From d934c16d8a1e0fb82fd4abfa54dcb5217430577c Mon Sep 17 00:00:00 2001 From: Taylor Simpson Date: Thu, 8 Apr 2021 20:07:45 -0500 Subject: [PATCH] Hexagon (target/hexagon) add F2_sfrecipa instruction Rd32,Pe4 = sfrecipa(Rs32, Rt32) Recripocal approx Test cases in tests/tcg/hexagon/multi_result.c FP exception tests added to tests/tcg/hexagon/fpstuff.c Signed-off-by: Taylor Simpson Reviewed-by: Richard Henderson Message-Id: <1617930474-31979-18-git-send-email-tsimpson@quicinc.com> Signed-off-by: Richard Henderson --- target/hexagon/arch.c | 26 +++++++++-- target/hexagon/arch.h | 2 + target/hexagon/gen_tcg.h | 21 +++++++++ target/hexagon/helper.h | 1 + target/hexagon/imported/encode_pp.def | 1 + target/hexagon/imported/float.idef | 16 +++++++ target/hexagon/op_helper.c | 37 ++++++++++++++++ tests/tcg/hexagon/Makefile.target | 1 + tests/tcg/hexagon/fpstuff.c | 82 +++++++++++++++++++++++++++++++++++ tests/tcg/hexagon/multi_result.c | 68 +++++++++++++++++++++++++++++ 10 files changed, 252 insertions(+), 3 deletions(-) create mode 100644 tests/tcg/hexagon/multi_result.c diff --git a/target/hexagon/arch.c b/target/hexagon/arch.c index 40b6e3d0c0..46edf45b13 100644 --- a/target/hexagon/arch.c +++ b/target/hexagon/arch.c @@ -181,12 +181,13 @@ int arch_sf_recip_common(float32 *Rs, float32 *Rt, float32 *Rd, int *adjust, /* or put Inf in num fixup? */ uint8_t RsV_sign = float32_is_neg(RsV); uint8_t RtV_sign = float32_is_neg(RtV); + /* Check that RsV is NOT infinite before we overwrite it */ + if (!float32_is_infinity(RsV)) { + float_raise(float_flag_divbyzero, fp_status); + } RsV = infinite_float32(RsV_sign ^ RtV_sign); RtV = float32_one; RdV = float32_one; - if (float32_is_infinity(RsV)) { - float_raise(float_flag_divbyzero, fp_status); - } } else if (float32_is_infinity(RtV)) { RsV = make_float32(0x80000000 & (RsV ^ RtV)); RtV = float32_one; @@ -279,3 +280,22 @@ int arch_sf_invsqrt_common(float32 *Rs, float32 *Rd, int *adjust, *adjust = PeV; return ret; } + +const uint8_t recip_lookup_table[128] = { + 0x0fe, 0x0fa, 0x0f6, 0x0f2, 0x0ef, 0x0eb, 0x0e7, 0x0e4, + 0x0e0, 0x0dd, 0x0d9, 0x0d6, 0x0d2, 0x0cf, 0x0cc, 0x0c9, + 0x0c6, 0x0c2, 0x0bf, 0x0bc, 0x0b9, 0x0b6, 0x0b3, 0x0b1, + 0x0ae, 0x0ab, 0x0a8, 0x0a5, 0x0a3, 0x0a0, 0x09d, 0x09b, + 0x098, 0x096, 0x093, 0x091, 0x08e, 0x08c, 0x08a, 0x087, + 0x085, 0x083, 0x080, 0x07e, 0x07c, 0x07a, 0x078, 0x075, + 0x073, 0x071, 0x06f, 0x06d, 0x06b, 0x069, 0x067, 0x065, + 0x063, 0x061, 0x05f, 0x05e, 0x05c, 0x05a, 0x058, 0x056, + 0x054, 0x053, 0x051, 0x04f, 0x04e, 0x04c, 0x04a, 0x049, + 0x047, 0x045, 0x044, 0x042, 0x040, 0x03f, 0x03d, 0x03c, + 0x03a, 0x039, 0x037, 0x036, 0x034, 0x033, 0x032, 0x030, + 0x02f, 0x02d, 0x02c, 0x02b, 0x029, 0x028, 0x027, 0x025, + 0x024, 0x023, 0x021, 0x020, 0x01f, 0x01e, 0x01c, 0x01b, + 0x01a, 0x019, 0x017, 0x016, 0x015, 0x014, 0x013, 0x012, + 0x011, 0x00f, 0x00e, 0x00d, 0x00c, 0x00b, 0x00a, 0x009, + 0x008, 0x007, 0x006, 0x005, 0x004, 0x003, 0x002, 0x000, +}; diff --git a/target/hexagon/arch.h b/target/hexagon/arch.h index 6e0b0d9a24..b6634e9615 100644 --- a/target/hexagon/arch.h +++ b/target/hexagon/arch.h @@ -30,4 +30,6 @@ int arch_sf_recip_common(float32 *Rs, float32 *Rt, float32 *Rd, int arch_sf_invsqrt_common(float32 *Rs, float32 *Rd, int *adjust, float_status *fp_status); +extern const uint8_t recip_lookup_table[128]; + #endif diff --git a/target/hexagon/gen_tcg.h b/target/hexagon/gen_tcg.h index a30048ee57..428a670281 100644 --- a/target/hexagon/gen_tcg.h +++ b/target/hexagon/gen_tcg.h @@ -195,6 +195,27 @@ #define fGEN_TCG_S4_stored_locked(SHORTCODE) \ do { SHORTCODE; READ_PREG(PdV, PdN); } while (0) +/* + * Mathematical operations with more than one definition require + * special handling + */ + +/* + * Approximate reciprocal + * r3,p1 = sfrecipa(r0, r1) + * + * The helper packs the 2 32-bit results into a 64-bit value, + * so unpack them into the proper results. + */ +#define fGEN_TCG_F2_sfrecipa(SHORTCODE) \ + do { \ + TCGv_i64 tmp = tcg_temp_new_i64(); \ + gen_helper_sfrecipa(tmp, cpu_env, RsV, RtV); \ + tcg_gen_extrh_i64_i32(RdV, tmp); \ + tcg_gen_extrl_i64_i32(PeV, tmp); \ + tcg_temp_free_i64(tmp); \ + } while (0) + /* Floating point */ #define fGEN_TCG_F2_conv_sf2df(SHORTCODE) \ gen_helper_conv_sf2df(RddV, cpu_env, RsV) diff --git a/target/hexagon/helper.h b/target/hexagon/helper.h index efe6069118..b377293dd3 100644 --- a/target/hexagon/helper.h +++ b/target/hexagon/helper.h @@ -24,6 +24,7 @@ DEF_HELPER_FLAGS_3(debug_check_store_width, TCG_CALL_NO_WG, void, env, int, int) DEF_HELPER_FLAGS_3(debug_commit_end, TCG_CALL_NO_WG, void, env, int, int) DEF_HELPER_2(commit_store, void, env, int) DEF_HELPER_FLAGS_4(fcircadd, TCG_CALL_NO_RWG_SE, s32, s32, s32, s32, s32) +DEF_HELPER_3(sfrecipa, i64, env, f32, f32) /* Floating point */ DEF_HELPER_2(conv_sf2df, f64, env, f32) diff --git a/target/hexagon/imported/encode_pp.def b/target/hexagon/imported/encode_pp.def index c21cb730af..b01b4d7aa7 100644 --- a/target/hexagon/imported/encode_pp.def +++ b/target/hexagon/imported/encode_pp.def @@ -1028,6 +1028,7 @@ MPY_ENC(F2_sfmin, "1011","ddddd","0","0","0","1","01") MPY_ENC(F2_sfmpy, "1011","ddddd","0","0","1","0","00") MPY_ENC(F2_sffixupn, "1011","ddddd","0","0","1","1","00") MPY_ENC(F2_sffixupd, "1011","ddddd","0","0","1","1","01") +MPY_ENC(F2_sfrecipa, "1011","ddddd","1","1","1","1","ee") DEF_FIELDROW_DESC32(ICLASS_M" 1100 -------- PP------ --------","[#12] Rd=(Rs,Rt)") DEF_FIELD32(ICLASS_M" 1100 -------- PP------ --!-----",Mc_tH,"Rt is High") /*Rt high */ diff --git a/target/hexagon/imported/float.idef b/target/hexagon/imported/float.idef index 76cecfebf5..eb5415801a 100644 --- a/target/hexagon/imported/float.idef +++ b/target/hexagon/imported/float.idef @@ -146,6 +146,22 @@ Q6INSN(F2_sfimm_n,"Rd32=sfmake(#u10):neg",ATTRIBS(), }) +Q6INSN(F2_sfrecipa,"Rd32,Pe4=sfrecipa(Rs32,Rt32)",ATTRIBS(), +"Reciprocal Approximation for Division", +{ + fHIDE(int idx;) + fHIDE(int adjust;) + fHIDE(int mant;) + fHIDE(int exp;) + if (fSF_RECIP_COMMON(RsV,RtV,RdV,adjust)) { + PeV = adjust; + idx = (RtV >> 16) & 0x7f; + mant = (fSF_RECIP_LOOKUP(idx) << 15) | 1; + exp = fSF_BIAS() - (fSF_GETEXP(RtV) - fSF_BIAS()) - 1; + RdV = fMAKESF(fGETBIT(31,RtV),exp,mant); + } +}) + Q6INSN(F2_sffixupn,"Rd32=sffixupn(Rs32,Rt32)",ATTRIBS(), "Fix Up Numerator", { diff --git a/target/hexagon/op_helper.c b/target/hexagon/op_helper.c index 33b67138ca..75861e26c4 100644 --- a/target/hexagon/op_helper.c +++ b/target/hexagon/op_helper.c @@ -289,6 +289,43 @@ int32_t HELPER(fcircadd)(int32_t RxV, int32_t offset, int32_t M, int32_t CS) return new_ptr; } +static float32 build_float32(uint8_t sign, uint32_t exp, uint32_t mant) +{ + return make_float32( + ((sign & 1) << 31) | + ((exp & 0xff) << SF_MANTBITS) | + (mant & ((1 << SF_MANTBITS) - 1))); +} + +/* + * sfrecipa, sfinvsqrta have two 32-bit results + * r0,p0=sfrecipa(r1,r2) + * r0,p0=sfinvsqrta(r1) + * + * Since helpers can only return a single value, we pack the two results + * into a 64-bit value. + */ +uint64_t HELPER(sfrecipa)(CPUHexagonState *env, float32 RsV, float32 RtV) +{ + int32_t PeV = 0; + float32 RdV; + int idx; + int adjust; + int mant; + int exp; + + arch_fpop_start(env); + if (arch_sf_recip_common(&RsV, &RtV, &RdV, &adjust, &env->fp_status)) { + PeV = adjust; + idx = (RtV >> 16) & 0x7f; + mant = (recip_lookup_table[idx] << 15) | 1; + exp = SF_BIAS - (float32_getexp(RtV) - SF_BIAS) - 1; + RdV = build_float32(extract32(RtV, 31, 1), exp, mant); + } + arch_fpop_end(env); + return ((uint64_t)RdV << 32) | PeV; +} + /* * mem_noshuf * Section 5.5 of the Hexagon V67 Programmer's Reference Manual diff --git a/tests/tcg/hexagon/Makefile.target b/tests/tcg/hexagon/Makefile.target index 616af697fe..18218ad05a 100644 --- a/tests/tcg/hexagon/Makefile.target +++ b/tests/tcg/hexagon/Makefile.target @@ -39,6 +39,7 @@ HEX_TESTS = first HEX_TESTS += misc HEX_TESTS += preg_alias HEX_TESTS += dual_stores +HEX_TESTS += multi_result HEX_TESTS += mem_noshuf HEX_TESTS += atomics HEX_TESTS += fpstuff diff --git a/tests/tcg/hexagon/fpstuff.c b/tests/tcg/hexagon/fpstuff.c index 6b60f92d18..8e3ba780d2 100644 --- a/tests/tcg/hexagon/fpstuff.c +++ b/tests/tcg/hexagon/fpstuff.c @@ -250,6 +250,87 @@ static void check_dfminmax(void) check_fpstatus(usr, FPINVF); } +static void check_recip_exception(void) +{ + int result; + int usr; + + /* + * Check that sfrecipa doesn't set status bits when + * a NaN with bit 22 non-zero is passed + */ + asm (CLEAR_FPSTATUS + "%0,p0 = sfrecipa(%2, %3)\n\t" + "%1 = usr\n\t" + : "=r"(result), "=r"(usr) : "r"(SF_NaN), "r"(SF_ANY) + : "r2", "p0", "usr"); + check32(result, SF_HEX_NAN); + check_fpstatus(usr, 0); + + asm (CLEAR_FPSTATUS + "%0,p0 = sfrecipa(%2, %3)\n\t" + "%1 = usr\n\t" + : "=r"(result), "=r"(usr) : "r"(SF_ANY), "r"(SF_NaN) + : "r2", "p0", "usr"); + check32(result, SF_HEX_NAN); + check_fpstatus(usr, 0); + + asm (CLEAR_FPSTATUS + "%0,p0 = sfrecipa(%2, %2)\n\t" + "%1 = usr\n\t" + : "=r"(result), "=r"(usr) : "r"(SF_NaN) + : "r2", "p0", "usr"); + check32(result, SF_HEX_NAN); + check_fpstatus(usr, 0); + + /* + * Check that sfrecipa doesn't set status bits when + * a NaN with bit 22 zero is passed + */ + asm (CLEAR_FPSTATUS + "%0,p0 = sfrecipa(%2, %3)\n\t" + "%1 = usr\n\t" + : "=r"(result), "=r"(usr) : "r"(SF_NaN_special), "r"(SF_ANY) + : "r2", "p0", "usr"); + check32(result, SF_HEX_NAN); + check_fpstatus(usr, FPINVF); + + asm (CLEAR_FPSTATUS + "%0,p0 = sfrecipa(%2, %3)\n\t" + "%1 = usr\n\t" + : "=r"(result), "=r"(usr) : "r"(SF_ANY), "r"(SF_NaN_special) + : "r2", "p0", "usr"); + check32(result, SF_HEX_NAN); + check_fpstatus(usr, FPINVF); + + asm (CLEAR_FPSTATUS + "%0,p0 = sfrecipa(%2, %2)\n\t" + "%1 = usr\n\t" + : "=r"(result), "=r"(usr) : "r"(SF_NaN_special) + : "r2", "p0", "usr"); + check32(result, SF_HEX_NAN); + check_fpstatus(usr, FPINVF); + + /* + * Check that sfrecipa properly sets divid-by-zero + */ + asm (CLEAR_FPSTATUS + "%0,p0 = sfrecipa(%2, %3)\n\t" + "%1 = usr\n\t" + : "=r"(result), "=r"(usr) : "r"(0x885dc960), "r"(0x80000000) + : "r2", "p0", "usr"); + check32(result, 0x3f800000); + check_fpstatus(usr, FPDBZF); + + asm (CLEAR_FPSTATUS + "%0,p0 = sfrecipa(%2, %3)\n\t" + "%1 = usr\n\t" + : "=r"(result), "=r"(usr) : "r"(0x7f800000), "r"(SF_ZERO) + : "r2", "p0", "usr"); + check32(result, 0x3f800000); + check_fpstatus(usr, 0); +} + static void check_canonical_NaN(void) { int sf_result; @@ -507,6 +588,7 @@ int main() check_compare_exception(); check_sfminmax(); check_dfminmax(); + check_recip_exception(); check_canonical_NaN(); check_float2int_convs(); diff --git a/tests/tcg/hexagon/multi_result.c b/tests/tcg/hexagon/multi_result.c new file mode 100644 index 0000000000..cb7dd313d4 --- /dev/null +++ b/tests/tcg/hexagon/multi_result.c @@ -0,0 +1,68 @@ +/* + * Copyright(c) 2019-2021 Qualcomm Innovation Center, Inc. All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, see . + */ + +#include + +static int sfrecipa(int Rs, int Rt, int *pred_result) +{ + int result; + int predval; + + asm volatile("%0,p0 = sfrecipa(%2, %3)\n\t" + "%1 = p0\n\t" + : "+r"(result), "=r"(predval) + : "r"(Rs), "r"(Rt) + : "p0"); + *pred_result = predval; + return result; +} + +int err; + +static void check(int val, int expect) +{ + if (val != expect) { + printf("ERROR: 0x%08x != 0x%08x\n", val, expect); + err++; + } +} + +static void check_p(int val, int expect) +{ + if (val != expect) { + printf("ERROR: 0x%02x != 0x%02x\n", val, expect); + err++; + } +} + +static void test_sfrecipa() +{ + int res; + int pred_result; + + res = sfrecipa(0x04030201, 0x05060708, &pred_result); + check(res, 0x59f38001); + check_p(pred_result, 0x00); +} + +int main() +{ + test_sfrecipa(); + + puts(err ? "FAIL" : "PASS"); + return err; +} -- 2.11.4.GIT