On Fri, Jun 16, 2023 at 4:12 AM liuhongt <hongtao....@intel.com> wrote:
>
> The packing in vpacksswb/vpackssdw is not a simple concat, it's an
> interweave from src1 and src2 for every 128 bit(or 64-bit for the
> ss_truncate result).
>
> .i.e.
>
> dst[192-255] = ss_truncate (src2[128-255])
> dst[128-191] = ss_truncate (src1[128-255])
> dst[64-127] = ss_truncate (src2[0-127])
> dst[0-63] = ss_truncate (src1[0-127]
>
> The patch refined those patterns with an extra vec_select for the
> interweave.
>
> The patch will fix below testcase which failed after
> g:921b841350c4fc298d09f6c5674663e0f4208610 added constant-folding for
> SS_TRUNCATE
> FAIL: gcc.target/i386/avx2-vpackssdw-2.c execution test.
>
> Bootstrapped and regtested on x86_64-pc-linux-gnu.
> Ok for trunk?
>
> gcc/ChangeLog:
>
> PR target/110235
> * config/i386/sse.md (<sse2_avx2>_packsswb<mask_name>): Split
> to below 3 new define_insns.
> (sse2_packsswb<mask_name>): New define_insn.
> (avx2_packsswb<mask_name>): Ditto.
> (avx512bw_packsswb<mask_name>): Ditto.
> (<sse2_avx2>_packssdw<mask_name>): Split to below 3 new define_insns.
> (sse2_packssdw<mask_name>): New define_insn.
> (avx2_packssdw<mask_name>): Ditto.
> (avx512bw_packssdw<mask_name>): Ditto.
>
> gcc/testsuite/ChangeLog:
>
> * gcc.target/i386/avx512bw-vpackssdw-3.c: New test.
> * gcc.target/i386/avx512bw-vpacksswb-3.c: New test.
Please proofread and fix ChangeLog entry, in the same way as your
previous patch.
Otherwise LGTM.
Thanks,
Uros.
> ---
> gcc/config/i386/sse.md | 165 ++++++++++++++++--
> .../gcc.target/i386/avx512bw-vpackssdw-3.c | 55 ++++++
> .../gcc.target/i386/avx512bw-vpacksswb-3.c | 50 ++++++
> 3 files changed, 252 insertions(+), 18 deletions(-)
> create mode 100644 gcc/testsuite/gcc.target/i386/avx512bw-vpackssdw-3.c
> create mode 100644 gcc/testsuite/gcc.target/i386/avx512bw-vpacksswb-3.c
>
> diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
> index 83e3f534fd2..cc4e4620257 100644
> --- a/gcc/config/i386/sse.md
> +++ b/gcc/config/i386/sse.md
> @@ -17762,14 +17762,14 @@ (define_expand "vec_pack_sbool_trunc_qi"
> DONE;
> })
>
> -(define_insn "<sse2_avx2>_packsswb<mask_name>"
> - [(set (match_operand:VI1_AVX512 0 "register_operand" "=x,<v_Yw>")
> - (vec_concat:VI1_AVX512
> - (ss_truncate:<ssehalfvecmode>
> - (match_operand:<sseunpackmode> 1 "register_operand" "0,<v_Yw>"))
> - (ss_truncate:<ssehalfvecmode>
> - (match_operand:<sseunpackmode> 2 "vector_operand"
> "xBm,<v_Yw>m"))))]
> - "TARGET_SSE2 && <mask_mode512bit_condition> && <mask_avx512bw_condition>"
> +(define_insn "sse2_packsswb<mask_name>"
> + [(set (match_operand:V16QI 0 "register_operand" "=x,Yw")
> + (vec_concat:V16QI
> + (ss_truncate:V8QI
> + (match_operand:V8HI 1 "register_operand" "0,Yw"))
> + (ss_truncate:V8QI
> + (match_operand:V8HI 2 "vector_operand" "xBm,Ywm"))))]
> + "TARGET_SSE2 && <mask_avx512vl_condition> && <mask_avx512bw_condition>"
> "@
> packsswb\t{%2, %0|%0, %2}
> vpacksswb\t{%2, %1, %0<mask_operand3>|%0<mask_operand3>, %1, %2}"
> @@ -17777,16 +17777,93 @@ (define_insn "<sse2_avx2>_packsswb<mask_name>"
> (set_attr "type" "sselog")
> (set_attr "prefix_data16" "1,*")
> (set_attr "prefix" "orig,<mask_prefix>")
> - (set_attr "mode" "<sseinsnmode>")])
> + (set_attr "mode" "TI")])
>
> -(define_insn "<sse2_avx2>_packssdw<mask_name>"
> - [(set (match_operand:VI2_AVX2 0 "register_operand" "=x,<v_Yw>")
> - (vec_concat:VI2_AVX2
> - (ss_truncate:<ssehalfvecmode>
> - (match_operand:<sseunpackmode> 1 "register_operand" "0,<v_Yw>"))
> - (ss_truncate:<ssehalfvecmode>
> - (match_operand:<sseunpackmode> 2 "vector_operand"
> "xBm,<v_Yw>m"))))]
> - "TARGET_SSE2 && <mask_mode512bit_condition> && <mask_avx512bw_condition>"
> +(define_insn "avx2_packsswb<mask_name>"
> + [(set (match_operand:V32QI 0 "register_operand" "=Yw")
> + (vec_select:V32QI
> + (vec_concat:V32QI
> + (ss_truncate:V16QI
> + (match_operand:V16HI 1 "register_operand" "Yw"))
> + (ss_truncate:V16QI
> + (match_operand:V16HI 2 "vector_operand" "Ywm")))
> + (parallel [(const_int 0) (const_int 1)
> + (const_int 2) (const_int 3)
> + (const_int 4) (const_int 5)
> + (const_int 6) (const_int 7)
> + (const_int 16) (const_int 17)
> + (const_int 18) (const_int 19)
> + (const_int 20) (const_int 21)
> + (const_int 22) (const_int 23)
> + (const_int 8) (const_int 9)
> + (const_int 10) (const_int 11)
> + (const_int 12) (const_int 13)
> + (const_int 14) (const_int 15)
> + (const_int 24) (const_int 25)
> + (const_int 26) (const_int 27)
> + (const_int 28) (const_int 29)
> + (const_int 30) (const_int 31)])))]
> + "TARGET_AVX2 && <mask_avx512vl_condition> && <mask_avx512bw_condition>"
> + "vpacksswb\t{%2, %1, %0<mask_operand3>|%0<mask_operand3>, %1, %2}"
> + [(set_attr "type" "sselog")
> + (set_attr "prefix" "<mask_prefix>")
> + (set_attr "mode" "OI")])
> +
> +(define_insn "avx512bw_packsswb<mask_name>"
> + [(set (match_operand:V64QI 0 "register_operand" "=v")
> + (vec_select:V64QI
> + (vec_concat:V64QI
> + (ss_truncate:V32QI
> + (match_operand:V32HI 1 "register_operand" "v"))
> + (ss_truncate:V32QI
> + (match_operand:V32HI 2 "vector_operand" "vm")))
> + (parallel [(const_int 0) (const_int 1)
> + (const_int 2) (const_int 3)
> + (const_int 4) (const_int 5)
> + (const_int 6) (const_int 7)
> + (const_int 32) (const_int 33)
> + (const_int 34) (const_int 35)
> + (const_int 36) (const_int 37)
> + (const_int 38) (const_int 39)
> + (const_int 8) (const_int 9)
> + (const_int 10) (const_int 11)
> + (const_int 12) (const_int 13)
> + (const_int 14) (const_int 15)
> + (const_int 40) (const_int 41)
> + (const_int 42) (const_int 43)
> + (const_int 44) (const_int 45)
> + (const_int 46) (const_int 47)
> + (const_int 16) (const_int 17)
> + (const_int 18) (const_int 19)
> + (const_int 20) (const_int 21)
> + (const_int 22) (const_int 23)
> + (const_int 48) (const_int 49)
> + (const_int 50) (const_int 51)
> + (const_int 52) (const_int 53)
> + (const_int 54) (const_int 55)
> + (const_int 24) (const_int 25)
> + (const_int 26) (const_int 27)
> + (const_int 28) (const_int 29)
> + (const_int 30) (const_int 31)
> + (const_int 56) (const_int 57)
> + (const_int 58) (const_int 59)
> + (const_int 60) (const_int 61)
> + (const_int 62) (const_int 63)])))]
> +
> + "TARGET_AVX512BW"
> + "vpacksswb\t{%2, %1, %0<mask_operand3>|%0<mask_operand3>, %1, %2}"
> + [(set_attr "type" "sselog")
> + (set_attr "prefix" "<mask_prefix>")
> + (set_attr "mode" "XI")])
> +
> +(define_insn "sse2_packssdw<mask_name>"
> + [(set (match_operand:V8HI 0 "register_operand" "=x,Yw")
> + (vec_concat:V8HI
> + (ss_truncate:V4HI
> + (match_operand:V4SI 1 "register_operand" "0,Yw"))
> + (ss_truncate:V4HI
> + (match_operand:V4SI 2 "vector_operand" "xBm,Ywm"))))]
> + "TARGET_SSE2 && <mask_avx512vl_condition> && <mask_avx512bw_condition>"
> "@
> packssdw\t{%2, %0|%0, %2}
> vpackssdw\t{%2, %1, %0<mask_operand3>|%0<mask_operand3>, %1, %2}"
> @@ -17794,7 +17871,59 @@ (define_insn "<sse2_avx2>_packssdw<mask_name>"
> (set_attr "type" "sselog")
> (set_attr "prefix_data16" "1,*")
> (set_attr "prefix" "orig,<mask_prefix>")
> - (set_attr "mode" "<sseinsnmode>")])
> + (set_attr "mode" "TI")])
> +
> +(define_insn "avx2_packssdw<mask_name>"
> + [(set (match_operand:V16HI 0 "register_operand" "=Yw")
> + (vec_select:V16HI
> + (vec_concat:V16HI
> + (ss_truncate:V8HI
> + (match_operand:V8SI 1 "register_operand" "Yw"))
> + (ss_truncate:V8HI
> + (match_operand:V8SI 2 "vector_operand" "Ywm")))
> + (parallel [(const_int 0) (const_int 1)
> + (const_int 2) (const_int 3)
> + (const_int 8) (const_int 9)
> + (const_int 10) (const_int 11)
> + (const_int 4) (const_int 5)
> + (const_int 6) (const_int 7)
> + (const_int 12) (const_int 13)
> + (const_int 14) (const_int 15)])))]
> + "TARGET_AVX2 && <mask_avx512vl_condition> && <mask_avx512bw_condition>"
> + "vpackssdw\t{%2, %1, %0<mask_operand3>|%0<mask_operand3>, %1, %2}"
> + [(set_attr "type" "sselog")
> + (set_attr "prefix" "<mask_prefix>")
> + (set_attr "mode" "OI")])
> +
> +(define_insn "avx512bw_packssdw<mask_name>"
> + [(set (match_operand:V32HI 0 "register_operand" "=v")
> + (vec_select:V32HI
> + (vec_concat:V32HI
> + (ss_truncate:V16HI
> + (match_operand:V16SI 1 "register_operand" "v"))
> + (ss_truncate:V16HI
> + (match_operand:V16SI 2 "vector_operand" "vm")))
> + (parallel [(const_int 0) (const_int 1)
> + (const_int 2) (const_int 3)
> + (const_int 16) (const_int 17)
> + (const_int 18) (const_int 19)
> + (const_int 4) (const_int 5)
> + (const_int 6) (const_int 7)
> + (const_int 20) (const_int 21)
> + (const_int 22) (const_int 23)
> + (const_int 8) (const_int 9)
> + (const_int 10) (const_int 11)
> + (const_int 24) (const_int 25)
> + (const_int 26) (const_int 27)
> + (const_int 12) (const_int 13)
> + (const_int 14) (const_int 15)
> + (const_int 28) (const_int 29)
> + (const_int 30) (const_int 31)])))]
> + "TARGET_AVX512BW"
> + "vpackssdw\t{%2, %1, %0<mask_operand3>|%0<mask_operand3>, %1, %2}"
> + [(set_attr "type" "sselog")
> + (set_attr "prefix" "<mask_prefix>")
> + (set_attr "mode" "XI")])
>
> ;; This is different from rtl unsigned saturation, the instruction does
> ;; unsigned saturation for signed value.
> diff --git a/gcc/testsuite/gcc.target/i386/avx512bw-vpackssdw-3.c
> b/gcc/testsuite/gcc.target/i386/avx512bw-vpackssdw-3.c
> new file mode 100644
> index 00000000000..ae839e8d2c6
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/avx512bw-vpackssdw-3.c
> @@ -0,0 +1,55 @@
> +/* { dg-do run } */
> +/* { dg-options "-O2 -mavx512bw" } */
> +/* { dg-require-effective-target avx512bw } */
> +
> +#define AVX512BW
> +#include "avx512f-helper.h"
> +
> +#define DST_SIZE (AVX512F_LEN / 16)
> +#define SRC_SIZE (AVX512F_LEN / 32)
> +
> +#include "limits.h"
> +
> +#include "avx512f-mask-type.h"
> +
> +static short
> +int_to_short (int iVal)
> +{
> + short sVal;
> +
> + if (iVal < -32768)
> + sVal = -32768;
> + else if (iVal > 32767)
> + sVal = 32767;
> + else
> + sVal = iVal;
> +
> + return sVal;
> +}
> +
> +void
> +TEST (void)
> +{
> + union512i_d s1, s2;
> + union512i_w res1;
> + short dst_ref[32];
> + int i;
> +
> + s1.x = _mm512_set_epi32 (1, 2, 3, 4, 65000, 20, 30, 90, 88, 44, 33, 22,
> 11, 98, 76, -65000);
> + s2.x = _mm512_set_epi32 (80, 40, 31, 21, 10, 99, 74, -65000, 2, 3, 4, 5,
> 65010, 21, 31, 91);
> + res1.x = _mm512_packs_epi32 (s1.x, s2.x);
> + for (int i = 0; i != 4; i++)
> + {
> + dst_ref[i] = int_to_short (s1.a[i]);
> + dst_ref[i + 4] = int_to_short (s2.a[i]);
> + dst_ref[i + 8] = int_to_short (s1.a[i + 4]);
> + dst_ref[i + 12] = int_to_short (s2.a[i + 4]);
> + dst_ref[i + 16] = int_to_short (s1.a[i + 8]);
> + dst_ref[i + 20] = int_to_short (s2.a[i + 8]);
> + dst_ref[i + 24] = int_to_short (s1.a[i + 12]);
> + dst_ref[i + 28] = int_to_short (s2.a[i + 12]);
> + }
> +
> + if (check_union512i_w (res1, dst_ref))
> + abort ();
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/avx512bw-vpacksswb-3.c
> b/gcc/testsuite/gcc.target/i386/avx512bw-vpacksswb-3.c
> new file mode 100644
> index 00000000000..056c735ae0e
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/avx512bw-vpacksswb-3.c
> @@ -0,0 +1,50 @@
> +/* { dg-do run } */
> +/* { dg-options "-O2 -mavx512bw" } */
> +/* { dg-require-effective-target avx512bw } */
> +
> +#define AVX512BW
> +#include "avx512f-helper.h"
> +
> +static char
> +short_to_byte (short iVal)
> +{
> + short sVal;
> +
> + if (iVal < -128)
> + sVal = -128;
> + else if (iVal > 127)
> + sVal = 127;
> + else
> + sVal = iVal;
> +
> + return sVal;
> +}
> +
> +void
> +TEST (void)
> +{
> + union512i_w s1, s2;
> + union512i_b res1;
> + char dst_ref[64];
> + int i;
> +
> + s1.x = _mm512_set_epi16 (1, 2, 3, 4, 650, 20, 30, 90, 88, 44, 33, 22, 11,
> 98, 76, -650,
> + 128, 230, -112, -128, -3, -4, -7, 9, 10, 11, 12,
> 13, -223, 10, 8, 11);
> + s2.x = _mm512_set_epi16 (80, 40, 31, 21, 10, 99, 74, -650, 2, 3, 4, 5,
> 650, 21, 31, 91,
> + 280, -140, 310, 20, 9, 98, 73, -651, 3, 4, 5, 6,
> 651, 22, 32, 92);
> + res1.x = _mm512_packs_epi16 (s1.x, s2.x);
> + for (int i = 0; i != 8; i++)
> + {
> + dst_ref[i] = short_to_byte (s1.a[i]);
> + dst_ref[i + 8] = short_to_byte (s2.a[i]);
> + dst_ref[i + 16] = short_to_byte (s1.a[i + 8]);
> + dst_ref[i + 24] = short_to_byte (s2.a[i + 8]);
> + dst_ref[i + 32] = short_to_byte (s1.a[i + 16]);
> + dst_ref[i + 40] = short_to_byte (s2.a[i + 16]);
> + dst_ref[i + 48] = short_to_byte (s1.a[i + 24]);
> + dst_ref[i + 56] = short_to_byte (s2.a[i + 24]);
> + }
> +
> + if (check_union512i_b (res1, dst_ref))
> + abort ();
> +}
> --
> 2.39.1.388.g2fc9e9ca3c
>
