Add vfdiv

README.md

@@ -192,8 +192,6 @@ Vector Multiplication High
 
 ### vfmul.s/d
 
-### vfdiv.s/d
-
 ### vfmax.s/d
 
 ### vfmin.s/d

code/gen_impl.py

@@ -48,6 +48,11 @@
     "qu": "qword",
 }
 
+members_fp = {
+    "s": "fp32",
+    "d": "fp64",
+}
+
 for width in ["b", "bu", "h", "hu", "w", "wu", "d", "du"]:
     w = widths[width]
     m = members[width]
@@ -200,4 +205,14 @@
         )
         print(f"}}", file=f)
 
+for width in ["s", "d"]:
+    m = members_fp[width]
+    with open(f"vfdiv_{width}.h", "w") as f:
+        print(f"for (int i = 0;i < {128 // w};i++) {{", file=f)
+        print(
+            f"  dst.{m}[i] = a.{m}[i] / b.{m}[i];",
+            file=f,
+        )
+        print(f"}}", file=f)
+
 os.system("clang-format -i *.cpp *.h")

code/gen_tb.py

@@ -1,6 +1,7 @@
 import os
 
 widths_signed = ["b", "h", "w", "d"]
+widths_fp = ["s", "d"]
 widths_all = ["b", "bu", "h", "hu", "w", "wu", "d", "du"]
 widths_vexth = ["h_b", "hu_bu", "w_h", "wu_hu", "d_w", "du_wu", "q_d", "qu_du"]
 widths_vaddw = [

code/vfdiv_d.h

@@ -0,0 +1,3 @@
+for (int i = 0; i < 2; i++) {
+  dst.fp64[i] = a.fp64[i] / b.fp64[i];
+}

code/vfdiv_s.h

@@ -0,0 +1,3 @@
+for (int i = 0; i < 2; i++) {
+  dst.fp32[i] = a.fp32[i] / b.fp32[i];
+}

docs/lsx/float_computation.md

@@ -45,3 +45,6 @@ for (int i = 0;i < 4;i++) {
     dst.fp32[i] = a.fp32[i] + b.fp32[i];
 }
 ```
+
+{{ vfdiv('s') }}
+{{ vfdiv('d') }}

main.py

@@ -27,6 +27,16 @@ def define_env(env):
         "qu": "unsigned",
     }
 
+    precisions = {
+        "s": "single",
+        "d": "double"
+    }
+
+    fp_types = {
+        "s": "__m128",
+        "d": "__m128d"
+    }
+
     def include(file):
         return open(f"code/{file}").read().strip()
 
@@ -323,6 +333,16 @@ def vfcmp(cond):
             """
         )
 
+    @env.macro
+    def vfdiv(name):
+        precision = precisions[name]
+        fp_type = fp_types[name]
+        return instruction(
+            intrinsic=f"{fp_type} __lsx_vfdiv_{name} ({fp_type} a, {fp_type} b)",
+            instr=f"vfdiv.{name} vr, vr, vr",
+            desc=f"Divide {fp_type} precision floating point elements in `a` by elements in `b`.",
+        )
+
     @env.macro
     def vshuf_hwd(name):
         width = widths[name]