Cache registers in more instructions

It turns out to be useful to preserve the register cache in the `bs_get_position` and `bs_set_position` instructions.
bjorng · Nov 4, 2024 · 8d92b40 · 8d92b40
1 parent ca6bc56
commit 8d92b40
Show file tree

Hide file tree

Showing 2 changed files with 39 additions and 17 deletions.
diff --git a/erts/emulator/beam/jit/arm/instr_bs.cpp b/erts/emulator/beam/jit/arm/instr_bs.cpp
@@ -230,16 +230,22 @@ void BeamModuleAssembler::emit_i_bs_match_string(const ArgRegister &Ctx,
 void BeamModuleAssembler::emit_i_bs_get_position(const ArgRegister &Ctx,
                                                  const ArgRegister &Dst) {
     const int start_offset = offsetof(ErlSubBits, start);
-    auto ctx_reg = load_source(Ctx);
-    auto dst_reg = init_destination(Dst, TMP2);
+    auto ctx = load_source(Ctx);
+    auto dst = init_destination(Dst, TMP2);
+    a64::Gp ctx_reg = ctx.reg;
+    a64::Gp dst_reg = dst.reg;
 
     /* Match contexts can never be literals, so we can skip clearing literal
      * tags. */
-    a.ldur(dst_reg.reg, emit_boxed_val(ctx_reg.reg, start_offset));
-    a.lsl(dst_reg.reg, dst_reg.reg, imm(_TAG_IMMED1_SIZE));
-    a.orr(dst_reg.reg, dst_reg.reg, imm(_TAG_IMMED1_SMALL));
+    preserve_cache(
+            [&]() {
+                a.ldur(dst_reg, emit_boxed_val(ctx_reg, start_offset));
+                a.lsl(dst_reg, dst_reg, imm(_TAG_IMMED1_SIZE));
+                a.orr(dst_reg, dst_reg, imm(_TAG_IMMED1_SMALL));
+            },
+            dst_reg);
 
-    flush_var(dst_reg);
+    flush_var(dst);
 }
 
 void BeamModuleAssembler::emit_bs_get_small(const Label &fail,
@@ -368,9 +374,15 @@ void BeamModuleAssembler::emit_bs_set_position(const ArgRegister &Ctx,
                                                const ArgRegister &Pos) {
     const int start_offset = offsetof(ErlSubBits, start);
     auto [ctx, pos] = load_sources(Ctx, TMP1, Pos, TMP2);
-
-    a.lsr(TMP2, pos.reg, imm(_TAG_IMMED1_SIZE));
-    a.stur(TMP2, emit_boxed_val(ctx.reg, start_offset));
+    a64::Gp ctx_reg = ctx.reg;
+    a64::Gp pos_reg = pos.reg;
+
+    preserve_cache(
+            [&]() {
+                a.lsr(TMP2, pos_reg, imm(_TAG_IMMED1_SIZE));
+                a.stur(TMP2, emit_boxed_val(ctx_reg, start_offset));
+            },
+            TMP2);
 }
 
 void BeamModuleAssembler::emit_i_bs_get_binary_all2(const ArgRegister &Ctx,
@@ -558,7 +570,6 @@ void BeamModuleAssembler::emit_i_bs_get_bin_and_tail(const ArgRegister &Ctx,
     mov_arg(Dst2, ARG1);
 }
 
-
 void BeamModuleAssembler::emit_i_bs_get_float2(const ArgRegister &Ctx,
                                                const ArgLabel &Fail,
                                                const ArgWord &Live,

diff --git a/erts/emulator/beam/jit/x86/instr_bs.cpp b/erts/emulator/beam/jit/x86/instr_bs.cpp
@@ -242,15 +242,22 @@ void BeamModuleAssembler::emit_i_bs_match_string(const ArgRegister &Ctx,
 
 void BeamModuleAssembler::emit_i_bs_get_position(const ArgRegister &Ctx,
                                                  const ArgRegister &Dst) {
+    x86::Gp tmp_reg = alloc_temp_reg();
+
     mov_arg(ARG1, Ctx);
 
     /* Match contexts can never be literals, so we can skip clearing literal
      * tags. */
-    a.mov(ARG1, emit_boxed_val(ARG1, offsetof(ErlSubBits, start)));
-    a.sal(ARG1, imm(_TAG_IMMED1_SIZE));
-    a.or_(ARG1, imm(_TAG_IMMED1_SMALL));
-
-    mov_arg(Dst, ARG1);
+    mov_preserve_cache(tmp_reg,
+                       emit_boxed_val(ARG1, offsetof(ErlSubBits, start)));
+    preserve_cache(
+            [&]() {
+                a.sal(tmp_reg, imm(_TAG_IMMED1_SIZE));
+                a.or_(tmp_reg, imm(_TAG_IMMED1_SMALL));
+            },
+            tmp_reg);
+
+    mov_arg(Dst, tmp_reg);
 }
 
 void BeamModuleAssembler::emit_bs_get_small(const Label &fail,
@@ -385,8 +392,12 @@ void BeamModuleAssembler::emit_bs_set_position(const ArgRegister &Ctx,
     mov_arg(ARG1, Ctx);
     mov_arg(ARG2, Pos);
 
-    a.sar(ARG2, imm(_TAG_IMMED1_SIZE));
-    a.mov(emit_boxed_val(ARG1, offsetof(ErlSubBits, start)), ARG2);
+    preserve_cache(
+            [&]() {
+                a.sar(ARG2, imm(_TAG_IMMED1_SIZE));
+            },
+            ARG2);
+    mov_preserve_cache(emit_boxed_val(ARG1, offsetof(ErlSubBits, start)), ARG2);
 }
 
 void BeamModuleAssembler::emit_i_bs_get_binary_all2(const ArgRegister &Ctx,