVAL' instruction we prefer to generate the Altivec instructions +if we can. In addition, because 'XXEVAL' is a prefixed instruction, it +possibly might generate an extra NOP instruction to align the 'XXEVAL' +instruction. + +I have tested these patches on both big endian and little endian +PowerPC servers, with no regressions. Can I check these patchs into +the trunk? + +2026-06-23 Michael Meissner <[email protected]> + +gcc/ + + PR target/117251 + * config/rs6000/fusion.md: Regenerate. + * config/rs6000/genfusion.pl (gen_logical_addsubf): Add support + to generate vector/vector fusion if XXEVAL is supported. + +==================== Branch work248-sha, patch #404 ==================== + +PR target/117251: Improve vector fusion #4 + +See the following post for a complete explanation of what the patches +for PR target/117251: + + * https://gcc.gnu.org/pipermail/gcc-patches/2025-June/686474.html + +This is patch #2 of 45 to generate the 'XXEVAL' instruction on power10 +and power11 instead of using the Altivec 'VANDC' instruction feeding +into 'VAND'. The 'XXEVAL' instruction can use all 64 vector registers, +instead of the 32 registers that traditional Altivec vector +instructions use. By allowing all of the vector registers to be used, +it reduces the amount of spilling that a large benchmark generated. + +For vectors such as: + + vector int a, b, c, d; + +This patch will generate a call to xxeval if any of the registers are allocated +to traditional floating point registers: + + Code: Old: New: + ===== ==== ==== + a = ~ ((~ (c ^ d)) | b); veqv t,c,d xxeval a,b,c,d,96 + vnor a,t,b + + a = (c ^ d) ^ b; vxor t,c,d xxeval a,b,c,d,105 + vxor a,t,b + + a = (c ^ d) | b; vxor t,c,d xxeval a,b,c,d,111 + vor a,t,b + + a = ~ ((~ (c | d)) | b); vnor t,c,d xxeval a,b,c,d,112 + vnor a,t,b + + a = (c | d) ^ b; vor t,c,d xxeval a,b,c,d,120 + vxor a,t,b + + a = (c | d) | b; vor t,c,d xxeval a,b,c,d,127 + vor a,t,b + +Since fusion using 2 Altivec instructions is slightly faster than using +the 'XXEVAL' instruction we prefer to generate the Altivec instructions +if we can. In addition, because 'XXEVAL' is a prefixed instruction, it +possibly might generate an extra NOP instruction to align the 'XXEVAL' +instruction. + +I have tested these patches on both big endian and little endian +PowerPC servers, with no regressions. Can I check these patchs into +the trunk? + +2026-06-23 Michael Meissner <[email protected]> + +gcc/ + + PR target/117251 + * config/rs6000/fusion.md: Regenerate. + * config/rs6000/genfusion.pl (gen_logical_addsubf): Add support + to generate vector/vector fusion if XXEVAL is supported. + +==================== Branch work248-sha, patch #403 ==================== + +PR target/117251: Improve vector fusion #3 + +See the following post for a complete explanation of what the patches +for PR target/117251: + + * https://gcc.gnu.org/pipermail/gcc-patches/2025-June/686474.html + +This is patch #2 of 45 to generate the 'XXEVAL' instruction on power10 +and power11 instead of using the Altivec 'VANDC' instruction feeding +into 'VAND'. The 'XXEVAL' instruction can use all 64 vector registers, +instead of the 32 registers that traditional Altivec vector +instructions use. By allowing all of the vector registers to be used, +it reduces the amount of spilling that a large benchmark generated. + +For vectors such as: + + vector int a, b, c, d; + +This patch will generate a call to xxeval if any of the registers are allocated +to traditional floating point registers: + + Code: Old: New: + ===== ==== ==== + a = ~ ((~ (c & d)) | b); vnan xxeval a,b,c,d,16d t,c,d + vnor a,t,b + + a = (c & d) ^ b; vand t,c,d xxeval a,b,c,d,30 + vxor a,t,b + + a = (c & d) | b; vand t,c,d xxeval a,b,c,d,31 + vor a,t,b + + a = (c & ~ d) ^ b; vandc t,c,d xxeval a,b,c,d,45 + vxor a,t,b + + a = (c & ~ d) | b; vandc t,c,d xxeval a,b,c,d,47 + vor a,t,b + + a = ~ ((c | ~ d) | b); vorc t,c,d xxeval a,b,c,d,64 + vnor a,t,b + + a = ~ ((c | ~ d) ^ b); vorc t,c,d xxeval a,b,c,d,75 + veqv a,t,b + + a = (c | ~ d) | ~ b; vorc t,c,d xxeval a,b,c,d,79 + vorc a,t,b + +Since fusion using 2 Altivec instructions is slightly faster than using +the 'XXEVAL' instruction we prefer to generate the Altivec instructions +if we can. In addition, because 'XXEVAL' is a prefixed instruction, it +possibly might generate an extra NOP instruction to align the 'XXEVAL' +instruction. + +I have tested these patches on both big endian and little endian +PowerPC servers, with no regressions. Can I check these patchs into +the trunk? + +2026-06-23 Michael Meissner <[email protected]> + +gcc/ + + PR target/117251 + * config/rs6000/fusion.md: Regenerate. + * config/rs6000/genfusion.pl (gen_logical_addsubf): Add support + to generate vector/vector fusion if XXEVAL is supported. + +==================== Branch work248-sha, patch #402 ==================== + +PR target/117251: Improve vector fusion #2 + +See the following post for a complete explanation of what the patches +for PR target/117251: + + * https://gcc.gnu.org/pipermail/gcc-patches/2025-June/686474.html + +This is patch #2 of 45 to generate the 'XXEVAL' instruction on power10 +and power11 instead of using the Altivec 'VANDC' instruction feeding +into 'VAND'. The 'XXEVAL' instruction can use all 64 vector registers, +instead of the 32 registers that traditional Altivec vector +instructions use. By allowing all of the vector registers to be used, +it reduces the amount of spilling that a large benchmark generated. + +For vectors such as: + + vector int a, b, c, d; + +This patch will generate a call to xxeval if any of the registers are allocated +to traditional floating point registers: + + Code: Old: New: + ===== ==== ==== + a = (c & ~ d) & b; vandc t,c,d xxeval a,b,c,d,2 + vand a,t,b + + a = (c ^ d) & b; vxor t,c,d xxeval a,b,c,d,6 + vand a,t,b + + a = (c | d) & b; vor t,c,d xxeval a,b,c,d,7 + vand a,t,b + + a = (~ (c | d)) & b; vnor t,c,d xxeval a,b,c,d,8 + vand a,t,b + + a = (~ (c ^ d)) & b; veqv t,c,d xxeval a,b,c,d,9 + vand a,t,b + + a = (c | ~ d) & b; vorc t,c,d xxeval a,b,c,d,11 + vand a,t,b + + a = (c & ~ d) & ~ b; vandc t,c,d xxeval a,b,c,d,13 + vandc a,t,b + + a = (~ (c & d)) & b; vnand t,c,d xxeval a,b,c,d,14 + vand a,t,b + +Since fusion using 2 Altivec instructions is slightly faster than using +the 'XXEVAL' instruction we prefer to generate the Altivec instructions +if we can. In addition, because 'XXEVAL' is a prefixed instruction, it +possibly might generate an extra NOP instruction to align the 'XXEVAL' +instruction. + +I have tested these patches on both big endian and little endian +PowerPC servers, with no regressions. Can I check these patchs into +the trunk? + +2026-06-23 Michael Meissner <[email protected]> + +gcc/ + + PR target/117251 + * config/rs6000/fusion.md: Regenerate. + * config/rs6000/genfusion.pl (gen_logical_addsubf): Add support + to generate vector/vector fusion if XXEVAL is supported. + +==================== Branch work248-sha, patch #401 ==================== + +PR target/117251: Improve vector fusion #1 + +See the following post for a complete explanation of what the patches +for PR target/117251: + + * https://gcc.gnu.org/pipermail/gcc-patches/2025-June/686474.html + +This is patch #1 of 45 to generate the 'XXEVAL' instruction on power10 +and power11 instead of using the Altivec 'VAND' instruction feeding +into 'VAND'. The 'XXEVAL' instruction can use all 64 vector registers, +instead of the 32 registers that traditional Altivec vector +instructions use. By allowing all of the vector registers to be used, +it reduces the amount of spilling that a large benchmark generated. + +Currently the following code: + + vector int a, b, c, d; + a = (c & d) & b; + +Generates: + + vand t,c,d + vand a,t,b + +Now in addition with this patch, if the arguments or result is +allocated to a traditional FPR register, the GCC compiler will now +generate the following code instead of adding vector move instructions: + + xxeval a,b,c,1 + +Since fusion using 2 Altivec instructions is slightly faster than using +the 'XXEVAL' instruction we prefer to generate the Altivec instructions +if we can. In addition, because 'XXEVAL' is a prefixed instruction, it +possibly might generate an extra NOP instruction to align the 'XXEVAL' +instruction. + +I have tested these patches on both big endian and little endian +PowerPC servers, with no regressions. Can I check these patchs into +the trunk? + +2026-06-23 Michael Meissner <[email protected]> + +gcc/ + + PR target/117251 + * config/rs6000/fusion.md: Regenerate. + * config/rs6000/genfusion.pl (gen_logical_addsubf): Add + support to generate vector/vector and/and fusion if XXEVAL is + supported. + * config/rs6000/predicates.md (vector_fusion_operand): New + predicate. + * config/rs6000/rs6000.h (TARGET_XXEVAL): New macro. + * config/rs6000/rs6000.md (isa attribute): Add xxeval. + (enabled attribute): Add support for XXEVAL support. + +==================== Branch work248-sha, explanation ==================== + +PR target/117251: Add PowerPC XXEVAL support to speed up SHA3 calculations + +History: This is version 2 of the patch. In the original patch, all 44 +fusion opportunities were lumped together in one patch. Outside of +fusion.md, these changes are fairly small, in that it adds one +alternative to each of the fusion patterns to add xxeval support. +Fusion.md is a generated file (created from genfusion.md) that does all +of the fusion combinations. Because of these automated changes, +fusion.md had 265 lines that were deleted and 397 lines that were +added. + +In version 2 of the patch, I broke the original patch into 45 separate +patches. The first patch adds the basic support to genfusion.pl, +predicates.md, rs6000.h, and rs6000.md. The first patch adds the first +fusion case (vector 'AND' fusing into vector 'AND'). The next 43 +patches each add one more fusion case. Then the last case adds the two +test cases. + +The multibuff.c benchmark attached to the PR target/117251 compiled for +Power10 PowerPC that implement SHA3 has a slowdown in the current trunk +and GCC 14 compared to GCC 11 - GCC 13, due to excessive amounts of +spilling. + +The main function for the multibuf.c file has 3,747 lines, all of which +are using vector unsigned long long. There are 696 vector rotates (all +rotates are constant), 1,824 vector xor's and 600 vector andc's. + +In looking at it, the main thing that steps out is the reason for +either spilling or moving variables is the support in fusion.md +(generated by genfusion.pl) that tries to fuse the vec_andc feeding +into vec_xor, and other vec_xor's feeding into vec_xor. + +On the powerpc for power10, there is a special fusion mode that happens +if the machine has a VANDC or VXOR instruction that is adjacent to a +VXOR instruction and the VANDC/VXOR feeds into the 2nd VXOR +instruction. + +While the Power10 has 64 vector registers (which uses the XXL prefix to +do logical operations), the fusion only works with the older Altivec +instruction set (which uses the V prefix). The Altivec instruction +only has 32 vector registers (which are overlaid over the VSX vector +registers 32-63). + +By having the combiner patterns fuse_vandc_vxor and fuse_vxor_vxor to +do this fusion, it means that the register allocator has more register +pressure for the traditional Altivec registers instead of the VSX +registers. + +In addition, since there are vector rotates, these rotates only work on +the traditional Altivec registers, which adds to the Altivec register +pressure. + +Finally in addition to doing the explicit xor, andc, and rotates using +the Altivec registers, we have to also load vector constants for the +rotate amount and these registers also are allocated as Altivec +registers. + +Current trunk and GCC 12-14 have more vector spills than GCC 11, but +GCC 11 has many more vector moves that the later compilers. Thus even +though it has way less spills, the vector moves are why GCC 11 have the +slowest results. + +There is an instruction that was added in power10 (XXEVAL) that does +provide fusion between VSX vectors that includes ANDC->XOR and XOR->XOR +fusion. + +The latency of XXEVAL is slightly more than the fused VANDC/VXOR or +VXOR/VXOR, so I have written the patch to prefer doing the Altivec +instructions if they don't need a temporary register. + +Here are the results for adding support for XXEVAL for the multibuff.c +benchmark attached to the PR. Note that we essentially recover the +speed with this patch that were lost with GCC 14 and the current trunk: + + XXEVAL Trunk GCC15 GCC14 GCC13 + ------ ----- ----- ----- ----- +Multibuf time in seconds 5.600 6.151 6.129 6.053 5.539 +XXEVAL improvement percentage --- +9.8% +9.4% +8.1% -1.1% + +Fuse VANDC -> VXOR 209 600 600 600 600 +Fuse VXOR -> VXOR 0 241 241 240 120 +XXEVAL to fuse ANDC -> XOR (#45) 391 0 0 0 0 +XXEVAL to fuse XOR -> XOR (#105) 240 0 0 0 0 + +Spill vector to stack 140 417 417 403 226 +Load spilled vector from stack 490 1,012 1,012 1,000 766 +Vector moves 8 93 100 70 72 + +XXLANDC or VANDC 209 600 600 600 600 +XXLXOR or VXOR 953 1,824 1,824 1,824 1,824 +XXEVAL 631 0 0 0 0 + + +Here are the results for adding support for XXEVAL for the singlebuff.c +benchmark attached to the PR. Note that adding XXEVAL greatly speeds +up this particular benchmark: + + XXEVAL Trunk GCC15 GCC14 GCC13 + ------ ----- ----- ----- ----- +Singlebuf time in seconds 4.429 5.330 5.333 5.315 5.270 +XXEVAL improvement percentage --- +20.3% +20.4% +20.0% +19.0% + +Fuse VANDC -> VXOR 210 600 600 600 600 +Fuse VXOR -> VXOR 0 240 240 240 120 +XXEVAL to fuse ANDC -> XOR (#45) 390 0 0 0 0 +XXEVAL to fuse XOR -> XOR (#105) 240 0 0 0 0 + +Spill vector to stack 134 388 388 388 391 +Load spilled vector from stack 357 808 808 808 769 +Vector moves 34 80 80 80 119 + +XXLANDC or VANDC 210 600 600 600 600 +XXLXOR or VXOR 954 1,824 1,824 1,824 1,824 +XXEVAL 630 0 0 0 0 + + +These patches add the following fusion patterns: + + xxland => xxland xxlandc => xxland + xxlxor => xxland xxlor => xxland + xxlnor => xxland xxleqv => xxland + xxlorc => xxland xxlandc => xxlandc + xxlnand => xxland xxlnand => xxlnor + xxland => xxlxor xxland => xxlor + xxlandc => xxlxor xxlandc => xxlor + xxlorc => xxlnor xxlorc => xxleqv + xxlorc => xxlorc xxleqv => xxlnor + xxlxor => xxlxor xxlxor => xxlor + xxlnor => xxlnor xxlor => xxlxor + xxlor => xxlor xxlor => xxlnor + xxlnor => xxlxor xxlnor => xxlor + xxlxor => xxlnor xxleqv => xxlxor + xxleqv => xxlor xxlorc => xxlxor + xxlorc => xxlor xxlandc => xxlnor + xxlandc => xxleqv xxland => xxlnor + xxlnand => xxlxor xxlnand => xxlor + xxlnand => xxlnand xxlorc => xxlnand + xxleqv => xxlnand xxlnor => xxlnand + xxlor => xxlnand xxlxor => xxlnand + xxlandc => xxlnand xxland => xxlnand + ==================== Branch work248-sha, baseline ==================== 2026-06-23 Michael Meissner <[email protected]>
