From: Tong Shen <endlessroad@google.com>
Date: Thu, 21 Jan 2021 00:11:01 +0000 (-0800)
Subject: arch-x86: implement PSHUFB SSE instruction.
X-Git-Tag: develop-gem5-snapshot~252
X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=3d257c287397ae5370209bd05ca55d8e6a6dd3eb;p=gem5.git

arch-x86: implement PSHUFB SSE instruction.

Change-Id: I9398f9ecb26b6aabf4015e0e285fdc2f4c2487dd
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39495
Reviewed-by: Tong Shen <endlessroad@google.com>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
---

diff --git a/src/arch/x86/isa/decoder/three_byte_0f38_opcodes.isa b/src/arch/x86/isa/decoder/three_byte_0f38_opcodes.isa
index 3165eb708..0f4330bf7 100644
--- a/src/arch/x86/isa/decoder/three_byte_0f38_opcodes.isa
+++ b/src/arch/x86/isa/decoder/three_byte_0f38_opcodes.isa
@@ -31,7 +31,7 @@
 'X86ISA::ThreeByte0F38Opcode': decode LEGACY_OP {
     format WarnUnimpl {
         1: decode OPCODE_OP {
-            0x00: pshufb_Vdq_Wdq();
+            0x00: Inst::PSHUFB(Vo, Wo);
             0x01: phaddw_Vdq_Wdq();
             0x02: phaddd_Vdq_Wdq();
             0x03: phaddsw_Vdq_Wdq();
diff --git a/src/arch/x86/isa/insts/simd128/integer/data_reordering/shuffle.py b/src/arch/x86/isa/insts/simd128/integer/data_reordering/shuffle.py
index 6651d8758..4187c4f59 100644
--- a/src/arch/x86/isa/insts/simd128/integer/data_reordering/shuffle.py
+++ b/src/arch/x86/isa/insts/simd128/integer/data_reordering/shuffle.py
@@ -84,4 +84,32 @@ def macroop PSHUFLW_XMM_P_I {
     ldfp ufp1, seg, riprel, "DISPLACEMENT", dataSize=8
     shuffle xmml, ufp1, ufp1, size=2, ext=imm
 };
-'''
+
+def macroop PSHUFB_XMM_XMM {
+    movfp ufp1, xmmlm, dataSize=8
+    movfp ufp2, xmmhm, dataSize=8
+    shuffle ufp1, xmml, xmmh, size=1, ext=0
+    shuffle ufp2, xmml, xmmh, size=1, ext=0
+    movfp xmml, ufp1, dataSize=8
+    movfp xmmh, ufp2, dataSize=8
+};
+
+def macroop PSHUFB_XMM_M {
+    ldfp ufp1, seg, sib, "DISPLACEMENT", dataSize=8
+    ldfp ufp2, seg, sib, "DISPLACEMENT + 8", dataSize=8
+    shuffle ufp1, xmml, xmmh, size=1, ext=0
+    shuffle ufp2, xmml, xmmh, size=1, ext=0
+    movfp xmml, ufp1, dataSize=8
+    movfp xmmh, ufp2, dataSize=8
+};
+
+def macroop PSHUFB_XMM_P {
+    rdip t7
+    ldfp ufp1, seg, riprel, "DISPLACEMENT", dataSize=8
+    ldfp ufp2, seg, riprel, "DISPLACEMENT + 8", dataSize=8
+    shuffle ufp1, xmml, xmmh, size=1, ext=0
+    shuffle ufp2, xmml, xmmh, size=1, ext=0
+    movfp xmml, ufp1, dataSize=8
+    movfp xmmh, ufp2, dataSize=8
+};
+'''
\ No newline at end of file
diff --git a/src/arch/x86/isa/microops/mediaop.isa b/src/arch/x86/isa/microops/mediaop.isa
index bf5fc6766..7e5fd1011 100644
--- a/src/arch/x86/isa/microops/mediaop.isa
+++ b/src/arch/x86/isa/microops/mediaop.isa
@@ -337,23 +337,32 @@ let {{
             if (size == 8) {
                 options = 2;
                 optionBits = 1;
+            } else if (size == 1) {
+                options = 16;
+                optionBits = 8;
             } else {
                 options = 4;
                 optionBits = 2;
             }
 
             uint64_t result = 0;
-            uint8_t sel = ext;
+            // PSHUFB stores shuffle encoding in destination XMM register
+            // directly (instead of passed in by ext).
+            uint64_t sel = (size == 1) ? FpDestReg_uqw : ext;
 
             for (int i = 0; i < items; i++) {
                 uint64_t resBits;
                 uint8_t lsel = sel & mask(optionBits);
-                if (lsel * size >= sizeof(double)) {
+                if (size == 1 && bits(lsel, 7)) {
+                    // PSHUFB sets result byte to zero when highest bit of the
+                    // corresponding shuffle encoding is 1.
+                    resBits = 0;
+                } else if (lsel * size >= sizeof(double)) {
                     lsel -= options / 2;
                     resBits = bits(FpSrcReg2_uqw,
                             (lsel + 1) * sizeBits - 1,
                             (lsel + 0) * sizeBits);
-                }  else {
+                } else {
                     resBits = bits(FpSrcReg1_uqw,
                             (lsel + 1) * sizeBits - 1,
                             (lsel + 0) * sizeBits);