[PATCH] i386: Optimization for mm512_set1_pch.

[Kong, Lingling]

Hi,

This patch is to support fold _mm512_fmadd_pch (a, _mm512_set1_pch(*(b)), c) to 1 instruction vfmaddcph (%rsp){1to16}, %zmm1, %zmm2.
OK for master?

gcc/ChangeLog:

	* config/i386/sse.md (fma_<complexpairopname>_<mode>_pair):
	Add new define_insn.
	(fma_<mode>_fmaddc_bcst): Add new define_insn_and_split.
	(fma_<mode>_fcmaddc_bcst): Likewise

gcc/testsuite/ChangeLog:

	* gcc.target/i386/avx512fp16vl-complex-broadcast-1.c: New test.
---
 gcc/config/i386/sse.md                        | 62 +++++++++++++++++++
 .../i386/avx512fp16vl-complex-broadcast-1.c   | 25 ++++++++
 2 files changed, 87 insertions(+)
 create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16vl-complex-broadcast-1.c

diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md index 0a7f5b178f9..eba8e77515f 100644
--- a/gcc/config/i386/sse.md
+++ b/gcc/config/i386/sse.md
@@ -193,7 +193,9 @@
 
   ;; For AVX512FP16 suppport
   UNSPEC_COMPLEX_FMA
+  UNSPEC_COMPLEX_FMA_PAIR
   UNSPEC_COMPLEX_FCMA
+  UNSPEC_COMPLEX_FCMA_PAIR
   UNSPEC_COMPLEX_FMUL
   UNSPEC_COMPLEX_FCMUL
   UNSPEC_COMPLEX_MASK
@@ -5913,6 +5915,9 @@
 (define_int_iterator UNSPEC_COMPLEX_F_C_MA
 	[UNSPEC_COMPLEX_FMA UNSPEC_COMPLEX_FCMA])
 
+(define_int_iterator UNSPEC_COMPLEX_F_C_MA_PAIR
+	[UNSPEC_COMPLEX_FMA_PAIR UNSPEC_COMPLEX_FCMA_PAIR])
+
 (define_int_iterator UNSPEC_COMPLEX_F_C_MUL
 	[UNSPEC_COMPLEX_FMUL UNSPEC_COMPLEX_FCMUL])
 
@@ -5922,6 +5927,10 @@
 	 (UNSPEC_COMPLEX_FMUL "fmulc")
 	 (UNSPEC_COMPLEX_FCMUL "fcmulc")])
 
+(define_int_attr complexpairopname
+	[(UNSPEC_COMPLEX_FMA_PAIR "fmaddc")
+	 (UNSPEC_COMPLEX_FCMA_PAIR "fcmaddc")])
+
 (define_mode_attr complexmove
   [(V32HF "avx512f_loadv16sf")
    (V16HF "avx512vl_loadv8sf")
@@ -6067,6 +6076,59 @@
 	  [(match_dup 1) (match_dup 2) (match_dup 4)]
 	   UNSPEC_COMPLEX_F_C_MA))])
 
+(define_insn "fma_<complexpairopname>_<mode>_pair"
+ [(set (match_operand:VF1_AVX512VL 0 "register_operand" "=&v")
+       (unspec:VF1_AVX512VL
+	 [(match_operand:VF1_AVX512VL 1 "vector_operand" "%v")
+	  (match_operand:VF1_AVX512VL 2 "bcst_vector_operand" "vmBr")
+	  (match_operand:VF1_AVX512VL 3 "vector_operand" "0")]
+	  UNSPEC_COMPLEX_F_C_MA_PAIR))]
+ "TARGET_AVX512FP16"
+ "v<complexpairopname>ph\t{%2, %1, %0|%0, %1, %2}"
+ [(set_attr "type" "ssemuladd")])
+
+(define_insn_and_split "fma_<mode>_fmaddc_bcst"
+  [(set (match_operand:VF_AVX512FP16VL 0 "register_operand")
+	(unspec:VF_AVX512FP16VL
+	  [(match_operand:VF_AVX512FP16VL 1 "vector_operand")
+	   (subreg:VF_AVX512FP16VL
+	     (match_operand:<ssePSmode> 2 "bcst_vector_operand") 0)
+	   (match_operand:VF_AVX512FP16VL 3 "vector_operand")]
+	   UNSPEC_COMPLEX_FMA))]
+  "TARGET_AVX512FP16"
+  "#"
+  "&& 1"
+  [(set (match_dup 0)
+	(unspec:<ssePSmode>
+	  [(match_dup 1) (match_dup 2) (match_dup 3)]
+	   UNSPEC_COMPLEX_FMA_PAIR))]
+  {
+    operands[0] = lowpart_subreg (<ssePSmode>mode, operands[0], <MODE>mode);
+    operands[1] = lowpart_subreg (<ssePSmode>mode, operands[1], <MODE>mode);
+    operands[3] = lowpart_subreg (<ssePSmode>mode, operands[3], 
+<MODE>mode);
+  })
+
+(define_insn_and_split "fma_<mode>_fcmaddc_bcst"
+  [(set (match_operand:VF_AVX512FP16VL 0 "register_operand")
+	(unspec:VF_AVX512FP16VL
+	  [(match_operand:VF_AVX512FP16VL 1 "vector_operand")
+	   (subreg:VF_AVX512FP16VL
+	     (match_operand:<ssePSmode> 2 "bcst_vector_operand") 0)
+	   (match_operand:VF_AVX512FP16VL 3 "vector_operand")]
+	   UNSPEC_COMPLEX_FCMA))]
+  "TARGET_AVX512FP16"
+  "#"
+  "&& 1"
+  [(set (match_dup 0)
+	(unspec:<ssePSmode>
+	  [(match_dup 1) (match_dup 2) (match_dup 3)]
+	   UNSPEC_COMPLEX_FCMA_PAIR))]
+  {
+    operands[0] = lowpart_subreg (<ssePSmode>mode, operands[0], <MODE>mode);
+    operands[1] = lowpart_subreg (<ssePSmode>mode, operands[1], <MODE>mode);
+    operands[3] = lowpart_subreg (<ssePSmode>mode, operands[3], 
+<MODE>mode);
+  })
+
 (define_insn "<avx512>_<complexopname>_<mode>_mask<round_name>"
   [(set (match_operand:VF_AVX512FP16VL 0 "register_operand" "=&v")
 	(vec_merge:VF_AVX512FP16VL
diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16vl-complex-broadcast-1.c b/gcc/testsuite/gcc.target/i386/avx512fp16vl-complex-broadcast-1.c
new file mode 100644
index 00000000000..3c8e84230f3
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/avx512fp16vl-complex-broadcast-1.c
@@ -0,0 +1,25 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512fp16 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "\[^\n\]*\\\{1to4\\\}" 2 } }  */
+/* { dg-final { scan-assembler-times "\[^\n\]*\\\{1to8\\\}" 2 } }  */
+/* { dg-final { scan-assembler-times "\[^\n\]*\\\{1to16\\\}" 2 } }  */
+
+#include <immintrin.h>
+
+volatile __m512h res0, a0, c0;
+volatile __m256h res1, a1, c1;
+volatile __m128h res2, a2, c2;
+volatile _Float16 *b;
+
+void extern
+avx_test(void)
+{
+  res0 = _mm512_fmadd_pch (a0, _mm512_set1_pch(*(b + 2 * 6)), c0);
+  res0 = _mm512_fcmadd_pch (a0, _mm512_set1_pch(*(b + 2 * 6)), c0);
+
+  res1 = _mm256_fmadd_pch (a1, _mm256_set1_pch(*(b + 2 * 6)), c1);
+  res1 = _mm256_fcmadd_pch (a1, _mm256_set1_pch(*(b + 2 * 6)), c1);
+
+  res2 =  _mm_fmadd_pch (a2, _mm_set1_pch(*(b + 2 * 6)), c2);
+  res2 =  _mm_fcmadd_pch (a2, _mm_set1_pch(*(b + 2 * 6)), c2); }
--
2.18.1