mirror/ghidra
1
0
Fork 0
mirror of https://github.com/NationalSecurityAgency/ghidra.git synced 2026-07-30 07:18:37 -09:00
Code Issues Packages Projects Releases Wiki Activity

AARCH64: zero-extend dest on ADDV/SMAXV/SMINV/UMAXV/UMINV

Browse Source 
Across-lane reductions wrote only the result element and left the upper
   bits of the destination SIMD&FP register stale, violating the ARM
   zero-extension-on-write semantics. Add zext_z{b,h,s} per element size,
   and fix ADDV.8B/.16B which used zext_zq (16-byte clear) for an 8-bit
   result.
This commit is contained in:
valium007
2026-06-15 02:13:25 +05:30
committed by ghidorahrex
parent 08e8f0b6d8
commit b4c3c95490
1 changed files with 22 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
Ghidra/Processors/AARCH64/data/languages/AARCH64neon.sinc
View File

@@ -734,7 +734,7 @@ is b_3131=0 & q=1 & u=0 & b_2428=0xe & advSIMD3.size=0 & b_1721=0x18 & b_1216=0x
Rd_FPR8 = Rd_FPR8 + Rn_VPR128.16B[104,8];
Rd_FPR8 = Rd_FPR8 + Rn_VPR128.16B[112,8];
Rd_FPR8 = Rd_FPR8 + Rn_VPR128.16B[120,8];
zext_zq(Zd); # zero upper 31 bytes of Zd
zext_zb(Zd); # zero upper bytes of dest (ARM SIMD&FP write semantics)
}
# C7.2.6 ADDV page C7-2027 line 118452 MATCH x0e31b800/mask=xbf3ffc00
@@ -753,7 +753,7 @@ is b_3131=0 & q=0 & u=0 & b_2428=0xe & advSIMD3.size=0 & b_1721=0x18 & b_1216=0x
Rd_FPR8 = Rd_FPR8 + Rn_VPR64.8B[40,8];
Rd_FPR8 = Rd_FPR8 + Rn_VPR64.8B[48,8];
Rd_FPR8 = Rd_FPR8 + Rn_VPR64.8B[56,8];
zext_zq(Zd); # zero upper 31 bytes of Zd
zext_zb(Zd); # zero upper bytes of dest (ARM SIMD&FP write semantics)
}
# C7.2.6 ADDV page C7-2027 line 118452 MATCH x0e31b800/mask=xbf3ffc00
@@ -17093,6 +17093,7 @@ is b_3131=0 & q=1 & u=0 & b_2428=0xe & advSIMD3.size=1 & b_2121=1 & Rm_VPR128.8H
is b_3131=0 & q=1 & u=0 & b_2428=0xe & advSIMD3.size=0 & b_1721=0x18 & b_1216=0xa & b_1011=2 & Rn_VPR128.16B & Rd_FPR8 & Zd
{
Rd_FPR8 = NEON_smaxv(Rn_VPR128.16B, 1:1);
zext_zb(Zd); # zero upper bytes of dest (ARM SIMD&FP write semantics)
}
# C7.2.270 SMAXV page C7-2616 line 152715 MATCH x0e30a800/mask=xbf3ffc00
@@ -17104,6 +17105,7 @@ is b_3131=0 & q=1 & u=0 & b_2428=0xe & advSIMD3.size=0 & b_1721=0x18 & b_1216=0x
is b_3131=0 & q=0 & u=0 & b_2428=0xe & advSIMD3.size=0 & b_1721=0x18 & b_1216=0xa & b_1011=2 & Rn_VPR64.8B & Rd_FPR8 & Zd
{
Rd_FPR8 = NEON_smaxv(Rn_VPR64.8B, 1:1);
zext_zb(Zd); # zero upper bytes of dest (ARM SIMD&FP write semantics)
}
# C7.2.270 SMAXV page C7-2616 line 152715 MATCH x0e30a800/mask=xbf3ffc00
@@ -17115,6 +17117,7 @@ is b_3131=0 & q=0 & u=0 & b_2428=0xe & advSIMD3.size=0 & b_1721=0x18 & b_1216=0x
is b_3131=0 & q=0 & u=0 & b_2428=0xe & advSIMD3.size=1 & b_1721=0x18 & b_1216=0xa & b_1011=2 & Rn_VPR64.4H & Rd_FPR16 & Zd
{
Rd_FPR16 = NEON_smaxv(Rn_VPR64.4H, 2:1);
zext_zh(Zd); # zero upper bytes of dest (ARM SIMD&FP write semantics)
}
# C7.2.270 SMAXV page C7-2616 line 152715 MATCH x0e30a800/mask=xbf3ffc00
@@ -17126,6 +17129,7 @@ is b_3131=0 & q=0 & u=0 & b_2428=0xe & advSIMD3.size=1 & b_1721=0x18 & b_1216=0x
is b_3131=0 & q=1 & u=0 & b_2428=0xe & advSIMD3.size=1 & b_1721=0x18 & b_1216=0xa & b_1011=2 & Rn_VPR128.8H & Rd_FPR16 & Zd
{
Rd_FPR16 = NEON_smaxv(Rn_VPR128.8H, 2:1);
zext_zh(Zd); # zero upper bytes of dest (ARM SIMD&FP write semantics)
}
# C7.2.270 SMAXV page C7-2616 line 152715 MATCH x0e30a800/mask=xbf3ffc00
@@ -17137,6 +17141,7 @@ is b_3131=0 & q=1 & u=0 & b_2428=0xe & advSIMD3.size=1 & b_1721=0x18 & b_1216=0x
is b_3131=0 & q=1 & u=0 & b_2428=0xe & advSIMD3.size=2 & b_1721=0x18 & b_1216=0xa & b_1011=2 & Rn_VPR128.4S & Rd_FPR32 & Zd
{
Rd_FPR32 = NEON_smaxv(Rn_VPR128.4S, 4:1);
zext_zs(Zd); # zero upper bytes of dest (ARM SIMD&FP write semantics)
}
# C7.2.271 SMIN page C7-2618 line 152818 MATCH x0e206c00/mask=xbf20fc00
@@ -17280,6 +17285,7 @@ is b_3131=0 & q=1 & u=0 & b_2428=0xe & advSIMD3.size=1 & b_2121=1 & Rm_VPR128.8H
is b_3131=0 & q=1 & u=0 & b_2428=0xe & advSIMD3.size=0 & b_1721=0x18 & b_1216=0x1a & b_1011=2 & Rn_VPR128.16B & Rd_FPR8 & Zd
{
Rd_FPR8 = NEON_sminv(Rn_VPR128.16B, 1:1);
zext_zb(Zd); # zero upper bytes of dest (ARM SIMD&FP write semantics)
}
# C7.2.273 SMINV page C7-2622 line 153024 MATCH x0e31a800/mask=xbf3ffc00
@@ -17291,6 +17297,7 @@ is b_3131=0 & q=1 & u=0 & b_2428=0xe & advSIMD3.size=0 & b_1721=0x18 & b_1216=0x
is b_3131=0 & q=0 & u=0 & b_2428=0xe & advSIMD3.size=0 & b_1721=0x18 & b_1216=0x1a & b_1011=2 & Rn_VPR64.8B & Rd_FPR8 & Zd
{
Rd_FPR8 = NEON_sminv(Rn_VPR64.8B, 1:1);
zext_zb(Zd); # zero upper bytes of dest (ARM SIMD&FP write semantics)
}
# C7.2.273 SMINV page C7-2622 line 153024 MATCH x0e31a800/mask=xbf3ffc00
@@ -17302,6 +17309,7 @@ is b_3131=0 & q=0 & u=0 & b_2428=0xe & advSIMD3.size=0 & b_1721=0x18 & b_1216=0x
is b_3131=0 & q=0 & u=0 & b_2428=0xe & advSIMD3.size=1 & b_1721=0x18 & b_1216=0x1a & b_1011=2 & Rn_VPR64.4H & Rd_FPR16 & Zd
{
Rd_FPR16 = NEON_sminv(Rn_VPR64.4H, 2:1);
zext_zh(Zd); # zero upper bytes of dest (ARM SIMD&FP write semantics)
}
# C7.2.273 SMINV page C7-2622 line 153024 MATCH x0e31a800/mask=xbf3ffc00
@@ -17313,6 +17321,7 @@ is b_3131=0 & q=0 & u=0 & b_2428=0xe & advSIMD3.size=1 & b_1721=0x18 & b_1216=0x
is b_3131=0 & q=1 & u=0 & b_2428=0xe & advSIMD3.size=1 & b_1721=0x18 & b_1216=0x1a & b_1011=2 & Rn_VPR128.8H & Rd_FPR16 & Zd
{
Rd_FPR16 = NEON_sminv(Rn_VPR128.8H, 2:1);
zext_zh(Zd); # zero upper bytes of dest (ARM SIMD&FP write semantics)
}
# C7.2.273 SMINV page C7-2622 line 153024 MATCH x0e31a800/mask=xbf3ffc00
@@ -17324,6 +17333,7 @@ is b_3131=0 & q=1 & u=0 & b_2428=0xe & advSIMD3.size=1 & b_1721=0x18 & b_1216=0x
is b_3131=0 & q=1 & u=0 & b_2428=0xe & advSIMD3.size=2 & b_1721=0x18 & b_1216=0x1a & b_1011=2 & Rn_VPR128.4S & Rd_FPR32 & Zd
{
Rd_FPR32 = NEON_sminv(Rn_VPR128.4S, 4:1);
zext_zs(Zd); # zero upper bytes of dest (ARM SIMD&FP write semantics)
}
# C7.2.274 SMLAL, SMLAL2 (by element) page C7-2624 line 153127 MATCH x0f002000/mask=xbf00f400
@@ -26673,6 +26683,7 @@ is b_3131=0 & q=1 & u=1 & b_2428=0xe & advSIMD3.size=1 & b_2121=1 & Rm_VPR128.8H
is b_3131=0 & q=1 & u=1 & b_2428=0xe & advSIMD3.size=0 & b_1721=0x18 & b_1216=0xa & b_1011=2 & Rn_VPR128.16B & Rd_FPR8 & Zd
{
Rd_FPR8 = NEON_umaxv(Rn_VPR128.16B, 1:1);
zext_zb(Zd); # zero upper bytes of dest (ARM SIMD&FP write semantics)
}
# C7.2.362 UMAXV page C7-2849 line 166369 MATCH x2e30a800/mask=xbf3ffc00
@@ -26684,6 +26695,7 @@ is b_3131=0 & q=1 & u=1 & b_2428=0xe & advSIMD3.size=0 & b_1721=0x18 & b_1216=0x
is b_3131=0 & q=0 & u=1 & b_2428=0xe & advSIMD3.size=0 & b_1721=0x18 & b_1216=0xa & b_1011=2 & Rn_VPR64.8B & Rd_FPR8 & Zd
{
Rd_FPR8 = NEON_umaxv(Rn_VPR64.8B, 1:1);
zext_zb(Zd); # zero upper bytes of dest (ARM SIMD&FP write semantics)
}
# C7.2.362 UMAXV page C7-2849 line 166369 MATCH x2e30a800/mask=xbf3ffc00
@@ -26695,6 +26707,7 @@ is b_3131=0 & q=0 & u=1 & b_2428=0xe & advSIMD3.size=0 & b_1721=0x18 & b_1216=0x
is b_3131=0 & q=0 & u=1 & b_2428=0xe & advSIMD3.size=1 & b_1721=0x18 & b_1216=0xa & b_1011=2 & Rn_VPR64.4H & Rd_FPR16 & Zd
{
Rd_FPR16 = NEON_umaxv(Rn_VPR64.4H, 2:1);
zext_zh(Zd); # zero upper bytes of dest (ARM SIMD&FP write semantics)
}
# C7.2.362 UMAXV page C7-2849 line 166369 MATCH x2e30a800/mask=xbf3ffc00
@@ -26706,6 +26719,7 @@ is b_3131=0 & q=0 & u=1 & b_2428=0xe & advSIMD3.size=1 & b_1721=0x18 & b_1216=0x
is b_3131=0 & q=1 & u=1 & b_2428=0xe & advSIMD3.size=1 & b_1721=0x18 & b_1216=0xa & b_1011=2 & Rn_VPR128.8H & Rd_FPR16 & Zd
{
Rd_FPR16 = NEON_umaxv(Rn_VPR128.8H, 2:1);
zext_zh(Zd); # zero upper bytes of dest (ARM SIMD&FP write semantics)
}
# C7.2.362 UMAXV page C7-2849 line 166369 MATCH x2e30a800/mask=xbf3ffc00
@@ -26717,6 +26731,7 @@ is b_3131=0 & q=1 & u=1 & b_2428=0xe & advSIMD3.size=1 & b_1721=0x18 & b_1216=0x
is b_3131=0 & q=1 & u=1 & b_2428=0xe & advSIMD3.size=2 & b_1721=0x18 & b_1216=0xa & b_1011=2 & Rn_VPR128.4S & Rd_FPR32 & Zd
{
Rd_FPR32 = NEON_umaxv(Rn_VPR128.4S, 4:1);
zext_zs(Zd); # zero upper bytes of dest (ARM SIMD&FP write semantics)
}
# C7.2.363 UMIN page C7-2851 line 166472 MATCH x2e206c00/mask=xbf20fc00
@@ -26860,6 +26875,7 @@ is b_3131=0 & q=1 & u=1 & b_2428=0xe & advSIMD3.size=1 & b_2121=1 & Rm_VPR128.8H
is b_3131=0 & q=1 & u=1 & b_2428=0xe & advSIMD3.size=0 & b_1721=0x18 & b_1216=0x1a & b_1011=2 & Rn_VPR128.16B & Rd_FPR8 & Zd
{
Rd_FPR8 = NEON_uminv(Rn_VPR128.16B, 1:1);
zext_zb(Zd); # zero upper bytes of dest (ARM SIMD&FP write semantics)
}
# C7.2.365 UMINV page C7-2855 line 166678 MATCH x2e31a800/mask=xbf3ffc00
@@ -26871,6 +26887,7 @@ is b_3131=0 & q=1 & u=1 & b_2428=0xe & advSIMD3.size=0 & b_1721=0x18 & b_1216=0x
is b_3131=0 & q=0 & u=1 & b_2428=0xe & advSIMD3.size=0 & b_1721=0x18 & b_1216=0x1a & b_1011=2 & Rn_VPR64.8B & Rd_FPR8 & Zd
{
Rd_FPR8 = NEON_uminv(Rn_VPR64.8B, 1:1);
zext_zb(Zd); # zero upper bytes of dest (ARM SIMD&FP write semantics)
}
# C7.2.365 UMINV page C7-2855 line 166678 MATCH x2e31a800/mask=xbf3ffc00
@@ -26882,6 +26899,7 @@ is b_3131=0 & q=0 & u=1 & b_2428=0xe & advSIMD3.size=0 & b_1721=0x18 & b_1216=0x
is b_3131=0 & q=0 & u=1 & b_2428=0xe & advSIMD3.size=1 & b_1721=0x18 & b_1216=0x1a & b_1011=2 & Rn_VPR64.4H & Rd_FPR16 & Zd
{
Rd_FPR16 = NEON_uminv(Rn_VPR64.4H, 2:1);
zext_zh(Zd); # zero upper bytes of dest (ARM SIMD&FP write semantics)
}
# C7.2.365 UMINV page C7-2855 line 166678 MATCH x2e31a800/mask=xbf3ffc00
@@ -26893,6 +26911,7 @@ is b_3131=0 & q=0 & u=1 & b_2428=0xe & advSIMD3.size=1 & b_1721=0x18 & b_1216=0x
is b_3131=0 & q=1 & u=1 & b_2428=0xe & advSIMD3.size=1 & b_1721=0x18 & b_1216=0x1a & b_1011=2 & Rn_VPR128.8H & Rd_FPR16 & Zd
{
Rd_FPR16 = NEON_uminv(Rn_VPR128.8H, 2:1);
zext_zh(Zd); # zero upper bytes of dest (ARM SIMD&FP write semantics)
}
# C7.2.365 UMINV page C7-2855 line 166678 MATCH x2e31a800/mask=xbf3ffc00
@@ -26904,6 +26923,7 @@ is b_3131=0 & q=1 & u=1 & b_2428=0xe & advSIMD3.size=1 & b_1721=0x18 & b_1216=0x
is b_3131=0 & q=1 & u=1 & b_2428=0xe & advSIMD3.size=2 & b_1721=0x18 & b_1216=0x1a & b_1011=2 & Rn_VPR128.4S & Rd_FPR32 & Zd
{
Rd_FPR32 = NEON_uminv(Rn_VPR128.4S, 4:1);
zext_zs(Zd); # zero upper bytes of dest (ARM SIMD&FP write semantics)
}
# C7.2.366 UMLAL, UMLAL2 (by element) page C7-2857 line 166781 MATCH x2f002000/mask=xbf00f400