[RFC PATCH v2 58/67] Hexagon HVX import macro definitions

[Taylor Simpson]

Imported from the Hexagon architecture library
    imported/allext_macros.def       Top level macro include for all extensions
    imported/mmvec/macros.def        HVX macro definitions
The macro definition files specify instruction attributes that are applied
to each instruction that reverences the macro.

Signed-off-by: Taylor Simpson <address@hidden>
---
 target/hexagon/imported/allext_macros.def |   25 +
 target/hexagon/imported/mmvec/macros.def  | 1110 +++++++++++++++++++++++++++++
 2 files changed, 1135 insertions(+)
 create mode 100644 target/hexagon/imported/allext_macros.def
 create mode 100755 target/hexagon/imported/mmvec/macros.def

diff --git a/target/hexagon/imported/allext_macros.def 
b/target/hexagon/imported/allext_macros.def
new file mode 100644
index 0000000..e1f16eb
--- /dev/null
+++ b/target/hexagon/imported/allext_macros.def
@@ -0,0 +1,25 @@
+/*
+ *  Copyright(c) 2019-2020 Qualcomm Innovation Center, Inc. All Rights 
Reserved.
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2 of the License, or
+ *  (at your option) any later version.
+ *
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *  GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License
+ *  along with this program; if not, see <http://www.gnu.org/licenses/>.
+ */
+
+/*
+ * Top level file for all instruction set extensions
+ */
+#define EXTNAME mmvec
+#define EXTSTR "mmvec"
+#include "mmvec/macros.def"
+#undef EXTNAME
+#undef EXTSTR
diff --git a/target/hexagon/imported/mmvec/macros.def 
b/target/hexagon/imported/mmvec/macros.def
new file mode 100755
index 0000000..fa0beb1
--- /dev/null
+++ b/target/hexagon/imported/mmvec/macros.def
@@ -0,0 +1,1110 @@
+/*
+ *  Copyright(c) 2019-2020 Qualcomm Innovation Center, Inc. All Rights 
Reserved.
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2 of the License, or
+ *  (at your option) any later version.
+ *
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *  GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License
+ *  along with this program; if not, see <http://www.gnu.org/licenses/>.
+ */
+
+DEF_MACRO(fDUMPQ,(STR,REG),
+       "dump REG",
+       "dump REG",
+       do {
+               printf(STR ":" #REG ": 0x%016llx\n",REG.ud[0]);
+       } while (0),
+       ()
+)
+
+DEF_MACRO(fUSE_LOOKUP_ADDRESS_BY_REV,(PROC),
+       "",
+       "Use full VA address for lookup and exception based on REV ",
+       PROC->arch_proc_options->mmvec_use_full_va_for_lookup,
+       ()
+)
+
+DEF_MACRO(fUSE_LOOKUP_ADDRESS,(),
+       "",
+       "Use full VA address for lookup and exception",
+       1,
+       ()
+)
+
+DEF_MACRO(fRT8NOTE, (),
+       "",
+       "",
+       ,
+       (A_NOTE_RT8)
+)
+
+DEF_MACRO(fNOTQ,(VAL),
+       "~VAL",
+       "~VAL",
+       /* Will break Visual Studio? */
+       ({mmqreg_t _ret = {0}; int _i_; for (_i_ = 0; _i_ < fVECSIZE()/64; 
_i_++) _ret.ud[_i_] = ~VAL.ud[_i_]; _ret;}),
+       ()
+)
+
+DEF_MACRO(fGETQBITS,(REG,WIDTH,MASK,BITNO),
+       "REG[BITNO+WIDTH-1:BITNO]",
+       "Get MASK bits at BITNO from REG",
+       ((MASK) & (REG.w[(BITNO)>>5] >> ((BITNO) & 0x1f))),
+       ()
+)
+
+DEF_MACRO(fGETQBIT,(REG,BITNO),
+       "REG[BITNO]",
+       "Get bit BITNO from REG",
+       fGETQBITS(REG,1,1,BITNO),
+       ()
+)
+
+DEF_MACRO(fGENMASKW,(QREG,IDX),
+       "maskw(QREG,IDX)",
+       "Generate mask from QREG for word IDX",
+       (((fGETQBIT(QREG,(IDX*4+0)) ? 0xFF : 0x0) << 0)
+       |((fGETQBIT(QREG,(IDX*4+1)) ? 0xFF : 0x0) << 8)
+       |((fGETQBIT(QREG,(IDX*4+2)) ? 0xFF : 0x0) << 16)
+       |((fGETQBIT(QREG,(IDX*4+3)) ? 0xFF : 0x0) << 24)),
+       ()
+)
+DEF_MACRO(fGET10BIT,(COE,VAL,POS),
+       "COE=(((((fGETUBYTE(3,VAL) >> (2 * POS)) & 3) << 8) | 
fGETUBYTE(POS,VAL)) << 6) >> 6;",
+       "Get 10-bit coefficient from current word value and byte position",
+       {
+               COE = (((((fGETUBYTE(3,VAL) >> (2 * POS)) & 3) << 8) | 
fGETUBYTE(POS,VAL)) << 6);
+               COE >>= 6;
+       },
+       ()
+)
+
+DEF_MACRO(fVMAX,(X,Y),
+       "max(X,Y)",
+       "",
+       (X>Y) ? X : Y,
+       ()
+)
+
+
+DEF_MACRO(fREAD_VEC,
+    (DST,IDX),
+    "DST=VREG[IDX]",   /* short desc */
+    "Read Vector IDX", /* long desc */
+       (DST = READ_VREG(fMODCIRCU((IDX),5))),
+    ()
+)
+
+DEF_MACRO(fGETNIBBLE,(IDX,SRC),
+    "SRC.s4[IDX]",
+    "Get nibble",
+    ( fSXTN(4,8,(SRC >> (4*IDX)) & 0xF) ),
+    ()
+)
+
+DEF_MACRO(fGETCRUMB,(IDX,SRC),
+    "SRC.s2[IDX]",
+    "Get 2bits",
+    ( fSXTN(2,8,(SRC >> (2*IDX)) & 0x3) ),
+    ()
+)
+
+DEF_MACRO(fGETCRUMB_SYMMETRIC,(IDX,SRC),
+    "SRC.s2[IDX] >= 0 ? (2-SRC.s2[IDX]) : SRC.s2[IDX]",
+    "Get 2bits",
+    ( (fGETCRUMB(IDX,SRC)>=0 ? (2-fGETCRUMB(IDX,SRC)) : fGETCRUMB(IDX,SRC) ) ),
+    ()
+)
+
+#define ZERO_OFFSET_2B +
+
+DEF_MACRO(fWRITE_VEC,
+    (IDX,VAR),
+    "VREG[IDX]=VAR",   /* short desc */
+    "Write Vector IDX", /* long desc */
+       (WRITE_VREG(fMODCIRCU((IDX),5),VAR)),
+    ()
+)
+
+DEF_MACRO(fGENMASKH,(QREG,IDX),
+       "maskh(QREG,IDX)",
+       "generate mask from QREG for halfword IDX",
+       (((fGETQBIT(QREG,(IDX*2+0)) ? 0xFF : 0x0) << 0)
+       |((fGETQBIT(QREG,(IDX*2+1)) ? 0xFF : 0x0) << 8)),
+       ()
+)
+
+DEF_MACRO(fGETMASKW,(VREG,QREG,IDX),
+       "VREG.w[IDX] & fGENMASKW(QREG,IDX)",
+       "Mask word IDX from VREG using QREG",
+       (VREG.w[IDX] & fGENMASKW((QREG),IDX)),
+       ()
+)
+
+DEF_MACRO(fGETMASKH,(VREG,QREG,IDX),
+       "VREG.h[IDX] & fGENMASKH(QREG,IDX)",
+       "Mask word IDX from VREG using QREG",
+       (VREG.h[IDX] & fGENMASKH((QREG),IDX)),
+       ()
+)
+
+DEF_MACRO(fCONDMASK8,(QREG,IDX,YESVAL,NOVAL),
+       "QREG.IDX ? YESVAL : NOVAL",
+       "QREG.IDX ? YESVAL : NOVAL",
+       (fGETQBIT(QREG,IDX) ? (YESVAL) : (NOVAL)),
+       ()
+)
+
+DEF_MACRO(fCONDMASK16,(QREG,IDX,YESVAL,NOVAL),
+       "select_bytes(QREG,IDX,YESVAL,NOVAL)",
+       "select_bytes(QREG,IDX,YESVAL,NOVAL)",
+       ((fGENMASKH(QREG,IDX) & (YESVAL)) | (fGENMASKH(fNOTQ(QREG),IDX) & 
(NOVAL))),
+       ()
+)
+
+DEF_MACRO(fCONDMASK32,(QREG,IDX,YESVAL,NOVAL),
+       "select_bytes(QREG,IDX,YESVAL,NOVAL)",
+       "select_bytes(QREG,IDX,YESVAL,NOVAL)",
+       ((fGENMASKW(QREG,IDX) & (YESVAL)) | (fGENMASKW(fNOTQ(QREG),IDX) & 
(NOVAL))),
+       ()
+)
+
+
+DEF_MACRO(fSETQBITS,(REG,WIDTH,MASK,BITNO,VAL),
+       "REG[BITNO+WIDTH-1:BITNO] = VAL",
+       "Put bits into REG",
+       do {
+               size4u_t __TMP = (VAL);
+               REG.w[(BITNO)>>5] &= ~((MASK) << ((BITNO) & 0x1f));
+               REG.w[(BITNO)>>5] |= (((__TMP) & (MASK)) << ((BITNO) & 0x1f));
+       } while (0),
+       ()
+)
+
+DEF_MACRO(fSETQBIT,(REG,BITNO,VAL),
+       "REG[BITNO]=VAL",
+       "Put bit into REG",
+       fSETQBITS(REG,1,1,BITNO,VAL),
+       ()
+)
+
+DEF_MACRO(fVBYTES,(),
+       "VWIDTH",
+       "Number of bytes in a vector",
+       (fVECSIZE()),
+       ()
+)
+
+DEF_MACRO(fVHALVES,(),
+       "VWIDTH/2",
+       "Number of halves in a vector",
+       (fVECSIZE()/2),
+       ()
+)
+
+DEF_MACRO(fVWORDS,(),
+       "VWIDTH/2",
+       "Number of words in a vector",
+       (fVECSIZE()/4),
+       ()
+)
+
+DEF_MACRO(fVDWORDS,(),
+       "VWIDTH/8",
+       "Number of double words in a vector",
+       (fVECSIZE()/8),
+       ()
+)
+
+DEF_MACRO(fVALIGN, (ADDR, LOG2_ALIGNMENT),
+    "ADDR = ADDR & ~(LOG2_ALIGNMENT-1)",
+    "Align to Element Size",
+    ( ADDR = ADDR & ~(LOG2_ALIGNMENT-1)),
+    ()
+)
+
+DEF_MACRO(fVLASTBYTE, (ADDR, LOG2_ALIGNMENT),
+    "ADDR = ADDR | (LOG2_ALIGNMENT-1)",
+    "Set LSB of length to last byte",
+    ( ADDR = ADDR | (LOG2_ALIGNMENT-1)),
+    ()
+)
+
+
+DEF_MACRO(fVELEM, (WIDTH),
+    "VBITS/WIDTH",
+    "Number of WIDTH-bit elements in a vector",
+    ((fVECSIZE()*8)/WIDTH),
+    ()
+)
+
+DEF_MACRO(fVECLOGSIZE,(),
+    "log2(VECTOR_SIZE)",
+    "Log base 2 of the number of bytes in a vector",
+    (mmvec_current_veclogsize(thread)),
+    ()
+)
+
+DEF_MACRO(fVBUF_IDX,(EA),
+       "(EA >> log2(VECTOR_SIZE)) & 0xFF",
+       "(EA >> log2(VECTOR_SIZE)) & 0xFF",
+       (((EA) >> fVECLOGSIZE()) & 0xFF),
+       (A_FAKEINSN)
+)
+
+DEF_MACRO(fREAD_VBUF,(IDX,WIDX),
+       "vbuf[IDX].w[WIDX]",
+       "vbuf[IDX].w[WIDX]",
+       READ_VBUF(IDX,WIDX),
+       (A_FAKEINSN)
+)
+
+DEF_MACRO(fLOG_VBUF,(IDX,VAL,WIDX),
+       "vbuf[IDX].w[WIDX] = VAL",
+       "vbuf[IDX].w[WIDX] = VAL",
+       LOG_VBUF(IDX,VAL,WIDX),
+       (A_FAKEINSN)
+)
+
+DEF_MACRO(fVECSIZE,(),
+    "VBYTES",
+    "Number of bytes in a vector currently",
+    (1<<fVECLOGSIZE()),
+    ()
+)
+
+DEF_MACRO(fSWAPB,(A, B),
+    "SWAP(A,B)",
+    "Swap bytes",
+    {
+               size1u_t tmp = A;
+               A = B;
+               B = tmp;
+       },
+    /* NOTHING */
+)
+
+DEF_MACRO(
+       fVZERO,(),
+       "0",
+       "0",
+       mmvec_zero_vector(),
+       ()
+)
+
+DEF_MACRO(
+    fNEWVREG,(VNUM),
+    "VNUM.new",
+    "Register value produced in this packet",
+    ((THREAD2STRUCT->VRegs_updated & (((VRegMask)1)<<VNUM)) ? 
THREAD2STRUCT->future_VRegs[VNUM] : mmvec_zero_vector()),
+    (A_DOTNEWVALUE,A_RESTRICT_SLOT0ONLY)
+)
+
+DEF_MACRO(
+       fV_AL_CHECK,
+       (EA,MASK),
+       "",
+       "",
+       if ((EA) & (MASK)) {
+               warn("aligning misaligned vector. PC=%08x 
EA=%08x",thread->Regs[REG_PC],(EA));
+       },
+       ()
+)
+DEF_MACRO(fSCATTER_INIT, ( REGION_START, LENGTH, ELEMENT_SIZE),
+    "",
+    "",
+    {
+    mem_vector_scatter_init(thread, insn,   REGION_START, LENGTH, 
ELEMENT_SIZE);
+       if (EXCEPTION_DETECTED) return;
+    },
+    (A_STORE,A_MEMLIKE,A_RESTRICT_SINGLE_MEM_FIRST,A_RESTRICT_SLOT0ONLY)
+)
+
+DEF_MACRO(fGATHER_INIT, ( REGION_START, LENGTH, ELEMENT_SIZE),
+    "",
+    "",
+    {
+    mem_vector_gather_init(thread, insn,   REGION_START, LENGTH, ELEMENT_SIZE);
+       if (EXCEPTION_DETECTED) return;
+    },
+    (A_LOAD,A_MEMLIKE,A_RESTRICT_SINGLE_MEM_FIRST,A_RESTRICT_SLOT1ONLY)
+)
+
+DEF_MACRO(fSCATTER_FINISH, (OP),
+    "",
+    "",
+    {
+       if (EXCEPTION_DETECTED) return;
+    mem_vector_scatter_finish(thread, insn, OP);
+    },
+    ()
+)
+
+DEF_MACRO(fGATHER_FINISH, (),
+    "",
+    "",
+    {
+       if (EXCEPTION_DETECTED) return;
+    mem_vector_gather_finish(thread, insn);
+    },
+    ()
+)
+
+
+DEF_MACRO(CHECK_VTCM_PAGE, (FLAG, BASE, LENGTH, OFFSET, ALIGNMENT),
+    "FLAG=((BASE+OFFSET) < (BASE+LENGTH))",
+    "FLAG=((BASE+OFFSET) < (BASE+LENGTH))",
+     {
+        int slot = insn->slot;
+        paddr_t pa = thread->mem_access[slot].paddr+OFFSET;
+        pa = pa & ~(ALIGNMENT-1);
+        FLAG = (pa < (thread->mem_access[slot].paddr+LENGTH));
+     },
+    ()
+)
+DEF_MACRO(COUNT_OUT_OF_BOUNDS, (FLAG, SIZE),
+    " ",
+    "",
+     {
+        if (!FLAG)
+        {
+               THREAD2STRUCT->vtcm_log.oob_access += SIZE;
+               warn("Scatter/Gather out of bounds of region");
+        }
+     },
+    ()
+)
+
+DEF_MACRO(fLOG_SCATTER_OP, (SIZE),
+    "  ",
+    "  ",
+    {
+        // Log the size and indicate that the extension ext.c file needs to 
increment right before memory write
+        THREAD2STRUCT->vtcm_log.op = 1;
+        THREAD2STRUCT->vtcm_log.op_size = SIZE;
+    },
+    ()
+)
+
+
+
+DEF_MACRO(fVLOG_VTCM_WORD_INCREMENT, (EA,OFFSET,INC,IDX,ALIGNMENT,LEN),
+    "if (RtV <= EA <= RtV + LEN) *EA += INC.uw[IDX] ",
+    "if (RtV <= EA <= RtV + LEN) *EA += INC.uw[IDX] ",
+    {
+        int slot = insn->slot;
+        int log_bank = 0;
+        int log_byte =0;
+        paddr_t pa = thread->mem_access[slot].paddr+(OFFSET & ~(ALIGNMENT-1));
+        paddr_t pa_high = thread->mem_access[slot].paddr+LEN;
+        for(int i0 = 0; i0 < 4; i0++)
+        {
+            log_byte =  ((OFFSET>=0)&&((pa+i0)<=pa_high));
+            log_bank |= (log_byte<<i0);
+            LOG_VTCM_BYTE(pa+i0,log_byte,INC.ub[4*IDX+i0],4*IDX+i0);
+        }
+        { LOG_VTCM_BANK(pa, log_bank, IDX); }
+    },
+    ()
+)
+
+DEF_MACRO(fVLOG_VTCM_HALFWORD_INCREMENT, (EA,OFFSET,INC,IDX,ALIGNMENT,LEN),
+    "if (RtV <= EA <= RtV + LEN) *EA += INC.uh[IDX] ",
+    "if (RtV <= EA <= RtV + LEN) *EA += INC.uh[IDX] ",
+    {
+        int slot = insn->slot;
+        int log_bank = 0;
+        int log_byte = 0;
+        paddr_t pa = thread->mem_access[slot].paddr+(OFFSET & ~(ALIGNMENT-1));
+        paddr_t pa_high = thread->mem_access[slot].paddr+LEN;
+        for(int i0 = 0; i0 < 2; i0++) {
+            log_byte =  ((OFFSET>=0)&&((pa+i0)<=pa_high));
+            log_bank |= (log_byte<<i0);
+            LOG_VTCM_BYTE(pa+i0,log_byte,INC.ub[2*IDX+i0],2*IDX+i0);
+        }
+        { LOG_VTCM_BANK(pa, log_bank,IDX); }
+    },
+    ()
+)
+
+DEF_MACRO(fVLOG_VTCM_HALFWORD_INCREMENT_DV, 
(EA,OFFSET,INC,IDX,IDX2,IDX_H,ALIGNMENT,LEN),
+    "if (RtV <= EA <= RtV + LEN) *EA += IN.uw[IDX2].uh[IDX_H] ",
+    "if (RtV <= EA <= RtV + LEN) *EA += IN.uw[IDX2].uh[IDX_H] ",
+    {
+        int slot = insn->slot;
+        int log_bank = 0;
+        int log_byte = 0;
+        paddr_t pa = thread->mem_access[slot].paddr+(OFFSET & ~(ALIGNMENT-1));
+        paddr_t pa_high = thread->mem_access[slot].paddr+LEN;
+        for(int i0 = 0; i0 < 2; i0++) {
+            log_byte =  ((OFFSET>=0)&&((pa+i0)<=pa_high));
+            log_bank |= (log_byte<<i0);
+            LOG_VTCM_BYTE(pa+i0,log_byte,INC.ub[2*IDX+i0],2*IDX+i0);
+        }
+        { LOG_VTCM_BANK(pa, log_bank,(2*IDX2+IDX_H));}
+    },
+    ()
+)
+
+
+
+DEF_MACRO(GATHER_FUNCTION, (EA,OFFSET,IDX, LEN, ELEMENT_SIZE, BANK_IDX, QVAL),
+"",
+"",
+{
+        int slot = insn->slot;
+        int i0;
+        paddr_t pa = thread->mem_access[slot].paddr+OFFSET;
+        paddr_t pa_high = thread->mem_access[slot].paddr+LEN;
+        int log_bank = 0;
+        int log_byte = 0;
+        for(i0 = 0; i0 < ELEMENT_SIZE; i0++)
+        {
+            log_byte =  ((OFFSET>=0)&&((pa+i0)<=pa_high)) && QVAL;
+            log_bank |= (log_byte<<i0);
+            size1u_t B  = sim_mem_read1(thread->system_ptr, thread->threadId, 
thread->mem_access[slot].paddr+OFFSET+i0);
+            THREAD2STRUCT->tmp_VRegs[0].ub[ELEMENT_SIZE*IDX+i0] = B;
+            LOG_VTCM_BYTE(pa+i0,log_byte,B,ELEMENT_SIZE*IDX+i0);
+        }
+        LOG_VTCM_BANK(pa, log_bank,BANK_IDX);
+},
+()
+)
+
+
+
+DEF_MACRO(fVLOG_VTCM_GATHER_WORD, (EA,OFFSET,IDX, LEN),
+    "if (RtV <= EA <= RtV + LEN) TEMP.uw[IDX] = *EA ",
+    "if (RtV <= EA <= RtV + LEN) TEMP.uw[IDX] = *EA ",
+    {
+               GATHER_FUNCTION(EA,OFFSET,IDX, LEN, 4, IDX, 1);
+    },
+    ()
+)
+DEF_MACRO(fVLOG_VTCM_GATHER_HALFWORD, (EA,OFFSET,IDX, LEN),
+    " if (RtV <= EA <= RtV + LEN) TEMP.uh[IDX] = *EA ",
+    " if (RtV <= EA <= RtV + LEN) TEMP.uh[IDX] = *EA ",
+    {
+               GATHER_FUNCTION(EA,OFFSET,IDX, LEN, 2, IDX, 1);
+    },
+    ()
+)
+DEF_MACRO(fVLOG_VTCM_GATHER_HALFWORD_DV, (EA,OFFSET,IDX,IDX2,IDX_H, LEN),
+    "if (RtV <= EA <= RtV + LEN) TEMP.uw[IDX2].uh[IDX_H] = *EA ",
+    "if (RtV <= EA <= RtV + LEN) TEMP.uw[IDX2].uh[IDX_H] = *EA ",
+    {
+               GATHER_FUNCTION(EA,OFFSET,IDX, LEN, 2, (2*IDX2+IDX_H), 1);
+    },
+    ()
+)
+DEF_MACRO(fVLOG_VTCM_GATHER_WORDQ, (EA,OFFSET,IDX, Q, LEN),
+    " if ( (RtV <= EA <= RtV + LEN) & Q) TEMP.uw[IDX] = *EA ",
+    " if ( (RtV <= EA <= RtV + LEN) & Q) TEMP.uw[IDX] = *EA ",
+    {
+               GATHER_FUNCTION(EA,OFFSET,IDX, LEN, 4, IDX, 
fGETQBIT(QsV,4*IDX+i0));
+    },
+    ()
+)
+DEF_MACRO(fVLOG_VTCM_GATHER_HALFWORDQ, (EA,OFFSET,IDX, Q, LEN),
+    " if ( (RtV <= EA <= RtV + LEN) & Q) TEMP.uh[IDX] = *EA ",
+    " if ( (RtV <= EA <= RtV + LEN) & Q) TEMP.uh[IDX] = *EA ",
+    {
+               GATHER_FUNCTION(EA,OFFSET,IDX, LEN, 2, IDX, 
fGETQBIT(QsV,2*IDX+i0));
+    },
+    ()
+)
+
+DEF_MACRO(fVLOG_VTCM_GATHER_HALFWORDQ_DV, (EA,OFFSET,IDX,IDX2,IDX_H, Q, LEN),
+    " if ( (RtV <= EA <= RtV + LEN) & Q) TEMP.uw[IDX2].uh[IDX_H] = *EA ",
+    " if ( (RtV <= EA <= RtV + LEN) & Q) TEMP.uw[IDX2].uh[IDX_H] = *EA ",
+    {
+               GATHER_FUNCTION(EA,OFFSET,IDX, LEN, 2, (2*IDX2+IDX_H), 
fGETQBIT(QsV,2*IDX+i0));
+    },
+    ()
+)
+
+
+DEF_MACRO(DEBUG_LOG_ADDR, (OFFSET),
+    "  ",
+    "  ",
+    {
+
+        if (thread->processor_ptr->arch_proc_options->mmvec_network_addr_log2)
+        {
+
+            int slot = insn->slot;
+            paddr_t pa = thread->mem_access[slot].paddr+OFFSET;
+        }
+    },
+    ()
+)
+
+
+
+
+
+
+
+DEF_MACRO(SCATTER_OP_WRITE_TO_MEM, (TYPE),
+    " Read, accumulate, and write to VTCM",
+    "  ",
+    {
+        for (int i = 0; i < mmvecx->vtcm_log.size; i+=sizeof(TYPE))
+        {
+            if ( mmvecx->vtcm_log.mask.ub[i] != 0) {
+                TYPE dst = 0;
+                TYPE inc = 0;
+                for(int j = 0; j < sizeof(TYPE); j++) {
+                    dst |= (sim_mem_read1(thread->system_ptr, 
thread->threadId, mmvecx->vtcm_log.pa[i+j]) << (8*j));
+                    inc |= mmvecx->vtcm_log.data.ub[j+i] << (8*j);
+
+                    mmvecx->vtcm_log.mask.ub[j+i] = 0;
+                    mmvecx->vtcm_log.data.ub[j+i] = 0;
+                    mmvecx->vtcm_log.offsets.ub[j+i] = 0;
+                }
+                dst += inc;
+                for(int j = 0; j < sizeof(TYPE); j++) {
+                    sim_mem_write1(thread->system_ptr,thread->threadId, 
mmvecx->vtcm_log.pa[i+j], (dst >> (8*j))& 0xFF );
+                }
+        }
+
+    }
+    },
+    ()
+)
+
+DEF_MACRO(SCATTER_FUNCTION, (EA,OFFSET,IDX, LEN, ELEMENT_SIZE, BANK_IDX, QVAL, 
IN),
+"",
+"",
+{
+        int slot = insn->slot;
+        int i0;
+        paddr_t pa = thread->mem_access[slot].paddr+OFFSET;
+        paddr_t pa_high = thread->mem_access[slot].paddr+LEN;
+        int log_bank = 0;
+        int log_byte = 0;
+        for(i0 = 0; i0 < ELEMENT_SIZE; i0++) {
+            log_byte = ((OFFSET>=0)&&((pa+i0)<=pa_high)) && QVAL;
+            log_bank |= (log_byte<<i0);
+            
LOG_VTCM_BYTE(pa+i0,log_byte,IN.ub[ELEMENT_SIZE*IDX+i0],ELEMENT_SIZE*IDX+i0);
+        }
+        LOG_VTCM_BANK(pa, log_bank,BANK_IDX);
+
+},
+()
+)
+
+DEF_MACRO(fVLOG_VTCM_HALFWORD, (EA,OFFSET,IN,IDX, LEN),
+    "if (RtV <= EA <= RtV + LEN) *EA = IN.uh[IDX] ",
+    "if (RtV <= EA <= RtV + LEN) *EA = IN.uh[IDX] ",
+    {
+               SCATTER_FUNCTION (EA,OFFSET,IDX, LEN, 2, IDX, 1, IN);
+    },
+    ()
+)
+DEF_MACRO(fVLOG_VTCM_WORD, (EA,OFFSET,IN,IDX,LEN),
+    "if (RtV <= EA <= RtV + LEN) *EA = IN.uw[IDX] ",
+    "if (RtV <= EA <= RtV + LEN) *EA = IN.uw[IDX] ",
+    {
+               SCATTER_FUNCTION (EA,OFFSET,IDX, LEN, 4, IDX, 1, IN);
+    },
+    ()
+)
+
+DEF_MACRO(fVLOG_VTCM_HALFWORDQ, (EA,OFFSET,IN,IDX,Q,LEN),
+    " if ( (RtV <= EA <= RtV + LEN) & Q) *EA = IN.uh[IDX] ",
+    " if ( (RtV <= EA <= RtV + LEN) & Q) *EA = IN.uh[IDX] ",
+    {
+               SCATTER_FUNCTION (EA,OFFSET,IDX, LEN, 2, IDX, 
fGETQBIT(QsV,2*IDX+i0), IN);
+    },
+    ()
+)
+DEF_MACRO(fVLOG_VTCM_WORDQ, (EA,OFFSET,IN,IDX,Q,LEN),
+    " if ( (RtV <= EA <= RtV + LEN) & Q) *EA = IN.uw[IDX] ",
+    " if ( (RtV <= EA <= RtV + LEN) & Q) *EA = IN.uw[IDX] ",
+    {
+               SCATTER_FUNCTION (EA,OFFSET,IDX, LEN, 4, IDX, 
fGETQBIT(QsV,4*IDX+i0), IN);
+    },
+    ()
+)
+
+
+
+
+
+DEF_MACRO(fVLOG_VTCM_HALFWORD_DV, (EA,OFFSET,IN,IDX,IDX2,IDX_H, LEN),
+    "if (RtV <= EA <= RtV + LEN) *EA = IN.hw[IDX2].uh[IDX_H] ",
+    "if (RtV <= EA <= RtV + LEN) *EA = IN.hw[IDX2].uh[IDX_H] ",
+    {
+               SCATTER_FUNCTION (EA,OFFSET,IDX, LEN, 2, (2*IDX2+IDX_H), 1, IN);
+    },
+    ()
+)
+
+DEF_MACRO(fVLOG_VTCM_HALFWORDQ_DV, (EA,OFFSET,IN,IDX,Q,IDX2,IDX_H, LEN),
+    " if ( (RtV <= EA <= RtV + LEN) & Q) *EA = IN.hw[IDX2].uh[IDX_H] ",
+    " if ( (RtV <= EA <= RtV + LEN) & Q) *EA = IN.hw[IDX2].uh[IDX_H] ",
+    {
+               SCATTER_FUNCTION (EA,OFFSET,IDX, LEN, 2, (2*IDX2+IDX_H), 
fGETQBIT(QsV,2*IDX+i0), IN);
+    },
+    ()
+)
+
+
+
+
+
+
+DEF_MACRO(fSTORERELEASE, (EA,TYPE),
+       "char* addr = EA&~(ALIGNMENT-1); Zero Byte Store Release (Non-blocking 
Sync)",
+       "Zero Byte Store Release (Sync)",
+    {
+        fV_AL_CHECK(EA,fVECSIZE()-1);
+
+        mem_store_release(thread, insn, fVECSIZE(), EA&~(fVECSIZE()-1), EA, 
TYPE, fUSE_LOOKUP_ADDRESS_BY_REV(thread->processor_ptr));
+    },
+       (A_STORE,A_MEMLIKE,A_RESTRICT_SINGLE_MEM_FIRST)
+)
+
+DEF_MACRO(fVFETCH_AL, (EA),
+    "Prefetch vector into L2 cache at EA",
+    "Prefetch vector into L2 cache at EA",
+    {
+    fV_AL_CHECK(EA,fVECSIZE()-1);
+    mem_fetch_vector(thread, insn, EA&~(fVECSIZE()-1), insn->slot, fVECSIZE());
+    },
+    (A_LOAD,A_MEMLIKE,A_RESTRICT_SINGLE_MEM_FIRST,A_RESTRICT_NOSLOT1_STORE)
+)
+
+
+DEF_MACRO(fLOADMMV_AL, (EA, ALIGNMENT, LEN, DST),
+    "char* addr = EA&~(ALIGNMENT-1); for (i=0; i<LEN; ++i) DST[i] = addr[i]",
+    "Load LEN bytes from memory at EA (forced alignment) to DST.",
+    {
+    fV_AL_CHECK(EA,ALIGNMENT-1);
+       thread->last_pkt->double_access_vec = 0;
+    mem_load_vector_oddva(thread, insn, EA&~(ALIGNMENT-1), EA, insn->slot, 
LEN, &DST.ub[0], LEN, fUSE_LOOKUP_ADDRESS_BY_REV(thread->processor_ptr));
+    },
+    (A_LOAD,A_MEMLIKE,A_RESTRICT_SINGLE_MEM_FIRST,A_RESTRICT_NOSLOT1_STORE)
+)
+
+DEF_MACRO(fLOADMMV, (EA, DST),
+       "DST = *(EA&~(ALIGNMENT-1))",
+       "Load vector from memory at EA (forced alignment) to DST.",
+       fLOADMMV_AL(EA,fVECSIZE(),fVECSIZE(),DST),
+       ()
+)
+
+DEF_MACRO(fLOADMMVQ, (EA,DST,QVAL),
+       "DST = vmux(QVAL,*(EA&~(ALIGNMENT-1)),0)",
+       "Load vector from memory at EA (forced alignment) to DST.",
+       do {
+               int __i;
+               fLOADMMV_AL(EA,fVECSIZE(),fVECSIZE(),DST);
+               fVFOREACH(8,__i) if (!fGETQBIT(QVAL,__i)) DST.b[__i] = 0;
+       } while (0),
+       ()
+)
+
+DEF_MACRO(fLOADMMVNQ, (EA,DST,QVAL),
+       "DST = vmux(QVAL,0,*(EA&~(ALIGNMENT-1)))",
+       "Load vector from memory at EA (forced alignment) to DST.",
+       do {
+               int __i;
+               fLOADMMV_AL(EA,fVECSIZE(),fVECSIZE(),DST);
+               fVFOREACH(8,__i) if (fGETQBIT(QVAL,__i)) DST.b[__i] = 0;
+       } while (0),
+       ()
+)
+
+DEF_MACRO(fLOADMMVU_AL, (EA, ALIGNMENT, LEN, DST),
+    "char* addr = EA; for (i=0; i<LEN; ++i) DST[i] = addr[i]",
+    "Load LEN bytes from memory at EA (unaligned) to DST.",
+    {
+    size4u_t size2 = (EA)&(ALIGNMENT-1);
+    size4u_t size1 = LEN-size2;
+       thread->last_pkt->double_access_vec = 1;
+    mem_load_vector_oddva(thread, insn, EA+size1, EA+fVECSIZE(), /* slot */ 1, 
size2, &DST.ub[size1], size2, fUSE_LOOKUP_ADDRESS());
+    mem_load_vector_oddva(thread, insn, EA, EA,/* slot */ 0, size1, 
&DST.ub[0], size1, fUSE_LOOKUP_ADDRESS_BY_REV(thread->processor_ptr));
+    },
+    (A_LOAD,A_MEMLIKE,A_RESTRICT_SINGLE_MEM_FIRST,A_RESTRICT_NOSLOT1_STORE)
+)
+
+DEF_MACRO(fLOADMMVU, (EA, DST),
+       "DST = *EA",
+       "Load vector from memory at EA (unaligned) to DST.",
+       {
+               /* if address happens to be aligned, only do aligned load */
+        thread->last_pkt->pkt_has_vtcm_access = 0;
+        thread->last_pkt->pkt_access_count = 0;
+               if ( (EA & (fVECSIZE()-1)) == 0) {
+            thread->last_pkt->pkt_has_vmemu_access = 0;
+                       thread->last_pkt->double_access = 0;
+
+                       fLOADMMV_AL(EA,fVECSIZE(),fVECSIZE(),DST);
+               } else {
+            thread->last_pkt->pkt_has_vmemu_access = 1;
+                       thread->last_pkt->double_access = 1;
+
+                       fLOADMMVU_AL(EA,fVECSIZE(),fVECSIZE(),DST);
+               }
+       },
+       ()
+)
+
+DEF_MACRO(fSTOREMMV_AL, (EA, ALIGNMENT, LEN, SRC),
+    "char* addr = EA&~(ALIGNMENT-1); for (i=0; i<LEN; ++i) addr[i] = SRC[i]",
+    "Store LEN bytes from SRC into memory at EA (forced alignment).",
+    {
+    fV_AL_CHECK(EA,ALIGNMENT-1);
+    mem_store_vector_oddva(thread, insn, EA&~(ALIGNMENT-1), EA, insn->slot, 
LEN, &SRC.ub[0], 0, 0, fUSE_LOOKUP_ADDRESS_BY_REV(thread->processor_ptr));
+    },
+    (A_STORE,A_MEMLIKE,A_RESTRICT_SINGLE_MEM_FIRST)
+)
+
+DEF_MACRO(fSTOREMMV, (EA, SRC),
+       "*(EA&~(ALIGNMENT-1)) = SRC",
+       "Store vector SRC to memory at EA (unaligned).",
+       fSTOREMMV_AL(EA,fVECSIZE(),fVECSIZE(),SRC),
+       ()
+)
+
+DEF_MACRO(fSTOREMMVQ_AL, (EA, ALIGNMENT, LEN, SRC, MASK),
+    "char* addr = EA&~(ALIGNMENT-1); for (i=0; i<LEN; ++i) if (MASK[i]) 
addr[i] = SRC[i]",
+    "Store LEN bytes from SRC into memory at EA (forced alignment).",
+    do {
+       mmvector_t maskvec;
+       int i;
+       for (i = 0; i < fVECSIZE(); i++) maskvec.ub[i] = fGETQBIT(MASK,i);
+       mem_store_vector_oddva(thread, insn, EA&~(ALIGNMENT-1), EA, insn->slot, 
LEN, &SRC.ub[0], &maskvec.ub[0], 0, 
fUSE_LOOKUP_ADDRESS_BY_REV(thread->processor_ptr));
+    } while (0),
+    (A_STORE,A_MEMLIKE,A_RESTRICT_SINGLE_MEM_FIRST)
+)
+
+DEF_MACRO(fSTOREMMVQ, (EA, SRC, MASK),
+       "*(EA&~(ALIGNMENT-1)) = SRC",
+       "Masked store vector SRC to memory at EA (forced alignment).",
+       fSTOREMMVQ_AL(EA,fVECSIZE(),fVECSIZE(),SRC,MASK),
+       ()
+)
+
+DEF_MACRO(fSTOREMMVNQ_AL, (EA, ALIGNMENT, LEN, SRC, MASK),
+    "char* addr = EA&~(ALIGNMENT-1); for (i=0; i<LEN; ++i) if (!MASK[i]) 
addr[i] = SRC[i]",
+    "Store LEN bytes from SRC into memory at EA (forced alignment).",
+    {
+       mmvector_t maskvec;
+       int i;
+       for (i = 0; i < fVECSIZE(); i++) maskvec.ub[i] = fGETQBIT(MASK,i);
+        fV_AL_CHECK(EA,ALIGNMENT-1);
+       mem_store_vector_oddva(thread, insn, EA&~(ALIGNMENT-1), EA, insn->slot, 
LEN, &SRC.ub[0], &maskvec.ub[0], 1, 
fUSE_LOOKUP_ADDRESS_BY_REV(thread->processor_ptr));
+    },
+    (A_STORE,A_MEMLIKE,A_RESTRICT_SINGLE_MEM_FIRST)
+)
+
+DEF_MACRO(fSTOREMMVNQ, (EA, SRC, MASK),
+       "*(EA&~(ALIGNMENT-1)) = SRC",
+       "Masked negated store vector SRC to memory at EA (forced alignment).",
+       fSTOREMMVNQ_AL(EA,fVECSIZE(),fVECSIZE(),SRC,MASK),
+       ()
+)
+
+DEF_MACRO(fSTOREMMVU_AL, (EA, ALIGNMENT, LEN, SRC),
+    "char* addr = EA; for (i=0; i<LEN; ++i) addr[i] = SRC[i]",
+    "Store LEN bytes from SRC into memory at EA (unaligned).",
+    {
+    size4u_t size1 = ALIGNMENT-((EA)&(ALIGNMENT-1));
+    size4u_t size2;
+    if (size1>LEN) size1 = LEN;
+    size2 = LEN-size1;
+    mem_store_vector_oddva(thread, insn, EA+size1, EA+fVECSIZE(), /* slot */ 
1, size2, &SRC.ub[size1], 0, 0, fUSE_LOOKUP_ADDRESS());
+    mem_store_vector_oddva(thread, insn, EA, EA, /* slot */ 0, size1, 
&SRC.ub[0], 0, 0, fUSE_LOOKUP_ADDRESS_BY_REV(thread->processor_ptr));
+    },
+    (A_STORE,A_MEMLIKE,A_RESTRICT_SINGLE_MEM_FIRST)
+)
+
+DEF_MACRO(fSTOREMMVU, (EA, SRC),
+       "*EA = SRC",
+       "Store vector SRC to memory at EA (unaligned).",
+       {
+        thread->last_pkt->pkt_has_vtcm_access = 0;
+        thread->last_pkt->pkt_access_count = 0;
+               if ( (EA & (fVECSIZE()-1)) == 0) {
+                       thread->last_pkt->double_access = 0;
+                       fSTOREMMV_AL(EA,fVECSIZE(),fVECSIZE(),SRC);
+               } else {
+                       thread->last_pkt->double_access = 1;
+            thread->last_pkt->pkt_has_vmemu_access = 1;
+                       fSTOREMMVU_AL(EA,fVECSIZE(),fVECSIZE(),SRC);
+               }
+       },
+       ()
+)
+
+DEF_MACRO(fSTOREMMVQU_AL, (EA, ALIGNMENT, LEN, SRC, MASK),
+    "char* addr = EA; for (i=0; i<LEN; ++i) if (MASK[i]) addr[i] = SRC[i]",
+    "Store LEN bytes from SRC into memory at EA (unaligned).",
+    {
+       size4u_t size1 = ALIGNMENT-((EA)&(ALIGNMENT-1));
+       size4u_t size2;
+       mmvector_t maskvec;
+       int i;
+       for (i = 0; i < fVECSIZE(); i++) maskvec.ub[i] = fGETQBIT(MASK,i);
+       if (size1>LEN) size1 = LEN;
+       size2 = LEN-size1;
+       mem_store_vector_oddva(thread, insn, EA+size1, EA+fVECSIZE(),/* slot */ 
1, size2, &SRC.ub[size1], &maskvec.ub[size1], 0, fUSE_LOOKUP_ADDRESS());
+       mem_store_vector_oddva(thread, insn, EA, /* slot */ 0, size1, 
&SRC.ub[0], &maskvec.ub[0], 0, 
fUSE_LOOKUP_ADDRESS_BY_REV(thread->processor_ptr));
+    },
+    (A_STORE,A_MEMLIKE,A_RESTRICT_SINGLE_MEM_FIRST)
+)
+
+DEF_MACRO(fSTOREMMVQU, (EA, SRC, MASK),
+       "*EA = SRC",
+       "Store vector SRC to memory at EA (unaligned).",
+       {
+        thread->last_pkt->pkt_has_vtcm_access = 0;
+        thread->last_pkt->pkt_access_count = 0;
+               if ( (EA & (fVECSIZE()-1)) == 0) {
+                       thread->last_pkt->double_access = 0;
+                       fSTOREMMVQ_AL(EA,fVECSIZE(),fVECSIZE(),SRC,MASK);
+               } else {
+                       thread->last_pkt->double_access = 1;
+            thread->last_pkt->pkt_has_vmemu_access = 1;
+                       fSTOREMMVQU_AL(EA,fVECSIZE(),fVECSIZE(),SRC,MASK);
+               }
+       },
+       ()
+)
+
+DEF_MACRO(fSTOREMMVNQU_AL, (EA, ALIGNMENT, LEN, SRC, MASK),
+    "char* addr = EA; for (i=0; i<LEN; ++i) if (!MASK[i]) addr[i] = SRC[i]",
+    "Store LEN bytes from SRC into memory at EA (unaligned).",
+    {
+       size4u_t size1 = ALIGNMENT-((EA)&(ALIGNMENT-1));
+       size4u_t size2;
+       mmvector_t maskvec;
+       int i;
+       for (i = 0; i < fVECSIZE(); i++) maskvec.ub[i] = fGETQBIT(MASK,i);
+       if (size1>LEN) size1 = LEN;
+       size2 = LEN-size1;
+       mem_store_vector_oddva(thread, insn, EA+size1, EA+fVECSIZE(), /* slot 
*/ 1, size2, &SRC.ub[size1], &maskvec.ub[size1], 1, fUSE_LOOKUP_ADDRESS());
+       mem_store_vector_oddva(thread, insn, EA, EA, /* slot */ 0, size1, 
&SRC.ub[0], &maskvec.ub[0], 1, 
fUSE_LOOKUP_ADDRESS_BY_REV(thread->processor_ptr));
+    },
+    (A_STORE,A_MEMLIKE,A_RESTRICT_SINGLE_MEM_FIRST)
+)
+
+DEF_MACRO(fSTOREMMVNQU, (EA, SRC, MASK),
+       "*EA = SRC",
+       "Store vector SRC to memory at EA (unaligned).",
+       {
+        thread->last_pkt->pkt_has_vtcm_access = 0;
+        thread->last_pkt->pkt_access_count = 0;
+               if ( (EA & (fVECSIZE()-1)) == 0) {
+                       thread->last_pkt->double_access = 0;
+                       fSTOREMMVNQ_AL(EA,fVECSIZE(),fVECSIZE(),SRC,MASK);
+               } else {
+                       thread->last_pkt->double_access = 1;
+            thread->last_pkt->pkt_has_vmemu_access = 1;
+                       fSTOREMMVNQU_AL(EA,fVECSIZE(),fVECSIZE(),SRC,MASK);
+               }
+       },
+       ()
+)
+
+
+
+
+DEF_MACRO(fVFOREACH,(WIDTH, VAR),
+    "for (VAR = 0; VAR < VELEM(WIDTH); VAR++)",
+    "For VAR in each WIDTH-bit vector index",
+    for (VAR = 0; VAR < fVELEM(WIDTH); VAR++),
+    /* NOTHING */
+)
+
+DEF_MACRO(fVARRAY_ELEMENT_ACCESS, (ARRAY, TYPE, INDEX),
+    "ARRAY.TYPE[INDEX]",
+    "Access element of type TYPE at position INDEX of flattened ARRAY",
+    ARRAY.v[(INDEX) / (fVECSIZE()/(sizeof(ARRAY.TYPE[0])))].TYPE[(INDEX) % 
(fVECSIZE()/(sizeof(ARRAY.TYPE[0])))],
+    ()
+)
+
+DEF_MACRO(fVNEWCANCEL,(REGNUM),
+       "Ignore current value for register REGNUM",
+       "Ignore current value for register REGNUM",
+       do { THREAD2STRUCT->VRegs_select &= ~(1<<(REGNUM)); } while (0),
+       ()
+)
+
+DEF_MACRO(fTMPVDATA,(),
+       "Data from .tmp load",
+       "Data from .tmp load and clear tmp status",
+       mmvec_vtmp_data(thread),
+       (A_CVI,A_CVI_REQUIRES_TMPLOAD)
+)
+
+DEF_MACRO(fVSATDW, (U,V),
+    "usat_32(U:V)",
+    "Use 32-bits of U as MSW and 32-bits of V as LSW and saturate the 
resultant 64-bits to 32 bits",
+    fVSATW( ( ( ((long long)U)<<32 ) | fZXTN(32,64,V) ) ),
+    /* attribs */
+)
+
+DEF_MACRO(fVASL_SATHI, (U,V),
+    "uasl_sathi(U:V)",
+    "Use 32-bits of U as MSW and 32-bits of V as LSW, left shift by 1 and 
saturate the result and take high word",
+    fVSATW(((U)<<1) | ((V)>>31)),
+    /* attribs */
+)
+
+DEF_MACRO(fVUADDSAT,(WIDTH,U,V),
+       "usat_##WIDTH(U+V)",
+       "Add WIDTH-bit values U and V with saturation",
+       fVSATUN( WIDTH, fZXTN(WIDTH, 2*WIDTH, U)  + fZXTN(WIDTH, 2*WIDTH, V)),
+       /* attribs */
+)
+
+DEF_MACRO(fVSADDSAT,(WIDTH,U,V),
+       "sat_##WIDTH(U+V)",
+       "Add WIDTH-bit values U and V with saturation",
+       fVSATN(  WIDTH, fSXTN(WIDTH, 2*WIDTH, U)  + fSXTN(WIDTH, 2*WIDTH, V)),
+       /* attribs */
+)
+
+DEF_MACRO(fVUSUBSAT,(WIDTH,U,V),
+       "usat_##WIDTH(U-V)",
+       "sub WIDTH-bit values U and V with saturation",
+       fVSATUN( WIDTH, fZXTN(WIDTH, 2*WIDTH, U)  - fZXTN(WIDTH, 2*WIDTH, V)),
+       /* attribs */
+)
+
+DEF_MACRO(fVSSUBSAT,(WIDTH,U,V),
+       "sat_##WIDTH(U-V)",
+       "sub WIDTH-bit values U and V with saturation",
+       fVSATN(  WIDTH, fSXTN(WIDTH, 2*WIDTH, U)  - fSXTN(WIDTH, 2*WIDTH, V)),
+       /* attribs */
+)
+
+DEF_MACRO(fVAVGU,(WIDTH,U,V),
+       "(U+V)/2",
+       "average WIDTH-bit values U and V with saturation",
+       ((fZXTN(WIDTH, 2*WIDTH, U) + fZXTN(WIDTH, 2*WIDTH, V))>>1),
+       /* attribs */
+)
+
+DEF_MACRO(fVAVGURND,(WIDTH,U,V),
+       "(U+V+1)/2",
+       "average WIDTH-bit values U and V with saturation",
+       ((fZXTN(WIDTH, 2*WIDTH, U) + fZXTN(WIDTH, 2*WIDTH, V)+1)>>1),
+       /* attribs */
+)
+
+DEF_MACRO(fVNAVGU,(WIDTH,U,V),
+       "(U-V)/2",
+       "average WIDTH-bit values U and V with saturation",
+       ((fZXTN(WIDTH, 2*WIDTH, U) - fZXTN(WIDTH, 2*WIDTH, V))>>1),
+       /* attribs */
+)
+
+DEF_MACRO(fVNAVGURNDSAT,(WIDTH,U,V),
+       "(U-V+1)/2",
+       "average WIDTH-bit values U and V with saturation",
+       fVSATUN(WIDTH,((fZXTN(WIDTH, 2*WIDTH, U) - fZXTN(WIDTH, 2*WIDTH, 
V)+1)>>1)),
+       /* attribs */
+)
+
+DEF_MACRO(fVAVGS,(WIDTH,U,V),
+       "(U+V)/2",
+       "average WIDTH-bit values U and V with saturation",
+       ((fSXTN(WIDTH, 2*WIDTH, U) + fSXTN(WIDTH, 2*WIDTH, V))>>1),
+       /* attribs */
+)
+
+DEF_MACRO(fVAVGSRND,(WIDTH,U,V),
+       "(U+V+1)/2",
+       "average WIDTH-bit values U and V with saturation",
+       ((fSXTN(WIDTH, 2*WIDTH, U) + fSXTN(WIDTH, 2*WIDTH, V)+1)>>1),
+       /* attribs */
+)
+
+DEF_MACRO(fVNAVGS,(WIDTH,U,V),
+       "(U-V)/2",
+       "average WIDTH-bit values U and V with saturation",
+       ((fSXTN(WIDTH, 2*WIDTH, U) - fSXTN(WIDTH, 2*WIDTH, V))>>1),
+       /* attribs */
+)
+
+DEF_MACRO(fVNAVGSRND,(WIDTH,U,V),
+       "(U-V+1)/2",
+       "average WIDTH-bit values U and negative V followed by rounding",
+       ((fSXTN(WIDTH, 2*WIDTH, U) - fSXTN(WIDTH, 2*WIDTH, V)+1)>>1),
+       /* attribs */
+)
+
+DEF_MACRO(fVNAVGSRNDSAT,(WIDTH,U,V),
+       "(U-V+1)/2",
+       "average WIDTH-bit values U and V with saturation",
+       fVSATN(WIDTH,((fSXTN(WIDTH, 2*WIDTH, U) - fSXTN(WIDTH, 2*WIDTH, 
V)+1)>>1)),
+       /* attribs */
+)
+
+
+DEF_MACRO(fVNOROUND,(VAL,SHAMT),
+       "VAL",
+       "VAL",
+       VAL,
+       /* NOTHING */
+)
+DEF_MACRO(fVNOSAT,(VAL),
+       "VAL",
+       "VAL",
+       VAL,
+       /* NOTHING */
+)
+
+DEF_MACRO(fVROUND,(VAL,SHAMT),
+       "VAL + (1<<(SHAMT-1))",
+       "VAL + RNDBIT",
+       ((VAL) + (((SHAMT)>0)?(1LL<<((SHAMT)-1)):0)),
+       /* NOTHING */
+)
+
+DEF_MACRO(fCARRY_FROM_ADD32,(A,B,C),
+       "carry_from(A,B,C)",
+       "carry_from(A,B,C)",
+       (((fZXTN(32,64,A)+fZXTN(32,64,B)+C) >> 32) & 1),
+       /* NOTHING */
+)
+
+DEF_MACRO(fUARCH_NOTE_PUMP_4X,(),
+       "",
+       "",
+       ,
+       (A_CVI_PUMP_4X)
+)
+
+DEF_MACRO(fUARCH_NOTE_PUMP_2X,(),
+       "",
+       "",
+       ,
+       (A_CVI_PUMP_2X)
+)
+
+DEF_MACRO(fVDOCHKPAGECROSS,(BASE,SUM),
+       "",
+       "",
+       if (UNLIKELY(thread->timing_on)) {
+               thread->mem_access[slot].check_page_crosses = 1;
+               thread->mem_access[slot].page_cross_base = BASE;
+               thread->mem_access[slot].page_cross_sum = SUM;
+       },
+       (A_EA_PAGECROSS)
+)
+
+
+
-- 
2.7.4