)]}'
{
  "commit": "730f52e05a1fb5c8cd92e352e9b191a6332be5c2",
  "tree": "02681e0721aae3013e14b323f38bbafaf26e4ceb",
  "parents": [
    "2c564e813c0626802e5bfb066c094933d5e6a774"
  ],
  "author": {
    "name": "Roger Sayle",
    "email": "roger@nextmovesoftware.com",
    "time": "Fri Oct 15 19:09:25 2021 +0100"
  },
  "committer": {
    "name": "Roger Sayle",
    "email": "roger@nextmovesoftware.com",
    "time": "Fri Oct 15 19:09:25 2021 +0100"
  },
  "message": "Allow early sets of SSE hard registers from standard_sse_constant_p.\n\nMy previous patch, which was intended to reduce the differences seen by\nthe combination of -march\u003dcascadelake and -m32, has additionally found\nsome more instances where this combination behaves differently to regular\nx86_64-pc-linux-gnu.  The middle-end always, and backends usually, use\nemit_move_insn to emit/expand move instructions allowing the backend\ncontrol over placing things in constant pools, adding REG_EQUAL notes,\nand so on.  Several of the AVX512 built-in expanders bypass this logic,\nand instead generate moves directly using emit_insn(gen_rtx_SET (dst,src)).\n\nFor example, i386-expand.c line 12004 contains:\n      for (i \u003d 0; i \u003c 8; i++)\n        emit_insn (gen_rtx_SET (xmm_regs[i], const0_rtx));\n\nI suspect that in this case, loading of standard_sse_constant_p, my\nchange to require loading of likely spilled hard registers via a\npseudo is perhaps overly strict, so this patch/fix reallows these\nimmediate constants values to be loaded directly prior to reload.\n\n2021-10-15  Roger Sayle  \u003croger@nextmovesoftware.com\u003e\n\ngcc/ChangeLog\n\t* config/i386/i386.c (ix86_hardreg_mov_ok): For vector modes,\n\tallow standard_sse_constant_p immediate constants.\n",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "fb656094e9ef9bd42912201c30c27431624cb275",
      "old_mode": 33188,
      "old_path": "gcc/config/i386/i386.c",
      "new_id": "9cc903e826b4bf13589fd935b52c6d3a3dc33bbc",
      "new_mode": 33188,
      "new_path": "gcc/config/i386/i386.c"
    }
  ]
}