From a473c9593bf50cd5d9ed2d36ea0b6210fcf45690 Mon Sep 17 00:00:00 2001 From: Eric Christopher Date: Mon, 25 Oct 2010 21:17:59 +0000 Subject: [PATCH] Add and describe a quick script to migrate gcc builtins that clang doesn't know about to the default APIs for x86 vector operations. git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@117313 91177308-0d34-0410-b5e6-96231b3b80d8 --- www/builtins.py | 160 +++++++++++++++++++++++++++++++++++++++++ www/compatibility.html | 10 +++ 2 files changed, 170 insertions(+) create mode 100755 www/builtins.py diff --git a/www/builtins.py b/www/builtins.py new file mode 100755 index 0000000000..5e8577054f --- /dev/null +++ b/www/builtins.py @@ -0,0 +1,160 @@ +#!/usr/bin/env python + +import sys, fileinput + +err=0 + +# Giant associative set of builtin->intrinsic mappings where clang doesn't +# implement the builtin since the vector operation works by default. + +repl_map = { +'__builtin_ia32_addps': '_mm_add_ps', +'__builtin_ia32_addsd': '_mm_add_sd', +'__builtin_ia32_addpd': '_mm_add_pd', +'__builtin_ia32_addss': '_mm_add_ss', +'__builtin_ia32_paddb128': '_mm_add_epi8', +'__builtin_ia32_paddw128': '_mm_add_epi16', +'__builtin_ia32_paddd128': '_mm_add_epi32', +'__builtin_ia32_paddq128': '_mm_add_epi64', +'__builtin_ia32_subps': '_mm_sub_ps', +'__builtin_ia32_subsd': '_mm_sub_sd', +'__builtin_ia32_subpd': '_mm_sub_pd', +'__builtin_ia32_subss': '_mm_sub_ss', +'__builtin_ia32_psubb128': '_mm_sub_epi8', +'__builtin_ia32_psubw128': '_mm_sub_epi16', +'__builtin_ia32_psubd128': '_mm_sub_epi32', +'__builtin_ia32_psubq128': '_mm_sub_epi64', +'__builtin_ia32_mulsd': '_mm_mul_sd', +'__builtin_ia32_mulpd': '_mm_mul_pd', +'__builtin_ia32_mulps': '_mm_mul_ps', +'__builtin_ia32_mulss': '_mm_mul_ss', +'__builtin_ia32_pmullw128': '_mm_mullo_epi16', +'__builtin_ia32_divsd': '_mm_div_sd', +'__builtin_ia32_divpd': '_mm_div_pd', +'__builtin_ia32_divps': '_mm_div_ps', +'__builtin_ia32_subss': '_mm_div_ss', +'__builtin_ia32_andpd': '_mm_and_pd', +'__builtin_ia32_andps': '_mm_and_ps', +'__builtin_ia32_pand128': '_mm_and_si128', +'__builtin_ia32_andnpd': '_mm_andnot_pd', +'__builtin_ia32_andnps': '_mm_andnot_ps', +'__builtin_ia32_pandn128': '_mm_andnot_si128', +'__builtin_ia32_orpd': '_mm_or_pd', +'__builtin_ia32_orps': '_mm_or_ps', +'__builtin_ia32_por128': '_mm_or_si128', +'__builtin_ia32_xorpd': '_mm_xor_pd', +'__builtin_ia32_xorps': '_mm_xor_ps', +'__builtin_ia32_pxor128': '_mm_xor_si128', +'__builtin_ia32_cvtsd2ss': '_mm_cvtsd_ss', +'__builtin_ia32_cvtsi2sd': '_mm_cvtsi32_sd', +'__builtin_ia32_cvtss2sd': '_mm_cvtss_sd', +'__builtin_ia32_cvttsd2si': '_mm_cvttsd_si32', +'__builtin_ia32_vec_ext_v2df': '_mm_cvtsd_f64', +'__builtin_ia32_loadhpd': '_mm_loadh_pd', +'__builtin_ia32_loadlpd': '_mm_loadl_pd', +'__builtin_ia32_loadlv4si': '_mm_loadl_epi64', +'__builtin_ia32_cmpeqps': '_mm_cmpeq_ps', +'__builtin_ia32_cmpltps': '_mm_cmplt_ps', +'__builtin_ia32_cmpleps': '_mm_cmple_ps', +'__builtin_ia32_cmpgtps': '_mm_cmpgt_ps', +'__builtin_ia32_cmpgeps': '_mm_cmpge_ps', +'__builtin_ia32_cmpunordps': '_mm_cmpunord_ps', +'__builtin_ia32_cmpneqps': '_mm_cmpneq_ps', +'__builtin_ia32_cmpnltps': '_mm_cmpnlt_ps', +'__builtin_ia32_cmpnleps': '_mm_cmpnle_ps', +'__builtin_ia32_cmpngtps': '_mm_cmpngt_ps', +'__builtin_ia32_cmpordps': '_mm_cmpord_ps', +'__builtin_ia32_cmpeqss': '_mm_cmpeq_ss', +'__builtin_ia32_cmpltss': '_mm_cmplt_ss', +'__builtin_ia32_cmpless': '_mm_cmple_ss', +'__builtin_ia32_cmpunordss': '_mm_cmpunord_ss', +'__builtin_ia32_cmpneqss': '_mm_cmpneq_ss', +'__builtin_ia32_cmpnltss': '_mm_cmpnlt_ss', +'__builtin_ia32_cmpnless': '_mm_cmpnle_ss', +'__builtin_ia32_cmpngtss': '_mm_cmpngt_ss', +'__builtin_ia32_cmpngess': '_mm_cmpnge_ss', +'__builtin_ia32_cmpordss': '_mm_cmpord_ss', +'__builtin_ia32_movss': '_mm_move_ss', +'__builtin_ia32_movsd': '_mm_move_sd', +'__builtin_ia32_movhlps': '_mm_movehl_ps', +'__builtin_ia32_movlhps': '_mm_movelh_ps', +'__builtin_ia32_movqv4si': '_mm_move_epi64', +'__builtin_ia32_unpckhps': '_mm_unpackhi_ps', +'__builtin_ia32_unpckhpd': '_mm_unpackhi_pd', +'__builtin_ia32_punpckhbw128': '_mm_unpackhi_epi8', +'__builtin_ia32_punpckhwd128': '_mm_unpackhi_epi16', +'__builtin_ia32_punpckhdq128': '_mm_unpackhi_epi32', +'__builtin_ia32_punpckhqdq128': '_mm_unpackhi_epi64', +'__builtin_ia32_unpcklps': '_mm_unpacklo_ps', +'__builtin_ia32_unpcklpd': '_mm_unpacklo_pd', +'__builtin_ia32_punpcklbw128': '_mm_unpacklo_epi8', +'__builtin_ia32_punpcklwd128': '_mm_unpacklo_epi16', +'__builtin_ia32_punpckldq128': '_mm_unpacklo_epi32', +'__builtin_ia32_punpcklqdq128': '_mm_unpacklo_epi64', +'__builtin_ia32_cmpeqpd': '_mm_cmpeq_pd', +'__builtin_ia32_cmpltpd': '_mm_cmplt_pd', +'__builtin_ia32_cmplepd': '_mm_cmple_pd', +'__builtin_ia32_cmpgtpd': '_mm_cmpgt_pd', +'__builtin_ia32_cmpgepd': '_mm_cmpge_pd', +'__builtin_ia32_cmpunordpd': '_mm_cmpunord_pd', +'__builtin_ia32_cmpneqpd': '_mm_cmpneq_pd', +'__builtin_ia32_cmpnltpd': '_mm_cmpnlt_pd', +'__builtin_ia32_cmpnlepd': '_mm_cmpnle_pd', +'__builtin_ia32_cmpngtpd': '_mm_cmpngt_pd', +'__builtin_ia32_cmpngepd': '_mm_cmpnge_pd', +'__builtin_ia32_cmpordpd': '_mm_cmpord_pd', +'__builtin_ia32_cmpeqsd': '_mm_cmpeq_sd', +'__builtin_ia32_cmpltsd': '_mm_cmplt_sd', +'__builtin_ia32_cmplesd': '_mm_cmple_sd', +'__builtin_ia32_cmpunordsd': '_mm_cmpunord_sd', +'__builtin_ia32_cmpneqsd': '_mm_cmpneq_sd', +'__builtin_ia32_cmpnltsd': '_mm_cmpnlt_sd', +'__builtin_ia32_cmpnlesd': '_mm_cmpnle_sd', +'__builtin_ia32_cmpordsd': '_mm_cmpord_sd', +'__builtin_ia32_cvtsi642ss': '_mm_cvtsi64_ss', +'__builtin_ia32_cvttss2si64': '_mm_cvtss_si64', +'__builtin_ia32_shufps': '_mm_shuffle_ps', +'__builtin_ia32_shufpd': '_mm_shuffle_pd', +'__builtin_ia32_pshufhw': '_mm_shufflehi_epi16', +'__builtin_ia32_pshuflw': '_mm_shufflelo_epi16', +'__builtin_ia32_pshufd': '_mm_shuffle_epi32', +'__builtin_ia32_movshdup': '_mm_movehdup_ps', +'__builtin_ia32_movsldup': '_mm_moveldup_ps', +'__builtin_ia32_vec_set_v16qi': '_mm_insert_epi8', +'__builtin_ia32_vec_set_v8hi': '_mm_insert_epi16', +'__builtin_ia32_vec_set_v4si': '_mm_insert_epi32', +'__builtin_ia32_vec_set_v2di': '_mm_insert_epi64', +'__builtin_ia32_vec_set_v4hi': '_mm_insert_pi16', +'__builtin_ia32_vec_ext_v16qi': '_mm_extract_epi8', +'__builtin_ia32_vec_ext_v8hi': '_mm_extract_epi16', +'__builtin_ia32_vec_ext_v4si': '_mm_extract_epi32', +'__builtin_ia32_vec_ext_v2di': '_mm_extract_epi64', +'__builtin_ia32_vec_ext_v4hi': '_mm_extract_pi16', +'__builtin_ia32_vec_ext_v4sf': '_mm_extract_ps' +} + +# Special unhandled cases: +# __builtin_ia32_vec_ext_*(__P, idx) -> _mm_store_sd/_mm_storeh_pd +# depending on index. No abstract insert/extract for these oddly. +unhandled = [ +'__builtin_ia32_vec_ext_v2df', +'__builtin_ia32_vec_ext_v2si', +] + +def report_repl(builtin, repl): + sys.stderr.write("%s:%d: x86 builtin %s used, replaced with %s\n" % (fileinput.filename(), fileinput.filelineno(), builtin, repl)) + +def report_cant(builtin): + sys.stderr.write("%s:%d: x86 builtin %s used, too many replacements\n" % (fileinput.filename(), fileinput.filelineno(), builtin)) + +for line in fileinput.input(inplace=1): + for builtin, repl in repl_map.iteritems(): + if builtin in line: + line = line.replace(builtin, repl) + report_repl(builtin, repl) + for unh in unhandled: + if unh in line: + report_cant(unh) + sys.stdout.write(line) + +sys.exit(err) \ No newline at end of file diff --git a/www/compatibility.html b/www/compatibility.html index bd22f9a8dc..68fd025929 100644 --- a/www/compatibility.html +++ b/www/compatibility.html @@ -141,6 +141,16 @@ mentioning these, the fix is simple: switch to the *mmintrin.h functions.

architectures respectively. For these, make sure to use the <arm_neon.h> and <altivec.h> headers.

+

For x86 architectures this script should help with +the manual migration process. It will rewrite your source files in place to +use the APIs instead of builtin function calls. Just call it like this:

+ +
+  builtins.py *.c *.h
+
+ +

and it will rewrite all of the .c and .h files in the current directory to +use the API calls instead of calls like __builtin_ia32_paddw128.

Lvalue casts

-- 2.40.0