8342103: C2 compiler support for Float16 type and associated scalar operations

src/hotspot/cpu/x86/x86.ad

@@ -10901,6 +10901,16 @@ instruct convF2HFAndS2HF(regF dst, regF src)
   ins_pipe(pipe_slow);
 %}
 
+instruct convHF2SAndHF2F(regF dst, regF src)
+%{
+  match(Set dst (ConvHF2F (ReinterpretHF2S src)));
+  format %{ "convHF2SAndHF2F $dst, $src" %}
+  ins_encode %{
+    __ vcvtph2ps($dst$$XMMRegister, $src$$XMMRegister, Assembler::AVX_128bit);
+  %}
+  ins_pipe(pipe_slow);
+%}
+
 instruct reinterpretHF2S(rRegI dst, regF src)
 %{
   match(Set dst (ReinterpretHF2S src));

test/hotspot/jtreg/compiler/c2/irTests/TestFloat16ScalarOperations.java

@@ -43,6 +43,8 @@ public class TestFloat16ScalarOperations {
     private short[] dst;
     private short res;
 
+    private static final Float16 ONE = valueOf(1.0f);
+    private static final Float16 MONE = valueOf(-1.0f);
     private static final Float16 POSITIVE_ZERO = valueOf(0.0f);
     private static final Float16 NEGATIVE_ZERO = valueOf(-0.0f);
     private static final Float16 MIN_NORMAL = valueOf(0x1.0P-14f);
@@ -81,6 +83,19 @@ static void assertResult(float actual, float expected, String msg, int iter) {
         }
     }
 
+    @Test
+    @IR(counts = {"convHF2SAndHF2F", " >0 "}, phase = {CompilePhase.FINAL_CODE},
+        applyIfCPUFeature = {"avx512_fp16", "true"})
+    public void testEliminateIntermediateHF2S() {
+        Float16 res = shortBitsToFloat16((short)0);
+        for (int i = 0; i < count; i++) {
+            // Intermediate HF2S + S2HF is eliminated in following transformation
+            // AddHF S2HF(HF2S (AddHF S2HF(src[i]), S2HF(0))), S2HF(src[i]) => AddHF (AddHF S2HF(src[i]), S2HF(0)), S2HF(src[i])
+            res = add(add(res, shortBitsToFloat16(src[i])), shortBitsToFloat16(src[i]));
+            dst[i] = (short)res.floatValue();
+        }
+    }
+
     @Test
     @IR(counts = {IRNode.ADD_HF, " >0 ", IRNode.REINTERPRET_S2HF, " >0 ", IRNode.REINTERPRET_HF2S, " >0 "},
         applyIfCPUFeature = {"avx512_fp16", "true"})
@@ -137,6 +152,17 @@ public void testDiv() {
         }
     }
 
+    @Test
+    @IR(counts = {IRNode.DIV_HF, " 0 ", IRNode.REINTERPRET_S2HF, " 0 ", IRNode.REINTERPRET_HF2S, " 0 "},
+        applyIfCPUFeature = {"avx512_fp16", "true"})
+    public void testDivByOne() {
+        Float16 res = shortBitsToFloat16((short)0);
+        for (int i = 0; i < count; i++) {
+            res = Float16.divide(shortBitsToFloat16(src[i]), ONE);
+            dst[i] = float16ToRawShortBits(res);
+        }
+    }
+
     @Test
     @IR(counts = {IRNode.MAX_HF, " >0 ", IRNode.REINTERPRET_S2HF, " >0 ", IRNode.REINTERPRET_HF2S, " >0 "},
         applyIfCPUFeature = {"avx512_fp16", "true"})
@@ -417,6 +443,24 @@ public void testMulConstantFolding() {
         assertResult(multiply(multiply(multiply(valueOf(1.0f), valueOf(2.0f)), valueOf(3.0f)), valueOf(4.0f)).floatValue(), 1.0f * 2.0f * 3.0f * 4.0f, "testMulConstantFolding");
     }
 
+    @Test
+    @IR(counts = {IRNode.SQRT_HF, " 0 ", IRNode.REINTERPRET_S2HF, " 0 ", IRNode.REINTERPRET_HF2S, " 0 "},
+        applyIfCPUFeature = {"avx512_fp16", "true"})
+    public void testSqrtConstantFolding() {
+        // If the argument is NaN or less than zero, then the result is NaN.
+        assertResult(sqrt(Float16.NaN).floatValue(), Float.NaN, "testSqrtConstantFolding");
+
+        // If the argument is positive infinity, then the result is positive infinity.
+        assertResult(sqrt(Float16.POSITIVE_INFINITY).floatValue(), Float.POSITIVE_INFINITY, "testSqrtConstantFolding");
+
+        // If the argument is positive zero or negative zero, then the result is the same as the argument.
+        assertResult(sqrt(POSITIVE_ZERO).floatValue(), 0.0f, "testSqrtConstantFolding");
+        assertResult(sqrt(NEGATIVE_ZERO).floatValue(), -0.0f, "testSqrtConstantFolding");
+
+        // Other cases.
+        assertResult(Math.round(sqrt(valueOf(0x1.ffcP+14f)).floatValue()), Math.round(Math.sqrt(0x1.ffcP+14f)), "testSqrtConstantFolding");
+    }
+
     @Test
     @IR(counts = {IRNode.FMA_HF, " 0 ", IRNode.REINTERPRET_S2HF, " 0 ", IRNode.REINTERPRET_HF2S, " 0 "},
         applyIfCPUFeature = {"avx512_fp16", "true"})

test/hotspot/jtreg/compiler/lib/ir_framework/test/IREncodingPrinter.java

@@ -108,8 +108,6 @@ public class IREncodingPrinter {
         "sha3",
         "asimd",
         "sve",
-        "fphp",
-        "asimdhp",
         // Riscv64
         "rvv",
         "zvbb",