PPC/s390: [wasm][arm][ia32] Push instance only twice in lazy-compile builtin

Port d5b3d8e9

Original Commit Message:

    This change already landed for x64, now come arm and ia32. The code
    already existed for arm64.

    The wasm instance got pushed three times in the lazy-compile builtin:
    1) as part of the parameters;
    2) as a parameter for the runtime function;
    3) to load the jump table address after the runtime function;

    The third push can be avoided by loading the jump table address after
    all parameters get loaded from the stack again.

R=ahaas@chromium.org, joransiu@ca.ibm.com, junyan@redhat.com, midawson@redhat.com
BUG=
LOG=N

Change-Id: I731473b2d5e08e7ea5841ef589dd3f896b5302db
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3769698
Commit-Queue: Milad Farazmand <mfarazma@redhat.com>
Reviewed-by: Junliang Yan <junyan@redhat.com>
Reviewed-by: Andreas Haas <ahaas@chromium.org>
Cr-Commit-Position: refs/heads/main@{#81789}

src/builtins/ppc/builtins-ppc.cc

@@ -2940,12 +2940,7 @@ void Builtins::Generate_WasmCompileLazy(MacroAssembler* masm) {
     __ MultiPush(gp_regs);
     __ MultiPushF64AndV128(fp_regs, simd_regs);
 
-    // Push the Wasm instance for loading the jump table address after the
-    // runtime call.
-    __ Push(kWasmInstanceRegister);
-
-    // Push the Wasm instance again as an explicit argument to the runtime
-    // function.
+    // Push the Wasm instance as an explicit argument to the runtime function.
     __ Push(kWasmInstanceRegister);
     // Push the function index as second argument.
     __ Push(kWasmCompileLazyFuncIndexRegister);
@@ -2955,19 +2950,21 @@ void Builtins::Generate_WasmCompileLazy(MacroAssembler* masm) {
     __ CallRuntime(Runtime::kWasmCompileLazy, 2);
     // The runtime function returns the jump table slot offset as a Smi. Use
     // that to compute the jump target in r11.
-    __ Pop(kWasmInstanceRegister);
-    __ LoadU64(
-        r11,
-        MemOperand(kWasmInstanceRegister,
-                   WasmInstanceObject::kJumpTableStartOffset - kHeapObjectTag),
-        r0);
     __ SmiUntag(kReturnRegister0);
-    __ AddS64(r11, r11, kReturnRegister0);
-    // r11 now holds the jump table slot where we want to jump to in the end.
+    __ mr(r11, kReturnRegister0);
 
     // Restore registers.
     __ MultiPopF64AndV128(fp_regs, simd_regs);
     __ MultiPop(gp_regs);
+
+    // After the instance register has been restored, we can add the jump table
+    // start to the jump table offset already stored in r8.
+    __ LoadU64(
+        ip,
+        MemOperand(kWasmInstanceRegister,
+                   WasmInstanceObject::kJumpTableStartOffset - kHeapObjectTag),
+        r0);
+    __ AddS64(r11, r11, ip);
   }
 
   // Finally, jump to the jump table slot for the function.

src/builtins/s390/builtins-s390.cc

@@ -2938,12 +2938,7 @@ void Builtins::Generate_WasmCompileLazy(MacroAssembler* masm) {
     __ MultiPush(gp_regs);
     __ MultiPushF64OrV128(fp_regs, ip);
 
-    // Push the Wasm instance for loading the jump table address after the
-    // runtime call.
-    __ Push(kWasmInstanceRegister);
-
-    // Push the Wasm instance again as an explicit argument to the runtime
-    // function.
+    // Push the Wasm instance as an explicit argument to the runtime function.
     __ Push(kWasmInstanceRegister);
     // Push the function index as second argument.
     __ Push(kWasmCompileLazyFuncIndexRegister);
@@ -2953,17 +2948,19 @@ void Builtins::Generate_WasmCompileLazy(MacroAssembler* masm) {
     __ CallRuntime(Runtime::kWasmCompileLazy, 2);
     // The runtime function returns the jump table slot offset as a Smi. Use
     // that to compute the jump target in ip.
-    __ Pop(kWasmInstanceRegister);
-    __ LoadU64(ip, MemOperand(kWasmInstanceRegister,
-                              WasmInstanceObject::kJumpTableStartOffset -
-                                  kHeapObjectTag));
     __ SmiUntag(kReturnRegister0);
-    __ AddS64(ip, ip, kReturnRegister0);
-    // ip now holds the jump table slot where we want to jump to in the end.
+    __ mov(ip, kReturnRegister0);
 
     // Restore registers.
     __ MultiPopF64OrV128(fp_regs, r1);
     __ MultiPop(gp_regs);
+
+    // After the instance register has been restored, we can add the jump table
+    // start to the jump table offset already stored in r8.
+    __ LoadU64(r0, MemOperand(kWasmInstanceRegister,
+                              WasmInstanceObject::kJumpTableStartOffset -
+                                  kHeapObjectTag));
+    __ AddS64(ip, ip, r0);
   }
 
   // Finally, jump to the jump table slot for the function.