[RFC/RFA][PATCH v3 06/12] aarch64: Implement new expander for efficient CRC computation

[RFC/RFA][PATCH v3 06/12] aarch64: Implement new expander for efficient CRC computation

[Mariam Arutunian]

[-- Attachment #1.1: Type: text/plain, Size: 2665 bytes --]

   This patch introduces two new expanders for the aarch64 backend,
dedicated to generate optimized code for CRC computations.
   The new expanders are designed to leverage specific hardware
capabilities to achieve faster CRC calculations,
   particularly using the crc32, crc32c and pmull instructions when
supported by the target architecture.

   Expander 1: Bit-Forward CRC (crc<ALLI:mode><ALLX:mode>4)
   For targets that support pmul instruction (TARGET_AES),
   the expander will generate code that uses the pmull (crypto_pmulldi)
instruction for CRC computation.

   Expander 2: Bit-Reversed CRC (crc_rev<ALLI:mode><ALLX:mode>4)
   The expander first checks if the target supports the CRC32* instruction
set (TARGET_CRC32)
   and the polynomial in use is 0x1EDC6F41 (iSCSI) or 0x04C11DB7 (HDLC). If
the conditions are met,
   it emits calls to the corresponding crc32* instruction (depending on the
data size and the polynomial).
   If the target does not support crc32* but supports pmull, it then uses
the pmull (crypto_pmulldi) instruction for bit-reversed CRC computation.
   Otherwise table-based CRC is generated.

     gcc/config/aarch64/

       * aarch64-protos.h (aarch64_expand_crc_using_pmull): New extern
function declaration.
       (aarch64_expand_reversed_crc_using_pmull):  Likewise.
       * aarch64.cc (aarch64_expand_crc_using_pmull): New function.
       (aarch64_expand_reversed_crc_using_pmull):  Likewise.
       * aarch64.md (crc_rev<ALLI:mode><ALLX:mode>4): New expander for
reversed CRC.
       (crc<ALLI:mode><ALLX:mode>4): New expander for bit-forward CRC.
       * iterators.md (crc_data_type): New mode attribute.

     gcc/testsuite/gcc.target/aarch64/

       * crc-1-pmul.c: New test.
       * crc-10-pmul.c: Likewise.
       * crc-12-pmul.c: Likewise.
       * crc-13-pmul.c: Likewise.
       * crc-14-pmul.c: Likewise.
       * crc-17-pmul.c: Likewise.
       * crc-18-pmul.c: Likewise.
       * crc-21-pmul.c: Likewise.
       * crc-22-pmul.c: Likewise.
       * crc-23-pmul.c: Likewise.
       * crc-4-pmul.c: Likewise.
       * crc-5-pmul.c: Likewise.
       * crc-6-pmul.c: Likewise.
       * crc-7-pmul.c: Likewise.
       * crc-8-pmul.c: Likewise.
       * crc-9-pmul.c: Likewise.
       * crc-CCIT-data16-pmul.c: Likewise.
       * crc-CCIT-data8-pmul.c: Likewise.
       * crc-coremark-16bitdata-pmul.c: Likewise.
       * crc-crc32-data16.c: Likewise.
       * crc-crc32-data32.c: Likewise.
       * crc-crc32-data8.c: Likewise.
       * crc-crc32c-data16.c: Likewise.
       * crc-crc32c-data32.c: Likewise.
       * crc-crc32c-data8.c: Likewise.

   Signed-off-by: Mariam Arutunian <mariamarutunian@gmail.com>

[-- Attachment #2: 0006-aarch64-Implement-new-expander-for-efficient-CRC-com.patch --]
[-- Type: text/x-patch, Size: 29569 bytes --]

diff --git a/gcc/config/aarch64/aarch64-protos.h b/gcc/config/aarch64/aarch64-protos.h
index 1d3f94c813e..7c6bf9d2c71 100644
--- a/gcc/config/aarch64/aarch64-protos.h
+++ b/gcc/config/aarch64/aarch64-protos.h
@@ -1117,5 +1117,8 @@ extern void mingw_pe_encode_section_info (tree, rtx, int);
 
 bool aarch64_optimize_mode_switching (aarch64_mode_entity);
 void aarch64_restore_za (rtx);
+void aarch64_expand_crc_using_pmull (scalar_mode, scalar_mode, rtx *);
+void aarch64_expand_reversed_crc_using_pmull (scalar_mode, scalar_mode, rtx *);
+
 
 #endif /* GCC_AARCH64_PROTOS_H */
diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc
index ee12d8897a8..546a379fd74 100644
--- a/gcc/config/aarch64/aarch64.cc
+++ b/gcc/config/aarch64/aarch64.cc
@@ -30265,6 +30265,126 @@ aarch64_retrieve_sysreg (const char *regname, bool write_p, bool is128op)
   return sysreg->encoding;
 }
 
+/* Generate assembly to calculate CRC
+   using carry-less multiplication instruction.
+   OPERANDS[1] is input CRC,
+   OPERANDS[2] is data (message),
+   OPERANDS[3] is the polynomial without the leading 1.  */
+
+void
+aarch64_expand_crc_using_pmull (scalar_mode crc_mode,
+				scalar_mode data_mode,
+				rtx *operands)
+{
+  /* Check and keep arguments.  */
+  gcc_assert (!CONST_INT_P (operands[0]));
+  gcc_assert (CONST_INT_P (operands[3]));
+  rtx crc = operands[1];
+  rtx data = operands[2];
+  rtx polynomial = operands[3];
+
+  unsigned HOST_WIDE_INT crc_size = GET_MODE_BITSIZE (crc_mode);
+  unsigned HOST_WIDE_INT data_size = GET_MODE_BITSIZE (data_mode);
+  gcc_assert (crc_size <= 32);
+  gcc_assert (data_size <= crc_size);
+
+  /* Calculate the quotient.  */
+  unsigned HOST_WIDE_INT
+      q = gf2n_poly_long_div_quotient (UINTVAL (polynomial), crc_size);
+  /* CRC calculation's main part.  */
+  if (crc_size > data_size)
+    crc = expand_shift (RSHIFT_EXPR, DImode, crc, crc_size - data_size,
+			NULL_RTX, 1);
+
+  rtx t0 = force_reg (DImode, gen_int_mode (q, DImode));
+  polynomial = simplify_gen_unary (ZERO_EXTEND, DImode, polynomial,
+				   GET_MODE (polynomial));
+  rtx t1 = force_reg (DImode, polynomial);
+
+  rtx a0 = expand_binop (DImode, xor_optab, crc, data, NULL_RTX, 1,
+			 OPTAB_WIDEN);
+
+  rtx clmul_res = gen_reg_rtx (TImode);
+  emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t0));
+  a0 = gen_lowpart (DImode, clmul_res);
+
+  a0 = expand_shift (RSHIFT_EXPR, DImode, a0, crc_size, NULL_RTX, 1);
+
+  emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t1));
+  a0 = gen_lowpart (DImode, clmul_res);
+
+  if (crc_size > data_size)
+    {
+      rtx crc_part = expand_shift (LSHIFT_EXPR, DImode, operands[1], data_size,
+				   NULL_RTX, 0);
+      a0 = expand_binop (DImode, xor_optab, a0, crc_part, NULL_RTX, 1,
+			 OPTAB_DIRECT);
+    }
+
+  /* Zero upper bits beyond crc_size.  */
+  aarch64_emit_move (operands[0], gen_lowpart (crc_mode, a0));
+}
+
+/* Generate assembly to calculate reversed CRC
+   using carry-less multiplication instruction.
+   OPERANDS[1] is input CRC,
+   OPERANDS[2] is data,
+   OPERANDS[3] is the polynomial without the leading 1.  */
+
+void
+aarch64_expand_reversed_crc_using_pmull (scalar_mode crc_mode,
+					 scalar_mode data_mode,
+					 rtx *operands)
+{
+  /* Check and keep arguments.  */
+  gcc_assert (!CONST_INT_P (operands[0]));
+  gcc_assert (CONST_INT_P (operands[3]));
+  rtx crc = operands[1];
+  rtx data = operands[2];
+  rtx polynomial = operands[3];
+
+  unsigned HOST_WIDE_INT crc_size = GET_MODE_BITSIZE (crc_mode);
+  unsigned HOST_WIDE_INT data_size = GET_MODE_BITSIZE (data_mode);
+  gcc_assert (crc_size <= 32);
+  gcc_assert (data_size <= crc_size);
+
+  /* Calculate the quotient.  */
+  unsigned HOST_WIDE_INT
+      q = gf2n_poly_long_div_quotient (UINTVAL (polynomial), crc_size);
+  /* Reflect the calculated quotient.  */
+  q = reflect (q);
+  rtx t0 = force_reg (DImode, gen_int_mode (q >> (data_size - 4), DImode));
+
+  /* Reflect the polynomial.  */
+  unsigned HOST_WIDE_INT ref_polynomial = reflect (UINTVAL (polynomial));
+  rtx t1 = force_reg (DImode, gen_int_mode (ref_polynomial << 1, DImode));
+
+  /* CRC calculation's main part.  */
+  rtx a0 = expand_binop (DImode, xor_optab, crc, data, NULL_RTX, 1,
+			 OPTAB_WIDEN);
+
+  /* Perform carry-less multiplication and get low part.  */
+  rtx clmul_res = gen_reg_rtx (TImode);
+  emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t0));
+  a0 = gen_lowpart (DImode, clmul_res);
+
+  a0 = expand_shift (LSHIFT_EXPR, DImode, a0, 64 - crc_size - 3, NULL_RTX, 0);
+
+  /* Perform carry-less multiplication and get high part.  */
+  emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t1));
+  a0 = gen_highpart (DImode, clmul_res);
+
+  if (crc_size > data_size)
+    {
+      rtx crc_part = expand_shift (RSHIFT_EXPR, DImode, crc, data_size,
+				   NULL_RTX, 1);
+      a0 = expand_binop (DImode, xor_optab, a0, crc_part, NULL_RTX, 1,
+			 OPTAB_DIRECT);
+    }
+
+  aarch64_emit_move (operands[0], gen_lowpart (crc_mode, a0));
+}
+
 /* Target-specific selftests.  */
 
 #if CHECKING_P
diff --git a/gcc/config/aarch64/aarch64.md b/gcc/config/aarch64/aarch64.md
index 9dff2d7a2b0..08c588bc475 100644
--- a/gcc/config/aarch64/aarch64.md
+++ b/gcc/config/aarch64/aarch64.md
@@ -4543,6 +4543,63 @@
   [(set_attr "type" "crc")]
 )
 
+;; Reversed CRC
+(define_expand "crc_rev<ALLI:mode><ALLX:mode>4"
+  [;; return value (calculated CRC)
+   (match_operand:ALLX 0 "register_operand" "=r")
+   ;; initial CRC
+   (match_operand:ALLX 1 "register_operand" "r")
+   ;; data
+   (match_operand:ALLI 2 "register_operand" "r")
+   ;; polynomial without leading 1
+   (match_operand:ALLX 3)]
+  ""
+  {
+    /* If the polynomial is the same as the polynomial of crc32c* instruction,
+       put that instruction.  crc32c uses iSCSI polynomial.  */
+    if (TARGET_CRC32 && INTVAL (operands[3]) == 0x1EDC6F41
+	&& <ALLX:MODE>mode == SImode)
+      emit_insn (gen_aarch64_crc32c<ALLI:crc_data_type> (operands[0],
+							 operands[1],
+							 operands[2]));
+    /* If the polynomial is the same as the polynomial of crc32* instruction,
+	put that instruction.  crc32 uses HDLC etc.  polynomial.  */
+    else if (TARGET_CRC32 && INTVAL (operands[3]) == 0x04C11DB7
+	     && <ALLX:MODE>mode == SImode)
+      emit_insn (gen_aarch64_crc32<ALLI:crc_data_type> (operands[0],
+							operands[1],
+							operands[2]));
+    else if (TARGET_AES && <ALLI:sizen> <= <ALLX:sizen>)
+      aarch64_expand_reversed_crc_using_pmull (<ALLX:MODE>mode,
+					       <ALLI:MODE>mode,
+					       operands);
+    else
+      /* Otherwise, generate table-based CRC.  */
+      expand_reversed_crc_table_based (operands[0], operands[1], operands[2],
+				       operands[3], <ALLI:MODE>mode,
+				       generate_reflecting_code_standard);
+    DONE;
+  }
+)
+
+;; Bit-forward CRC
+(define_expand "crc<ALLI:mode><ALLX:mode>4"
+  [;; return value (calculated CRC)
+   (match_operand:ALLX 0 "register_operand" "=r")
+   ;; initial CRC
+   (match_operand:ALLX 1 "register_operand" "r")
+   ;; data
+   (match_operand:ALLI 2 "register_operand" "r")
+   ;; polynomial without leading 1
+   (match_operand:ALLX 3)]
+  "TARGET_AES && <ALLI:sizen> <= <ALLX:sizen>"
+  {
+    aarch64_expand_crc_using_pmull (<ALLX:MODE>mode, <ALLI:MODE>mode,
+				    operands);
+    DONE;
+  }
+)
+
 (define_insn "*csinc2<mode>_insn"
   [(set (match_operand:GPI 0 "register_operand" "=r")
         (plus:GPI (match_operand 2 "aarch64_comparison_operation" "")
diff --git a/gcc/config/aarch64/iterators.md b/gcc/config/aarch64/iterators.md
index 99cde46f1ba..86e4863d684 100644
--- a/gcc/config/aarch64/iterators.md
+++ b/gcc/config/aarch64/iterators.md
@@ -1276,6 +1276,10 @@
 ;; Map a mode to a specific constraint character.
 (define_mode_attr cmode [(QI "q") (HI "h") (SI "s") (DI "d")])
 
+;; Map a mode to a specific constraint character for calling
+;; appropriate version of crc.
+(define_mode_attr crc_data_type [(QI "b") (HI "h") (SI "w") (DI "x")])
+
 ;; Map modes to Usg and Usj constraints for SISD right shifts
 (define_mode_attr cmode_simd [(SI "g") (DI "j")])
 
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-1-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-1-pmul.c
new file mode 100644
index 00000000000..4043251dbd8
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-1-pmul.c
@@ -0,0 +1,8 @@
+/* { dg-do run } */
+/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc -fdisable-tree-phiopt2 -fdisable-tree-phiopt3" } */
+
+#include "../../gcc.dg/torture/crc-1.c"
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
\ No newline at end of file
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-10-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-10-pmul.c
new file mode 100644
index 00000000000..dd866b38e83
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-10-pmul.c
@@ -0,0 +1,9 @@
+/* { dg-do run } */
+/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
+/* { dg-skip-if "" { *-*-* } { "-flto"} } */
+
+#include "../../gcc.dg/torture/crc-10.c"
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-12-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-12-pmul.c
new file mode 100644
index 00000000000..16d901eeaef
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-12-pmul.c
@@ -0,0 +1,9 @@
+/* { dg-do run } */
+/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc -fdisable-tree-phiopt2 -fdisable-tree-phiopt3" } */
+/* { dg-skip-if "" { *-*-* } { "-flto"} } */
+
+#include "../../gcc.dg/torture/crc-12.c"
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-13-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-13-pmul.c
new file mode 100644
index 00000000000..5f7741fad0f
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-13-pmul.c
@@ -0,0 +1,9 @@
+/* { dg-do run } */
+/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
+/* { dg-skip-if "" { *-*-* } { "-flto"} } */
+
+#include "../../gcc.dg/torture/crc-13.c"
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-14-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-14-pmul.c
new file mode 100644
index 00000000000..cdedbbd3db1
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-14-pmul.c
@@ -0,0 +1,9 @@
+/* { dg-do run } */
+/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
+/* { dg-skip-if "" { *-*-* } { "-flto"} } */
+
+#include "../../gcc.dg/torture/crc-14.c"
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-17-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-17-pmul.c
new file mode 100644
index 00000000000..c219e49a2b1
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-17-pmul.c
@@ -0,0 +1,9 @@
+/* { dg-do run } */
+/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
+/* { dg-skip-if "" { *-*-* } { "-flto"} } */
+
+#include "../../gcc.dg/torture/crc-17.c"
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-18-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-18-pmul.c
new file mode 100644
index 00000000000..124900a979b
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-18-pmul.c
@@ -0,0 +1,9 @@
+/* { dg-do run } */
+/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
+/* { dg-skip-if "" { *-*-* } { "-flto"} } */
+
+#include "../../gcc.dg/torture/crc-18.c"
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-21-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-21-pmul.c
new file mode 100644
index 00000000000..3cae1a7f57b
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-21-pmul.c
@@ -0,0 +1,9 @@
+/* { dg-do run } */
+/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
+/* { dg-skip-if "" { *-*-* } { "-flto"} } */
+
+#include "../../gcc.dg/torture/crc-21.c"
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-22-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-22-pmul.c
new file mode 100644
index 00000000000..0ec2e312f8f
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-22-pmul.c
@@ -0,0 +1,9 @@
+/* { dg-do run } */
+/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
+/* { dg-skip-if "" { *-*-* } { "-flto"} } */
+
+#include "../../gcc.dg/torture/crc-22.c"
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-23-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-23-pmul.c
new file mode 100644
index 00000000000..0c4542adb40
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-23-pmul.c
@@ -0,0 +1,9 @@
+/* { dg-do run } */
+/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
+/* { dg-skip-if "" { *-*-* } { "-flto"} } */
+
+#include "../../gcc.dg/torture/crc-23.c"
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-4-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-4-pmul.c
new file mode 100644
index 00000000000..08f1d3b69d7
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-4-pmul.c
@@ -0,0 +1,9 @@
+/* { dg-do run } */
+/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
+/* { dg-skip-if "" { *-*-* } { "-flto"} } */
+
+#include "../../gcc.dg/torture/crc-4.c"
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-5-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-5-pmul.c
new file mode 100644
index 00000000000..91bf5e6353d
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-5-pmul.c
@@ -0,0 +1,9 @@
+/* { dg-do run } */
+/* { dg-options "-march=armv8-a+crypto -O2 -w -fdump-rtl-dfinish -fdump-tree-crc" } */
+/* { dg-skip-if "" { *-*-* } { "-flto"} } */
+
+#include "../../gcc.dg/torture/crc-5.c"
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
\ No newline at end of file
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-6-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-6-pmul.c
new file mode 100644
index 00000000000..4680eafe758
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-6-pmul.c
@@ -0,0 +1,9 @@
+/* { dg-do run } */
+/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
+/* { dg-skip-if "" { *-*-* } { "-flto"} } */
+
+#include "../../gcc.dg/torture/crc-6.c"
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
\ No newline at end of file
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-7-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-7-pmul.c
new file mode 100644
index 00000000000..655484d10d4
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-7-pmul.c
@@ -0,0 +1,9 @@
+/* { dg-do run } */
+/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
+/* { dg-skip-if "" { *-*-* } { "-flto"} } */
+
+#include "../../gcc.dg/torture/crc-7.c"
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-8-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-8-pmul.c
new file mode 100644
index 00000000000..6c2acc84c32
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-8-pmul.c
@@ -0,0 +1,9 @@
+/* { dg-do run } */
+/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
+/* { dg-skip-if "" { *-*-* } { "-flto"} } */
+
+#include "../../gcc.dg/torture/crc-8.c"
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-9-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-9-pmul.c
new file mode 100644
index 00000000000..e76f3c77b59
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-9-pmul.c
@@ -0,0 +1,9 @@
+/* { dg-do run } */
+/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
+/* { dg-skip-if "" { *-*-* } { "-flto"} } */
+
+#include "../../gcc.dg/torture/crc-9.c"
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data16-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data16-pmul.c
new file mode 100644
index 00000000000..21520474564
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data16-pmul.c
@@ -0,0 +1,9 @@
+/* { dg-do run } */
+/* { dg-options "-w -march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
+/* { dg-skip-if "" { *-*-* } { "-flto"} } */
+
+#include "../../gcc.dg/torture/crc-CCIT-data16.c"
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
\ No newline at end of file
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data8-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data8-pmul.c
new file mode 100644
index 00000000000..3dcc92320f3
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data8-pmul.c
@@ -0,0 +1,9 @@
+/* { dg-do run } */
+/* { dg-options "-w -march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
+/* { dg-skip-if "" { *-*-* } { "-flto" } } */
+
+#include "../../gcc.dg/torture/crc-CCIT-data8.c"
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
\ No newline at end of file
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-coremark-16bitdata-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-coremark-16bitdata-pmul.c
new file mode 100644
index 00000000000..e5196aaafef
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-coremark-16bitdata-pmul.c
@@ -0,0 +1,9 @@
+/* { dg-do run } */
+/* { dg-options "-w -march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
+/* { dg-skip-if "" { *-*-* } { "-flto"} } */
+
+#include "../../gcc.dg/torture/crc-coremark16-data16.c"
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
\ No newline at end of file
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32-data16.c b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data16.c
new file mode 100644
index 00000000000..e82cb04fcc3
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data16.c
@@ -0,0 +1,53 @@
+/* { dg-do run } */
+/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
+/* { dg-skip-if "" { *-*-* } { "-flto"} } */
+
+#include <stdint.h>
+#include <stdlib.h>
+
+__attribute__ ((noinline,optimize(0)))
+uint32_t _crc32_O0 (uint32_t crc, uint16_t data) {
+  int i;
+  crc = crc ^ data;
+
+  for (i = 0; i < 8; i++) {
+      if (crc & 1)
+	crc = (crc >> 1) ^ 0xEDB88320;
+      else
+	crc = (crc >> 1);
+    }
+
+  return crc;
+}
+
+uint32_t _crc32 (uint32_t crc, uint16_t data) {
+  int i;
+  crc = crc ^ data;
+
+  for (i = 0; i < 8; i++) {
+      if (crc & 1)
+	crc = (crc >> 1) ^ 0xEDB88320;
+      else
+	crc = (crc >> 1);
+    }
+
+  return crc;
+}
+
+int main ()
+{
+  uint32_t crc = 0x0D800D80;
+  for (uint16_t i = 0; i < 0xffff; i++)
+    {
+      uint32_t res1 = _crc32_O0 (crc, i);
+      uint32_t res2 = _crc32 (crc, i);
+      if (res1 != res2)
+	 abort ();
+      crc = res1;
+    }
+}
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "UNSPEC_CRC32" "dfinish"} } */
+/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32-data32.c b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data32.c
new file mode 100644
index 00000000000..a7564a7e28a
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data32.c
@@ -0,0 +1,52 @@
+/* { dg-do run } */
+/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
+/* { dg-skip-if "" { *-*-* } { "-flto"} } */
+
+#include <stdint.h>
+#include <stdlib.h>
+__attribute__ ((noinline,optimize(0)))
+uint32_t _crc32_O0 (uint32_t crc, uint32_t data) {
+  int i;
+  crc = crc ^ data;
+
+  for (i = 0; i < 32; i++) {
+      if (crc & 1)
+	crc = (crc >> 1) ^ 0xEDB88320;
+      else
+	crc = (crc >> 1);
+    }
+
+  return crc;
+}
+
+uint32_t _crc32 (uint32_t crc, uint32_t data) {
+  int i;
+  crc = crc ^ data;
+
+  for (i = 0; i < 32; i++) {
+      if (crc & 1)
+	crc = (crc >> 1) ^ 0xEDB88320;
+      else
+	crc = (crc >> 1);
+    }
+
+  return crc;
+}
+
+int main ()
+{
+  uint32_t crc = 0x0D800D80;
+  for (uint8_t i = 0; i < 0xff; i++)
+    {
+      uint32_t res1 = _crc32_O0 (crc, i);
+      uint32_t res2 = _crc32 (crc, i);
+      if (res1 != res2)
+	 abort ();
+      crc = res1;
+    }
+}
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "UNSPEC_CRC32" "dfinish"} } */
+/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32-data8.c b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data8.c
new file mode 100644
index 00000000000..c88cafadedc
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data8.c
@@ -0,0 +1,53 @@
+/* { dg-do run } */
+/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
+/* { dg-skip-if "" { *-*-* } { "-flto"} } */
+
+#include <stdint.h>
+#include <stdlib.h>
+
+__attribute__ ((noinline,optimize(0)))
+uint32_t _crc32_O0 (uint32_t crc, uint8_t data) {
+  int i;
+  crc = crc ^ data;
+
+  for (i = 0; i < 8; i++) {
+      if (crc & 1)
+	crc = (crc >> 1) ^ 0xEDB88320;
+      else
+	crc = (crc >> 1);
+    }
+
+  return crc;
+}
+
+uint32_t _crc32 (uint32_t crc, uint8_t data) {
+  int i;
+  crc = crc ^ data;
+
+  for (i = 0; i < 8; i++) {
+      if (crc & 1)
+	crc = (crc >> 1) ^ 0xEDB88320;
+      else
+	crc = (crc >> 1);
+    }
+
+  return crc;
+}
+
+int main ()
+{
+  uint32_t crc = 0x0D800D80;
+  for (uint8_t i = 0; i < 0xff; i++)
+    {
+      uint32_t res1 = _crc32_O0 (crc, i);
+      uint32_t res2 = _crc32 (crc, i);
+      if (res1 != res2)
+	 abort ();
+      crc = res1;
+    }
+}
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "UNSPEC_CRC32" "dfinish"} } */
+/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data16.c b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data16.c
new file mode 100644
index 00000000000..d82e6252603
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data16.c
@@ -0,0 +1,53 @@
+/* { dg-do run } */
+/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
+/* { dg-skip-if "" { *-*-* } { "-flto"} } */
+
+#include <stdint.h>
+#include <stdlib.h>
+
+__attribute__ ((noinline,optimize(0)))
+uint32_t _crc32_O0 (uint32_t crc, uint16_t data) {
+  int i;
+  crc = crc ^ data;
+
+  for (i = 0; i < 8; i++) {
+      if (crc & 1)
+	crc = (crc >> 1) ^ 0x82F63B78;
+      else
+	crc = (crc >> 1);
+    }
+
+  return crc;
+}
+
+uint32_t _crc32 (uint32_t crc, uint16_t data) {
+  int i;
+  crc = crc ^ data;
+
+  for (i = 0; i < 8; i++) {
+      if (crc & 1)
+	crc = (crc >> 1) ^ 0x82F63B78;
+      else
+	crc = (crc >> 1);
+    }
+
+  return crc;
+}
+
+int main ()
+{
+  uint32_t crc = 0x0D800D80;
+  for (uint16_t i = 0; i < 0xffff; i++)
+    {
+      uint32_t res1 = _crc32_O0 (crc, i);
+      uint32_t res2 = _crc32 (crc, i);
+      if (res1 != res2)
+	 abort ();
+      crc = res1;
+    }
+}
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "UNSPEC_CRC32C" "dfinish"} } */
+/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data32.c b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data32.c
new file mode 100644
index 00000000000..7acb6fc239c
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data32.c
@@ -0,0 +1,52 @@
+/* { dg-do run } */
+/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
+/* { dg-skip-if "" { *-*-* } { "-flto"} } */
+
+#include <stdint.h>
+#include <stdlib.h>
+__attribute__ ((noinline,optimize(0)))
+uint32_t _crc32_O0 (uint32_t crc, uint32_t data) {
+  int i;
+  crc = crc ^ data;
+
+  for (i = 0; i < 32; i++) {
+      if (crc & 1)
+	crc = (crc >> 1) ^ 0x82F63B78;
+      else
+	crc = (crc >> 1);
+    }
+
+  return crc;
+}
+
+uint32_t _crc32 (uint32_t crc, uint32_t data) {
+  int i;
+  crc = crc ^ data;
+
+  for (i = 0; i < 32; i++) {
+      if (crc & 1)
+	crc = (crc >> 1) ^ 0x82F63B78;
+      else
+	crc = (crc >> 1);
+    }
+
+  return crc;
+}
+
+int main ()
+{
+  uint32_t crc = 0x0D800D80;
+  for (uint8_t i = 0; i < 0xff; i++)
+    {
+      uint32_t res1 = _crc32_O0 (crc, i);
+      uint32_t res2 = _crc32 (crc, i);
+      if (res1 != res2)
+	 abort ();
+      crc = res1;
+    }
+}
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "UNSPEC_CRC32C" "dfinish"} } */
+/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */
diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data8.c b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data8.c
new file mode 100644
index 00000000000..e8a8901e453
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data8.c
@@ -0,0 +1,53 @@
+/* { dg-do run } */
+/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
+/* { dg-skip-if "" { *-*-* } { "-flto"} } */
+
+#include <stdint.h>
+#include <stdlib.h>
+
+__attribute__ ((noinline,optimize(0)))
+uint32_t _crc32_O0 (uint32_t crc, uint8_t data) {
+  int i;
+  crc = crc ^ data;
+
+  for (i = 0; i < 8; i++) {
+      if (crc & 1)
+	crc = (crc >> 1) ^ 0x82F63B78;
+      else
+	crc = (crc >> 1);
+    }
+
+  return crc;
+}
+
+uint32_t _crc32 (uint32_t crc, uint8_t data) {
+  int i;
+  crc = crc ^ data;
+
+  for (i = 0; i < 8; i++) {
+      if (crc & 1)
+	crc = (crc >> 1) ^ 0x82F63B78;
+      else
+	crc = (crc >> 1);
+    }
+
+  return crc;
+}
+
+int main ()
+{
+  uint32_t crc = 0x0D800D80;
+  for (uint8_t i = 0; i < 0xff; i++)
+    {
+      uint32_t res1 = _crc32_O0 (crc, i);
+      uint32_t res2 = _crc32 (crc, i);
+      if (res1 != res2)
+	 abort ();
+      crc = res1;
+    }
+}
+
+/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
+/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
+/* { dg-final { scan-rtl-dump "UNSPEC_CRC32C" "dfinish"} } */
+/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */
-- 
2.25.1

Re: [RFC/RFA][PATCH v3 06/12] aarch64: Implement new expander for efficient CRC computation

[Richard Sandiford]

Sorry again for the slow review. :(

I only really looked at the unreversed version earlier, on the basis
that the comments would apply to both versions.  But I've got a couple
of comments about the reversed version below:

Mariam Arutunian <mariamarutunian@gmail.com> writes:
> [...]
> diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc
> index ee12d8897a8..546a379fd74 100644
> --- a/gcc/config/aarch64/aarch64.cc
> +++ b/gcc/config/aarch64/aarch64.cc
> @@ -30265,6 +30265,126 @@ aarch64_retrieve_sysreg (const char *regname, bool write_p, bool is128op)
>    return sysreg->encoding;
>  }
>  
> +/* Generate assembly to calculate CRC
> +   using carry-less multiplication instruction.
> +   OPERANDS[1] is input CRC,
> +   OPERANDS[2] is data (message),
> +   OPERANDS[3] is the polynomial without the leading 1.  */
> +
> +void
> +aarch64_expand_crc_using_pmull (scalar_mode crc_mode,
> +				scalar_mode data_mode,
> +				rtx *operands)
> +{
> +  /* Check and keep arguments.  */
> +  gcc_assert (!CONST_INT_P (operands[0]));
> +  gcc_assert (CONST_INT_P (operands[3]));
> +  rtx crc = operands[1];
> +  rtx data = operands[2];
> +  rtx polynomial = operands[3];
> +
> +  unsigned HOST_WIDE_INT crc_size = GET_MODE_BITSIZE (crc_mode);
> +  unsigned HOST_WIDE_INT data_size = GET_MODE_BITSIZE (data_mode);
> +  gcc_assert (crc_size <= 32);
> +  gcc_assert (data_size <= crc_size);
> +
> +  /* Calculate the quotient.  */
> +  unsigned HOST_WIDE_INT
> +      q = gf2n_poly_long_div_quotient (UINTVAL (polynomial), crc_size);
> +  /* CRC calculation's main part.  */
> +  if (crc_size > data_size)
> +    crc = expand_shift (RSHIFT_EXPR, DImode, crc, crc_size - data_size,
> +			NULL_RTX, 1);
> +
> +  rtx t0 = force_reg (DImode, gen_int_mode (q, DImode));
> +  polynomial = simplify_gen_unary (ZERO_EXTEND, DImode, polynomial,
> +				   GET_MODE (polynomial));
> +  rtx t1 = force_reg (DImode, polynomial);
> +
> +  rtx a0 = expand_binop (DImode, xor_optab, crc, data, NULL_RTX, 1,
> +			 OPTAB_WIDEN);
> +
> +  rtx clmul_res = gen_reg_rtx (TImode);
> +  emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t0));
> +  a0 = gen_lowpart (DImode, clmul_res);
> +
> +  a0 = expand_shift (RSHIFT_EXPR, DImode, a0, crc_size, NULL_RTX, 1);
> +
> +  emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t1));
> +  a0 = gen_lowpart (DImode, clmul_res);
> +
> +  if (crc_size > data_size)
> +    {
> +      rtx crc_part = expand_shift (LSHIFT_EXPR, DImode, operands[1], data_size,
> +				   NULL_RTX, 0);
> +      a0 = expand_binop (DImode, xor_optab, a0, crc_part, NULL_RTX, 1,
> +			 OPTAB_DIRECT);
> +    }
> +
> +  /* Zero upper bits beyond crc_size.  */

The comment no longer applies.  Otherwise this function looks good to me.

> +  aarch64_emit_move (operands[0], gen_lowpart (crc_mode, a0));
> +}
> +
> +/* Generate assembly to calculate reversed CRC
> +   using carry-less multiplication instruction.
> +   OPERANDS[1] is input CRC,
> +   OPERANDS[2] is data,
> +   OPERANDS[3] is the polynomial without the leading 1.  */
> +
> +void
> +aarch64_expand_reversed_crc_using_pmull (scalar_mode crc_mode,
> +					 scalar_mode data_mode,
> +					 rtx *operands)
> +{
> +  /* Check and keep arguments.  */
> +  gcc_assert (!CONST_INT_P (operands[0]));
> +  gcc_assert (CONST_INT_P (operands[3]));
> +  rtx crc = operands[1];
> +  rtx data = operands[2];
> +  rtx polynomial = operands[3];
> +
> +  unsigned HOST_WIDE_INT crc_size = GET_MODE_BITSIZE (crc_mode);
> +  unsigned HOST_WIDE_INT data_size = GET_MODE_BITSIZE (data_mode);
> +  gcc_assert (crc_size <= 32);
> +  gcc_assert (data_size <= crc_size);
> +
> +  /* Calculate the quotient.  */
> +  unsigned HOST_WIDE_INT
> +      q = gf2n_poly_long_div_quotient (UINTVAL (polynomial), crc_size);
> +  /* Reflect the calculated quotient.  */
> +  q = reflect (q);
> +  rtx t0 = force_reg (DImode, gen_int_mode (q >> (data_size - 4), DImode));
> +
> +  /* Reflect the polynomial.  */
> +  unsigned HOST_WIDE_INT ref_polynomial = reflect (UINTVAL (polynomial));

It looks like reflect() autodetects the bitwidth based on the assumption
that the upper half will be nonzero.  But that might not be true for all
possible polynomials (when the implicit leading coefficient is absent)
E.g. it looks like the 64-bit HDLC CRC polynomial is 0x1b (just the
lowest byte nonzero), and although we don't support 64-bit polynomials
here, the approach wouldn't work for it.

I think it'd be safer to pass the bitwidth as an explicit parameter.
Also, maybe it could go in hwint.* instead of expr.* and be called
something like reflect_hwi.

> +  rtx t1 = force_reg (DImode, gen_int_mode (ref_polynomial << 1, DImode));
> +
> +  /* CRC calculation's main part.  */
> +  rtx a0 = expand_binop (DImode, xor_optab, crc, data, NULL_RTX, 1,
> +			 OPTAB_WIDEN);
> +
> +  /* Perform carry-less multiplication and get low part.  */
> +  rtx clmul_res = gen_reg_rtx (TImode);
> +  emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t0));
> +  a0 = gen_lowpart (DImode, clmul_res);
> +
> +  a0 = expand_shift (LSHIFT_EXPR, DImode, a0, 64 - crc_size - 3, NULL_RTX, 0);

I haven't really looked at this implementation strategy before, so this
is probably a silly question sorry :-) but is there a specific reason
for selecting the bias of 4 in the right shift above, cancelled out by
subtracting the same amount from this left shift?  It looks like we could
use any value in the range [1. crc_size - 1], is that right?

Just asking out of curiosity though.  I agree it works.

> +
> +  /* Perform carry-less multiplication and get high part.  */
> +  emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t1));
> +  a0 = gen_highpart (DImode, clmul_res);

Although this works, it's taking a DImode highpart of a single TImode
register, which requires a spill from the register allocator.  So we get
something like:

        pmull   v31.1q, v31.1d, v14.1d
        str     q31, [sp, 48]
        ldr     d31, [sp, 48]
        shl     d31, d31, 53
        pmull   v31.1q, v31.1d, v15.1d
        str     q31, [sp, 48]
        ldrb    w0, [sp, 56]

(I think the first STR and LDR come from reusing the same pseudo register
for the temporary results; it's the STR+LDRB that causes the spill.)

One way around that would be:

  emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t1));
  a0 = gen_reg_rtx (DImode);
  rtx v2di_res = gen_lowpart (V2DImode, clmul_res);
  rtx high_index = BYTES_BIG_ENDIAN ? const0_rtx : const1_rtx;
  emit_insn (gen_aarch64_get_lanev2di (a0, v2di_res, high_index));

That generates:

        pmull   v30.1q, v29.1d, v30.1d
        shl     d30, d30, 53
        pmull   v30.1q, v30.1d, v31.1d
        umov    x0, v30.d[1]

which is pretty nice.

There again, it might not be too bad if we ionstead used the flipped
version of the unreversed approach, such as the following minor adaption:

  /* Calculate the quotient.  */
  unsigned HOST_WIDE_INT
      q = gf2n_poly_long_div_quotient (UINTVAL (polynomial), crc_size);
  /* Reflect the calculated quotient.  */
  q = reflect_hwi (q, crc_size + 1);
  rtx t0 = force_reg (DImode, gen_int_mode (q, DImode));
 
  /* Reflect the polynomial.  */
  unsigned HOST_WIDE_INT ref_polynomial = reflect_hwi (UINTVAL (polynomial),
						       crc_size);
  /* An unshifted multiplier would require the final result to be extracted
     using a shift right by DATA_SIZE - 1 bits.  Shift the multiplier left
     so that the shift right can be by CRC_SIZE bits instead.  */
  ref_polynomial <<= crc_size - data_size + 1;
  rtx t1 = force_reg (DImode, gen_int_mode (ref_polynomial, DImode));

  /* CRC calculation's main part.  */
  rtx a0 = expand_binop (DImode, xor_optab, crc, data, NULL_RTX, 1,
			 OPTAB_WIDEN);

  /* Perform carry-less multiplication and get low part.  */
  rtx clmul_res = gen_reg_rtx (TImode);
  emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t0));
  a0 = gen_lowpart (DImode, clmul_res);

  a0 = expand_binop (DImode, and_optab, a0,
		     gen_int_mode (GET_MODE_MASK (data_mode), DImode),
		     NULL_RTX, 1, OPTAB_WIDEN);

  /* Perform carry-less multiplication.  */
  emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t1));

  /* Perform a shift right by CRC_SIZE as an extraction of lane 1.  */
  machine_mode crc_vmode = aarch64_vq_mode (crc_mode).require ();
  a0 = (crc_size > data_size ? gen_reg_rtx (crc_mode) : operands[0]);
  emit_insn (gen_aarch64_get_lane (crc_vmode, a0,
				   gen_lowpart (crc_vmode, clmul_res),
				   aarch64_endian_lane_rtx (crc_vmode, 1)));

  if (crc_size > data_size)
    {
      rtx crc_part = expand_shift (RSHIFT_EXPR, crc_mode, crc, data_size,
				   NULL_RTX, 1);
      a0 = expand_binop (crc_mode, xor_optab, a0, crc_part, operands[0], 1,
			 OPTAB_WIDEN);
      aarch64_emit_move (operands[0], a0);
    }

This gives:

        pmull   v30.1q, v30.1d, v31.1d
        movi    v31.2d, 0xff
        and     v30.8b, v30.8b, v31.8b
        pmull   v30.1q, v30.1d, v31.1d
        umov    w0, v30.b[1]

or, with SVE enabled:

        pmull   v30.1q, v31.1d, v30.1d
        and     z30.d, z30.d, #255
        pmull   v30.1q, v30.1d, v31.1d
        umov    w0, v30.b[1]

This is preferable since ANDs are generally cheaper than shifts.

That's just a suggestion though; the original version is ok too.

Minor comment: it'd be good to use "pmull_res" instead of "clmul_res"
for the variables: :)

LGTM otherwise.

Thanks,
Richard

> +
> +  if (crc_size > data_size)
> +    {
> +      rtx crc_part = expand_shift (RSHIFT_EXPR, DImode, crc, data_size,
> +				   NULL_RTX, 1);
> +      a0 = expand_binop (DImode, xor_optab, a0, crc_part, NULL_RTX, 1,
> +			 OPTAB_DIRECT);
> +    }
> +
> +  aarch64_emit_move (operands[0], gen_lowpart (crc_mode, a0));
> +}
> +
>  /* Target-specific selftests.  */
>  
>  #if CHECKING_P
> diff --git a/gcc/config/aarch64/aarch64.md b/gcc/config/aarch64/aarch64.md
> index 9dff2d7a2b0..08c588bc475 100644
> --- a/gcc/config/aarch64/aarch64.md
> +++ b/gcc/config/aarch64/aarch64.md
> @@ -4543,6 +4543,63 @@
>    [(set_attr "type" "crc")]
>  )
>  
> +;; Reversed CRC
> +(define_expand "crc_rev<ALLI:mode><ALLX:mode>4"
> +  [;; return value (calculated CRC)
> +   (match_operand:ALLX 0 "register_operand" "=r")
> +   ;; initial CRC
> +   (match_operand:ALLX 1 "register_operand" "r")
> +   ;; data
> +   (match_operand:ALLI 2 "register_operand" "r")
> +   ;; polynomial without leading 1
> +   (match_operand:ALLX 3)]
> +  ""
> +  {
> +    /* If the polynomial is the same as the polynomial of crc32c* instruction,
> +       put that instruction.  crc32c uses iSCSI polynomial.  */
> +    if (TARGET_CRC32 && INTVAL (operands[3]) == 0x1EDC6F41
> +	&& <ALLX:MODE>mode == SImode)
> +      emit_insn (gen_aarch64_crc32c<ALLI:crc_data_type> (operands[0],
> +							 operands[1],
> +							 operands[2]));
> +    /* If the polynomial is the same as the polynomial of crc32* instruction,
> +	put that instruction.  crc32 uses HDLC etc.  polynomial.  */
> +    else if (TARGET_CRC32 && INTVAL (operands[3]) == 0x04C11DB7
> +	     && <ALLX:MODE>mode == SImode)
> +      emit_insn (gen_aarch64_crc32<ALLI:crc_data_type> (operands[0],
> +							operands[1],
> +							operands[2]));
> +    else if (TARGET_AES && <ALLI:sizen> <= <ALLX:sizen>)
> +      aarch64_expand_reversed_crc_using_pmull (<ALLX:MODE>mode,
> +					       <ALLI:MODE>mode,
> +					       operands);
> +    else
> +      /* Otherwise, generate table-based CRC.  */
> +      expand_reversed_crc_table_based (operands[0], operands[1], operands[2],
> +				       operands[3], <ALLI:MODE>mode,
> +				       generate_reflecting_code_standard);
> +    DONE;
> +  }
> +)
> +
> +;; Bit-forward CRC
> +(define_expand "crc<ALLI:mode><ALLX:mode>4"
> +  [;; return value (calculated CRC)
> +   (match_operand:ALLX 0 "register_operand" "=r")
> +   ;; initial CRC
> +   (match_operand:ALLX 1 "register_operand" "r")
> +   ;; data
> +   (match_operand:ALLI 2 "register_operand" "r")
> +   ;; polynomial without leading 1
> +   (match_operand:ALLX 3)]
> +  "TARGET_AES && <ALLI:sizen> <= <ALLX:sizen>"
> +  {
> +    aarch64_expand_crc_using_pmull (<ALLX:MODE>mode, <ALLI:MODE>mode,
> +				    operands);
> +    DONE;
> +  }
> +)
> +
>  (define_insn "*csinc2<mode>_insn"
>    [(set (match_operand:GPI 0 "register_operand" "=r")
>          (plus:GPI (match_operand 2 "aarch64_comparison_operation" "")
> diff --git a/gcc/config/aarch64/iterators.md b/gcc/config/aarch64/iterators.md
> index 99cde46f1ba..86e4863d684 100644
> --- a/gcc/config/aarch64/iterators.md
> +++ b/gcc/config/aarch64/iterators.md
> @@ -1276,6 +1276,10 @@
>  ;; Map a mode to a specific constraint character.
>  (define_mode_attr cmode [(QI "q") (HI "h") (SI "s") (DI "d")])
>  
> +;; Map a mode to a specific constraint character for calling
> +;; appropriate version of crc.
> +(define_mode_attr crc_data_type [(QI "b") (HI "h") (SI "w") (DI "x")])
> +
>  ;; Map modes to Usg and Usj constraints for SISD right shifts
>  (define_mode_attr cmode_simd [(SI "g") (DI "j")])
>  
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-1-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-1-pmul.c
> new file mode 100644
> index 00000000000..4043251dbd8
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-1-pmul.c
> @@ -0,0 +1,8 @@
> +/* { dg-do run } */
> +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc -fdisable-tree-phiopt2 -fdisable-tree-phiopt3" } */
> +
> +#include "../../gcc.dg/torture/crc-1.c"
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> \ No newline at end of file
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-10-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-10-pmul.c
> new file mode 100644
> index 00000000000..dd866b38e83
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-10-pmul.c
> @@ -0,0 +1,9 @@
> +/* { dg-do run } */
> +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> +
> +#include "../../gcc.dg/torture/crc-10.c"
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-12-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-12-pmul.c
> new file mode 100644
> index 00000000000..16d901eeaef
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-12-pmul.c
> @@ -0,0 +1,9 @@
> +/* { dg-do run } */
> +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc -fdisable-tree-phiopt2 -fdisable-tree-phiopt3" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> +
> +#include "../../gcc.dg/torture/crc-12.c"
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-13-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-13-pmul.c
> new file mode 100644
> index 00000000000..5f7741fad0f
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-13-pmul.c
> @@ -0,0 +1,9 @@
> +/* { dg-do run } */
> +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> +
> +#include "../../gcc.dg/torture/crc-13.c"
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-14-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-14-pmul.c
> new file mode 100644
> index 00000000000..cdedbbd3db1
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-14-pmul.c
> @@ -0,0 +1,9 @@
> +/* { dg-do run } */
> +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> +
> +#include "../../gcc.dg/torture/crc-14.c"
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-17-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-17-pmul.c
> new file mode 100644
> index 00000000000..c219e49a2b1
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-17-pmul.c
> @@ -0,0 +1,9 @@
> +/* { dg-do run } */
> +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> +
> +#include "../../gcc.dg/torture/crc-17.c"
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-18-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-18-pmul.c
> new file mode 100644
> index 00000000000..124900a979b
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-18-pmul.c
> @@ -0,0 +1,9 @@
> +/* { dg-do run } */
> +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> +
> +#include "../../gcc.dg/torture/crc-18.c"
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-21-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-21-pmul.c
> new file mode 100644
> index 00000000000..3cae1a7f57b
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-21-pmul.c
> @@ -0,0 +1,9 @@
> +/* { dg-do run } */
> +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> +
> +#include "../../gcc.dg/torture/crc-21.c"
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-22-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-22-pmul.c
> new file mode 100644
> index 00000000000..0ec2e312f8f
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-22-pmul.c
> @@ -0,0 +1,9 @@
> +/* { dg-do run } */
> +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> +
> +#include "../../gcc.dg/torture/crc-22.c"
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-23-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-23-pmul.c
> new file mode 100644
> index 00000000000..0c4542adb40
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-23-pmul.c
> @@ -0,0 +1,9 @@
> +/* { dg-do run } */
> +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> +
> +#include "../../gcc.dg/torture/crc-23.c"
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-4-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-4-pmul.c
> new file mode 100644
> index 00000000000..08f1d3b69d7
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-4-pmul.c
> @@ -0,0 +1,9 @@
> +/* { dg-do run } */
> +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> +
> +#include "../../gcc.dg/torture/crc-4.c"
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-5-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-5-pmul.c
> new file mode 100644
> index 00000000000..91bf5e6353d
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-5-pmul.c
> @@ -0,0 +1,9 @@
> +/* { dg-do run } */
> +/* { dg-options "-march=armv8-a+crypto -O2 -w -fdump-rtl-dfinish -fdump-tree-crc" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> +
> +#include "../../gcc.dg/torture/crc-5.c"
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> \ No newline at end of file
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-6-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-6-pmul.c
> new file mode 100644
> index 00000000000..4680eafe758
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-6-pmul.c
> @@ -0,0 +1,9 @@
> +/* { dg-do run } */
> +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> +
> +#include "../../gcc.dg/torture/crc-6.c"
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> \ No newline at end of file
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-7-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-7-pmul.c
> new file mode 100644
> index 00000000000..655484d10d4
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-7-pmul.c
> @@ -0,0 +1,9 @@
> +/* { dg-do run } */
> +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> +
> +#include "../../gcc.dg/torture/crc-7.c"
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-8-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-8-pmul.c
> new file mode 100644
> index 00000000000..6c2acc84c32
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-8-pmul.c
> @@ -0,0 +1,9 @@
> +/* { dg-do run } */
> +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> +
> +#include "../../gcc.dg/torture/crc-8.c"
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-9-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-9-pmul.c
> new file mode 100644
> index 00000000000..e76f3c77b59
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-9-pmul.c
> @@ -0,0 +1,9 @@
> +/* { dg-do run } */
> +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> +
> +#include "../../gcc.dg/torture/crc-9.c"
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data16-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data16-pmul.c
> new file mode 100644
> index 00000000000..21520474564
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data16-pmul.c
> @@ -0,0 +1,9 @@
> +/* { dg-do run } */
> +/* { dg-options "-w -march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> +
> +#include "../../gcc.dg/torture/crc-CCIT-data16.c"
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> \ No newline at end of file
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data8-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data8-pmul.c
> new file mode 100644
> index 00000000000..3dcc92320f3
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data8-pmul.c
> @@ -0,0 +1,9 @@
> +/* { dg-do run } */
> +/* { dg-options "-w -march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto" } } */
> +
> +#include "../../gcc.dg/torture/crc-CCIT-data8.c"
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> \ No newline at end of file
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-coremark-16bitdata-pmul.c b/gcc/testsuite/gcc.target/aarch64/crc-coremark-16bitdata-pmul.c
> new file mode 100644
> index 00000000000..e5196aaafef
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-coremark-16bitdata-pmul.c
> @@ -0,0 +1,9 @@
> +/* { dg-do run } */
> +/* { dg-options "-w -march=armv8-a+crypto -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> +
> +#include "../../gcc.dg/torture/crc-coremark16-data16.c"
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> \ No newline at end of file
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32-data16.c b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data16.c
> new file mode 100644
> index 00000000000..e82cb04fcc3
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data16.c
> @@ -0,0 +1,53 @@
> +/* { dg-do run } */
> +/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> +
> +#include <stdint.h>
> +#include <stdlib.h>
> +
> +__attribute__ ((noinline,optimize(0)))
> +uint32_t _crc32_O0 (uint32_t crc, uint16_t data) {
> +  int i;
> +  crc = crc ^ data;
> +
> +  for (i = 0; i < 8; i++) {
> +      if (crc & 1)
> +	crc = (crc >> 1) ^ 0xEDB88320;
> +      else
> +	crc = (crc >> 1);
> +    }
> +
> +  return crc;
> +}
> +
> +uint32_t _crc32 (uint32_t crc, uint16_t data) {
> +  int i;
> +  crc = crc ^ data;
> +
> +  for (i = 0; i < 8; i++) {
> +      if (crc & 1)
> +	crc = (crc >> 1) ^ 0xEDB88320;
> +      else
> +	crc = (crc >> 1);
> +    }
> +
> +  return crc;
> +}
> +
> +int main ()
> +{
> +  uint32_t crc = 0x0D800D80;
> +  for (uint16_t i = 0; i < 0xffff; i++)
> +    {
> +      uint32_t res1 = _crc32_O0 (crc, i);
> +      uint32_t res2 = _crc32 (crc, i);
> +      if (res1 != res2)
> +	 abort ();
> +      crc = res1;
> +    }
> +}
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "UNSPEC_CRC32" "dfinish"} } */
> +/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32-data32.c b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data32.c
> new file mode 100644
> index 00000000000..a7564a7e28a
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data32.c
> @@ -0,0 +1,52 @@
> +/* { dg-do run } */
> +/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> +
> +#include <stdint.h>
> +#include <stdlib.h>
> +__attribute__ ((noinline,optimize(0)))
> +uint32_t _crc32_O0 (uint32_t crc, uint32_t data) {
> +  int i;
> +  crc = crc ^ data;
> +
> +  for (i = 0; i < 32; i++) {
> +      if (crc & 1)
> +	crc = (crc >> 1) ^ 0xEDB88320;
> +      else
> +	crc = (crc >> 1);
> +    }
> +
> +  return crc;
> +}
> +
> +uint32_t _crc32 (uint32_t crc, uint32_t data) {
> +  int i;
> +  crc = crc ^ data;
> +
> +  for (i = 0; i < 32; i++) {
> +      if (crc & 1)
> +	crc = (crc >> 1) ^ 0xEDB88320;
> +      else
> +	crc = (crc >> 1);
> +    }
> +
> +  return crc;
> +}
> +
> +int main ()
> +{
> +  uint32_t crc = 0x0D800D80;
> +  for (uint8_t i = 0; i < 0xff; i++)
> +    {
> +      uint32_t res1 = _crc32_O0 (crc, i);
> +      uint32_t res2 = _crc32 (crc, i);
> +      if (res1 != res2)
> +	 abort ();
> +      crc = res1;
> +    }
> +}
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "UNSPEC_CRC32" "dfinish"} } */
> +/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32-data8.c b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data8.c
> new file mode 100644
> index 00000000000..c88cafadedc
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data8.c
> @@ -0,0 +1,53 @@
> +/* { dg-do run } */
> +/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> +
> +#include <stdint.h>
> +#include <stdlib.h>
> +
> +__attribute__ ((noinline,optimize(0)))
> +uint32_t _crc32_O0 (uint32_t crc, uint8_t data) {
> +  int i;
> +  crc = crc ^ data;
> +
> +  for (i = 0; i < 8; i++) {
> +      if (crc & 1)
> +	crc = (crc >> 1) ^ 0xEDB88320;
> +      else
> +	crc = (crc >> 1);
> +    }
> +
> +  return crc;
> +}
> +
> +uint32_t _crc32 (uint32_t crc, uint8_t data) {
> +  int i;
> +  crc = crc ^ data;
> +
> +  for (i = 0; i < 8; i++) {
> +      if (crc & 1)
> +	crc = (crc >> 1) ^ 0xEDB88320;
> +      else
> +	crc = (crc >> 1);
> +    }
> +
> +  return crc;
> +}
> +
> +int main ()
> +{
> +  uint32_t crc = 0x0D800D80;
> +  for (uint8_t i = 0; i < 0xff; i++)
> +    {
> +      uint32_t res1 = _crc32_O0 (crc, i);
> +      uint32_t res2 = _crc32 (crc, i);
> +      if (res1 != res2)
> +	 abort ();
> +      crc = res1;
> +    }
> +}
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "UNSPEC_CRC32" "dfinish"} } */
> +/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data16.c b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data16.c
> new file mode 100644
> index 00000000000..d82e6252603
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data16.c
> @@ -0,0 +1,53 @@
> +/* { dg-do run } */
> +/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> +
> +#include <stdint.h>
> +#include <stdlib.h>
> +
> +__attribute__ ((noinline,optimize(0)))
> +uint32_t _crc32_O0 (uint32_t crc, uint16_t data) {
> +  int i;
> +  crc = crc ^ data;
> +
> +  for (i = 0; i < 8; i++) {
> +      if (crc & 1)
> +	crc = (crc >> 1) ^ 0x82F63B78;
> +      else
> +	crc = (crc >> 1);
> +    }
> +
> +  return crc;
> +}
> +
> +uint32_t _crc32 (uint32_t crc, uint16_t data) {
> +  int i;
> +  crc = crc ^ data;
> +
> +  for (i = 0; i < 8; i++) {
> +      if (crc & 1)
> +	crc = (crc >> 1) ^ 0x82F63B78;
> +      else
> +	crc = (crc >> 1);
> +    }
> +
> +  return crc;
> +}
> +
> +int main ()
> +{
> +  uint32_t crc = 0x0D800D80;
> +  for (uint16_t i = 0; i < 0xffff; i++)
> +    {
> +      uint32_t res1 = _crc32_O0 (crc, i);
> +      uint32_t res2 = _crc32 (crc, i);
> +      if (res1 != res2)
> +	 abort ();
> +      crc = res1;
> +    }
> +}
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "UNSPEC_CRC32C" "dfinish"} } */
> +/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data32.c b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data32.c
> new file mode 100644
> index 00000000000..7acb6fc239c
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data32.c
> @@ -0,0 +1,52 @@
> +/* { dg-do run } */
> +/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> +
> +#include <stdint.h>
> +#include <stdlib.h>
> +__attribute__ ((noinline,optimize(0)))
> +uint32_t _crc32_O0 (uint32_t crc, uint32_t data) {
> +  int i;
> +  crc = crc ^ data;
> +
> +  for (i = 0; i < 32; i++) {
> +      if (crc & 1)
> +	crc = (crc >> 1) ^ 0x82F63B78;
> +      else
> +	crc = (crc >> 1);
> +    }
> +
> +  return crc;
> +}
> +
> +uint32_t _crc32 (uint32_t crc, uint32_t data) {
> +  int i;
> +  crc = crc ^ data;
> +
> +  for (i = 0; i < 32; i++) {
> +      if (crc & 1)
> +	crc = (crc >> 1) ^ 0x82F63B78;
> +      else
> +	crc = (crc >> 1);
> +    }
> +
> +  return crc;
> +}
> +
> +int main ()
> +{
> +  uint32_t crc = 0x0D800D80;
> +  for (uint8_t i = 0; i < 0xff; i++)
> +    {
> +      uint32_t res1 = _crc32_O0 (crc, i);
> +      uint32_t res2 = _crc32 (crc, i);
> +      if (res1 != res2)
> +	 abort ();
> +      crc = res1;
> +    }
> +}
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "UNSPEC_CRC32C" "dfinish"} } */
> +/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */
> diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data8.c b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data8.c
> new file mode 100644
> index 00000000000..e8a8901e453
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data8.c
> @@ -0,0 +1,53 @@
> +/* { dg-do run } */
> +/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish -fdump-tree-crc" } */
> +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> +
> +#include <stdint.h>
> +#include <stdlib.h>
> +
> +__attribute__ ((noinline,optimize(0)))
> +uint32_t _crc32_O0 (uint32_t crc, uint8_t data) {
> +  int i;
> +  crc = crc ^ data;
> +
> +  for (i = 0; i < 8; i++) {
> +      if (crc & 1)
> +	crc = (crc >> 1) ^ 0x82F63B78;
> +      else
> +	crc = (crc >> 1);
> +    }
> +
> +  return crc;
> +}
> +
> +uint32_t _crc32 (uint32_t crc, uint8_t data) {
> +  int i;
> +  crc = crc ^ data;
> +
> +  for (i = 0; i < 8; i++) {
> +      if (crc & 1)
> +	crc = (crc >> 1) ^ 0x82F63B78;
> +      else
> +	crc = (crc >> 1);
> +    }
> +
> +  return crc;
> +}
> +
> +int main ()
> +{
> +  uint32_t crc = 0x0D800D80;
> +  for (uint8_t i = 0; i < 0xff; i++)
> +    {
> +      uint32_t res1 = _crc32_O0 (crc, i);
> +      uint32_t res2 = _crc32 (crc, i);
> +      if (res1 != res2)
> +	 abort ();
> +      crc = res1;
> +    }
> +}
> +
> +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC code." 0 "crc"} } */
> +/* { dg-final { scan-rtl-dump "UNSPEC_CRC32C" "dfinish"} } */
> +/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */

Re: [RFC/RFA][PATCH v3 06/12] aarch64: Implement new expander for efficient CRC computation

[Richard Sandiford]

Richard Sandiford <richard.sandiford@arm.com> writes:
> [...]
>   /* Perform a shift right by CRC_SIZE as an extraction of lane 1.  */
>   machine_mode crc_vmode = aarch64_vq_mode (crc_mode).require ();
>   a0 = (crc_size > data_size ? gen_reg_rtx (crc_mode) : operands[0]);
>   emit_insn (gen_aarch64_get_lane (crc_vmode, a0,
> 				   gen_lowpart (crc_vmode, clmul_res),
> 				   aarch64_endian_lane_rtx (crc_vmode, 1)));

Sorry, I forgot to say that I'd locally patched:

diff --git a/gcc/config/aarch64/aarch64-simd.md b/gcc/config/aarch64/aarch64-simd.md
index 816f499e963..af7beecb735 100644
--- a/gcc/config/aarch64/aarch64-simd.md
+++ b/gcc/config/aarch64/aarch64-simd.md
@@ -4301,7 +4301,7 @@ (define_insn "*aarch64_get_lane_zero_extend<GPI:mode><VDQQH:mode>"
 ;; RTL uses GCC vector extension indices throughout so flip only for assembly.
 ;; Extracting lane zero is split into a simple move when it is between SIMD
 ;; registers or a store.
-(define_insn_and_split "aarch64_get_lane<mode>"
+(define_insn_and_split "@aarch64_get_lane<mode>"
   [(set (match_operand:<VEL> 0 "aarch64_simd_nonimmediate_operand" "=?r, w, Utv")
 	(vec_select:<VEL>
 	  (match_operand:VALL_F16 1 "register_operand" "w, w, w")

Richard

Re: [RFC/RFA][PATCH v3 06/12] aarch64: Implement new expander for efficient CRC computation

[Mariam Arutunian]

[-- Attachment #1: Type: text/plain, Size: 41289 bytes --]

On Fri, Aug 9, 2024 at 7:22 PM Richard Sandiford <richard.sandiford@arm.com>
wrote:

> Sorry again for the slow review. :(
>
> I only really looked at the unreversed version earlier, on the basis
> that the comments would apply to both versions.  But I've got a couple
> of comments about the reversed version below:
>
> Mariam Arutunian <mariamarutunian@gmail.com> writes:
> > [...]
> > diff --git a/gcc/config/aarch64/aarch64.cc
> b/gcc/config/aarch64/aarch64.cc
> > index ee12d8897a8..546a379fd74 100644
> > --- a/gcc/config/aarch64/aarch64.cc
> > +++ b/gcc/config/aarch64/aarch64.cc
> > @@ -30265,6 +30265,126 @@ aarch64_retrieve_sysreg (const char *regname,
> bool write_p, bool is128op)
> >    return sysreg->encoding;
> >  }
> >
> > +/* Generate assembly to calculate CRC
> > +   using carry-less multiplication instruction.
> > +   OPERANDS[1] is input CRC,
> > +   OPERANDS[2] is data (message),
> > +   OPERANDS[3] is the polynomial without the leading 1.  */
> > +
> > +void
> > +aarch64_expand_crc_using_pmull (scalar_mode crc_mode,
> > +                             scalar_mode data_mode,
> > +                             rtx *operands)
> > +{
> > +  /* Check and keep arguments.  */
> > +  gcc_assert (!CONST_INT_P (operands[0]));
> > +  gcc_assert (CONST_INT_P (operands[3]));
> > +  rtx crc = operands[1];
> > +  rtx data = operands[2];
> > +  rtx polynomial = operands[3];
> > +
> > +  unsigned HOST_WIDE_INT crc_size = GET_MODE_BITSIZE (crc_mode);
> > +  unsigned HOST_WIDE_INT data_size = GET_MODE_BITSIZE (data_mode);
> > +  gcc_assert (crc_size <= 32);
> > +  gcc_assert (data_size <= crc_size);
> > +
> > +  /* Calculate the quotient.  */
> > +  unsigned HOST_WIDE_INT
> > +      q = gf2n_poly_long_div_quotient (UINTVAL (polynomial), crc_size);
> > +  /* CRC calculation's main part.  */
> > +  if (crc_size > data_size)
> > +    crc = expand_shift (RSHIFT_EXPR, DImode, crc, crc_size - data_size,
> > +                     NULL_RTX, 1);
> > +
> > +  rtx t0 = force_reg (DImode, gen_int_mode (q, DImode));
> > +  polynomial = simplify_gen_unary (ZERO_EXTEND, DImode, polynomial,
> > +                                GET_MODE (polynomial));
> > +  rtx t1 = force_reg (DImode, polynomial);
> > +
> > +  rtx a0 = expand_binop (DImode, xor_optab, crc, data, NULL_RTX, 1,
> > +                      OPTAB_WIDEN);
> > +
> > +  rtx clmul_res = gen_reg_rtx (TImode);
> > +  emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t0));
> > +  a0 = gen_lowpart (DImode, clmul_res);
> > +
> > +  a0 = expand_shift (RSHIFT_EXPR, DImode, a0, crc_size, NULL_RTX, 1);
> > +
> > +  emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t1));
> > +  a0 = gen_lowpart (DImode, clmul_res);
> > +
> > +  if (crc_size > data_size)
> > +    {
> > +      rtx crc_part = expand_shift (LSHIFT_EXPR, DImode, operands[1],
> data_size,
> > +                                NULL_RTX, 0);
> > +      a0 = expand_binop (DImode, xor_optab, a0, crc_part, NULL_RTX, 1,
> > +                      OPTAB_DIRECT);
> > +    }
> > +
> > +  /* Zero upper bits beyond crc_size.  */
>
> The comment no longer applies.  Otherwise this function looks good to me.
>
>
Ok.)


> > +  aarch64_emit_move (operands[0], gen_lowpart (crc_mode, a0));
> > +}
> > +
> > +/* Generate assembly to calculate reversed CRC
> > +   using carry-less multiplication instruction.
> > +   OPERANDS[1] is input CRC,
> > +   OPERANDS[2] is data,
> > +   OPERANDS[3] is the polynomial without the leading 1.  */
> > +
> > +void
> > +aarch64_expand_reversed_crc_using_pmull (scalar_mode crc_mode,
> > +                                      scalar_mode data_mode,
> > +                                      rtx *operands)
> > +{
> > +  /* Check and keep arguments.  */
> > +  gcc_assert (!CONST_INT_P (operands[0]));
> > +  gcc_assert (CONST_INT_P (operands[3]));
> > +  rtx crc = operands[1];
> > +  rtx data = operands[2];
> > +  rtx polynomial = operands[3];
> > +
> > +  unsigned HOST_WIDE_INT crc_size = GET_MODE_BITSIZE (crc_mode);
> > +  unsigned HOST_WIDE_INT data_size = GET_MODE_BITSIZE (data_mode);
> > +  gcc_assert (crc_size <= 32);
> > +  gcc_assert (data_size <= crc_size);
> > +
> > +  /* Calculate the quotient.  */
> > +  unsigned HOST_WIDE_INT
> > +      q = gf2n_poly_long_div_quotient (UINTVAL (polynomial), crc_size);
> > +  /* Reflect the calculated quotient.  */
> > +  q = reflect (q);
> > +  rtx t0 = force_reg (DImode, gen_int_mode (q >> (data_size - 4),
> DImode));
> > +
> > +  /* Reflect the polynomial.  */
> > +  unsigned HOST_WIDE_INT ref_polynomial = reflect (UINTVAL
> (polynomial));
>
> It looks like reflect() autodetects the bitwidth based on the assumption
> that the upper half will be nonzero.  But that might not be true for all
> possible polynomials (when the implicit leading coefficient is absent)
> E.g. it looks like the 64-bit HDLC CRC polynomial is 0x1b (just the
> lowest byte nonzero), and although we don't support 64-bit polynomials
> here, the approach wouldn't work for it.
>
> I think it'd be safer to pass the bitwidth as an explicit parameter.
> Also, maybe it could go in hwint.* instead of expr.* and be called
> something like reflect_hwi.
>

Yes. In the CRC loop recognition part, I only support those polynomials
whose upper half is nonzero. I.e required memories for the polynomial and
the CRC are equal. The reason for this is that, as the polynomial's leading
bit is emitted, it is not possible to precisely assume what CRC is being
calculated (e.g., 64, 32, etc.). Because there are some implementations,
where more memory than needed is used to store the CRC. For example, in
some implementations, a 64-bit memory may be used to calculate a 32-bit CRC.
But, I agree, it's safer to add an explicit parameter.
I'll make the suggested changes.


> > +  rtx t1 = force_reg (DImode, gen_int_mode (ref_polynomial << 1,
> DImode));
> > +
> > +  /* CRC calculation's main part.  */
> > +  rtx a0 = expand_binop (DImode, xor_optab, crc, data, NULL_RTX, 1,
> > +                      OPTAB_WIDEN);
> > +
> > +  /* Perform carry-less multiplication and get low part.  */
> > +  rtx clmul_res = gen_reg_rtx (TImode);
> > +  emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t0));
> > +  a0 = gen_lowpart (DImode, clmul_res);
> > +
> > +  a0 = expand_shift (LSHIFT_EXPR, DImode, a0, 64 - crc_size - 3,
> NULL_RTX, 0);
>
> I haven't really looked at this implementation strategy before, so this
> is probably a silly question sorry :-) but is there a specific reason
> for selecting the bias of 4 in the right shift above, cancelled out by
> subtracting the same amount from this left shift?  It looks like we could
> use any value in the range [1. crc_size - 1], is that right?
>
> Just asking out of curiosity though.  I agree it works.
>
>
Sorry, I don’t remember exactly why I used a bias of 4. Upon reviewing it,
I think it's related to the reflection process, where zeros are added to
the quotient.
For example, in the case of CRC-8, the quotient is 9 bits but is stored in
a 16-bit value, so 7 bits are zeros.
It might be better to use q >> (crc_size - 1) and then do the left shift
with 64 - data_size instead of 64 - crc_size - 3.
However, I noticed that you provided a better solution. Thank you very
much.)

> +
> > +  /* Perform carry-less multiplication and get high part.  */
> > +  emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t1));
> > +  a0 = gen_highpart (DImode, clmul_res);
>
> Although this works, it's taking a DImode highpart of a single TImode
> register, which requires a spill from the register allocator.  So we get
> something like:
>
>         pmull   v31.1q, v31.1d, v14.1d
>         str     q31, [sp, 48]
>         ldr     d31, [sp, 48]
>         shl     d31, d31, 53
>         pmull   v31.1q, v31.1d, v15.1d
>         str     q31, [sp, 48]
>         ldrb    w0, [sp, 56]
>
> (I think the first STR and LDR come from reusing the same pseudo register
> for the temporary results; it's the STR+LDRB that causes the spill.)
>
> One way around that would be:
>
>   emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t1));
>   a0 = gen_reg_rtx (DImode);
>   rtx v2di_res = gen_lowpart (V2DImode, clmul_res);
>   rtx high_index = BYTES_BIG_ENDIAN ? const0_rtx : const1_rtx;
>   emit_insn (gen_aarch64_get_lanev2di (a0, v2di_res, high_index));
>
> That generates:
>
>         pmull   v30.1q, v29.1d, v30.1d
>         shl     d30, d30, 53
>         pmull   v30.1q, v30.1d, v31.1d
>         umov    x0, v30.d[1]
>
> which is pretty nice.
>

There again, it might not be too bad if we ionstead used the flipped
> version of the unreversed approach, such as the following minor adaption:
>
>   /* Calculate the quotient.  */
>   unsigned HOST_WIDE_INT
>       q = gf2n_poly_long_div_quotient (UINTVAL (polynomial), crc_size);
>   /* Reflect the calculated quotient.  */
>   q = reflect_hwi (q, crc_size + 1);
>   rtx t0 = force_reg (DImode, gen_int_mode (q, DImode));
>
>   /* Reflect the polynomial.  */
>   unsigned HOST_WIDE_INT ref_polynomial = reflect_hwi (UINTVAL
> (polynomial),
>                                                        crc_size);
>   /* An unshifted multiplier would require the final result to be extracted
>      using a shift right by DATA_SIZE - 1 bits.  Shift the multiplier left
>      so that the shift right can be by CRC_SIZE bits instead.  */
>   ref_polynomial <<= crc_size - data_size + 1;
>   rtx t1 = force_reg (DImode, gen_int_mode (ref_polynomial, DImode));
>
>   /* CRC calculation's main part.  */
>   rtx a0 = expand_binop (DImode, xor_optab, crc, data, NULL_RTX, 1,
>                          OPTAB_WIDEN);
>
>   /* Perform carry-less multiplication and get low part.  */
>   rtx clmul_res = gen_reg_rtx (TImode);
>   emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t0));
>   a0 = gen_lowpart (DImode, clmul_res);
>
>   a0 = expand_binop (DImode, and_optab, a0,
>                      gen_int_mode (GET_MODE_MASK (data_mode), DImode),
>                      NULL_RTX, 1, OPTAB_WIDEN);
>
>   /* Perform carry-less multiplication.  */
>   emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t1));
>
>   /* Perform a shift right by CRC_SIZE as an extraction of lane 1.  */
>   machine_mode crc_vmode = aarch64_vq_mode (crc_mode).require ();
>   a0 = (crc_size > data_size ? gen_reg_rtx (crc_mode) : operands[0]);
>   emit_insn (gen_aarch64_get_lane (crc_vmode, a0,
>                                    gen_lowpart (crc_vmode, clmul_res),
>                                    aarch64_endian_lane_rtx (crc_vmode,
> 1)));
>
>   if (crc_size > data_size)
>     {
>       rtx crc_part = expand_shift (RSHIFT_EXPR, crc_mode, crc, data_size,
>                                    NULL_RTX, 1);
>       a0 = expand_binop (crc_mode, xor_optab, a0, crc_part, operands[0], 1,
>                          OPTAB_WIDEN);
>       aarch64_emit_move (operands[0], a0);
>     }
>
> This gives:
>
>         pmull   v30.1q, v30.1d, v31.1d
>         movi    v31.2d, 0xff
>         and     v30.8b, v30.8b, v31.8b
>         pmull   v30.1q, v30.1d, v31.1d
>         umov    w0, v30.b[1]
>
> or, with SVE enabled:
>
>         pmull   v30.1q, v31.1d, v30.1d
>         and     z30.d, z30.d, #255
>         pmull   v30.1q, v30.1d, v31.1d
>         umov    w0, v30.b[1]
>
> This is preferable since ANDs are generally cheaper than shifts.
>
> That's just a suggestion though; the original version is ok too.
>
>
Thanks for the suggestions. I don't have much experience with gcc, so this
is a huge help for me.
I'll apply all the changes and send the new version.


> Minor comment: it'd be good to use "pmull_res" instead of "clmul_res"
> for the variables: :)
>
>
Ok)


> LGTM otherwise.
>
>
Thanks,
Mariam


> Thanks,
> Richard
>
> > +
> > +  if (crc_size > data_size)
> > +    {
> > +      rtx crc_part = expand_shift (RSHIFT_EXPR, DImode, crc, data_size,
> > +                                NULL_RTX, 1);
> > +      a0 = expand_binop (DImode, xor_optab, a0, crc_part, NULL_RTX, 1,
> > +                      OPTAB_DIRECT);
> > +    }
> > +
> > +  aarch64_emit_move (operands[0], gen_lowpart (crc_mode, a0));
> > +}
> > +
> >  /* Target-specific selftests.  */
> >
> >  #if CHECKING_P
> > diff --git a/gcc/config/aarch64/aarch64.md
> b/gcc/config/aarch64/aarch64.md
> > index 9dff2d7a2b0..08c588bc475 100644
> > --- a/gcc/config/aarch64/aarch64.md
> > +++ b/gcc/config/aarch64/aarch64.md
> > @@ -4543,6 +4543,63 @@
> >    [(set_attr "type" "crc")]
> >  )
> >
> > +;; Reversed CRC
> > +(define_expand "crc_rev<ALLI:mode><ALLX:mode>4"
> > +  [;; return value (calculated CRC)
> > +   (match_operand:ALLX 0 "register_operand" "=r")
> > +   ;; initial CRC
> > +   (match_operand:ALLX 1 "register_operand" "r")
> > +   ;; data
> > +   (match_operand:ALLI 2 "register_operand" "r")
> > +   ;; polynomial without leading 1
> > +   (match_operand:ALLX 3)]
> > +  ""
> > +  {
> > +    /* If the polynomial is the same as the polynomial of crc32c*
> instruction,
> > +       put that instruction.  crc32c uses iSCSI polynomial.  */
> > +    if (TARGET_CRC32 && INTVAL (operands[3]) == 0x1EDC6F41
> > +     && <ALLX:MODE>mode == SImode)
> > +      emit_insn (gen_aarch64_crc32c<ALLI:crc_data_type> (operands[0],
> > +                                                      operands[1],
> > +                                                      operands[2]));
> > +    /* If the polynomial is the same as the polynomial of crc32*
> instruction,
> > +     put that instruction.  crc32 uses HDLC etc.  polynomial.  */
> > +    else if (TARGET_CRC32 && INTVAL (operands[3]) == 0x04C11DB7
> > +          && <ALLX:MODE>mode == SImode)
> > +      emit_insn (gen_aarch64_crc32<ALLI:crc_data_type> (operands[0],
> > +                                                     operands[1],
> > +                                                     operands[2]));
> > +    else if (TARGET_AES && <ALLI:sizen> <= <ALLX:sizen>)
> > +      aarch64_expand_reversed_crc_using_pmull (<ALLX:MODE>mode,
> > +                                            <ALLI:MODE>mode,
> > +                                            operands);
> > +    else
> > +      /* Otherwise, generate table-based CRC.  */
> > +      expand_reversed_crc_table_based (operands[0], operands[1],
> operands[2],
> > +                                    operands[3], <ALLI:MODE>mode,
> > +                                    generate_reflecting_code_standard);
> > +    DONE;
> > +  }
> > +)
> > +
> > +;; Bit-forward CRC
> > +(define_expand "crc<ALLI:mode><ALLX:mode>4"
> > +  [;; return value (calculated CRC)
> > +   (match_operand:ALLX 0 "register_operand" "=r")
> > +   ;; initial CRC
> > +   (match_operand:ALLX 1 "register_operand" "r")
> > +   ;; data
> > +   (match_operand:ALLI 2 "register_operand" "r")
> > +   ;; polynomial without leading 1
> > +   (match_operand:ALLX 3)]
> > +  "TARGET_AES && <ALLI:sizen> <= <ALLX:sizen>"
> > +  {
> > +    aarch64_expand_crc_using_pmull (<ALLX:MODE>mode, <ALLI:MODE>mode,
> > +                                 operands);
> > +    DONE;
> > +  }
> > +)
> > +
> >  (define_insn "*csinc2<mode>_insn"
> >    [(set (match_operand:GPI 0 "register_operand" "=r")
> >          (plus:GPI (match_operand 2 "aarch64_comparison_operation" "")
> > diff --git a/gcc/config/aarch64/iterators.md
> b/gcc/config/aarch64/iterators.md
> > index 99cde46f1ba..86e4863d684 100644
> > --- a/gcc/config/aarch64/iterators.md
> > +++ b/gcc/config/aarch64/iterators.md
> > @@ -1276,6 +1276,10 @@
> >  ;; Map a mode to a specific constraint character.
> >  (define_mode_attr cmode [(QI "q") (HI "h") (SI "s") (DI "d")])
> >
> > +;; Map a mode to a specific constraint character for calling
> > +;; appropriate version of crc.
> > +(define_mode_attr crc_data_type [(QI "b") (HI "h") (SI "w") (DI "x")])
> > +
> >  ;; Map modes to Usg and Usj constraints for SISD right shifts
> >  (define_mode_attr cmode_simd [(SI "g") (DI "j")])
> >
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-1-pmul.c
> b/gcc/testsuite/gcc.target/aarch64/crc-1-pmul.c
> > new file mode 100644
> > index 00000000000..4043251dbd8
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-1-pmul.c
> > @@ -0,0 +1,8 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
> -fdump-tree-crc -fdisable-tree-phiopt2 -fdisable-tree-phiopt3" } */
> > +
> > +#include "../../gcc.dg/torture/crc-1.c"
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> > \ No newline at end of file
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-10-pmul.c
> b/gcc/testsuite/gcc.target/aarch64/crc-10-pmul.c
> > new file mode 100644
> > index 00000000000..dd866b38e83
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-10-pmul.c
> > @@ -0,0 +1,9 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
> -fdump-tree-crc" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> > +
> > +#include "../../gcc.dg/torture/crc-10.c"
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-12-pmul.c
> b/gcc/testsuite/gcc.target/aarch64/crc-12-pmul.c
> > new file mode 100644
> > index 00000000000..16d901eeaef
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-12-pmul.c
> > @@ -0,0 +1,9 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
> -fdump-tree-crc -fdisable-tree-phiopt2 -fdisable-tree-phiopt3" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> > +
> > +#include "../../gcc.dg/torture/crc-12.c"
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-13-pmul.c
> b/gcc/testsuite/gcc.target/aarch64/crc-13-pmul.c
> > new file mode 100644
> > index 00000000000..5f7741fad0f
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-13-pmul.c
> > @@ -0,0 +1,9 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
> -fdump-tree-crc" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> > +
> > +#include "../../gcc.dg/torture/crc-13.c"
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-14-pmul.c
> b/gcc/testsuite/gcc.target/aarch64/crc-14-pmul.c
> > new file mode 100644
> > index 00000000000..cdedbbd3db1
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-14-pmul.c
> > @@ -0,0 +1,9 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
> -fdump-tree-crc" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> > +
> > +#include "../../gcc.dg/torture/crc-14.c"
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-17-pmul.c
> b/gcc/testsuite/gcc.target/aarch64/crc-17-pmul.c
> > new file mode 100644
> > index 00000000000..c219e49a2b1
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-17-pmul.c
> > @@ -0,0 +1,9 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
> -fdump-tree-crc" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> > +
> > +#include "../../gcc.dg/torture/crc-17.c"
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-18-pmul.c
> b/gcc/testsuite/gcc.target/aarch64/crc-18-pmul.c
> > new file mode 100644
> > index 00000000000..124900a979b
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-18-pmul.c
> > @@ -0,0 +1,9 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
> -fdump-tree-crc" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> > +
> > +#include "../../gcc.dg/torture/crc-18.c"
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-21-pmul.c
> b/gcc/testsuite/gcc.target/aarch64/crc-21-pmul.c
> > new file mode 100644
> > index 00000000000..3cae1a7f57b
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-21-pmul.c
> > @@ -0,0 +1,9 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
> -fdump-tree-crc" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> > +
> > +#include "../../gcc.dg/torture/crc-21.c"
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-22-pmul.c
> b/gcc/testsuite/gcc.target/aarch64/crc-22-pmul.c
> > new file mode 100644
> > index 00000000000..0ec2e312f8f
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-22-pmul.c
> > @@ -0,0 +1,9 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
> -fdump-tree-crc" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> > +
> > +#include "../../gcc.dg/torture/crc-22.c"
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-23-pmul.c
> b/gcc/testsuite/gcc.target/aarch64/crc-23-pmul.c
> > new file mode 100644
> > index 00000000000..0c4542adb40
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-23-pmul.c
> > @@ -0,0 +1,9 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
> -fdump-tree-crc" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> > +
> > +#include "../../gcc.dg/torture/crc-23.c"
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-4-pmul.c
> b/gcc/testsuite/gcc.target/aarch64/crc-4-pmul.c
> > new file mode 100644
> > index 00000000000..08f1d3b69d7
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-4-pmul.c
> > @@ -0,0 +1,9 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
> -fdump-tree-crc" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> > +
> > +#include "../../gcc.dg/torture/crc-4.c"
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-5-pmul.c
> b/gcc/testsuite/gcc.target/aarch64/crc-5-pmul.c
> > new file mode 100644
> > index 00000000000..91bf5e6353d
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-5-pmul.c
> > @@ -0,0 +1,9 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-march=armv8-a+crypto -O2 -w -fdump-rtl-dfinish
> -fdump-tree-crc" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> > +
> > +#include "../../gcc.dg/torture/crc-5.c"
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> > \ No newline at end of file
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-6-pmul.c
> b/gcc/testsuite/gcc.target/aarch64/crc-6-pmul.c
> > new file mode 100644
> > index 00000000000..4680eafe758
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-6-pmul.c
> > @@ -0,0 +1,9 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
> -fdump-tree-crc" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> > +
> > +#include "../../gcc.dg/torture/crc-6.c"
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> > \ No newline at end of file
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-7-pmul.c
> b/gcc/testsuite/gcc.target/aarch64/crc-7-pmul.c
> > new file mode 100644
> > index 00000000000..655484d10d4
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-7-pmul.c
> > @@ -0,0 +1,9 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
> -fdump-tree-crc" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> > +
> > +#include "../../gcc.dg/torture/crc-7.c"
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-8-pmul.c
> b/gcc/testsuite/gcc.target/aarch64/crc-8-pmul.c
> > new file mode 100644
> > index 00000000000..6c2acc84c32
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-8-pmul.c
> > @@ -0,0 +1,9 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
> -fdump-tree-crc" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> > +
> > +#include "../../gcc.dg/torture/crc-8.c"
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-9-pmul.c
> b/gcc/testsuite/gcc.target/aarch64/crc-9-pmul.c
> > new file mode 100644
> > index 00000000000..e76f3c77b59
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-9-pmul.c
> > @@ -0,0 +1,9 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
> -fdump-tree-crc" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> > +
> > +#include "../../gcc.dg/torture/crc-9.c"
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data16-pmul.c
> b/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data16-pmul.c
> > new file mode 100644
> > index 00000000000..21520474564
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data16-pmul.c
> > @@ -0,0 +1,9 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-w -march=armv8-a+crypto -O2 -fdump-rtl-dfinish
> -fdump-tree-crc" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> > +
> > +#include "../../gcc.dg/torture/crc-CCIT-data16.c"
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> > \ No newline at end of file
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data8-pmul.c
> b/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data8-pmul.c
> > new file mode 100644
> > index 00000000000..3dcc92320f3
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data8-pmul.c
> > @@ -0,0 +1,9 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-w -march=armv8-a+crypto -O2 -fdump-rtl-dfinish
> -fdump-tree-crc" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto" } } */
> > +
> > +#include "../../gcc.dg/torture/crc-CCIT-data8.c"
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> > \ No newline at end of file
> > diff --git
> a/gcc/testsuite/gcc.target/aarch64/crc-coremark-16bitdata-pmul.c
> b/gcc/testsuite/gcc.target/aarch64/crc-coremark-16bitdata-pmul.c
> > new file mode 100644
> > index 00000000000..e5196aaafef
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-coremark-16bitdata-pmul.c
> > @@ -0,0 +1,9 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-w -march=armv8-a+crypto -O2 -fdump-rtl-dfinish
> -fdump-tree-crc" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> > +
> > +#include "../../gcc.dg/torture/crc-coremark16-data16.c"
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
> > \ No newline at end of file
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32-data16.c
> b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data16.c
> > new file mode 100644
> > index 00000000000..e82cb04fcc3
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data16.c
> > @@ -0,0 +1,53 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish
> -fdump-tree-crc" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> > +
> > +#include <stdint.h>
> > +#include <stdlib.h>
> > +
> > +__attribute__ ((noinline,optimize(0)))
> > +uint32_t _crc32_O0 (uint32_t crc, uint16_t data) {
> > +  int i;
> > +  crc = crc ^ data;
> > +
> > +  for (i = 0; i < 8; i++) {
> > +      if (crc & 1)
> > +     crc = (crc >> 1) ^ 0xEDB88320;
> > +      else
> > +     crc = (crc >> 1);
> > +    }
> > +
> > +  return crc;
> > +}
> > +
> > +uint32_t _crc32 (uint32_t crc, uint16_t data) {
> > +  int i;
> > +  crc = crc ^ data;
> > +
> > +  for (i = 0; i < 8; i++) {
> > +      if (crc & 1)
> > +     crc = (crc >> 1) ^ 0xEDB88320;
> > +      else
> > +     crc = (crc >> 1);
> > +    }
> > +
> > +  return crc;
> > +}
> > +
> > +int main ()
> > +{
> > +  uint32_t crc = 0x0D800D80;
> > +  for (uint16_t i = 0; i < 0xffff; i++)
> > +    {
> > +      uint32_t res1 = _crc32_O0 (crc, i);
> > +      uint32_t res2 = _crc32 (crc, i);
> > +      if (res1 != res2)
> > +      abort ();
> > +      crc = res1;
> > +    }
> > +}
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "UNSPEC_CRC32" "dfinish"} } */
> > +/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32-data32.c
> b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data32.c
> > new file mode 100644
> > index 00000000000..a7564a7e28a
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data32.c
> > @@ -0,0 +1,52 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish
> -fdump-tree-crc" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> > +
> > +#include <stdint.h>
> > +#include <stdlib.h>
> > +__attribute__ ((noinline,optimize(0)))
> > +uint32_t _crc32_O0 (uint32_t crc, uint32_t data) {
> > +  int i;
> > +  crc = crc ^ data;
> > +
> > +  for (i = 0; i < 32; i++) {
> > +      if (crc & 1)
> > +     crc = (crc >> 1) ^ 0xEDB88320;
> > +      else
> > +     crc = (crc >> 1);
> > +    }
> > +
> > +  return crc;
> > +}
> > +
> > +uint32_t _crc32 (uint32_t crc, uint32_t data) {
> > +  int i;
> > +  crc = crc ^ data;
> > +
> > +  for (i = 0; i < 32; i++) {
> > +      if (crc & 1)
> > +     crc = (crc >> 1) ^ 0xEDB88320;
> > +      else
> > +     crc = (crc >> 1);
> > +    }
> > +
> > +  return crc;
> > +}
> > +
> > +int main ()
> > +{
> > +  uint32_t crc = 0x0D800D80;
> > +  for (uint8_t i = 0; i < 0xff; i++)
> > +    {
> > +      uint32_t res1 = _crc32_O0 (crc, i);
> > +      uint32_t res2 = _crc32 (crc, i);
> > +      if (res1 != res2)
> > +      abort ();
> > +      crc = res1;
> > +    }
> > +}
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "UNSPEC_CRC32" "dfinish"} } */
> > +/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32-data8.c
> b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data8.c
> > new file mode 100644
> > index 00000000000..c88cafadedc
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data8.c
> > @@ -0,0 +1,53 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish
> -fdump-tree-crc" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> > +
> > +#include <stdint.h>
> > +#include <stdlib.h>
> > +
> > +__attribute__ ((noinline,optimize(0)))
> > +uint32_t _crc32_O0 (uint32_t crc, uint8_t data) {
> > +  int i;
> > +  crc = crc ^ data;
> > +
> > +  for (i = 0; i < 8; i++) {
> > +      if (crc & 1)
> > +     crc = (crc >> 1) ^ 0xEDB88320;
> > +      else
> > +     crc = (crc >> 1);
> > +    }
> > +
> > +  return crc;
> > +}
> > +
> > +uint32_t _crc32 (uint32_t crc, uint8_t data) {
> > +  int i;
> > +  crc = crc ^ data;
> > +
> > +  for (i = 0; i < 8; i++) {
> > +      if (crc & 1)
> > +     crc = (crc >> 1) ^ 0xEDB88320;
> > +      else
> > +     crc = (crc >> 1);
> > +    }
> > +
> > +  return crc;
> > +}
> > +
> > +int main ()
> > +{
> > +  uint32_t crc = 0x0D800D80;
> > +  for (uint8_t i = 0; i < 0xff; i++)
> > +    {
> > +      uint32_t res1 = _crc32_O0 (crc, i);
> > +      uint32_t res2 = _crc32 (crc, i);
> > +      if (res1 != res2)
> > +      abort ();
> > +      crc = res1;
> > +    }
> > +}
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "UNSPEC_CRC32" "dfinish"} } */
> > +/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data16.c
> b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data16.c
> > new file mode 100644
> > index 00000000000..d82e6252603
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data16.c
> > @@ -0,0 +1,53 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish
> -fdump-tree-crc" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> > +
> > +#include <stdint.h>
> > +#include <stdlib.h>
> > +
> > +__attribute__ ((noinline,optimize(0)))
> > +uint32_t _crc32_O0 (uint32_t crc, uint16_t data) {
> > +  int i;
> > +  crc = crc ^ data;
> > +
> > +  for (i = 0; i < 8; i++) {
> > +      if (crc & 1)
> > +     crc = (crc >> 1) ^ 0x82F63B78;
> > +      else
> > +     crc = (crc >> 1);
> > +    }
> > +
> > +  return crc;
> > +}
> > +
> > +uint32_t _crc32 (uint32_t crc, uint16_t data) {
> > +  int i;
> > +  crc = crc ^ data;
> > +
> > +  for (i = 0; i < 8; i++) {
> > +      if (crc & 1)
> > +     crc = (crc >> 1) ^ 0x82F63B78;
> > +      else
> > +     crc = (crc >> 1);
> > +    }
> > +
> > +  return crc;
> > +}
> > +
> > +int main ()
> > +{
> > +  uint32_t crc = 0x0D800D80;
> > +  for (uint16_t i = 0; i < 0xffff; i++)
> > +    {
> > +      uint32_t res1 = _crc32_O0 (crc, i);
> > +      uint32_t res2 = _crc32 (crc, i);
> > +      if (res1 != res2)
> > +      abort ();
> > +      crc = res1;
> > +    }
> > +}
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "UNSPEC_CRC32C" "dfinish"} } */
> > +/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data32.c
> b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data32.c
> > new file mode 100644
> > index 00000000000..7acb6fc239c
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data32.c
> > @@ -0,0 +1,52 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish
> -fdump-tree-crc" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> > +
> > +#include <stdint.h>
> > +#include <stdlib.h>
> > +__attribute__ ((noinline,optimize(0)))
> > +uint32_t _crc32_O0 (uint32_t crc, uint32_t data) {
> > +  int i;
> > +  crc = crc ^ data;
> > +
> > +  for (i = 0; i < 32; i++) {
> > +      if (crc & 1)
> > +     crc = (crc >> 1) ^ 0x82F63B78;
> > +      else
> > +     crc = (crc >> 1);
> > +    }
> > +
> > +  return crc;
> > +}
> > +
> > +uint32_t _crc32 (uint32_t crc, uint32_t data) {
> > +  int i;
> > +  crc = crc ^ data;
> > +
> > +  for (i = 0; i < 32; i++) {
> > +      if (crc & 1)
> > +     crc = (crc >> 1) ^ 0x82F63B78;
> > +      else
> > +     crc = (crc >> 1);
> > +    }
> > +
> > +  return crc;
> > +}
> > +
> > +int main ()
> > +{
> > +  uint32_t crc = 0x0D800D80;
> > +  for (uint8_t i = 0; i < 0xff; i++)
> > +    {
> > +      uint32_t res1 = _crc32_O0 (crc, i);
> > +      uint32_t res2 = _crc32 (crc, i);
> > +      if (res1 != res2)
> > +      abort ();
> > +      crc = res1;
> > +    }
> > +}
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "UNSPEC_CRC32C" "dfinish"} } */
> > +/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */
> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data8.c
> b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data8.c
> > new file mode 100644
> > index 00000000000..e8a8901e453
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data8.c
> > @@ -0,0 +1,53 @@
> > +/* { dg-do run } */
> > +/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish
> -fdump-tree-crc" } */
> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
> > +
> > +#include <stdint.h>
> > +#include <stdlib.h>
> > +
> > +__attribute__ ((noinline,optimize(0)))
> > +uint32_t _crc32_O0 (uint32_t crc, uint8_t data) {
> > +  int i;
> > +  crc = crc ^ data;
> > +
> > +  for (i = 0; i < 8; i++) {
> > +      if (crc & 1)
> > +     crc = (crc >> 1) ^ 0x82F63B78;
> > +      else
> > +     crc = (crc >> 1);
> > +    }
> > +
> > +  return crc;
> > +}
> > +
> > +uint32_t _crc32 (uint32_t crc, uint8_t data) {
> > +  int i;
> > +  crc = crc ^ data;
> > +
> > +  for (i = 0; i < 8; i++) {
> > +      if (crc & 1)
> > +     crc = (crc >> 1) ^ 0x82F63B78;
> > +      else
> > +     crc = (crc >> 1);
> > +    }
> > +
> > +  return crc;
> > +}
> > +
> > +int main ()
> > +{
> > +  uint32_t crc = 0x0D800D80;
> > +  for (uint8_t i = 0; i < 0xff; i++)
> > +    {
> > +      uint32_t res1 = _crc32_O0 (crc, i);
> > +      uint32_t res2 = _crc32 (crc, i);
> > +      if (res1 != res2)
> > +      abort ();
> > +      crc = res1;
> > +    }
> > +}
> > +
> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
> code." 0 "crc"} } */
> > +/* { dg-final { scan-rtl-dump "UNSPEC_CRC32C" "dfinish"} } */
> > +/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */
>

Re: [RFC/RFA][PATCH v3 06/12] aarch64: Implement new expander for efficient CRC computation

[Mariam Arutunian]

[-- Attachment #1: Type: text/plain, Size: 42640 bytes --]

I sent the new version and added you as a co-author.

Thanks,
Mariam

On Fri, Aug 16, 2024 at 7:25 PM Mariam Arutunian <mariamarutunian@gmail.com>
wrote:

>
>
> On Fri, Aug 9, 2024 at 7:22 PM Richard Sandiford <
> richard.sandiford@arm.com> wrote:
>
>> Sorry again for the slow review. :(
>>
>> I only really looked at the unreversed version earlier, on the basis
>> that the comments would apply to both versions.  But I've got a couple
>> of comments about the reversed version below:
>>
>> Mariam Arutunian <mariamarutunian@gmail.com> writes:
>> > [...]
>> > diff --git a/gcc/config/aarch64/aarch64.cc
>> b/gcc/config/aarch64/aarch64.cc
>> > index ee12d8897a8..546a379fd74 100644
>> > --- a/gcc/config/aarch64/aarch64.cc
>> > +++ b/gcc/config/aarch64/aarch64.cc
>> > @@ -30265,6 +30265,126 @@ aarch64_retrieve_sysreg (const char *regname,
>> bool write_p, bool is128op)
>> >    return sysreg->encoding;
>> >  }
>> >
>> > +/* Generate assembly to calculate CRC
>> > +   using carry-less multiplication instruction.
>> > +   OPERANDS[1] is input CRC,
>> > +   OPERANDS[2] is data (message),
>> > +   OPERANDS[3] is the polynomial without the leading 1.  */
>> > +
>> > +void
>> > +aarch64_expand_crc_using_pmull (scalar_mode crc_mode,
>> > +                             scalar_mode data_mode,
>> > +                             rtx *operands)
>> > +{
>> > +  /* Check and keep arguments.  */
>> > +  gcc_assert (!CONST_INT_P (operands[0]));
>> > +  gcc_assert (CONST_INT_P (operands[3]));
>> > +  rtx crc = operands[1];
>> > +  rtx data = operands[2];
>> > +  rtx polynomial = operands[3];
>> > +
>> > +  unsigned HOST_WIDE_INT crc_size = GET_MODE_BITSIZE (crc_mode);
>> > +  unsigned HOST_WIDE_INT data_size = GET_MODE_BITSIZE (data_mode);
>> > +  gcc_assert (crc_size <= 32);
>> > +  gcc_assert (data_size <= crc_size);
>> > +
>> > +  /* Calculate the quotient.  */
>> > +  unsigned HOST_WIDE_INT
>> > +      q = gf2n_poly_long_div_quotient (UINTVAL (polynomial), crc_size);
>> > +  /* CRC calculation's main part.  */
>> > +  if (crc_size > data_size)
>> > +    crc = expand_shift (RSHIFT_EXPR, DImode, crc, crc_size - data_size,
>> > +                     NULL_RTX, 1);
>> > +
>> > +  rtx t0 = force_reg (DImode, gen_int_mode (q, DImode));
>> > +  polynomial = simplify_gen_unary (ZERO_EXTEND, DImode, polynomial,
>> > +                                GET_MODE (polynomial));
>> > +  rtx t1 = force_reg (DImode, polynomial);
>> > +
>> > +  rtx a0 = expand_binop (DImode, xor_optab, crc, data, NULL_RTX, 1,
>> > +                      OPTAB_WIDEN);
>> > +
>> > +  rtx clmul_res = gen_reg_rtx (TImode);
>> > +  emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t0));
>> > +  a0 = gen_lowpart (DImode, clmul_res);
>> > +
>> > +  a0 = expand_shift (RSHIFT_EXPR, DImode, a0, crc_size, NULL_RTX, 1);
>> > +
>> > +  emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t1));
>> > +  a0 = gen_lowpart (DImode, clmul_res);
>> > +
>> > +  if (crc_size > data_size)
>> > +    {
>> > +      rtx crc_part = expand_shift (LSHIFT_EXPR, DImode, operands[1],
>> data_size,
>> > +                                NULL_RTX, 0);
>> > +      a0 = expand_binop (DImode, xor_optab, a0, crc_part, NULL_RTX, 1,
>> > +                      OPTAB_DIRECT);
>> > +    }
>> > +
>> > +  /* Zero upper bits beyond crc_size.  */
>>
>> The comment no longer applies.  Otherwise this function looks good to me.
>>
>>
> Ok.)
>
>
>> > +  aarch64_emit_move (operands[0], gen_lowpart (crc_mode, a0));
>> > +}
>> > +
>> > +/* Generate assembly to calculate reversed CRC
>> > +   using carry-less multiplication instruction.
>> > +   OPERANDS[1] is input CRC,
>> > +   OPERANDS[2] is data,
>> > +   OPERANDS[3] is the polynomial without the leading 1.  */
>> > +
>> > +void
>> > +aarch64_expand_reversed_crc_using_pmull (scalar_mode crc_mode,
>> > +                                      scalar_mode data_mode,
>> > +                                      rtx *operands)
>> > +{
>> > +  /* Check and keep arguments.  */
>> > +  gcc_assert (!CONST_INT_P (operands[0]));
>> > +  gcc_assert (CONST_INT_P (operands[3]));
>> > +  rtx crc = operands[1];
>> > +  rtx data = operands[2];
>> > +  rtx polynomial = operands[3];
>> > +
>> > +  unsigned HOST_WIDE_INT crc_size = GET_MODE_BITSIZE (crc_mode);
>> > +  unsigned HOST_WIDE_INT data_size = GET_MODE_BITSIZE (data_mode);
>> > +  gcc_assert (crc_size <= 32);
>> > +  gcc_assert (data_size <= crc_size);
>> > +
>> > +  /* Calculate the quotient.  */
>> > +  unsigned HOST_WIDE_INT
>> > +      q = gf2n_poly_long_div_quotient (UINTVAL (polynomial), crc_size);
>> > +  /* Reflect the calculated quotient.  */
>> > +  q = reflect (q);
>> > +  rtx t0 = force_reg (DImode, gen_int_mode (q >> (data_size - 4),
>> DImode));
>> > +
>> > +  /* Reflect the polynomial.  */
>> > +  unsigned HOST_WIDE_INT ref_polynomial = reflect (UINTVAL
>> (polynomial));
>>
>> It looks like reflect() autodetects the bitwidth based on the assumption
>> that the upper half will be nonzero.  But that might not be true for all
>> possible polynomials (when the implicit leading coefficient is absent)
>> E.g. it looks like the 64-bit HDLC CRC polynomial is 0x1b (just the
>> lowest byte nonzero), and although we don't support 64-bit polynomials
>> here, the approach wouldn't work for it.
>>
>> I think it'd be safer to pass the bitwidth as an explicit parameter.
>> Also, maybe it could go in hwint.* instead of expr.* and be called
>> something like reflect_hwi.
>>
>
> Yes. In the CRC loop recognition part, I only support those polynomials
> whose upper half is nonzero. I.e required memories for the polynomial and
> the CRC are equal. The reason for this is that, as the polynomial's
> leading bit is emitted, it is not possible to precisely assume what CRC is
> being calculated (e.g., 64, 32, etc.). Because there are some
> implementations, where more memory than needed is used to store the CRC.
> For example, in some implementations, a 64-bit memory may be used to
> calculate a 32-bit CRC.
> But, I agree, it's safer to add an explicit parameter.
> I'll make the suggested changes.
>
>
>> > +  rtx t1 = force_reg (DImode, gen_int_mode (ref_polynomial << 1,
>> DImode));
>> > +
>> > +  /* CRC calculation's main part.  */
>> > +  rtx a0 = expand_binop (DImode, xor_optab, crc, data, NULL_RTX, 1,
>> > +                      OPTAB_WIDEN);
>> > +
>> > +  /* Perform carry-less multiplication and get low part.  */
>> > +  rtx clmul_res = gen_reg_rtx (TImode);
>> > +  emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t0));
>> > +  a0 = gen_lowpart (DImode, clmul_res);
>> > +
>> > +  a0 = expand_shift (LSHIFT_EXPR, DImode, a0, 64 - crc_size - 3,
>> NULL_RTX, 0);
>>
>> I haven't really looked at this implementation strategy before, so this
>> is probably a silly question sorry :-) but is there a specific reason
>> for selecting the bias of 4 in the right shift above, cancelled out by
>> subtracting the same amount from this left shift?  It looks like we could
>> use any value in the range [1. crc_size - 1], is that right?
>>
>> Just asking out of curiosity though.  I agree it works.
>>
>>
> Sorry, I don’t remember exactly why I used a bias of 4. Upon reviewing it,
> I think it's related to the reflection process, where zeros are added to
> the quotient.
> For example, in the case of CRC-8, the quotient is 9 bits but is stored in
> a 16-bit value, so 7 bits are zeros.
> It might be better to use q >> (crc_size - 1) and then do the left shift
> with 64 - data_size instead of 64 - crc_size - 3.
> However, I noticed that you provided a better solution. Thank you very
> much.)
>
> > +
>> > +  /* Perform carry-less multiplication and get high part.  */
>> > +  emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t1));
>> > +  a0 = gen_highpart (DImode, clmul_res);
>>
>> Although this works, it's taking a DImode highpart of a single TImode
>> register, which requires a spill from the register allocator.  So we get
>> something like:
>>
>>         pmull   v31.1q, v31.1d, v14.1d
>>         str     q31, [sp, 48]
>>         ldr     d31, [sp, 48]
>>         shl     d31, d31, 53
>>         pmull   v31.1q, v31.1d, v15.1d
>>         str     q31, [sp, 48]
>>         ldrb    w0, [sp, 56]
>>
>> (I think the first STR and LDR come from reusing the same pseudo register
>> for the temporary results; it's the STR+LDRB that causes the spill.)
>>
>> One way around that would be:
>>
>>   emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t1));
>>   a0 = gen_reg_rtx (DImode);
>>   rtx v2di_res = gen_lowpart (V2DImode, clmul_res);
>>   rtx high_index = BYTES_BIG_ENDIAN ? const0_rtx : const1_rtx;
>>   emit_insn (gen_aarch64_get_lanev2di (a0, v2di_res, high_index));
>>
>> That generates:
>>
>>         pmull   v30.1q, v29.1d, v30.1d
>>         shl     d30, d30, 53
>>         pmull   v30.1q, v30.1d, v31.1d
>>         umov    x0, v30.d[1]
>>
>> which is pretty nice.
>>
>
> There again, it might not be too bad if we ionstead used the flipped
>> version of the unreversed approach, such as the following minor adaption:
>>
>>   /* Calculate the quotient.  */
>>   unsigned HOST_WIDE_INT
>>       q = gf2n_poly_long_div_quotient (UINTVAL (polynomial), crc_size);
>>   /* Reflect the calculated quotient.  */
>>   q = reflect_hwi (q, crc_size + 1);
>>   rtx t0 = force_reg (DImode, gen_int_mode (q, DImode));
>>
>>   /* Reflect the polynomial.  */
>>   unsigned HOST_WIDE_INT ref_polynomial = reflect_hwi (UINTVAL
>> (polynomial),
>>                                                        crc_size);
>>   /* An unshifted multiplier would require the final result to be
>> extracted
>>      using a shift right by DATA_SIZE - 1 bits.  Shift the multiplier left
>>      so that the shift right can be by CRC_SIZE bits instead.  */
>>   ref_polynomial <<= crc_size - data_size + 1;
>>   rtx t1 = force_reg (DImode, gen_int_mode (ref_polynomial, DImode));
>>
>>   /* CRC calculation's main part.  */
>>   rtx a0 = expand_binop (DImode, xor_optab, crc, data, NULL_RTX, 1,
>>                          OPTAB_WIDEN);
>>
>>   /* Perform carry-less multiplication and get low part.  */
>>   rtx clmul_res = gen_reg_rtx (TImode);
>>   emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t0));
>>   a0 = gen_lowpart (DImode, clmul_res);
>>
>>   a0 = expand_binop (DImode, and_optab, a0,
>>                      gen_int_mode (GET_MODE_MASK (data_mode), DImode),
>>                      NULL_RTX, 1, OPTAB_WIDEN);
>>
>>   /* Perform carry-less multiplication.  */
>>   emit_insn (gen_aarch64_crypto_pmulldi (clmul_res, a0, t1));
>>
>>   /* Perform a shift right by CRC_SIZE as an extraction of lane 1.  */
>>   machine_mode crc_vmode = aarch64_vq_mode (crc_mode).require ();
>>   a0 = (crc_size > data_size ? gen_reg_rtx (crc_mode) : operands[0]);
>>   emit_insn (gen_aarch64_get_lane (crc_vmode, a0,
>>                                    gen_lowpart (crc_vmode, clmul_res),
>>                                    aarch64_endian_lane_rtx (crc_vmode,
>> 1)));
>>
>>   if (crc_size > data_size)
>>     {
>>       rtx crc_part = expand_shift (RSHIFT_EXPR, crc_mode, crc, data_size,
>>                                    NULL_RTX, 1);
>>       a0 = expand_binop (crc_mode, xor_optab, a0, crc_part, operands[0],
>> 1,
>>                          OPTAB_WIDEN);
>>       aarch64_emit_move (operands[0], a0);
>>     }
>>
>> This gives:
>>
>>         pmull   v30.1q, v30.1d, v31.1d
>>         movi    v31.2d, 0xff
>>         and     v30.8b, v30.8b, v31.8b
>>         pmull   v30.1q, v30.1d, v31.1d
>>         umov    w0, v30.b[1]
>>
>> or, with SVE enabled:
>>
>>         pmull   v30.1q, v31.1d, v30.1d
>>         and     z30.d, z30.d, #255
>>         pmull   v30.1q, v30.1d, v31.1d
>>         umov    w0, v30.b[1]
>>
>> This is preferable since ANDs are generally cheaper than shifts.
>>
>> That's just a suggestion though; the original version is ok too.
>>
>>
> Thanks for the suggestions. I don't have much experience with gcc, so this
> is a huge help for me.
> I'll apply all the changes and send the new version.
>
>
>> Minor comment: it'd be good to use "pmull_res" instead of "clmul_res"
>> for the variables: :)
>>
>>
> Ok)
>
>
>> LGTM otherwise.
>>
>>
> Thanks,
> Mariam
>
>
>> Thanks,
>> Richard
>>
>> > +
>> > +  if (crc_size > data_size)
>> > +    {
>> > +      rtx crc_part = expand_shift (RSHIFT_EXPR, DImode, crc, data_size,
>> > +                                NULL_RTX, 1);
>> > +      a0 = expand_binop (DImode, xor_optab, a0, crc_part, NULL_RTX, 1,
>> > +                      OPTAB_DIRECT);
>> > +    }
>> > +
>> > +  aarch64_emit_move (operands[0], gen_lowpart (crc_mode, a0));
>> > +}
>> > +
>> >  /* Target-specific selftests.  */
>> >
>> >  #if CHECKING_P
>> > diff --git a/gcc/config/aarch64/aarch64.md
>> b/gcc/config/aarch64/aarch64.md
>> > index 9dff2d7a2b0..08c588bc475 100644
>> > --- a/gcc/config/aarch64/aarch64.md
>> > +++ b/gcc/config/aarch64/aarch64.md
>> > @@ -4543,6 +4543,63 @@
>> >    [(set_attr "type" "crc")]
>> >  )
>> >
>> > +;; Reversed CRC
>> > +(define_expand "crc_rev<ALLI:mode><ALLX:mode>4"
>> > +  [;; return value (calculated CRC)
>> > +   (match_operand:ALLX 0 "register_operand" "=r")
>> > +   ;; initial CRC
>> > +   (match_operand:ALLX 1 "register_operand" "r")
>> > +   ;; data
>> > +   (match_operand:ALLI 2 "register_operand" "r")
>> > +   ;; polynomial without leading 1
>> > +   (match_operand:ALLX 3)]
>> > +  ""
>> > +  {
>> > +    /* If the polynomial is the same as the polynomial of crc32c*
>> instruction,
>> > +       put that instruction.  crc32c uses iSCSI polynomial.  */
>> > +    if (TARGET_CRC32 && INTVAL (operands[3]) == 0x1EDC6F41
>> > +     && <ALLX:MODE>mode == SImode)
>> > +      emit_insn (gen_aarch64_crc32c<ALLI:crc_data_type> (operands[0],
>> > +                                                      operands[1],
>> > +                                                      operands[2]));
>> > +    /* If the polynomial is the same as the polynomial of crc32*
>> instruction,
>> > +     put that instruction.  crc32 uses HDLC etc.  polynomial.  */
>> > +    else if (TARGET_CRC32 && INTVAL (operands[3]) == 0x04C11DB7
>> > +          && <ALLX:MODE>mode == SImode)
>> > +      emit_insn (gen_aarch64_crc32<ALLI:crc_data_type> (operands[0],
>> > +                                                     operands[1],
>> > +                                                     operands[2]));
>> > +    else if (TARGET_AES && <ALLI:sizen> <= <ALLX:sizen>)
>> > +      aarch64_expand_reversed_crc_using_pmull (<ALLX:MODE>mode,
>> > +                                            <ALLI:MODE>mode,
>> > +                                            operands);
>> > +    else
>> > +      /* Otherwise, generate table-based CRC.  */
>> > +      expand_reversed_crc_table_based (operands[0], operands[1],
>> operands[2],
>> > +                                    operands[3], <ALLI:MODE>mode,
>> > +                                    generate_reflecting_code_standard);
>> > +    DONE;
>> > +  }
>> > +)
>> > +
>> > +;; Bit-forward CRC
>> > +(define_expand "crc<ALLI:mode><ALLX:mode>4"
>> > +  [;; return value (calculated CRC)
>> > +   (match_operand:ALLX 0 "register_operand" "=r")
>> > +   ;; initial CRC
>> > +   (match_operand:ALLX 1 "register_operand" "r")
>> > +   ;; data
>> > +   (match_operand:ALLI 2 "register_operand" "r")
>> > +   ;; polynomial without leading 1
>> > +   (match_operand:ALLX 3)]
>> > +  "TARGET_AES && <ALLI:sizen> <= <ALLX:sizen>"
>> > +  {
>> > +    aarch64_expand_crc_using_pmull (<ALLX:MODE>mode, <ALLI:MODE>mode,
>> > +                                 operands);
>> > +    DONE;
>> > +  }
>> > +)
>> > +
>> >  (define_insn "*csinc2<mode>_insn"
>> >    [(set (match_operand:GPI 0 "register_operand" "=r")
>> >          (plus:GPI (match_operand 2 "aarch64_comparison_operation" "")
>> > diff --git a/gcc/config/aarch64/iterators.md
>> b/gcc/config/aarch64/iterators.md
>> > index 99cde46f1ba..86e4863d684 100644
>> > --- a/gcc/config/aarch64/iterators.md
>> > +++ b/gcc/config/aarch64/iterators.md
>> > @@ -1276,6 +1276,10 @@
>> >  ;; Map a mode to a specific constraint character.
>> >  (define_mode_attr cmode [(QI "q") (HI "h") (SI "s") (DI "d")])
>> >
>> > +;; Map a mode to a specific constraint character for calling
>> > +;; appropriate version of crc.
>> > +(define_mode_attr crc_data_type [(QI "b") (HI "h") (SI "w") (DI "x")])
>> > +
>> >  ;; Map modes to Usg and Usj constraints for SISD right shifts
>> >  (define_mode_attr cmode_simd [(SI "g") (DI "j")])
>> >
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-1-pmul.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-1-pmul.c
>> > new file mode 100644
>> > index 00000000000..4043251dbd8
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-1-pmul.c
>> > @@ -0,0 +1,8 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc -fdisable-tree-phiopt2 -fdisable-tree-phiopt3" } */
>> > +
>> > +#include "../../gcc.dg/torture/crc-1.c"
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
>> > \ No newline at end of file
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-10-pmul.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-10-pmul.c
>> > new file mode 100644
>> > index 00000000000..dd866b38e83
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-10-pmul.c
>> > @@ -0,0 +1,9 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
>> > +
>> > +#include "../../gcc.dg/torture/crc-10.c"
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-12-pmul.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-12-pmul.c
>> > new file mode 100644
>> > index 00000000000..16d901eeaef
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-12-pmul.c
>> > @@ -0,0 +1,9 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc -fdisable-tree-phiopt2 -fdisable-tree-phiopt3" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
>> > +
>> > +#include "../../gcc.dg/torture/crc-12.c"
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-13-pmul.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-13-pmul.c
>> > new file mode 100644
>> > index 00000000000..5f7741fad0f
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-13-pmul.c
>> > @@ -0,0 +1,9 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
>> > +
>> > +#include "../../gcc.dg/torture/crc-13.c"
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-14-pmul.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-14-pmul.c
>> > new file mode 100644
>> > index 00000000000..cdedbbd3db1
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-14-pmul.c
>> > @@ -0,0 +1,9 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
>> > +
>> > +#include "../../gcc.dg/torture/crc-14.c"
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-17-pmul.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-17-pmul.c
>> > new file mode 100644
>> > index 00000000000..c219e49a2b1
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-17-pmul.c
>> > @@ -0,0 +1,9 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
>> > +
>> > +#include "../../gcc.dg/torture/crc-17.c"
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-18-pmul.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-18-pmul.c
>> > new file mode 100644
>> > index 00000000000..124900a979b
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-18-pmul.c
>> > @@ -0,0 +1,9 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
>> > +
>> > +#include "../../gcc.dg/torture/crc-18.c"
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-21-pmul.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-21-pmul.c
>> > new file mode 100644
>> > index 00000000000..3cae1a7f57b
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-21-pmul.c
>> > @@ -0,0 +1,9 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
>> > +
>> > +#include "../../gcc.dg/torture/crc-21.c"
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-22-pmul.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-22-pmul.c
>> > new file mode 100644
>> > index 00000000000..0ec2e312f8f
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-22-pmul.c
>> > @@ -0,0 +1,9 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
>> > +
>> > +#include "../../gcc.dg/torture/crc-22.c"
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-23-pmul.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-23-pmul.c
>> > new file mode 100644
>> > index 00000000000..0c4542adb40
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-23-pmul.c
>> > @@ -0,0 +1,9 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
>> > +
>> > +#include "../../gcc.dg/torture/crc-23.c"
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-4-pmul.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-4-pmul.c
>> > new file mode 100644
>> > index 00000000000..08f1d3b69d7
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-4-pmul.c
>> > @@ -0,0 +1,9 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
>> > +
>> > +#include "../../gcc.dg/torture/crc-4.c"
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-5-pmul.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-5-pmul.c
>> > new file mode 100644
>> > index 00000000000..91bf5e6353d
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-5-pmul.c
>> > @@ -0,0 +1,9 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-march=armv8-a+crypto -O2 -w -fdump-rtl-dfinish
>> -fdump-tree-crc" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
>> > +
>> > +#include "../../gcc.dg/torture/crc-5.c"
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
>> > \ No newline at end of file
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-6-pmul.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-6-pmul.c
>> > new file mode 100644
>> > index 00000000000..4680eafe758
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-6-pmul.c
>> > @@ -0,0 +1,9 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
>> > +
>> > +#include "../../gcc.dg/torture/crc-6.c"
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
>> > \ No newline at end of file
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-7-pmul.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-7-pmul.c
>> > new file mode 100644
>> > index 00000000000..655484d10d4
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-7-pmul.c
>> > @@ -0,0 +1,9 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
>> > +
>> > +#include "../../gcc.dg/torture/crc-7.c"
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-8-pmul.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-8-pmul.c
>> > new file mode 100644
>> > index 00000000000..6c2acc84c32
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-8-pmul.c
>> > @@ -0,0 +1,9 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
>> > +
>> > +#include "../../gcc.dg/torture/crc-8.c"
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-9-pmul.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-9-pmul.c
>> > new file mode 100644
>> > index 00000000000..e76f3c77b59
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-9-pmul.c
>> > @@ -0,0 +1,9 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-march=armv8-a+crypto -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
>> > +
>> > +#include "../../gcc.dg/torture/crc-9.c"
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data16-pmul.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data16-pmul.c
>> > new file mode 100644
>> > index 00000000000..21520474564
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data16-pmul.c
>> > @@ -0,0 +1,9 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-w -march=armv8-a+crypto -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
>> > +
>> > +#include "../../gcc.dg/torture/crc-CCIT-data16.c"
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
>> > \ No newline at end of file
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data8-pmul.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data8-pmul.c
>> > new file mode 100644
>> > index 00000000000..3dcc92320f3
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-CCIT-data8-pmul.c
>> > @@ -0,0 +1,9 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-w -march=armv8-a+crypto -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto" } } */
>> > +
>> > +#include "../../gcc.dg/torture/crc-CCIT-data8.c"
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
>> > \ No newline at end of file
>> > diff --git
>> a/gcc/testsuite/gcc.target/aarch64/crc-coremark-16bitdata-pmul.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-coremark-16bitdata-pmul.c
>> > new file mode 100644
>> > index 00000000000..e5196aaafef
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-coremark-16bitdata-pmul.c
>> > @@ -0,0 +1,9 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-w -march=armv8-a+crypto -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
>> > +
>> > +#include "../../gcc.dg/torture/crc-coremark16-data16.c"
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "pmull" "dfinish"} } */
>> > \ No newline at end of file
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32-data16.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data16.c
>> > new file mode 100644
>> > index 00000000000..e82cb04fcc3
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data16.c
>> > @@ -0,0 +1,53 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
>> > +
>> > +#include <stdint.h>
>> > +#include <stdlib.h>
>> > +
>> > +__attribute__ ((noinline,optimize(0)))
>> > +uint32_t _crc32_O0 (uint32_t crc, uint16_t data) {
>> > +  int i;
>> > +  crc = crc ^ data;
>> > +
>> > +  for (i = 0; i < 8; i++) {
>> > +      if (crc & 1)
>> > +     crc = (crc >> 1) ^ 0xEDB88320;
>> > +      else
>> > +     crc = (crc >> 1);
>> > +    }
>> > +
>> > +  return crc;
>> > +}
>> > +
>> > +uint32_t _crc32 (uint32_t crc, uint16_t data) {
>> > +  int i;
>> > +  crc = crc ^ data;
>> > +
>> > +  for (i = 0; i < 8; i++) {
>> > +      if (crc & 1)
>> > +     crc = (crc >> 1) ^ 0xEDB88320;
>> > +      else
>> > +     crc = (crc >> 1);
>> > +    }
>> > +
>> > +  return crc;
>> > +}
>> > +
>> > +int main ()
>> > +{
>> > +  uint32_t crc = 0x0D800D80;
>> > +  for (uint16_t i = 0; i < 0xffff; i++)
>> > +    {
>> > +      uint32_t res1 = _crc32_O0 (crc, i);
>> > +      uint32_t res2 = _crc32 (crc, i);
>> > +      if (res1 != res2)
>> > +      abort ();
>> > +      crc = res1;
>> > +    }
>> > +}
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "UNSPEC_CRC32" "dfinish"} } */
>> > +/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32-data32.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data32.c
>> > new file mode 100644
>> > index 00000000000..a7564a7e28a
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data32.c
>> > @@ -0,0 +1,52 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
>> > +
>> > +#include <stdint.h>
>> > +#include <stdlib.h>
>> > +__attribute__ ((noinline,optimize(0)))
>> > +uint32_t _crc32_O0 (uint32_t crc, uint32_t data) {
>> > +  int i;
>> > +  crc = crc ^ data;
>> > +
>> > +  for (i = 0; i < 32; i++) {
>> > +      if (crc & 1)
>> > +     crc = (crc >> 1) ^ 0xEDB88320;
>> > +      else
>> > +     crc = (crc >> 1);
>> > +    }
>> > +
>> > +  return crc;
>> > +}
>> > +
>> > +uint32_t _crc32 (uint32_t crc, uint32_t data) {
>> > +  int i;
>> > +  crc = crc ^ data;
>> > +
>> > +  for (i = 0; i < 32; i++) {
>> > +      if (crc & 1)
>> > +     crc = (crc >> 1) ^ 0xEDB88320;
>> > +      else
>> > +     crc = (crc >> 1);
>> > +    }
>> > +
>> > +  return crc;
>> > +}
>> > +
>> > +int main ()
>> > +{
>> > +  uint32_t crc = 0x0D800D80;
>> > +  for (uint8_t i = 0; i < 0xff; i++)
>> > +    {
>> > +      uint32_t res1 = _crc32_O0 (crc, i);
>> > +      uint32_t res2 = _crc32 (crc, i);
>> > +      if (res1 != res2)
>> > +      abort ();
>> > +      crc = res1;
>> > +    }
>> > +}
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "UNSPEC_CRC32" "dfinish"} } */
>> > +/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32-data8.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data8.c
>> > new file mode 100644
>> > index 00000000000..c88cafadedc
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32-data8.c
>> > @@ -0,0 +1,53 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
>> > +
>> > +#include <stdint.h>
>> > +#include <stdlib.h>
>> > +
>> > +__attribute__ ((noinline,optimize(0)))
>> > +uint32_t _crc32_O0 (uint32_t crc, uint8_t data) {
>> > +  int i;
>> > +  crc = crc ^ data;
>> > +
>> > +  for (i = 0; i < 8; i++) {
>> > +      if (crc & 1)
>> > +     crc = (crc >> 1) ^ 0xEDB88320;
>> > +      else
>> > +     crc = (crc >> 1);
>> > +    }
>> > +
>> > +  return crc;
>> > +}
>> > +
>> > +uint32_t _crc32 (uint32_t crc, uint8_t data) {
>> > +  int i;
>> > +  crc = crc ^ data;
>> > +
>> > +  for (i = 0; i < 8; i++) {
>> > +      if (crc & 1)
>> > +     crc = (crc >> 1) ^ 0xEDB88320;
>> > +      else
>> > +     crc = (crc >> 1);
>> > +    }
>> > +
>> > +  return crc;
>> > +}
>> > +
>> > +int main ()
>> > +{
>> > +  uint32_t crc = 0x0D800D80;
>> > +  for (uint8_t i = 0; i < 0xff; i++)
>> > +    {
>> > +      uint32_t res1 = _crc32_O0 (crc, i);
>> > +      uint32_t res2 = _crc32 (crc, i);
>> > +      if (res1 != res2)
>> > +      abort ();
>> > +      crc = res1;
>> > +    }
>> > +}
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "UNSPEC_CRC32" "dfinish"} } */
>> > +/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data16.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data16.c
>> > new file mode 100644
>> > index 00000000000..d82e6252603
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data16.c
>> > @@ -0,0 +1,53 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
>> > +
>> > +#include <stdint.h>
>> > +#include <stdlib.h>
>> > +
>> > +__attribute__ ((noinline,optimize(0)))
>> > +uint32_t _crc32_O0 (uint32_t crc, uint16_t data) {
>> > +  int i;
>> > +  crc = crc ^ data;
>> > +
>> > +  for (i = 0; i < 8; i++) {
>> > +      if (crc & 1)
>> > +     crc = (crc >> 1) ^ 0x82F63B78;
>> > +      else
>> > +     crc = (crc >> 1);
>> > +    }
>> > +
>> > +  return crc;
>> > +}
>> > +
>> > +uint32_t _crc32 (uint32_t crc, uint16_t data) {
>> > +  int i;
>> > +  crc = crc ^ data;
>> > +
>> > +  for (i = 0; i < 8; i++) {
>> > +      if (crc & 1)
>> > +     crc = (crc >> 1) ^ 0x82F63B78;
>> > +      else
>> > +     crc = (crc >> 1);
>> > +    }
>> > +
>> > +  return crc;
>> > +}
>> > +
>> > +int main ()
>> > +{
>> > +  uint32_t crc = 0x0D800D80;
>> > +  for (uint16_t i = 0; i < 0xffff; i++)
>> > +    {
>> > +      uint32_t res1 = _crc32_O0 (crc, i);
>> > +      uint32_t res2 = _crc32 (crc, i);
>> > +      if (res1 != res2)
>> > +      abort ();
>> > +      crc = res1;
>> > +    }
>> > +}
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "UNSPEC_CRC32C" "dfinish"} } */
>> > +/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data32.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data32.c
>> > new file mode 100644
>> > index 00000000000..7acb6fc239c
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data32.c
>> > @@ -0,0 +1,52 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
>> > +
>> > +#include <stdint.h>
>> > +#include <stdlib.h>
>> > +__attribute__ ((noinline,optimize(0)))
>> > +uint32_t _crc32_O0 (uint32_t crc, uint32_t data) {
>> > +  int i;
>> > +  crc = crc ^ data;
>> > +
>> > +  for (i = 0; i < 32; i++) {
>> > +      if (crc & 1)
>> > +     crc = (crc >> 1) ^ 0x82F63B78;
>> > +      else
>> > +     crc = (crc >> 1);
>> > +    }
>> > +
>> > +  return crc;
>> > +}
>> > +
>> > +uint32_t _crc32 (uint32_t crc, uint32_t data) {
>> > +  int i;
>> > +  crc = crc ^ data;
>> > +
>> > +  for (i = 0; i < 32; i++) {
>> > +      if (crc & 1)
>> > +     crc = (crc >> 1) ^ 0x82F63B78;
>> > +      else
>> > +     crc = (crc >> 1);
>> > +    }
>> > +
>> > +  return crc;
>> > +}
>> > +
>> > +int main ()
>> > +{
>> > +  uint32_t crc = 0x0D800D80;
>> > +  for (uint8_t i = 0; i < 0xff; i++)
>> > +    {
>> > +      uint32_t res1 = _crc32_O0 (crc, i);
>> > +      uint32_t res2 = _crc32 (crc, i);
>> > +      if (res1 != res2)
>> > +      abort ();
>> > +      crc = res1;
>> > +    }
>> > +}
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "UNSPEC_CRC32C" "dfinish"} } */
>> > +/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */
>> > diff --git a/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data8.c
>> b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data8.c
>> > new file mode 100644
>> > index 00000000000..e8a8901e453
>> > --- /dev/null
>> > +++ b/gcc/testsuite/gcc.target/aarch64/crc-crc32c-data8.c
>> > @@ -0,0 +1,53 @@
>> > +/* { dg-do run } */
>> > +/* { dg-options "-march=armv8-a+crc -O2 -fdump-rtl-dfinish
>> -fdump-tree-crc" } */
>> > +/* { dg-skip-if "" { *-*-* } { "-flto"} } */
>> > +
>> > +#include <stdint.h>
>> > +#include <stdlib.h>
>> > +
>> > +__attribute__ ((noinline,optimize(0)))
>> > +uint32_t _crc32_O0 (uint32_t crc, uint8_t data) {
>> > +  int i;
>> > +  crc = crc ^ data;
>> > +
>> > +  for (i = 0; i < 8; i++) {
>> > +      if (crc & 1)
>> > +     crc = (crc >> 1) ^ 0x82F63B78;
>> > +      else
>> > +     crc = (crc >> 1);
>> > +    }
>> > +
>> > +  return crc;
>> > +}
>> > +
>> > +uint32_t _crc32 (uint32_t crc, uint8_t data) {
>> > +  int i;
>> > +  crc = crc ^ data;
>> > +
>> > +  for (i = 0; i < 8; i++) {
>> > +      if (crc & 1)
>> > +     crc = (crc >> 1) ^ 0x82F63B78;
>> > +      else
>> > +     crc = (crc >> 1);
>> > +    }
>> > +
>> > +  return crc;
>> > +}
>> > +
>> > +int main ()
>> > +{
>> > +  uint32_t crc = 0x0D800D80;
>> > +  for (uint8_t i = 0; i < 0xff; i++)
>> > +    {
>> > +      uint32_t res1 = _crc32_O0 (crc, i);
>> > +      uint32_t res2 = _crc32 (crc, i);
>> > +      if (res1 != res2)
>> > +      abort ();
>> > +      crc = res1;
>> > +    }
>> > +}
>> > +
>> > +/* { dg-final { scan-tree-dump "calculates CRC!" "crc"} } */
>> > +/* { dg-final { scan-tree-dump-times "Couldn't generate faster CRC
>> code." 0 "crc"} } */
>> > +/* { dg-final { scan-rtl-dump "UNSPEC_CRC32C" "dfinish"} } */
>> > +/* { dg-final { scan-rtl-dump-times "pmull" 0 "dfinish"} } */
>>
>