BTW, I'd fear that with system GCC 10, one won't be able to build any earlier GCCs unless patching libcpp/lex.c :(.

Created attachment 46362 [details]
A patch

Please try this.

With that patch what previously failed bootstrap passes it now.
Compared to a build from 24 hours ago and ignoring FAILs introduces during that time also on x86_64-linux, I see:
+FAIL: gcc.target/i386/3dnow-1.c (internal compiler error)
+FAIL: gcc.target/i386/3dnow-1.c (test for excess errors)
+FAIL: gcc.target/i386/3dnow-2.c (internal compiler error)
+FAIL: gcc.target/i386/3dnow-2.c (test for excess errors)
that is most likely related either to this patch or more likely to the MMX in SSE changes.

spawn -ignore SIGHUP /home/jakub/src/gcc/obj45/gcc/xgcc -B/home/jakub/src/gcc/obj45/gcc/ -fno-diagnostics-show-caret -fno-diagnostics-show-line-numbers -fdiagnostics-color=never -O2 -Werror-implicit-function-declaration -m3dnow -c -o 3dnow-1.o /home/jakub/src/gcc/gcc/testsuite/gcc.target/i386/3dnow-1.c
In file included from /home/jakub/src/gcc/gcc/testsuite/gcc.target/i386/3dnow-1.c:14:
/home/jakub/src/gcc/obj45/gcc/include/mm3dnow.h: In function '_m_pavgusb':
/home/jakub/src/gcc/obj45/gcc/include/mm3dnow.h:53:1: error: unrecognizable insn:
(insn 7 4 8 2 (set (reg:V8QI 88)
        (truncate:V8QI (lshiftrt:V8HI (plus:V8HI (plus:V8HI (zero_extend:V8HI (subreg:V8QI (reg/v:V2SI 86 [ __A ]) 0))
                        (zero_extend:V8HI (subreg:V8QI (reg/v:V2SI 87 [ __B ]) 0)))
                    (const_vector:V8HI [
                            (const_int 1 [0x1]) repeated x8
                        ]))
                (const_int 1 [0x1])))) "/home/jakub/src/gcc/obj45/gcc/include/mm3dnow.h":52:17 -1
     (nil))
during RTL pass: vregs
/home/jakub/src/gcc/obj45/gcc/include/mm3dnow.h:53:1: internal compiler error: in extract_insn, at recog.c:2310
0x82dac28 _fatal_insn(char const*, rtx_def const*, char const*, int, char const*)
        ../../gcc/rtl-error.c:108
0x82daca5 _fatal_insn_not_found(rtx_def const*, char const*, int, char const*)
        ../../gcc/rtl-error.c:116
0x82cb6d5 extract_insn(rtx_insn*)
        ../../gcc/recog.c:2310
0x8848ebd instantiate_virtual_regs_in_insn
        ../../gcc/function.c:1605
0x8848ebd instantiate_virtual_regs
        ../../gcc/function.c:1975
0x8848ebd execute
        ../../gcc/function.c:2024
Please submit a full bug report,
with preprocessed source if appropriate.
Please include the complete backtrace with any bug report.
See <https://gcc.gnu.org/bugs/> for instructions.
compiler exited with status 1
FAIL: gcc.target/i386/3dnow-1.c (internal compiler error)
FAIL: gcc.target/i386/3dnow-1.c (test for excess errors)

Created attachment 46366 [details]
A new patch

Please try this.

Looks good on i686-linux, results in
FAIL: gcc.target/i386/pr90497-2.c (test for excess errors)
Excess errors:
/home/jakub/src/gcc/gcc/testsuite/gcc.target/i386/pr90497-2.c:11:10: warning: implicit declaration of function '__builtin_ia32_pavgusb'; did you mean '__builtin_ia32_paddusb'? [-Wimplicit-function-declaration]
/home/jakub/src/gcc/gcc/testsuite/gcc.target/i386/pr90497-2.c:11:10: error: incompatible types when returning type 'int' but '__v8qi' {aka '__vector(8) char'} was expected
on x86_64-linux, so that test needs some work.

Author: hjl
Date: Fri May 17 14:48:37 2019
New Revision: 271328

URL: https://gcc.gnu.org/viewcvs?rev=271328&root=gcc&view=rev
Log:
i386: Enable MMX intrinsics without SSE/SSE2/SSSE3

Since MMX intrinsics are marked with SSE/SSE2/SSSE3 for SSE emulation,
enable them without SSE/SSE2/SSSE3 if MMX is enabled.

Restore TARGET_3DNOW check, which was changed to TARGET_3DNOW_A by
revision 271235.

gcc/

	PR target/90497
	* config/i386/i386-expand.c (ix86_expand_builtin): Enable MMX
	intrinsics without SSE/SSE2/SSSE3.
	* config/i386/mmx.md (mmx_uavgv8qi3): Restore TARGET_3DNOW
	check.
	(*mmx_uavgv8qi3): Likewise.

gcc/testsuite/

	PR target/90497
	* gcc.target/i386/pr90497-1.c: New test.
	* gcc.target/i386/pr90497-2.c: Likewise.

Added:
    trunk/gcc/testsuite/gcc.target/i386/pr90497-1.c
    trunk/gcc/testsuite/gcc.target/i386/pr90497-2.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/i386/i386-expand.c
    trunk/gcc/config/i386/mmx.md
    trunk/gcc/testsuite/ChangeLog

Fixed.

Why do we have

define_expand "sse2_umulv1siv1di3"
  [(set (match_operand:V1DI 0 "register_operand")
        (mult:V1DI
          (zero_extend:V1DI
            (vec_select:V1SI
              (match_operand:V2SI 1 "nonimmediate_operand")
              (parallel [(const_int 0)])))
          (zero_extend:V1DI
            (vec_select:V1SI
              (match_operand:V2SI 2 "nonimmediate_operand")
              (parallel [(const_int 0)])))))]
  "TARGET_SSE2"
  "ix86_fixup_binary_operands_no_copy (MULT, V2SImode, operands);")

i386-builtin.def:BDESC (OPTION_MASK_ISA_SSE2 | OPTION_MASK_ISA_MMX, 0, CODE_FOR_sse2_umulv1siv1di3, "__builtin_ia32_pmuludq", IX86_BUILTIN_PMULUDQ, UNKNOWN, (int) V1DI_FTYPE_V2SI_V2SI)

sse2_umulv1siv1di3 isn't available with MMX.

That changed in PR79565 / PR82483, but I'm afraid I don't remember that change much.  Guess it has something to do with the patterns using V2SImode which wasn't enabled when MMX was disabled.

See https://gcc.gnu.org/legacy-ml/gcc-patches/2017-10/msg00576.html and
https://gcc.gnu.org/legacy-ml/gcc-patches/2017-10/msg00522.html for details.