From 69de50a8786bc817e013e77a17ff68d7e1dd58a5 Mon Sep 17 00:00:00 2001
From: MerryMage <MerryMage@users.noreply.github.com>
Date: Sat, 10 Feb 2018 16:24:43 +0000
Subject: [PATCH] emit_x64_vector: Add SSE4.1 implementations for
 VectorZeroExtend

---
 src/backend_x64/emit_x64_vector.cpp | 57 ++++++++++++++++-------------
 1 file changed, 32 insertions(+), 25 deletions(-)

diff --git a/src/backend_x64/emit_x64_vector.cpp b/src/backend_x64/emit_x64_vector.cpp
index 7c206005..adc78b61 100644
--- a/src/backend_x64/emit_x64_vector.cpp
+++ b/src/backend_x64/emit_x64_vector.cpp
@@ -703,45 +703,52 @@ void EmitX64::EmitVectorSub64(EmitContext& ctx, IR::Inst* inst) {
     EmitVectorOperation(code, ctx, inst, &Xbyak::CodeGenerator::psubq);
 }
 
-static void EmitVectorZeroExtend(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst, int size) {
+void EmitX64::EmitVectorZeroExtend8(EmitContext& ctx, IR::Inst* inst) {
     auto args = ctx.reg_alloc.GetArgumentInfo(inst);
-
     const Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
-    const Xbyak::Xmm zeros = ctx.reg_alloc.ScratchXmm();
-
-    code.pxor(zeros, zeros);
-    switch (size) {
-    case 8:
+    if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
+        code.pmovzxbw(a, a);
+    } else {
+        const Xbyak::Xmm zeros = ctx.reg_alloc.ScratchXmm();
+        code.pxor(zeros, zeros);
         code.punpcklbw(a, zeros);
-        break;
-    case 16:
-        code.punpcklwd(a, zeros);
-        break;
-    case 32:
-        code.punpckldq(a, zeros);
-        break;
-    case 64:
-        code.punpcklqdq(a, zeros);
-        break;
     }
-
     ctx.reg_alloc.DefineValue(inst, a);
 }
 
-void EmitX64::EmitVectorZeroExtend8(EmitContext& ctx, IR::Inst* inst) {
-    EmitVectorZeroExtend(code, ctx, inst, 8);
-}
-
 void EmitX64::EmitVectorZeroExtend16(EmitContext& ctx, IR::Inst* inst) {
-    EmitVectorZeroExtend(code, ctx, inst, 16);
+    auto args = ctx.reg_alloc.GetArgumentInfo(inst);
+    const Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
+    if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
+        code.pmovzxwd(a, a);
+    } else {
+        const Xbyak::Xmm zeros = ctx.reg_alloc.ScratchXmm();
+        code.pxor(zeros, zeros);
+        code.punpcklwd(a, zeros);
+    }
+    ctx.reg_alloc.DefineValue(inst, a);
 }
 
 void EmitX64::EmitVectorZeroExtend32(EmitContext& ctx, IR::Inst* inst) {
-    EmitVectorZeroExtend(code, ctx, inst, 32);
+    auto args = ctx.reg_alloc.GetArgumentInfo(inst);
+    const Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
+    if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
+        code.pmovzxdq(a, a);
+    } else {
+        const Xbyak::Xmm zeros = ctx.reg_alloc.ScratchXmm();
+        code.pxor(zeros, zeros);
+        code.punpckldq(a, zeros);
+    }
+    ctx.reg_alloc.DefineValue(inst, a);
 }
 
 void EmitX64::EmitVectorZeroExtend64(EmitContext& ctx, IR::Inst* inst) {
-    EmitVectorZeroExtend(code, ctx, inst, 64);
+    auto args = ctx.reg_alloc.GetArgumentInfo(inst);
+    const Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
+    const Xbyak::Xmm zeros = ctx.reg_alloc.ScratchXmm();
+    code.pxor(zeros, zeros);
+    code.punpcklqdq(a, zeros);
+    ctx.reg_alloc.DefineValue(inst, a);
 }
 
 void EmitX64::EmitVectorZeroUpper(EmitContext& ctx, IR::Inst* inst) {