[PATCH] Support cmul{_conj}v4hf3/cmla{_conj}v4hf4 with AVX512FP16 instruction.

[PATCH] Support cmul{_conj}v4hf3/cmla{_conj}v4hf4 with AVX512FP16 instruction.

[liuhongt]

Bootstrapped and regtested on x86_64-pc-linux-gnu{-m32,}
Ready push to trunk.

gcc/ChangeLog:

	* config/i386/mmx.md (cmlav4hf4): New expander.
	(cmla_conjv4hf4): Ditto.
	(cmulv4hf3): Ditto.
	(cmul_conjv4hf3): Ditto.

gcc/testsuite/ChangeLog:

	* gcc.target/i386/part-vect-complexhf.c: New test.
---
 gcc/config/i386/mmx.md                        | 86 +++++++++++++++++++
 .../gcc.target/i386/part-vect-complexhf.c     | 40 +++++++++
 2 files changed, 126 insertions(+)
 create mode 100644 gcc/testsuite/gcc.target/i386/part-vect-complexhf.c

diff --git a/gcc/config/i386/mmx.md b/gcc/config/i386/mmx.md
index 2b97bb8fa98..ba81ff72551 100644
--- a/gcc/config/i386/mmx.md
+++ b/gcc/config/i386/mmx.md
@@ -2622,6 +2622,92 @@ (define_expand "vec_fmsubaddv4hf4"
   DONE;
 })
 
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;
+;; Parallel half-precision floating point complex type operations
+;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+(define_expand "cmlav4hf4"
+  [(match_operand:V4HF 0 "register_operand")
+   (match_operand:V4HF 1 "vector_operand")
+   (match_operand:V4HF 2 "vector_operand")
+   (match_operand:V4HF 3 "vector_operand")]
+  "TARGET_AVX512FP16 && TARGET_AVX512VL"
+{
+  rtx op3 = gen_reg_rtx (V8HFmode);
+  rtx op2 = gen_reg_rtx (V8HFmode);
+  rtx op1 = gen_reg_rtx (V8HFmode);
+  rtx op0 = gen_reg_rtx (V8HFmode);
+
+  emit_insn (gen_movq_v4hf_to_sse (op3, operands[3]));
+  emit_insn (gen_movq_v4hf_to_sse (op2, operands[2]));
+  emit_insn (gen_movq_v4hf_to_sse (op1, operands[1]));
+
+  emit_insn (gen_cmlav8hf4 (op0, op1, op2, op3));
+
+  emit_move_insn (operands[0], lowpart_subreg (V4HFmode, op0, V8HFmode));
+  DONE;
+})
+
+(define_expand "cmla_conjv4hf4"
+  [(match_operand:V4HF 0 "register_operand")
+   (match_operand:V4HF 1 "vector_operand")
+   (match_operand:V4HF 2 "vector_operand")
+   (match_operand:V4HF 3 "vector_operand")]
+  "TARGET_AVX512FP16 && TARGET_AVX512VL"
+{
+  rtx op3 = gen_reg_rtx (V8HFmode);
+  rtx op2 = gen_reg_rtx (V8HFmode);
+  rtx op1 = gen_reg_rtx (V8HFmode);
+  rtx op0 = gen_reg_rtx (V8HFmode);
+
+  emit_insn (gen_movq_v4hf_to_sse (op3, operands[3]));
+  emit_insn (gen_movq_v4hf_to_sse (op2, operands[2]));
+  emit_insn (gen_movq_v4hf_to_sse (op1, operands[1]));
+
+  emit_insn (gen_cmla_conjv8hf4 (op0, op1, op2, op3));
+
+  emit_move_insn (operands[0], lowpart_subreg (V4HFmode, op0, V8HFmode));
+  DONE;
+})
+
+(define_expand "cmulv4hf3"
+  [(match_operand:V4HF 0 "register_operand")
+   (match_operand:V4HF 1 "vector_operand")
+   (match_operand:V4HF 2 "vector_operand")]
+  "TARGET_AVX512FP16 && TARGET_AVX512VL"
+{
+  rtx op2 = gen_reg_rtx (V8HFmode);
+  rtx op1 = gen_reg_rtx (V8HFmode);
+  rtx op0 = gen_reg_rtx (V8HFmode);
+
+  emit_insn (gen_movq_v4hf_to_sse (op2, operands[2]));
+  emit_insn (gen_movq_v4hf_to_sse (op1, operands[1]));
+
+  emit_insn (gen_cmulv8hf3 (op0, op1, op2));
+  emit_move_insn (operands[0], lowpart_subreg (V4HFmode, op0, V8HFmode));
+  DONE;
+})
+
+(define_expand "cmul_conjv4hf3"
+  [(match_operand:V4HF 0 "register_operand")
+   (match_operand:V4HF 1 "vector_operand")
+   (match_operand:V4HF 2 "vector_operand")]
+  "TARGET_AVX512FP16 && TARGET_AVX512VL"
+{
+  rtx op2 = gen_reg_rtx (V8HFmode);
+  rtx op1 = gen_reg_rtx (V8HFmode);
+  rtx op0 = gen_reg_rtx (V8HFmode);
+
+  emit_insn (gen_movq_v4hf_to_sse (op2, operands[2]));
+  emit_insn (gen_movq_v4hf_to_sse (op1, operands[1]));
+
+  emit_insn (gen_cmul_conjv8hf3 (op0, op1, op2));
+  emit_move_insn (operands[0], lowpart_subreg (V4HFmode, op0, V8HFmode));
+  DONE;
+})
+
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ;;
 ;; Parallel half-precision floating point conversion operations
diff --git a/gcc/testsuite/gcc.target/i386/part-vect-complexhf.c b/gcc/testsuite/gcc.target/i386/part-vect-complexhf.c
new file mode 100644
index 00000000000..b9f4ba2f4cf
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/part-vect-complexhf.c
@@ -0,0 +1,40 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-options "-O1 -ftree-vectorize -ffast-math -mavx512fp16 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vfmaddcph\[ \\t\]" 1 } } */
+/* { dg-final { scan-assembler-not "vfmadd\[123]*ph\[ \\t\]"} } */
+/* { dg-final { scan-assembler-not "vfmadd\[123]*sh\[ \\t\]"} } */
+/* { dg-final { scan-assembler-times "vfcmaddcph\[ \\t\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfmulcph\[ \\t\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfcmulcph\[ \\t\]" 1 } } */
+
+#include<complex.h>
+#define TYPE _Float16
+#define N 2
+
+void fma0 (_Complex TYPE *a, _Complex TYPE *b,
+           _Complex TYPE * __restrict c)
+{
+  for (int i = 0; i < N; i++)
+    c[i] += a[i] * b[i];
+}
+
+void fmaconj (_Complex TYPE a[restrict N], _Complex TYPE b[restrict N],
+	      _Complex TYPE c[restrict N])
+{
+  for (int i = 0; i < N; i++)
+    c[i] += a[i] * ~b[i];
+}
+
+void fmul (_Complex TYPE a[restrict N], _Complex TYPE b[restrict N],
+	   _Complex TYPE c[restrict N])
+{
+  for (int i = 0; i < N; i++)
+    c[i] = a[i] * b[i];
+}
+
+void fmulconj (_Complex TYPE a[restrict N], _Complex TYPE b[restrict N],
+	       _Complex TYPE c[restrict N])
+{
+  for (int i = 0; i < N; i++)
+    c[i] = a[i] * ~b[i];
+}
-- 
2.31.1