From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-vk1-xa32.google.com (mail-vk1-xa32.google.com [IPv6:2607:f8b0:4864:20::a32]) by sourceware.org (Postfix) with ESMTPS id 334D23848035 for ; Tue, 22 Jun 2021 02:39:36 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org 334D23848035 Received: by mail-vk1-xa32.google.com with SMTP id k16so880100vke.10 for ; Mon, 21 Jun 2021 19:39:36 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:mime-version:references:in-reply-to:from:date :message-id:subject:to:cc; bh=ojo/AM92AsQPpYelwilYgSuAw4FGGGxSm1g7aTqQGQs=; b=Rnfev7OlB+pCa1+AYNsMFbWcAFO57PX4RXZnqgaf6+psaoPkCy4lTQ1cNZPm28R3wG hbQoYc1w7IpVb0hp3ze9aSAcXVuUmTUY1V4NM/hjtxONVG07bSe58tQ1s3tMlUKTiXv1 1tTdL2gM3ge/goVenyIyymfOd+1wf8eqf6Fb/XpYdAfNN1x0AfdSD0W7ou0DUGC3HW+O SdPy7l7ie/NvSKgLYHVg55CzA3qsvCiH+43fKs7N8XVEKapNzHpzCAoIvPPnn4B1A9+I /smLNeOAJHzF2m0UG+Zf9R93sUUuWiYe71ICdXzKx/2IcxyQpUTtTl5aXrKoyVtzar40 f/kA== X-Gm-Message-State: AOAM533ftpERFIJvgVYL3KR0APE1qGdm/mBvmTH8fQzS5XWR8jCS6t2l oqBa9Hvn0CpW+hT+AYt6AM60PIcVuhdr6Z7dCmE= X-Google-Smtp-Source: ABdhPJyKUW/C3FjC9QYW8ocC8mPIfAA4t39/1WoMOKVzhogxmzo06aasde+0XMc9quK2rBhtc6ZiUSJnDp7U4gBX1Lg= X-Received: by 2002:a1f:ed85:: with SMTP id l127mr17605244vkh.16.1624329575773; Mon, 21 Jun 2021 19:39:35 -0700 (PDT) MIME-Version: 1.0 References: <20210617062912.89506-1-hongtao.liu@intel.com> In-Reply-To: From: Hongtao Liu Date: Tue, 22 Jun 2021 10:44:24 +0800 Message-ID: Subject: Re: [PATCH] Add vect_recog_popcount_pattern to handle mismatch between the vectorized popcount IFN and scalar popcount builtin. To: Richard Biener Cc: liuhongt , GCC Patches , "H. J. Lu" Content-Type: text/plain; charset="UTF-8" X-Spam-Status: No, score=-10.2 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, FREEMAIL_FROM, GIT_PATCH_0, RCVD_IN_DNSWL_NONE, SPF_HELO_NONE, SPF_PASS, TXREP autolearn=ham autolearn_force=no version=3.4.2 X-Spam-Checker-Version: SpamAssassin 3.4.2 (2018-09-13) on server2.sourceware.org X-BeenThere: gcc-patches@gcc.gnu.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Gcc-patches mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Tue, 22 Jun 2021 02:39:38 -0000 On Tue, Jun 22, 2021 at 10:43 AM Hongtao Liu wrote: > > On Mon, Jun 21, 2021 at 6:05 PM Richard Biener > wrote: > > > > On Thu, Jun 17, 2021 at 8:29 AM liuhongt wrote: > > > > > > The patch remove those pro- and demotions when backend support direct > > > optab. > > > > > > For i386: it enables vectorization for vpopcntb/vpopcntw and optimized > > > for vpopcntq. > > > > > > gcc/ChangeLog: > > > > > > PR tree-optimization/97770 > > > * tree-vect-patterns.c (vect_recog_popcount_pattern): > > > New. > > > (vect_recog_func vect_vect_recog_func_ptrs): Add new pattern. > > > > > > gcc/testsuite/ChangeLog: > > > > > > PR tree-optimization/97770 > > > * gcc.target/i386/avx512bitalg-pr97770-1.c: Remove xfail. > > > * gcc.target/i386/avx512vpopcntdq-pr97770-1.c: Remove xfail. > > > --- > > > .../gcc.target/i386/avx512bitalg-pr97770-1.c | 27 +++-- > > > .../i386/avx512vpopcntdq-pr97770-1.c | 9 +- > > > gcc/tree-vect-patterns.c | 110 ++++++++++++++++++ > > > 3 files changed, 127 insertions(+), 19 deletions(-) > > > > > > diff --git a/gcc/testsuite/gcc.target/i386/avx512bitalg-pr97770-1.c b/gcc/testsuite/gcc.target/i386/avx512bitalg-pr97770-1.c > > > index c83a477045c..d1beec4cdb4 100644 > > > --- a/gcc/testsuite/gcc.target/i386/avx512bitalg-pr97770-1.c > > > +++ b/gcc/testsuite/gcc.target/i386/avx512bitalg-pr97770-1.c > > > @@ -1,19 +1,18 @@ > > > /* PR target/97770 */ > > > /* { dg-do compile } */ > > > -/* { dg-options "-O2 -mavx512bitalg -mavx512vl -mprefer-vector-width=512" } */ > > > -/* Add xfail since no IFN for QI/HImode popcount */ > > > -/* { dg-final { scan-assembler-times "vpopcntb\[ \\t\]+\[^\\n\\r\]*xmm" 1 {xfail *-*-*} } } */ > > > -/* { dg-final { scan-assembler-times "vpopcntw\[ \\t\]+\[^\\n\\r\]*xmm" 1 {xfail *-*-*} } } */ > > > -/* { dg-final { scan-assembler-times "vpopcntb\[ \\t\]+\[^\\n\\r\]*ymm" 1 {xfail *-*-*} } } */ > > > -/* { dg-final { scan-assembler-times "vpopcntw\[ \\t\]+\[^\\n\\r\]*ymm" 1 {xfail *-*-*} } } */ > > > -/* { dg-final { scan-assembler-times "vpopcntb\[ \\t\]+\[^\\n\\r\]*zmm" 1 {xfail *-*-*} } } */ > > > -/* { dg-final { scan-assembler-times "vpopcntw\[ \\t\]+\[^\\n\\r\]*zmm" 1 {xfail *-*-*} } } */ > > > +/* { dg-options "-O2 -march=icelake-server -mprefer-vector-width=512" } */ > > > +/* { dg-final { scan-assembler-times "vpopcntb\[ \\t\]+\[^\\n\\r\]*xmm" 1 } } */ > > > +/* { dg-final { scan-assembler-times "vpopcntw\[ \\t\]+\[^\\n\\r\]*xmm" 1 } } */ > > > +/* { dg-final { scan-assembler-times "vpopcntb\[ \\t\]+\[^\\n\\r\]*ymm" 1 } } */ > > > +/* { dg-final { scan-assembler-times "vpopcntw\[ \\t\]+\[^\\n\\r\]*ymm" 1 } } */ > > > +/* { dg-final { scan-assembler-times "vpopcntb\[ \\t\]+\[^\\n\\r\]*zmm" 1 } } */ > > > +/* { dg-final { scan-assembler-times "vpopcntw\[ \\t\]+\[^\\n\\r\]*zmm" 1 } } */ > > > > > > #include > > > > > > void > > > __attribute__ ((noipa, optimize("-O3"))) > > > -popcountb_128 (char * __restrict dest, char* src) > > > +popcountb_128 (unsigned char * __restrict dest, unsigned char* src) > > > { > > > for (int i = 0; i != 16; i++) > > > dest[i] = __builtin_popcount (src[i]); > > > @@ -21,7 +20,7 @@ popcountb_128 (char * __restrict dest, char* src) > > > > > > void > > > __attribute__ ((noipa, optimize("-O3"))) > > > -popcountw_128 (short* __restrict dest, short* src) > > > +popcountw_128 (unsigned short* __restrict dest, unsigned short* src) > > > { > > > for (int i = 0; i != 8; i++) > > > dest[i] = __builtin_popcount (src[i]); > > > @@ -29,7 +28,7 @@ popcountw_128 (short* __restrict dest, short* src) > > > > > > void > > > __attribute__ ((noipa, optimize("-O3"))) > > > -popcountb_256 (char * __restrict dest, char* src) > > > +popcountb_256 (unsigned char * __restrict dest, unsigned char* src) > > > { > > > for (int i = 0; i != 32; i++) > > > dest[i] = __builtin_popcount (src[i]); > > > @@ -37,7 +36,7 @@ popcountb_256 (char * __restrict dest, char* src) > > > > > > void > > > __attribute__ ((noipa, optimize("-O3"))) > > > -popcountw_256 (short* __restrict dest, short* src) > > > +popcountw_256 (unsigned short* __restrict dest, unsigned short* src) > > > { > > > for (int i = 0; i != 16; i++) > > > dest[i] = __builtin_popcount (src[i]); > > > @@ -45,7 +44,7 @@ popcountw_256 (short* __restrict dest, short* src) > > > > > > void > > > __attribute__ ((noipa, optimize("-O3"))) > > > -popcountb_512 (char * __restrict dest, char* src) > > > +popcountb_512 (unsigned char * __restrict dest, unsigned char* src) > > > { > > > for (int i = 0; i != 64; i++) > > > dest[i] = __builtin_popcount (src[i]); > > > @@ -53,7 +52,7 @@ popcountb_512 (char * __restrict dest, char* src) > > > > > > void > > > __attribute__ ((noipa, optimize("-O3"))) > > > -popcountw_512 (short* __restrict dest, short* src) > > > +popcountw_512 (unsigned short* __restrict dest, unsigned short* src) > > > { > > > for (int i = 0; i != 32; i++) > > > dest[i] = __builtin_popcount (src[i]); > > > diff --git a/gcc/testsuite/gcc.target/i386/avx512vpopcntdq-pr97770-1.c b/gcc/testsuite/gcc.target/i386/avx512vpopcntdq-pr97770-1.c > > > index 63bb00d9b4a..dedd2e4c3d6 100644 > > > --- a/gcc/testsuite/gcc.target/i386/avx512vpopcntdq-pr97770-1.c > > > +++ b/gcc/testsuite/gcc.target/i386/avx512vpopcntdq-pr97770-1.c > > > @@ -1,13 +1,12 @@ > > > /* PR target/97770 */ > > > /* { dg-do compile } */ > > > -/* { dg-options "-O2 -mavx512vpopcntdq -mavx512vl -mprefer-vector-width=512" } */ > > > +/* { dg-options "-O2 -march=icelake-server -mprefer-vector-width=512" } */ > > > /* { dg-final { scan-assembler-times "vpopcntd\[ \\t\]+\[^\\n\\r\]*xmm" 1 } } */ > > > /* { dg-final { scan-assembler-times "vpopcntd\[ \\t\]+\[^\\n\\r\]*ymm" 1 } } */ > > > /* { dg-final { scan-assembler-times "vpopcntd\[ \\t\]+\[^\\n\\r\]*zmm" 1 } } */ > > > -/* Add xfail since current vectorizor cannot generate expected code for DImode popcount */ > > > -/* { dg-final { scan-assembler-times "vpopcntq\[ \\t\]+\[^\\n\\r\]*xmm" 1 { xfail *-*-* } } } */ > > > -/* { dg-final { scan-assembler-times "vpopcntq\[ \\t\]+\[^\\n\\r\]*ymm" 1 { xfail *-*-* } } } */ > > > -/* { dg-final { scan-assembler-times "vpopcntq\[ \\t\]+\[^\\n\\r\]*zmm" 1 { xfail *-*-* } } } */ > > > +/* { dg-final { scan-assembler-times "vpopcntq\[ \\t\]+\[^\\n\\r\]*xmm" 1 } } */ > > > +/* { dg-final { scan-assembler-times "vpopcntq\[ \\t\]+\[^\\n\\r\]*ymm" 1 } } */ > > > +/* { dg-final { scan-assembler-times "vpopcntq\[ \\t\]+\[^\\n\\r\]*zmm" 1 } } */ > > > #ifndef AVX512VPOPCNTQ_H_INCLUDED > > > #define AVX512VPOPCNTQ_H_INCLUDED > > > > > > diff --git a/gcc/tree-vect-patterns.c b/gcc/tree-vect-patterns.c > > > index 177d44ebb5e..5c80800efbb 100644 > > > --- a/gcc/tree-vect-patterns.c > > > +++ b/gcc/tree-vect-patterns.c > > > @@ -1292,6 +1292,115 @@ vect_recog_widen_minus_pattern (vec_info *vinfo, stmt_vec_info last_stmt_info, > > > "vect_recog_widen_minus_pattern"); > > > } > > > > > > +/* Function vect_recog_popcount_pattern > > > + > > > + Try to find the following pattern: > > > + > > > + UTYPE1 A; > > > + TYPE1 B; > > > + UTYPE2 temp_in; > > > + TYPE3 temp_out; > > > + temp_in = (TYPE2)A; > > > + > > > + temp_out = __builtin_popcount{,l,ll} (temp_in); > > > + B = (TYPE1) temp_out; > > > + > > > + TYPE2 may or may not be equal to TYPE3. > > > + i.e. TYPE2 is equal to TYPE3 for __builtin_popcount > > > + i.e. TYPE2 is not equal to TYPE3 for __builtin_popcountll > > > + > > > + Input: > > > + > > > + * STMT_VINFO: The stmt from which the pattern search begins. > > > + here it starts with B = (TYPE1) temp_out; > > > + > > > + Output: > > > + > > > + * TYPE_OUT: The vector type of the output of this pattern. > > > + > > > + * Return value: A new stmt that will be used to replace the sequence of > > > + stmts that constitute the pattern. In this case it will be: > > > + B = .POPCOUNT (A); > > > +*/ > > > + > > > +static gimple * > > > +vect_recog_popcount_pattern (vec_info *vinfo, > > > + stmt_vec_info stmt_vinfo, tree *type_out) > > > +{ > > > + gassign *last_stmt = dyn_cast (stmt_vinfo->stmt); > > > + gimple *popcount_stmt, *pattern_stmt; > > > + tree rhs_oprnd, rhs_origin, lhs_oprnd, lhs_type, vec_type, new_var; > > > + auto_vec vargs; > > > + > > > + /* Find B = (TYPE1) temp_out. */ > > > + if (!last_stmt) > > > + return NULL; > > > + tree_code code = gimple_assign_rhs_code (last_stmt); > > > + if (!CONVERT_EXPR_CODE_P (code)) > > > + return NULL; > > > + > > > + lhs_oprnd = gimple_assign_lhs (last_stmt); > > > + lhs_type = TREE_TYPE (lhs_oprnd); > > > + if (TREE_CODE (lhs_type) != INTEGER_TYPE) > > > + return NULL; > > > > INTEGRAL_TYPE_P > > > Changed. > > > + rhs_oprnd = gimple_assign_rhs1 (last_stmt); > > > + if (TREE_CODE (rhs_oprnd) != SSA_NAME > > > + || !has_single_use (rhs_oprnd)) > > > + return NULL; > > > + popcount_stmt = SSA_NAME_DEF_STMT (rhs_oprnd); > > > + > > > + /* Find temp_out = __builtin_popcount{,l,ll} (temp_in); */ > > > + if (!is_gimple_call (popcount_stmt) > > > + || !gimple_call_lhs (popcount_stmt)) > > > > Since you're arriving here via use-def chain the LHS will > > never be NULL. > > Forgot to mention this part is also changed. > > > + return NULL; > > > + switch (gimple_call_combined_fn (popcount_stmt)) > > > + { > > > + CASE_CFN_POPCOUNT: > > > + break; > > > + default: > > > + return NULL; > > > + } > > > + > > > > for safety: > > > > if (gimple_call_num_args (popcount_stmt) != 1) > > return NULL; > > > Changed. > > > + rhs_oprnd = gimple_call_arg (popcount_stmt, 0); > > > + vect_unpromoted_value unprom_diff; > > > + rhs_origin = vect_look_through_possible_promotion (vinfo, rhs_oprnd, > > > + &unprom_diff); > > > + > > > + if (!rhs_origin) > > > + return NULL; > > > + > > > + /* Input and outout of .POPCOUNT should be same-precision integer. > > > + Also A should be unsigned or same presion as temp_in, > > > + otherwise there would be sign_extend from A to temp_in. */ > > > + if (TYPE_PRECISION (unprom_diff.type) != TYPE_PRECISION (lhs_type) > > > + || !(TYPE_UNSIGNED (unprom_diff.type) > > > + || (TYPE_PRECISION (unprom_diff.type) > > > + == TYPE_PRECISION (TREE_TYPE (rhs_oprnd))))) > > > > Note I find a if (A || !(B || C)) hard to read, please write if (A || > > (!B && !C)) instead. > > > Changed. > > OK otherwise. > > > > Thanks, > > Richard. > > > > > + return NULL; > > > + vargs.safe_push (unprom_diff.op); > > > + > > > + vect_pattern_detected ("vec_regcog_popcount_pattern", popcount_stmt); > > > + vec_type = get_vectype_for_scalar_type (vinfo, lhs_type); > > > + /* Do it only the backend existed popcount2. */ > > > + if (!direct_internal_fn_supported_p (IFN_POPCOUNT, > > > + vec_type, > > > + OPTIMIZE_FOR_SPEED)) > > > + return NULL; > > > + > > > + /* Create B = .POPCOUNT (A). */ > > > + new_var = vect_recog_temp_ssa_var (lhs_type, NULL); > > > + pattern_stmt = gimple_build_call_internal_vec (IFN_POPCOUNT, vargs); > > > + gimple_call_set_lhs (pattern_stmt, new_var); > > > + gimple_set_location (pattern_stmt, gimple_location (last_stmt)); > > > + *type_out = vec_type; > > > + > > > + if (dump_enabled_p ()) > > > + dump_printf_loc (MSG_NOTE, vect_location, > > > + "created pattern stmt: %G", pattern_stmt); > > > + return pattern_stmt; > > > +} > > > + > > > /* Function vect_recog_pow_pattern > > > > > > Try to find the following pattern: > > > @@ -5283,6 +5392,7 @@ static vect_recog_func vect_vect_recog_func_ptrs[] = { > > > { vect_recog_sad_pattern, "sad" }, > > > { vect_recog_widen_sum_pattern, "widen_sum" }, > > > { vect_recog_pow_pattern, "pow" }, > > > + { vect_recog_popcount_pattern, "popcount" }, > > > { vect_recog_widen_shift_pattern, "widen_shift" }, > > > { vect_recog_rotate_pattern, "rotate" }, > > > { vect_recog_vector_vector_shift_pattern, "vector_vector_shift" }, > > > -- > > > 2.18.1 > > > > > Thanks for the review, here is the patch I'm checking in. > > -- > BR, > Hongtao -- BR, Hongtao