Remove nonexistent lasx instructions

jiegec · Dec 14, 2023 · 3c78221 · 3c78221
1 parent 0965215
commit 3c78221
Show file tree

Hide file tree

Showing 14 changed files with 12 additions and 41 deletions.
diff --git a/code/gen_impl.py b/code/gen_impl.py
@@ -753,8 +753,10 @@
                 file=f,
             )
             print(f"}}", file=f)
-        with open(f"{prefix}pickve2gr_{width}.h", "w") as f:
-            print(f"dst = a.{m}[idx];", file=f)
+        # skip xvpickve2gr_b/bu/h/hu
+        if prefix != "xv" or (width[0] != "b" and width[0] != "h"):
+            with open(f"{prefix}pickve2gr_{width}.h", "w") as f:
+                print(f"dst = a.{m}[idx];", file=f)
 
     for width in ["b", "bu", "h", "hu", "w", "wu", "d", "du"]:
         double_width = double_widths[width]

diff --git a/code/xvinsgr2vr_b.h b/code/xvinsgr2vr_b.h
diff --git a/code/xvinsgr2vr_h.h b/code/xvinsgr2vr_h.h
diff --git a/code/xvpickve2gr_b.h b/code/xvpickve2gr_b.h
diff --git a/code/xvpickve2gr_bu.h b/code/xvpickve2gr_bu.h
diff --git a/code/xvpickve2gr_h.h b/code/xvpickve2gr_h.h
diff --git a/code/xvpickve2gr_hu.h b/code/xvpickve2gr_hu.h
diff --git a/code/xvreplvei_b.h b/code/xvreplvei_b.h
diff --git a/code/xvreplvei_d.h b/code/xvreplvei_d.h
diff --git a/code/xvreplvei_h.h b/code/xvreplvei_h.h
diff --git a/code/xvreplvei_w.h b/code/xvreplvei_w.h
diff --git a/docs/lasx/misc.md b/docs/lasx/misc.md
@@ -10,8 +10,6 @@
 {{ xvilvl('w') }}
 {{ xvilvl('d') }}
 
-{{ xvinsgr2vr('b') }}
-{{ xvinsgr2vr('h') }}
 {{ xvinsgr2vr('w') }}
 {{ xvinsgr2vr('d') }}
 
@@ -45,10 +43,6 @@
 {{ xvpickev('w') }}
 {{ xvpickev('d') }}
 
-{{ xvpickve2gr('b') }}
-{{ xvpickve2gr('bu') }}
-{{ xvpickve2gr('h') }}
-{{ xvpickve2gr('hu') }}
 {{ xvpickve2gr('w') }}
 {{ xvpickve2gr('wu') }}
 {{ xvpickve2gr('d') }}
@@ -74,11 +68,6 @@
 {{ xvreplve('w') }}
 {{ xvreplve('d') }}
 
-{{ xvreplvei('b') }}
-{{ xvreplvei('h') }}
-{{ xvreplvei('w') }}
-{{ xvreplvei('d') }}
-
 {{ xvsat('b') }}
 {{ xvsat('bu') }}
 {{ xvsat('h') }}

diff --git a/gcc_lasxintrin.h b/gcc_lasxintrin.h
@@ -702,10 +702,10 @@ __m256i __lasx_xvssub_hu (__m256i, __m256i);
 __m256i __lasx_xvssub_w (__m256i, __m256i);
 __m256i __lasx_xvssub_wu (__m256i, __m256i);
 void __lasx_xvst (__m256i, void *, imm_n2048_2047);
-void __lasx_xvstelm_b (__m256i, void *, imm_n128_127, imm0_15);
-void __lasx_xvstelm_d (__m256i, void *, imm_n128_127, imm0_1);
-void __lasx_xvstelm_h (__m256i, void *, imm_n128_127, imm0_7);
-void __lasx_xvstelm_w (__m256i, void *, imm_n128_127, imm0_3);
+void __lasx_xvstelm_b (__m256i, void *, imm_n128_127, imm0_31);
+void __lasx_xvstelm_d (__m256i, void *, imm_n128_127, imm0_3);
+void __lasx_xvstelm_h (__m256i, void *, imm_n128_127, imm0_15);
+void __lasx_xvstelm_w (__m256i, void *, imm_n128_127, imm0_7);
 void __lasx_xvstx (__m256i, void *, long int);
 __m256i __lasx_xvsub_b (__m256i, __m256i);
 __m256i __lasx_xvsub_d (__m256i, __m256i);

diff --git a/main.py b/main.py
@@ -499,8 +499,9 @@ def vilvl(name):
 
     @my_macro(env)
     def vinsgr2vr(name):
+        global cur_vlen
         width = widths[name]
-        imm_upper = 128 // width - 1
+        imm_upper = cur_vlen // width - 1
         if name == "d":
             long = "long "
         else:
@@ -685,8 +686,9 @@ def vpcnt(name):
 
     @my_macro(env)
     def vstelm(name):
+        global cur_vlen
         width = widths[name]
-        imm_upper = 128 // width - 1
+        imm_upper = cur_vlen // width - 1
         return instruction(
             intrinsic=f"void __lsx_vstelm_{name} (__m128i data, void * addr, imm_n128_127 offset, imm0_{imm_upper} lane)",
             instr=f"vstelm.{name} vr, r, imm, imm",