Add vftint.l.d/vftint.w.s

README.md

@@ -56,14 +56,6 @@ TODO List:
 
 ### vffinth.d.w
 
-### vftint.w.s/l.d
-
-### vftintrm.w.s/l.d
-
-### vftintrz.w.s/l.d
-
-### vftintrne.w.s/l.d
-
 ### vftint.wu.s/lu.d
 
 ### vftintrz.wu.s/lu.d

code/gen_impl.py

@@ -797,6 +797,18 @@
                     file=f,
                 )
                 print(f"}}", file=f)
+    for rounding in ["", "rm", "rp", "rz", "rne"]:
+        if width == "s":
+            int_width = "w"
+        else:
+            int_width = "l"
+        with open(f"vftint{rounding}_{int_width}_{width}.h", "w") as f:
+            print(f"for (int i = 0;i < {128 // w};i++) {{", file=f)
+            print(
+                f"  dst.{int_m}[i] = a.{m}[i]; // rounding mode is not expressed in C",
+                file=f,
+            )
+            print(f"}}", file=f)
 
     for name in ["max", "min"]:
         with open(f"vf{name}_{width}.h", "w") as f:

code/vftint_l_d.h

@@ -0,0 +1,3 @@
+for (int i = 0; i < 2; i++) {
+  dst.dword[i] = a.fp64[i]; // rounding mode is not expressed in C
+}

code/vftint_w_s.h

@@ -0,0 +1,3 @@
+for (int i = 0; i < 4; i++) {
+  dst.word[i] = a.fp32[i]; // rounding mode is not expressed in C
+}

code/vftintrm_l_d.h

@@ -0,0 +1,3 @@
+for (int i = 0; i < 2; i++) {
+  dst.dword[i] = a.fp64[i]; // rounding mode is not expressed in C
+}

code/vftintrm_w_s.h

@@ -0,0 +1,3 @@
+for (int i = 0; i < 4; i++) {
+  dst.word[i] = a.fp32[i]; // rounding mode is not expressed in C
+}

code/vftintrne_l_d.h

@@ -0,0 +1,3 @@
+for (int i = 0; i < 2; i++) {
+  dst.dword[i] = a.fp64[i]; // rounding mode is not expressed in C
+}

code/vftintrne_w_s.h

@@ -0,0 +1,3 @@
+for (int i = 0; i < 4; i++) {
+  dst.word[i] = a.fp32[i]; // rounding mode is not expressed in C
+}

code/vftintrp_l_d.h

@@ -0,0 +1,3 @@
+for (int i = 0; i < 2; i++) {
+  dst.dword[i] = a.fp64[i]; // rounding mode is not expressed in C
+}

code/vftintrp_w_s.h

@@ -0,0 +1,3 @@
+for (int i = 0; i < 4; i++) {
+  dst.word[i] = a.fp32[i]; // rounding mode is not expressed in C
+}

code/vftintrz_l_d.h

@@ -0,0 +1,3 @@
+for (int i = 0; i < 2; i++) {
+  dst.dword[i] = a.fp64[i]; // rounding mode is not expressed in C
+}

code/vftintrz_w_s.h

@@ -0,0 +1,3 @@
+for (int i = 0; i < 4; i++) {
+  dst.word[i] = a.fp32[i]; // rounding mode is not expressed in C
+}

docs/lsx/float_conversion.md

@@ -74,3 +74,14 @@ Convert double precision floating point elements in `a` and `b` to double precis
 {{ vftint_l_s('rz', 'h') }}
 {{ vftint_l_s('rne', 'l') }}
 {{ vftint_l_s('rne', 'h') }}
+
+{{ vftint('', 'l', 'd') }}
+{{ vftint('', 'w', 's') }}
+{{ vftint('rm', 'l', 'd') }}
+{{ vftint('rm', 'w', 's') }}
+{{ vftint('rp', 'l', 'd') }}
+{{ vftint('rp', 'w', 's') }}
+{{ vftint('rz', 'l', 'd') }}
+{{ vftint('rz', 'w', 's') }}
+{{ vftint('rne', 'l', 'd') }}
+{{ vftint('rne', 'w', 's') }}

main.py

@@ -1257,12 +1257,7 @@ def vsat(name):
             desc=f"Clamp {signedness} {width}-bit elements in `a` to range specified by `imm`.",
         )
 
-    @env.macro
-    def vftint_l_s(rounding, low_high):
-        if low_high == "l":
-            half = "lower"
-        else:
-            half = "higher"
+    def get_rounding_mode(rounding):
         if rounding == "":
             rounding_mode = "using current rounding mode specified in `fscr`"
         elif rounding == "rm":
@@ -1275,8 +1270,34 @@ def vftint_l_s(rounding, low_high):
             rounding_mode = "rounding towards nearest even"
         else:
             assert False
+        return rounding_mode
+
+    @env.macro
+    def vftint_l_s(rounding, low_high):
+        if low_high == "l":
+            half = "lower"
+        else:
+            half = "higher"
+        rounding_mode = get_rounding_mode(rounding)
         return instruction(
-            intrinsic=f"__m128i __lsx_vftint{rounding}{low_high}_l_s (__m128i a)",
+            intrinsic=f"__m128i __lsx_vftint{rounding}{low_high}_l_s (__m128 a)",
             instr=f"vftint{rounding}{low_high}.l.s vr, vr",
             desc=f"Convert single-precision floating point elements in {half} part of `a` to 64-bit integer, {rounding_mode}.",
+        )
+
+    @env.macro
+    def vftint(rounding, name, name2):
+        if name2 == "d":
+            arg_type = "__m128d"
+            precision = "double"
+            int_width = 64
+        else:
+            arg_type = "__m128"
+            precision = "single"
+            int_width = 32
+        rounding_mode = get_rounding_mode(rounding)
+        return instruction(
+            intrinsic=f"__m128i __lsx_vftint{rounding}_{name}_{name2} ({arg_type} a)",
+            instr=f"vftint{rounding}.{name}.{name2} vr, vr",
+            desc=f"Convert {precision}-precision floating point elements in `a` to {int_width}-bit integer, {rounding_mode}.",
         )