[llvm] Main merge true16 mc clean up 3 (PR #135588)
Brox Chen via llvm-commits
llvm-commits at lists.llvm.org
Tue Apr 15 07:17:24 PDT 2025
https://github.com/broxigarchen updated https://github.com/llvm/llvm-project/pull/135588
>From f1181cf3e6886fcda4e96776dad2408bedf28078 Mon Sep 17 00:00:00 2001
From: guochen2 <guochen2 at amd.com>
Date: Mon, 14 Apr 2025 00:52:36 -0400
Subject: [PATCH] vop1 missed mc test
---
.../MC/AMDGPU/gfx11_asm_vop1_t16_promote.s | 1078 ++++++++++++-----
.../test/MC/AMDGPU/gfx11_asm_vop3_from_vop1.s | 483 ++++----
llvm/test/MC/AMDGPU/gfx12_asm_vop1.s | 84 ++
llvm/test/MC/AMDGPU/gfx12_asm_vop1_dpp16.s | 434 ++++---
llvm/test/MC/AMDGPU/gfx12_asm_vop1_dpp8.s | 126 +-
llvm/test/MC/AMDGPU/gfx12_asm_vop1_t16_err.s | 294 +++++
.../MC/AMDGPU/gfx12_asm_vop1_t16_promote.s | 1078 ++++++++++++-----
.../test/MC/AMDGPU/gfx12_asm_vop3_from_vop1.s | 482 ++++----
.../AMDGPU/gfx12_asm_vop3_from_vop1_dpp16.s | 455 ++++---
.../MC/AMDGPU/gfx12_asm_vop3_from_vop1_dpp8.s | 175 ++-
10 files changed, 3165 insertions(+), 1524 deletions(-)
diff --git a/llvm/test/MC/AMDGPU/gfx11_asm_vop1_t16_promote.s b/llvm/test/MC/AMDGPU/gfx11_asm_vop1_t16_promote.s
index 5cb81c640f413..6facd4dca01bf 100644
--- a/llvm/test/MC/AMDGPU/gfx11_asm_vop1_t16_promote.s
+++ b/llvm/test/MC/AMDGPU/gfx11_asm_vop1_t16_promote.s
@@ -2,71 +2,137 @@
// RUN: llvm-mc -triple=amdgcn -mcpu=gfx1100 -mattr=+wavefrontsize32,+real-true16 -show-encoding %s | FileCheck --check-prefix=GFX11 --implicit-check-not=_e32 %s
// RUN: llvm-mc -triple=amdgcn -mcpu=gfx1100 -mattr=+wavefrontsize64,+real-true16 -show-encoding %s | FileCheck --check-prefix=GFX11 --implicit-check-not=_e32 %s
-v_ceil_f16 v128, 0xfe0b
-// GFX11: v_ceil_f16_e64 v128, 0xfe0b ; encoding: [0x80,0x00,0xdc,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+v_ceil_f16 v128.h, 0xfe0b
+// GFX11: v_ceil_f16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xdc,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
-v_ceil_f16 v255, -1
-// GFX11: v_ceil_f16_e64 v255, -1 ; encoding: [0xff,0x00,0xdc,0xd5,0xc1,0x00,0x00,0x00]
+v_ceil_f16 v128.l, 0xfe0b
+// GFX11: v_ceil_f16_e64 v128.l, 0xfe0b ; encoding: [0x80,0x00,0xdc,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
-v_ceil_f16 v255, 0.5
-// GFX11: v_ceil_f16_e64 v255, 0.5 ; encoding: [0xff,0x00,0xdc,0xd5,0xf0,0x00,0x00,0x00]
+v_ceil_f16 v255.h, -1
+// GFX11: v_ceil_f16_e64 v255.h, -1 op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0xc1,0x00,0x00,0x00]
-v_ceil_f16 v255, exec_hi
-// GFX11: v_ceil_f16_e64 v255, exec_hi ; encoding: [0xff,0x00,0xdc,0xd5,0x7f,0x00,0x00,0x00]
+v_ceil_f16 v255.h, 0.5
+// GFX11: v_ceil_f16_e64 v255.h, 0.5 op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0xf0,0x00,0x00,0x00]
-v_ceil_f16 v255, exec_lo
-// GFX11: v_ceil_f16_e64 v255, exec_lo ; encoding: [0xff,0x00,0xdc,0xd5,0x7e,0x00,0x00,0x00]
+v_ceil_f16 v255.h, exec_hi
+// GFX11: v_ceil_f16_e64 v255.h, exec_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0x7f,0x00,0x00,0x00]
-v_ceil_f16 v255, m0
-// GFX11: v_ceil_f16_e64 v255, m0 ; encoding: [0xff,0x00,0xdc,0xd5,0x7d,0x00,0x00,0x00]
+v_ceil_f16 v255.h, exec_lo
+// GFX11: v_ceil_f16_e64 v255.h, exec_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0x7e,0x00,0x00,0x00]
-v_ceil_f16 v255, null
-// GFX11: v_ceil_f16_e64 v255, null ; encoding: [0xff,0x00,0xdc,0xd5,0x7c,0x00,0x00,0x00]
+v_ceil_f16 v255.h, m0
+// GFX11: v_ceil_f16_e64 v255.h, m0 op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0x7d,0x00,0x00,0x00]
-v_ceil_f16 v255, s1
-// GFX11: v_ceil_f16_e64 v255, s1 ; encoding: [0xff,0x00,0xdc,0xd5,0x01,0x00,0x00,0x00]
+v_ceil_f16 v255.h, null
+// GFX11: v_ceil_f16_e64 v255.h, null op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0x7c,0x00,0x00,0x00]
-v_ceil_f16 v255, s105
-// GFX11: v_ceil_f16_e64 v255, s105 ; encoding: [0xff,0x00,0xdc,0xd5,0x69,0x00,0x00,0x00]
+v_ceil_f16 v255.h, s1
+// GFX11: v_ceil_f16_e64 v255.h, s1 op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0x01,0x00,0x00,0x00]
-v_ceil_f16 v255, src_scc
-// GFX11: v_ceil_f16_e64 v255, src_scc ; encoding: [0xff,0x00,0xdc,0xd5,0xfd,0x00,0x00,0x00]
+v_ceil_f16 v255.h, s105
+// GFX11: v_ceil_f16_e64 v255.h, s105 op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0x69,0x00,0x00,0x00]
-v_ceil_f16 v255, ttmp15
-// GFX11: v_ceil_f16_e64 v255, ttmp15 ; encoding: [0xff,0x00,0xdc,0xd5,0x7b,0x00,0x00,0x00]
+v_ceil_f16 v255.h, src_scc
+// GFX11: v_ceil_f16_e64 v255.h, src_scc op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0xfd,0x00,0x00,0x00]
-v_ceil_f16 v255, v1
-// GFX11: v_ceil_f16_e64 v255, v1 ; encoding: [0xff,0x00,0xdc,0xd5,0x01,0x01,0x00,0x00]
+v_ceil_f16 v255.h, ttmp15
+// GFX11: v_ceil_f16_e64 v255.h, ttmp15 op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0x7b,0x00,0x00,0x00]
-v_ceil_f16 v255, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_ceil_f16_e64_dpp v255, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xdc,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+v_ceil_f16 v255.h, v1.h
+// GFX11: v_ceil_f16_e64 v255.h, v1.h op_sel:[1,1] ; encoding: [0xff,0x48,0xdc,0xd5,0x01,0x01,0x00,0x00]
-v_ceil_f16 v255, v1 quad_perm:[3,2,1,0]
-// GFX11: v_ceil_f16_e64_dpp v255, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+v_ceil_f16 v255.h, v1.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_ceil_f16_e64_dpp v255.h, v1.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xdc,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
-v_ceil_f16 v255, v127
-// GFX11: v_ceil_f16_e64 v255, v127 ; encoding: [0xff,0x00,0xdc,0xd5,0x7f,0x01,0x00,0x00]
+v_ceil_f16 v255.h, v1.h quad_perm:[3,2,1,0]
+// GFX11: v_ceil_f16_e64_dpp v255.h, v1.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
-v_ceil_f16 v255, v127 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_ceil_f16_e64_dpp v255, v127 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xdc,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+v_ceil_f16 v255.h, v127.h
+// GFX11: v_ceil_f16_e64 v255.h, v127.h op_sel:[1,1] ; encoding: [0xff,0x48,0xdc,0xd5,0x7f,0x01,0x00,0x00]
-v_ceil_f16 v255, v127 quad_perm:[3,2,1,0]
-// GFX11: v_ceil_f16_e64_dpp v255, v127 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+v_ceil_f16 v255.h, v127.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_ceil_f16_e64_dpp v255.h, v127.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xdc,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
-v_ceil_f16 v255, vcc_hi
-// GFX11: v_ceil_f16_e64 v255, vcc_hi ; encoding: [0xff,0x00,0xdc,0xd5,0x6b,0x00,0x00,0x00]
+v_ceil_f16 v255.h, v127.h quad_perm:[3,2,1,0]
+// GFX11: v_ceil_f16_e64_dpp v255.h, v127.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
-v_ceil_f16 v255, vcc_lo
-// GFX11: v_ceil_f16_e64 v255, vcc_lo ; encoding: [0xff,0x00,0xdc,0xd5,0x6a,0x00,0x00,0x00]
+v_ceil_f16 v255.h, vcc_hi
+// GFX11: v_ceil_f16_e64 v255.h, vcc_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0x6b,0x00,0x00,0x00]
-v_ceil_f16 v5, v199
-// GFX11: v_ceil_f16_e64 v5, v199 ; encoding: [0x05,0x00,0xdc,0xd5,0xc7,0x01,0x00,0x00]
+v_ceil_f16 v255.h, vcc_lo
+// GFX11: v_ceil_f16_e64 v255.h, vcc_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0x6a,0x00,0x00,0x00]
-v_ceil_f16 v5, v199 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_ceil_f16_e64_dpp v5, v199 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xdc,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+v_ceil_f16 v255.l, -1
+// GFX11: v_ceil_f16_e64 v255.l, -1 ; encoding: [0xff,0x00,0xdc,0xd5,0xc1,0x00,0x00,0x00]
-v_ceil_f16 v5, v199 quad_perm:[3,2,1,0]
-// GFX11: v_ceil_f16_e64_dpp v5, v199 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+v_ceil_f16 v255.l, 0.5
+// GFX11: v_ceil_f16_e64 v255.l, 0.5 ; encoding: [0xff,0x00,0xdc,0xd5,0xf0,0x00,0x00,0x00]
+
+v_ceil_f16 v255.l, exec_hi
+// GFX11: v_ceil_f16_e64 v255.l, exec_hi ; encoding: [0xff,0x00,0xdc,0xd5,0x7f,0x00,0x00,0x00]
+
+v_ceil_f16 v255.l, exec_lo
+// GFX11: v_ceil_f16_e64 v255.l, exec_lo ; encoding: [0xff,0x00,0xdc,0xd5,0x7e,0x00,0x00,0x00]
+
+v_ceil_f16 v255.l, m0
+// GFX11: v_ceil_f16_e64 v255.l, m0 ; encoding: [0xff,0x00,0xdc,0xd5,0x7d,0x00,0x00,0x00]
+
+v_ceil_f16 v255.l, null
+// GFX11: v_ceil_f16_e64 v255.l, null ; encoding: [0xff,0x00,0xdc,0xd5,0x7c,0x00,0x00,0x00]
+
+v_ceil_f16 v255.l, s1
+// GFX11: v_ceil_f16_e64 v255.l, s1 ; encoding: [0xff,0x00,0xdc,0xd5,0x01,0x00,0x00,0x00]
+
+v_ceil_f16 v255.l, s105
+// GFX11: v_ceil_f16_e64 v255.l, s105 ; encoding: [0xff,0x00,0xdc,0xd5,0x69,0x00,0x00,0x00]
+
+v_ceil_f16 v255.l, src_scc
+// GFX11: v_ceil_f16_e64 v255.l, src_scc ; encoding: [0xff,0x00,0xdc,0xd5,0xfd,0x00,0x00,0x00]
+
+v_ceil_f16 v255.l, ttmp15
+// GFX11: v_ceil_f16_e64 v255.l, ttmp15 ; encoding: [0xff,0x00,0xdc,0xd5,0x7b,0x00,0x00,0x00]
+
+v_ceil_f16 v255.l, v1.l
+// GFX11: v_ceil_f16_e64 v255.l, v1.l ; encoding: [0xff,0x00,0xdc,0xd5,0x01,0x01,0x00,0x00]
+
+v_ceil_f16 v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_ceil_f16_e64_dpp v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xdc,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+
+v_ceil_f16 v255.l, v1.l quad_perm:[3,2,1,0]
+// GFX11: v_ceil_f16_e64_dpp v255.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+
+v_ceil_f16 v255.l, v127.l
+// GFX11: v_ceil_f16_e64 v255.l, v127.l ; encoding: [0xff,0x00,0xdc,0xd5,0x7f,0x01,0x00,0x00]
+
+v_ceil_f16 v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_ceil_f16_e64_dpp v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xdc,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+
+v_ceil_f16 v255.l, v127.l quad_perm:[3,2,1,0]
+// GFX11: v_ceil_f16_e64_dpp v255.l, v127.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+
+v_ceil_f16 v255.l, vcc_hi
+// GFX11: v_ceil_f16_e64 v255.l, vcc_hi ; encoding: [0xff,0x00,0xdc,0xd5,0x6b,0x00,0x00,0x00]
+
+v_ceil_f16 v255.l, vcc_lo
+// GFX11: v_ceil_f16_e64 v255.l, vcc_lo ; encoding: [0xff,0x00,0xdc,0xd5,0x6a,0x00,0x00,0x00]
+
+v_ceil_f16 v5.h, v199.h
+// GFX11: v_ceil_f16_e64 v5.h, v199.h op_sel:[1,1] ; encoding: [0x05,0x48,0xdc,0xd5,0xc7,0x01,0x00,0x00]
+
+v_ceil_f16 v5.h, v199.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_ceil_f16_e64_dpp v5.h, v199.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xdc,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+
+v_ceil_f16 v5.h, v199.h quad_perm:[3,2,1,0]
+// GFX11: v_ceil_f16_e64_dpp v5.h, v199.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x48,0xdc,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+
+v_ceil_f16 v5.l, v199.l
+// GFX11: v_ceil_f16_e64 v5.l, v199.l ; encoding: [0x05,0x00,0xdc,0xd5,0xc7,0x01,0x00,0x00]
+
+v_ceil_f16 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_ceil_f16_e64_dpp v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xdc,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+
+v_ceil_f16 v5.l, v199.l quad_perm:[3,2,1,0]
+// GFX11: v_ceil_f16_e64_dpp v5.l, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
v_cos_f16 v128.h, 0xfe0b
// GFX11: v_cos_f16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xe1,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
@@ -1160,137 +1226,269 @@ v_cvt_u32_u16 v5, v199.l dpp8:[7,6,5,4,3,2,1,0]
v_cvt_u32_u16 v5, v199.l quad_perm:[3,2,1,0]
// GFX11: v_cvt_u32_u16_e64_dpp v5, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xeb,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
-v_exp_f16 v128, 0xfe0b
-// GFX11: v_exp_f16_e64 v128, 0xfe0b ; encoding: [0x80,0x00,0xd8,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+v_exp_f16 v128.h, 0xfe0b
+// GFX11: v_exp_f16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xd8,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+
+v_exp_f16 v128.l, 0xfe0b
+// GFX11: v_exp_f16_e64 v128.l, 0xfe0b ; encoding: [0x80,0x00,0xd8,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+
+v_exp_f16 v255.h, -1
+// GFX11: v_exp_f16_e64 v255.h, -1 op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0xc1,0x00,0x00,0x00]
+
+v_exp_f16 v255.h, 0.5
+// GFX11: v_exp_f16_e64 v255.h, 0.5 op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0xf0,0x00,0x00,0x00]
+
+v_exp_f16 v255.h, exec_hi
+// GFX11: v_exp_f16_e64 v255.h, exec_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0x7f,0x00,0x00,0x00]
+
+v_exp_f16 v255.h, exec_lo
+// GFX11: v_exp_f16_e64 v255.h, exec_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0x7e,0x00,0x00,0x00]
+
+v_exp_f16 v255.h, m0
+// GFX11: v_exp_f16_e64 v255.h, m0 op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0x7d,0x00,0x00,0x00]
+
+v_exp_f16 v255.h, null
+// GFX11: v_exp_f16_e64 v255.h, null op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0x7c,0x00,0x00,0x00]
+
+v_exp_f16 v255.h, s1
+// GFX11: v_exp_f16_e64 v255.h, s1 op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0x01,0x00,0x00,0x00]
+
+v_exp_f16 v255.h, s105
+// GFX11: v_exp_f16_e64 v255.h, s105 op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0x69,0x00,0x00,0x00]
+
+v_exp_f16 v255.h, src_scc
+// GFX11: v_exp_f16_e64 v255.h, src_scc op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0xfd,0x00,0x00,0x00]
+
+v_exp_f16 v255.h, ttmp15
+// GFX11: v_exp_f16_e64 v255.h, ttmp15 op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0x7b,0x00,0x00,0x00]
+
+v_exp_f16 v255.h, v1.h
+// GFX11: v_exp_f16_e64 v255.h, v1.h op_sel:[1,1] ; encoding: [0xff,0x48,0xd8,0xd5,0x01,0x01,0x00,0x00]
+
+v_exp_f16 v255.h, v1.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_exp_f16_e64_dpp v255.h, v1.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xd8,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+
+v_exp_f16 v255.h, v1.h quad_perm:[3,2,1,0]
+// GFX11: v_exp_f16_e64_dpp v255.h, v1.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+
+v_exp_f16 v255.h, v127.h
+// GFX11: v_exp_f16_e64 v255.h, v127.h op_sel:[1,1] ; encoding: [0xff,0x48,0xd8,0xd5,0x7f,0x01,0x00,0x00]
+
+v_exp_f16 v255.h, v127.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_exp_f16_e64_dpp v255.h, v127.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xd8,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+
+v_exp_f16 v255.h, v127.h quad_perm:[3,2,1,0]
+// GFX11: v_exp_f16_e64_dpp v255.h, v127.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
-v_exp_f16 v255, -1
-// GFX11: v_exp_f16_e64 v255, -1 ; encoding: [0xff,0x00,0xd8,0xd5,0xc1,0x00,0x00,0x00]
+v_exp_f16 v255.h, vcc_hi
+// GFX11: v_exp_f16_e64 v255.h, vcc_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0x6b,0x00,0x00,0x00]
-v_exp_f16 v255, 0.5
-// GFX11: v_exp_f16_e64 v255, 0.5 ; encoding: [0xff,0x00,0xd8,0xd5,0xf0,0x00,0x00,0x00]
+v_exp_f16 v255.h, vcc_lo
+// GFX11: v_exp_f16_e64 v255.h, vcc_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0x6a,0x00,0x00,0x00]
-v_exp_f16 v255, exec_hi
-// GFX11: v_exp_f16_e64 v255, exec_hi ; encoding: [0xff,0x00,0xd8,0xd5,0x7f,0x00,0x00,0x00]
+v_exp_f16 v255.l, -1
+// GFX11: v_exp_f16_e64 v255.l, -1 ; encoding: [0xff,0x00,0xd8,0xd5,0xc1,0x00,0x00,0x00]
-v_exp_f16 v255, exec_lo
-// GFX11: v_exp_f16_e64 v255, exec_lo ; encoding: [0xff,0x00,0xd8,0xd5,0x7e,0x00,0x00,0x00]
+v_exp_f16 v255.l, 0.5
+// GFX11: v_exp_f16_e64 v255.l, 0.5 ; encoding: [0xff,0x00,0xd8,0xd5,0xf0,0x00,0x00,0x00]
-v_exp_f16 v255, m0
-// GFX11: v_exp_f16_e64 v255, m0 ; encoding: [0xff,0x00,0xd8,0xd5,0x7d,0x00,0x00,0x00]
+v_exp_f16 v255.l, exec_hi
+// GFX11: v_exp_f16_e64 v255.l, exec_hi ; encoding: [0xff,0x00,0xd8,0xd5,0x7f,0x00,0x00,0x00]
-v_exp_f16 v255, null
-// GFX11: v_exp_f16_e64 v255, null ; encoding: [0xff,0x00,0xd8,0xd5,0x7c,0x00,0x00,0x00]
+v_exp_f16 v255.l, exec_lo
+// GFX11: v_exp_f16_e64 v255.l, exec_lo ; encoding: [0xff,0x00,0xd8,0xd5,0x7e,0x00,0x00,0x00]
-v_exp_f16 v255, s1
-// GFX11: v_exp_f16_e64 v255, s1 ; encoding: [0xff,0x00,0xd8,0xd5,0x01,0x00,0x00,0x00]
+v_exp_f16 v255.l, m0
+// GFX11: v_exp_f16_e64 v255.l, m0 ; encoding: [0xff,0x00,0xd8,0xd5,0x7d,0x00,0x00,0x00]
-v_exp_f16 v255, s105
-// GFX11: v_exp_f16_e64 v255, s105 ; encoding: [0xff,0x00,0xd8,0xd5,0x69,0x00,0x00,0x00]
+v_exp_f16 v255.l, null
+// GFX11: v_exp_f16_e64 v255.l, null ; encoding: [0xff,0x00,0xd8,0xd5,0x7c,0x00,0x00,0x00]
-v_exp_f16 v255, src_scc
-// GFX11: v_exp_f16_e64 v255, src_scc ; encoding: [0xff,0x00,0xd8,0xd5,0xfd,0x00,0x00,0x00]
+v_exp_f16 v255.l, s1
+// GFX11: v_exp_f16_e64 v255.l, s1 ; encoding: [0xff,0x00,0xd8,0xd5,0x01,0x00,0x00,0x00]
-v_exp_f16 v255, ttmp15
-// GFX11: v_exp_f16_e64 v255, ttmp15 ; encoding: [0xff,0x00,0xd8,0xd5,0x7b,0x00,0x00,0x00]
+v_exp_f16 v255.l, s105
+// GFX11: v_exp_f16_e64 v255.l, s105 ; encoding: [0xff,0x00,0xd8,0xd5,0x69,0x00,0x00,0x00]
-v_exp_f16 v255, v1
-// GFX11: v_exp_f16_e64 v255, v1 ; encoding: [0xff,0x00,0xd8,0xd5,0x01,0x01,0x00,0x00]
+v_exp_f16 v255.l, src_scc
+// GFX11: v_exp_f16_e64 v255.l, src_scc ; encoding: [0xff,0x00,0xd8,0xd5,0xfd,0x00,0x00,0x00]
-v_exp_f16 v255, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_exp_f16_e64_dpp v255, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd8,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+v_exp_f16 v255.l, ttmp15
+// GFX11: v_exp_f16_e64 v255.l, ttmp15 ; encoding: [0xff,0x00,0xd8,0xd5,0x7b,0x00,0x00,0x00]
-v_exp_f16 v255, v1 quad_perm:[3,2,1,0]
-// GFX11: v_exp_f16_e64_dpp v255, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+v_exp_f16 v255.l, v1.l
+// GFX11: v_exp_f16_e64 v255.l, v1.l ; encoding: [0xff,0x00,0xd8,0xd5,0x01,0x01,0x00,0x00]
-v_exp_f16 v255, v127
-// GFX11: v_exp_f16_e64 v255, v127 ; encoding: [0xff,0x00,0xd8,0xd5,0x7f,0x01,0x00,0x00]
+v_exp_f16 v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_exp_f16_e64_dpp v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd8,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
-v_exp_f16 v255, v127 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_exp_f16_e64_dpp v255, v127 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd8,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+v_exp_f16 v255.l, v1.l quad_perm:[3,2,1,0]
+// GFX11: v_exp_f16_e64_dpp v255.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
-v_exp_f16 v255, v127 quad_perm:[3,2,1,0]
-// GFX11: v_exp_f16_e64_dpp v255, v127 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+v_exp_f16 v255.l, v127.l
+// GFX11: v_exp_f16_e64 v255.l, v127.l ; encoding: [0xff,0x00,0xd8,0xd5,0x7f,0x01,0x00,0x00]
-v_exp_f16 v255, vcc_hi
-// GFX11: v_exp_f16_e64 v255, vcc_hi ; encoding: [0xff,0x00,0xd8,0xd5,0x6b,0x00,0x00,0x00]
+v_exp_f16 v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_exp_f16_e64_dpp v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd8,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
-v_exp_f16 v255, vcc_lo
-// GFX11: v_exp_f16_e64 v255, vcc_lo ; encoding: [0xff,0x00,0xd8,0xd5,0x6a,0x00,0x00,0x00]
+v_exp_f16 v255.l, v127.l quad_perm:[3,2,1,0]
+// GFX11: v_exp_f16_e64_dpp v255.l, v127.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
-v_exp_f16 v5, v199
-// GFX11: v_exp_f16_e64 v5, v199 ; encoding: [0x05,0x00,0xd8,0xd5,0xc7,0x01,0x00,0x00]
+v_exp_f16 v255.l, vcc_hi
+// GFX11: v_exp_f16_e64 v255.l, vcc_hi ; encoding: [0xff,0x00,0xd8,0xd5,0x6b,0x00,0x00,0x00]
-v_exp_f16 v5, v199 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_exp_f16_e64_dpp v5, v199 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd8,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+v_exp_f16 v255.l, vcc_lo
+// GFX11: v_exp_f16_e64 v255.l, vcc_lo ; encoding: [0xff,0x00,0xd8,0xd5,0x6a,0x00,0x00,0x00]
-v_exp_f16 v5, v199 quad_perm:[3,2,1,0]
-// GFX11: v_exp_f16_e64_dpp v5, v199 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+v_exp_f16 v5.h, v199.h
+// GFX11: v_exp_f16_e64 v5.h, v199.h op_sel:[1,1] ; encoding: [0x05,0x48,0xd8,0xd5,0xc7,0x01,0x00,0x00]
-v_floor_f16 v128, 0xfe0b
-// GFX11: v_floor_f16_e64 v128, 0xfe0b ; encoding: [0x80,0x00,0xdb,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+v_exp_f16 v5.h, v199.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_exp_f16_e64_dpp v5.h, v199.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xd8,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
-v_floor_f16 v255, -1
-// GFX11: v_floor_f16_e64 v255, -1 ; encoding: [0xff,0x00,0xdb,0xd5,0xc1,0x00,0x00,0x00]
+v_exp_f16 v5.h, v199.h quad_perm:[3,2,1,0]
+// GFX11: v_exp_f16_e64_dpp v5.h, v199.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x48,0xd8,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
-v_floor_f16 v255, 0.5
-// GFX11: v_floor_f16_e64 v255, 0.5 ; encoding: [0xff,0x00,0xdb,0xd5,0xf0,0x00,0x00,0x00]
+v_exp_f16 v5.l, v199.l
+// GFX11: v_exp_f16_e64 v5.l, v199.l ; encoding: [0x05,0x00,0xd8,0xd5,0xc7,0x01,0x00,0x00]
-v_floor_f16 v255, exec_hi
-// GFX11: v_floor_f16_e64 v255, exec_hi ; encoding: [0xff,0x00,0xdb,0xd5,0x7f,0x00,0x00,0x00]
+v_exp_f16 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_exp_f16_e64_dpp v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd8,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
-v_floor_f16 v255, exec_lo
-// GFX11: v_floor_f16_e64 v255, exec_lo ; encoding: [0xff,0x00,0xdb,0xd5,0x7e,0x00,0x00,0x00]
+v_exp_f16 v5.l, v199.l quad_perm:[3,2,1,0]
+// GFX11: v_exp_f16_e64_dpp v5.l, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
-v_floor_f16 v255, m0
-// GFX11: v_floor_f16_e64 v255, m0 ; encoding: [0xff,0x00,0xdb,0xd5,0x7d,0x00,0x00,0x00]
+v_floor_f16 v128.h, 0xfe0b
+// GFX11: v_floor_f16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xdb,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
-v_floor_f16 v255, null
-// GFX11: v_floor_f16_e64 v255, null ; encoding: [0xff,0x00,0xdb,0xd5,0x7c,0x00,0x00,0x00]
+v_floor_f16 v128.l, 0xfe0b
+// GFX11: v_floor_f16_e64 v128.l, 0xfe0b ; encoding: [0x80,0x00,0xdb,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
-v_floor_f16 v255, s1
-// GFX11: v_floor_f16_e64 v255, s1 ; encoding: [0xff,0x00,0xdb,0xd5,0x01,0x00,0x00,0x00]
+v_floor_f16 v255.h, -1
+// GFX11: v_floor_f16_e64 v255.h, -1 op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0xc1,0x00,0x00,0x00]
-v_floor_f16 v255, s105
-// GFX11: v_floor_f16_e64 v255, s105 ; encoding: [0xff,0x00,0xdb,0xd5,0x69,0x00,0x00,0x00]
+v_floor_f16 v255.h, 0.5
+// GFX11: v_floor_f16_e64 v255.h, 0.5 op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0xf0,0x00,0x00,0x00]
-v_floor_f16 v255, src_scc
-// GFX11: v_floor_f16_e64 v255, src_scc ; encoding: [0xff,0x00,0xdb,0xd5,0xfd,0x00,0x00,0x00]
+v_floor_f16 v255.h, exec_hi
+// GFX11: v_floor_f16_e64 v255.h, exec_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0x7f,0x00,0x00,0x00]
-v_floor_f16 v255, ttmp15
-// GFX11: v_floor_f16_e64 v255, ttmp15 ; encoding: [0xff,0x00,0xdb,0xd5,0x7b,0x00,0x00,0x00]
+v_floor_f16 v255.h, exec_lo
+// GFX11: v_floor_f16_e64 v255.h, exec_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0x7e,0x00,0x00,0x00]
-v_floor_f16 v255, v1
-// GFX11: v_floor_f16_e64 v255, v1 ; encoding: [0xff,0x00,0xdb,0xd5,0x01,0x01,0x00,0x00]
+v_floor_f16 v255.h, m0
+// GFX11: v_floor_f16_e64 v255.h, m0 op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0x7d,0x00,0x00,0x00]
-v_floor_f16 v255, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_floor_f16_e64_dpp v255, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xdb,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+v_floor_f16 v255.h, null
+// GFX11: v_floor_f16_e64 v255.h, null op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0x7c,0x00,0x00,0x00]
-v_floor_f16 v255, v1 quad_perm:[3,2,1,0]
-// GFX11: v_floor_f16_e64_dpp v255, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+v_floor_f16 v255.h, s1
+// GFX11: v_floor_f16_e64 v255.h, s1 op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0x01,0x00,0x00,0x00]
-v_floor_f16 v255, v127
-// GFX11: v_floor_f16_e64 v255, v127 ; encoding: [0xff,0x00,0xdb,0xd5,0x7f,0x01,0x00,0x00]
+v_floor_f16 v255.h, s105
+// GFX11: v_floor_f16_e64 v255.h, s105 op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0x69,0x00,0x00,0x00]
-v_floor_f16 v255, v127 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_floor_f16_e64_dpp v255, v127 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xdb,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+v_floor_f16 v255.h, src_scc
+// GFX11: v_floor_f16_e64 v255.h, src_scc op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0xfd,0x00,0x00,0x00]
-v_floor_f16 v255, v127 quad_perm:[3,2,1,0]
-// GFX11: v_floor_f16_e64_dpp v255, v127 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+v_floor_f16 v255.h, ttmp15
+// GFX11: v_floor_f16_e64 v255.h, ttmp15 op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0x7b,0x00,0x00,0x00]
-v_floor_f16 v255, vcc_hi
-// GFX11: v_floor_f16_e64 v255, vcc_hi ; encoding: [0xff,0x00,0xdb,0xd5,0x6b,0x00,0x00,0x00]
+v_floor_f16 v255.h, v1.h
+// GFX11: v_floor_f16_e64 v255.h, v1.h op_sel:[1,1] ; encoding: [0xff,0x48,0xdb,0xd5,0x01,0x01,0x00,0x00]
-v_floor_f16 v255, vcc_lo
-// GFX11: v_floor_f16_e64 v255, vcc_lo ; encoding: [0xff,0x00,0xdb,0xd5,0x6a,0x00,0x00,0x00]
+v_floor_f16 v255.h, v1.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_floor_f16_e64_dpp v255.h, v1.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xdb,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
-v_floor_f16 v5, v199
-// GFX11: v_floor_f16_e64 v5, v199 ; encoding: [0x05,0x00,0xdb,0xd5,0xc7,0x01,0x00,0x00]
+v_floor_f16 v255.h, v1.h quad_perm:[3,2,1,0]
+// GFX11: v_floor_f16_e64_dpp v255.h, v1.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
-v_floor_f16 v5, v199 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_floor_f16_e64_dpp v5, v199 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xdb,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+v_floor_f16 v255.h, v127.h
+// GFX11: v_floor_f16_e64 v255.h, v127.h op_sel:[1,1] ; encoding: [0xff,0x48,0xdb,0xd5,0x7f,0x01,0x00,0x00]
-v_floor_f16 v5, v199 quad_perm:[3,2,1,0]
-// GFX11: v_floor_f16_e64_dpp v5, v199 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+v_floor_f16 v255.h, v127.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_floor_f16_e64_dpp v255.h, v127.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xdb,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+
+v_floor_f16 v255.h, v127.h quad_perm:[3,2,1,0]
+// GFX11: v_floor_f16_e64_dpp v255.h, v127.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+
+v_floor_f16 v255.h, vcc_hi
+// GFX11: v_floor_f16_e64 v255.h, vcc_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0x6b,0x00,0x00,0x00]
+
+v_floor_f16 v255.h, vcc_lo
+// GFX11: v_floor_f16_e64 v255.h, vcc_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0x6a,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, -1
+// GFX11: v_floor_f16_e64 v255.l, -1 ; encoding: [0xff,0x00,0xdb,0xd5,0xc1,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, 0.5
+// GFX11: v_floor_f16_e64 v255.l, 0.5 ; encoding: [0xff,0x00,0xdb,0xd5,0xf0,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, exec_hi
+// GFX11: v_floor_f16_e64 v255.l, exec_hi ; encoding: [0xff,0x00,0xdb,0xd5,0x7f,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, exec_lo
+// GFX11: v_floor_f16_e64 v255.l, exec_lo ; encoding: [0xff,0x00,0xdb,0xd5,0x7e,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, m0
+// GFX11: v_floor_f16_e64 v255.l, m0 ; encoding: [0xff,0x00,0xdb,0xd5,0x7d,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, null
+// GFX11: v_floor_f16_e64 v255.l, null ; encoding: [0xff,0x00,0xdb,0xd5,0x7c,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, s1
+// GFX11: v_floor_f16_e64 v255.l, s1 ; encoding: [0xff,0x00,0xdb,0xd5,0x01,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, s105
+// GFX11: v_floor_f16_e64 v255.l, s105 ; encoding: [0xff,0x00,0xdb,0xd5,0x69,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, src_scc
+// GFX11: v_floor_f16_e64 v255.l, src_scc ; encoding: [0xff,0x00,0xdb,0xd5,0xfd,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, ttmp15
+// GFX11: v_floor_f16_e64 v255.l, ttmp15 ; encoding: [0xff,0x00,0xdb,0xd5,0x7b,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, v1.l
+// GFX11: v_floor_f16_e64 v255.l, v1.l ; encoding: [0xff,0x00,0xdb,0xd5,0x01,0x01,0x00,0x00]
+
+v_floor_f16 v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_floor_f16_e64_dpp v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xdb,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+
+v_floor_f16 v255.l, v1.l quad_perm:[3,2,1,0]
+// GFX11: v_floor_f16_e64_dpp v255.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+
+v_floor_f16 v255.l, v127.l
+// GFX11: v_floor_f16_e64 v255.l, v127.l ; encoding: [0xff,0x00,0xdb,0xd5,0x7f,0x01,0x00,0x00]
+
+v_floor_f16 v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_floor_f16_e64_dpp v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xdb,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+
+v_floor_f16 v255.l, v127.l quad_perm:[3,2,1,0]
+// GFX11: v_floor_f16_e64_dpp v255.l, v127.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+
+v_floor_f16 v255.l, vcc_hi
+// GFX11: v_floor_f16_e64 v255.l, vcc_hi ; encoding: [0xff,0x00,0xdb,0xd5,0x6b,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, vcc_lo
+// GFX11: v_floor_f16_e64 v255.l, vcc_lo ; encoding: [0xff,0x00,0xdb,0xd5,0x6a,0x00,0x00,0x00]
+
+v_floor_f16 v5.h, v199.h
+// GFX11: v_floor_f16_e64 v5.h, v199.h op_sel:[1,1] ; encoding: [0x05,0x48,0xdb,0xd5,0xc7,0x01,0x00,0x00]
+
+v_floor_f16 v5.h, v199.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_floor_f16_e64_dpp v5.h, v199.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xdb,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+
+v_floor_f16 v5.h, v199.h quad_perm:[3,2,1,0]
+// GFX11: v_floor_f16_e64_dpp v5.h, v199.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x48,0xdb,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+
+v_floor_f16 v5.l, v199.l
+// GFX11: v_floor_f16_e64 v5.l, v199.l ; encoding: [0x05,0x00,0xdb,0xd5,0xc7,0x01,0x00,0x00]
+
+v_floor_f16 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_floor_f16_e64_dpp v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xdb,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+
+v_floor_f16 v5.l, v199.l quad_perm:[3,2,1,0]
+// GFX11: v_floor_f16_e64_dpp v5.l, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
v_fract_f16 v128.h, 0xfe0b
// GFX11: v_fract_f16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xdf,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
@@ -1688,71 +1886,137 @@ v_frexp_mant_f16 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
v_frexp_mant_f16 v5.l, v199.l quad_perm:[3,2,1,0]
// GFX11: v_frexp_mant_f16_e64_dpp v5.l, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd9,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
-v_log_f16 v128, 0xfe0b
-// GFX11: v_log_f16_e64 v128, 0xfe0b ; encoding: [0x80,0x00,0xd7,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+v_log_f16 v128.h, 0xfe0b
+// GFX11: v_log_f16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xd7,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+
+v_log_f16 v128.l, 0xfe0b
+// GFX11: v_log_f16_e64 v128.l, 0xfe0b ; encoding: [0x80,0x00,0xd7,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+
+v_log_f16 v255.h, -1
+// GFX11: v_log_f16_e64 v255.h, -1 op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0xc1,0x00,0x00,0x00]
+
+v_log_f16 v255.h, 0.5
+// GFX11: v_log_f16_e64 v255.h, 0.5 op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0xf0,0x00,0x00,0x00]
+
+v_log_f16 v255.h, exec_hi
+// GFX11: v_log_f16_e64 v255.h, exec_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0x7f,0x00,0x00,0x00]
+
+v_log_f16 v255.h, exec_lo
+// GFX11: v_log_f16_e64 v255.h, exec_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0x7e,0x00,0x00,0x00]
+
+v_log_f16 v255.h, m0
+// GFX11: v_log_f16_e64 v255.h, m0 op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0x7d,0x00,0x00,0x00]
+
+v_log_f16 v255.h, null
+// GFX11: v_log_f16_e64 v255.h, null op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0x7c,0x00,0x00,0x00]
+
+v_log_f16 v255.h, s1
+// GFX11: v_log_f16_e64 v255.h, s1 op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0x01,0x00,0x00,0x00]
+
+v_log_f16 v255.h, s105
+// GFX11: v_log_f16_e64 v255.h, s105 op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0x69,0x00,0x00,0x00]
+
+v_log_f16 v255.h, src_scc
+// GFX11: v_log_f16_e64 v255.h, src_scc op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0xfd,0x00,0x00,0x00]
+
+v_log_f16 v255.h, ttmp15
+// GFX11: v_log_f16_e64 v255.h, ttmp15 op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0x7b,0x00,0x00,0x00]
-v_log_f16 v255, -1
-// GFX11: v_log_f16_e64 v255, -1 ; encoding: [0xff,0x00,0xd7,0xd5,0xc1,0x00,0x00,0x00]
+v_log_f16 v255.h, v1.h
+// GFX11: v_log_f16_e64 v255.h, v1.h op_sel:[1,1] ; encoding: [0xff,0x48,0xd7,0xd5,0x01,0x01,0x00,0x00]
-v_log_f16 v255, 0.5
-// GFX11: v_log_f16_e64 v255, 0.5 ; encoding: [0xff,0x00,0xd7,0xd5,0xf0,0x00,0x00,0x00]
+v_log_f16 v255.h, v1.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_log_f16_e64_dpp v255.h, v1.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xd7,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
-v_log_f16 v255, exec_hi
-// GFX11: v_log_f16_e64 v255, exec_hi ; encoding: [0xff,0x00,0xd7,0xd5,0x7f,0x00,0x00,0x00]
+v_log_f16 v255.h, v1.h quad_perm:[3,2,1,0]
+// GFX11: v_log_f16_e64_dpp v255.h, v1.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
-v_log_f16 v255, exec_lo
-// GFX11: v_log_f16_e64 v255, exec_lo ; encoding: [0xff,0x00,0xd7,0xd5,0x7e,0x00,0x00,0x00]
+v_log_f16 v255.h, v127.h
+// GFX11: v_log_f16_e64 v255.h, v127.h op_sel:[1,1] ; encoding: [0xff,0x48,0xd7,0xd5,0x7f,0x01,0x00,0x00]
-v_log_f16 v255, m0
-// GFX11: v_log_f16_e64 v255, m0 ; encoding: [0xff,0x00,0xd7,0xd5,0x7d,0x00,0x00,0x00]
+v_log_f16 v255.h, v127.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_log_f16_e64_dpp v255.h, v127.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xd7,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
-v_log_f16 v255, null
-// GFX11: v_log_f16_e64 v255, null ; encoding: [0xff,0x00,0xd7,0xd5,0x7c,0x00,0x00,0x00]
+v_log_f16 v255.h, v127.h quad_perm:[3,2,1,0]
+// GFX11: v_log_f16_e64_dpp v255.h, v127.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
-v_log_f16 v255, s1
-// GFX11: v_log_f16_e64 v255, s1 ; encoding: [0xff,0x00,0xd7,0xd5,0x01,0x00,0x00,0x00]
+v_log_f16 v255.h, vcc_hi
+// GFX11: v_log_f16_e64 v255.h, vcc_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0x6b,0x00,0x00,0x00]
-v_log_f16 v255, s105
-// GFX11: v_log_f16_e64 v255, s105 ; encoding: [0xff,0x00,0xd7,0xd5,0x69,0x00,0x00,0x00]
+v_log_f16 v255.h, vcc_lo
+// GFX11: v_log_f16_e64 v255.h, vcc_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0x6a,0x00,0x00,0x00]
-v_log_f16 v255, src_scc
-// GFX11: v_log_f16_e64 v255, src_scc ; encoding: [0xff,0x00,0xd7,0xd5,0xfd,0x00,0x00,0x00]
+v_log_f16 v255.l, -1
+// GFX11: v_log_f16_e64 v255.l, -1 ; encoding: [0xff,0x00,0xd7,0xd5,0xc1,0x00,0x00,0x00]
-v_log_f16 v255, ttmp15
-// GFX11: v_log_f16_e64 v255, ttmp15 ; encoding: [0xff,0x00,0xd7,0xd5,0x7b,0x00,0x00,0x00]
+v_log_f16 v255.l, 0.5
+// GFX11: v_log_f16_e64 v255.l, 0.5 ; encoding: [0xff,0x00,0xd7,0xd5,0xf0,0x00,0x00,0x00]
-v_log_f16 v255, v1
-// GFX11: v_log_f16_e64 v255, v1 ; encoding: [0xff,0x00,0xd7,0xd5,0x01,0x01,0x00,0x00]
+v_log_f16 v255.l, exec_hi
+// GFX11: v_log_f16_e64 v255.l, exec_hi ; encoding: [0xff,0x00,0xd7,0xd5,0x7f,0x00,0x00,0x00]
-v_log_f16 v255, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_log_f16_e64_dpp v255, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd7,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+v_log_f16 v255.l, exec_lo
+// GFX11: v_log_f16_e64 v255.l, exec_lo ; encoding: [0xff,0x00,0xd7,0xd5,0x7e,0x00,0x00,0x00]
-v_log_f16 v255, v1 quad_perm:[3,2,1,0]
-// GFX11: v_log_f16_e64_dpp v255, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+v_log_f16 v255.l, m0
+// GFX11: v_log_f16_e64 v255.l, m0 ; encoding: [0xff,0x00,0xd7,0xd5,0x7d,0x00,0x00,0x00]
-v_log_f16 v255, v127
-// GFX11: v_log_f16_e64 v255, v127 ; encoding: [0xff,0x00,0xd7,0xd5,0x7f,0x01,0x00,0x00]
+v_log_f16 v255.l, null
+// GFX11: v_log_f16_e64 v255.l, null ; encoding: [0xff,0x00,0xd7,0xd5,0x7c,0x00,0x00,0x00]
-v_log_f16 v255, v127 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_log_f16_e64_dpp v255, v127 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd7,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+v_log_f16 v255.l, s1
+// GFX11: v_log_f16_e64 v255.l, s1 ; encoding: [0xff,0x00,0xd7,0xd5,0x01,0x00,0x00,0x00]
-v_log_f16 v255, v127 quad_perm:[3,2,1,0]
-// GFX11: v_log_f16_e64_dpp v255, v127 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+v_log_f16 v255.l, s105
+// GFX11: v_log_f16_e64 v255.l, s105 ; encoding: [0xff,0x00,0xd7,0xd5,0x69,0x00,0x00,0x00]
-v_log_f16 v255, vcc_hi
-// GFX11: v_log_f16_e64 v255, vcc_hi ; encoding: [0xff,0x00,0xd7,0xd5,0x6b,0x00,0x00,0x00]
+v_log_f16 v255.l, src_scc
+// GFX11: v_log_f16_e64 v255.l, src_scc ; encoding: [0xff,0x00,0xd7,0xd5,0xfd,0x00,0x00,0x00]
-v_log_f16 v255, vcc_lo
-// GFX11: v_log_f16_e64 v255, vcc_lo ; encoding: [0xff,0x00,0xd7,0xd5,0x6a,0x00,0x00,0x00]
+v_log_f16 v255.l, ttmp15
+// GFX11: v_log_f16_e64 v255.l, ttmp15 ; encoding: [0xff,0x00,0xd7,0xd5,0x7b,0x00,0x00,0x00]
-v_log_f16 v5, v199
-// GFX11: v_log_f16_e64 v5, v199 ; encoding: [0x05,0x00,0xd7,0xd5,0xc7,0x01,0x00,0x00]
+v_log_f16 v255.l, v1.l
+// GFX11: v_log_f16_e64 v255.l, v1.l ; encoding: [0xff,0x00,0xd7,0xd5,0x01,0x01,0x00,0x00]
-v_log_f16 v5, v199 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_log_f16_e64_dpp v5, v199 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd7,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+v_log_f16 v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_log_f16_e64_dpp v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd7,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
-v_log_f16 v5, v199 quad_perm:[3,2,1,0]
-// GFX11: v_log_f16_e64_dpp v5, v199 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+v_log_f16 v255.l, v1.l quad_perm:[3,2,1,0]
+// GFX11: v_log_f16_e64_dpp v255.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+
+v_log_f16 v255.l, v127.l
+// GFX11: v_log_f16_e64 v255.l, v127.l ; encoding: [0xff,0x00,0xd7,0xd5,0x7f,0x01,0x00,0x00]
+
+v_log_f16 v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_log_f16_e64_dpp v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd7,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+
+v_log_f16 v255.l, v127.l quad_perm:[3,2,1,0]
+// GFX11: v_log_f16_e64_dpp v255.l, v127.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+
+v_log_f16 v255.l, vcc_hi
+// GFX11: v_log_f16_e64 v255.l, vcc_hi ; encoding: [0xff,0x00,0xd7,0xd5,0x6b,0x00,0x00,0x00]
+
+v_log_f16 v255.l, vcc_lo
+// GFX11: v_log_f16_e64 v255.l, vcc_lo ; encoding: [0xff,0x00,0xd7,0xd5,0x6a,0x00,0x00,0x00]
+
+v_log_f16 v5.h, v199.h
+// GFX11: v_log_f16_e64 v5.h, v199.h op_sel:[1,1] ; encoding: [0x05,0x48,0xd7,0xd5,0xc7,0x01,0x00,0x00]
+
+v_log_f16 v5.h, v199.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_log_f16_e64_dpp v5.h, v199.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xd7,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+
+v_log_f16 v5.h, v199.h quad_perm:[3,2,1,0]
+// GFX11: v_log_f16_e64_dpp v5.h, v199.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x48,0xd7,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+
+v_log_f16 v5.l, v199.l
+// GFX11: v_log_f16_e64 v5.l, v199.l ; encoding: [0x05,0x00,0xd7,0xd5,0xc7,0x01,0x00,0x00]
+
+v_log_f16 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_log_f16_e64_dpp v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd7,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+
+v_log_f16 v5.l, v199.l quad_perm:[3,2,1,0]
+// GFX11: v_log_f16_e64_dpp v5.l, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
v_not_b16 v128.h, 0xfe0b
// GFX11: v_not_b16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xe9,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
@@ -1886,71 +2150,137 @@ v_not_b16 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
v_not_b16 v5.l, v199.l quad_perm:[3,2,1,0]
// GFX11: v_not_b16_e64_dpp v5.l, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xe9,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
-v_rcp_f16 v128, 0xfe0b
-// GFX11: v_rcp_f16_e64 v128, 0xfe0b ; encoding: [0x80,0x00,0xd4,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+v_rcp_f16 v128.h, 0xfe0b
+// GFX11: v_rcp_f16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xd4,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+
+v_rcp_f16 v128.l, 0xfe0b
+// GFX11: v_rcp_f16_e64 v128.l, 0xfe0b ; encoding: [0x80,0x00,0xd4,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+
+v_rcp_f16 v255.h, -1
+// GFX11: v_rcp_f16_e64 v255.h, -1 op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0xc1,0x00,0x00,0x00]
+
+v_rcp_f16 v255.h, 0.5
+// GFX11: v_rcp_f16_e64 v255.h, 0.5 op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0xf0,0x00,0x00,0x00]
+
+v_rcp_f16 v255.h, exec_hi
+// GFX11: v_rcp_f16_e64 v255.h, exec_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0x7f,0x00,0x00,0x00]
+
+v_rcp_f16 v255.h, exec_lo
+// GFX11: v_rcp_f16_e64 v255.h, exec_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0x7e,0x00,0x00,0x00]
+
+v_rcp_f16 v255.h, m0
+// GFX11: v_rcp_f16_e64 v255.h, m0 op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0x7d,0x00,0x00,0x00]
+
+v_rcp_f16 v255.h, null
+// GFX11: v_rcp_f16_e64 v255.h, null op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0x7c,0x00,0x00,0x00]
+
+v_rcp_f16 v255.h, s1
+// GFX11: v_rcp_f16_e64 v255.h, s1 op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0x01,0x00,0x00,0x00]
+
+v_rcp_f16 v255.h, s105
+// GFX11: v_rcp_f16_e64 v255.h, s105 op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0x69,0x00,0x00,0x00]
-v_rcp_f16 v255, -1
-// GFX11: v_rcp_f16_e64 v255, -1 ; encoding: [0xff,0x00,0xd4,0xd5,0xc1,0x00,0x00,0x00]
+v_rcp_f16 v255.h, src_scc
+// GFX11: v_rcp_f16_e64 v255.h, src_scc op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0xfd,0x00,0x00,0x00]
-v_rcp_f16 v255, 0.5
-// GFX11: v_rcp_f16_e64 v255, 0.5 ; encoding: [0xff,0x00,0xd4,0xd5,0xf0,0x00,0x00,0x00]
+v_rcp_f16 v255.h, ttmp15
+// GFX11: v_rcp_f16_e64 v255.h, ttmp15 op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0x7b,0x00,0x00,0x00]
-v_rcp_f16 v255, exec_hi
-// GFX11: v_rcp_f16_e64 v255, exec_hi ; encoding: [0xff,0x00,0xd4,0xd5,0x7f,0x00,0x00,0x00]
+v_rcp_f16 v255.h, v1.h
+// GFX11: v_rcp_f16_e64 v255.h, v1.h op_sel:[1,1] ; encoding: [0xff,0x48,0xd4,0xd5,0x01,0x01,0x00,0x00]
-v_rcp_f16 v255, exec_lo
-// GFX11: v_rcp_f16_e64 v255, exec_lo ; encoding: [0xff,0x00,0xd4,0xd5,0x7e,0x00,0x00,0x00]
+v_rcp_f16 v255.h, v1.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_rcp_f16_e64_dpp v255.h, v1.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xd4,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
-v_rcp_f16 v255, m0
-// GFX11: v_rcp_f16_e64 v255, m0 ; encoding: [0xff,0x00,0xd4,0xd5,0x7d,0x00,0x00,0x00]
+v_rcp_f16 v255.h, v1.h quad_perm:[3,2,1,0]
+// GFX11: v_rcp_f16_e64_dpp v255.h, v1.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
-v_rcp_f16 v255, null
-// GFX11: v_rcp_f16_e64 v255, null ; encoding: [0xff,0x00,0xd4,0xd5,0x7c,0x00,0x00,0x00]
+v_rcp_f16 v255.h, v127.h
+// GFX11: v_rcp_f16_e64 v255.h, v127.h op_sel:[1,1] ; encoding: [0xff,0x48,0xd4,0xd5,0x7f,0x01,0x00,0x00]
-v_rcp_f16 v255, s1
-// GFX11: v_rcp_f16_e64 v255, s1 ; encoding: [0xff,0x00,0xd4,0xd5,0x01,0x00,0x00,0x00]
+v_rcp_f16 v255.h, v127.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_rcp_f16_e64_dpp v255.h, v127.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xd4,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
-v_rcp_f16 v255, s105
-// GFX11: v_rcp_f16_e64 v255, s105 ; encoding: [0xff,0x00,0xd4,0xd5,0x69,0x00,0x00,0x00]
+v_rcp_f16 v255.h, v127.h quad_perm:[3,2,1,0]
+// GFX11: v_rcp_f16_e64_dpp v255.h, v127.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
-v_rcp_f16 v255, src_scc
-// GFX11: v_rcp_f16_e64 v255, src_scc ; encoding: [0xff,0x00,0xd4,0xd5,0xfd,0x00,0x00,0x00]
+v_rcp_f16 v255.h, vcc_hi
+// GFX11: v_rcp_f16_e64 v255.h, vcc_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0x6b,0x00,0x00,0x00]
-v_rcp_f16 v255, ttmp15
-// GFX11: v_rcp_f16_e64 v255, ttmp15 ; encoding: [0xff,0x00,0xd4,0xd5,0x7b,0x00,0x00,0x00]
+v_rcp_f16 v255.h, vcc_lo
+// GFX11: v_rcp_f16_e64 v255.h, vcc_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0x6a,0x00,0x00,0x00]
-v_rcp_f16 v255, v1
-// GFX11: v_rcp_f16_e64 v255, v1 ; encoding: [0xff,0x00,0xd4,0xd5,0x01,0x01,0x00,0x00]
+v_rcp_f16 v255.l, -1
+// GFX11: v_rcp_f16_e64 v255.l, -1 ; encoding: [0xff,0x00,0xd4,0xd5,0xc1,0x00,0x00,0x00]
-v_rcp_f16 v255, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_rcp_f16_e64_dpp v255, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd4,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+v_rcp_f16 v255.l, 0.5
+// GFX11: v_rcp_f16_e64 v255.l, 0.5 ; encoding: [0xff,0x00,0xd4,0xd5,0xf0,0x00,0x00,0x00]
-v_rcp_f16 v255, v1 quad_perm:[3,2,1,0]
-// GFX11: v_rcp_f16_e64_dpp v255, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+v_rcp_f16 v255.l, exec_hi
+// GFX11: v_rcp_f16_e64 v255.l, exec_hi ; encoding: [0xff,0x00,0xd4,0xd5,0x7f,0x00,0x00,0x00]
-v_rcp_f16 v255, v127
-// GFX11: v_rcp_f16_e64 v255, v127 ; encoding: [0xff,0x00,0xd4,0xd5,0x7f,0x01,0x00,0x00]
+v_rcp_f16 v255.l, exec_lo
+// GFX11: v_rcp_f16_e64 v255.l, exec_lo ; encoding: [0xff,0x00,0xd4,0xd5,0x7e,0x00,0x00,0x00]
-v_rcp_f16 v255, v127 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_rcp_f16_e64_dpp v255, v127 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd4,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+v_rcp_f16 v255.l, m0
+// GFX11: v_rcp_f16_e64 v255.l, m0 ; encoding: [0xff,0x00,0xd4,0xd5,0x7d,0x00,0x00,0x00]
-v_rcp_f16 v255, v127 quad_perm:[3,2,1,0]
-// GFX11: v_rcp_f16_e64_dpp v255, v127 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+v_rcp_f16 v255.l, null
+// GFX11: v_rcp_f16_e64 v255.l, null ; encoding: [0xff,0x00,0xd4,0xd5,0x7c,0x00,0x00,0x00]
-v_rcp_f16 v255, vcc_hi
-// GFX11: v_rcp_f16_e64 v255, vcc_hi ; encoding: [0xff,0x00,0xd4,0xd5,0x6b,0x00,0x00,0x00]
+v_rcp_f16 v255.l, s1
+// GFX11: v_rcp_f16_e64 v255.l, s1 ; encoding: [0xff,0x00,0xd4,0xd5,0x01,0x00,0x00,0x00]
-v_rcp_f16 v255, vcc_lo
-// GFX11: v_rcp_f16_e64 v255, vcc_lo ; encoding: [0xff,0x00,0xd4,0xd5,0x6a,0x00,0x00,0x00]
+v_rcp_f16 v255.l, s105
+// GFX11: v_rcp_f16_e64 v255.l, s105 ; encoding: [0xff,0x00,0xd4,0xd5,0x69,0x00,0x00,0x00]
-v_rcp_f16 v5, v199
-// GFX11: v_rcp_f16_e64 v5, v199 ; encoding: [0x05,0x00,0xd4,0xd5,0xc7,0x01,0x00,0x00]
+v_rcp_f16 v255.l, src_scc
+// GFX11: v_rcp_f16_e64 v255.l, src_scc ; encoding: [0xff,0x00,0xd4,0xd5,0xfd,0x00,0x00,0x00]
-v_rcp_f16 v5, v199 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_rcp_f16_e64_dpp v5, v199 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd4,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+v_rcp_f16 v255.l, ttmp15
+// GFX11: v_rcp_f16_e64 v255.l, ttmp15 ; encoding: [0xff,0x00,0xd4,0xd5,0x7b,0x00,0x00,0x00]
-v_rcp_f16 v5, v199 quad_perm:[3,2,1,0]
-// GFX11: v_rcp_f16_e64_dpp v5, v199 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+v_rcp_f16 v255.l, v1.l
+// GFX11: v_rcp_f16_e64 v255.l, v1.l ; encoding: [0xff,0x00,0xd4,0xd5,0x01,0x01,0x00,0x00]
+
+v_rcp_f16 v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_rcp_f16_e64_dpp v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd4,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+
+v_rcp_f16 v255.l, v1.l quad_perm:[3,2,1,0]
+// GFX11: v_rcp_f16_e64_dpp v255.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+
+v_rcp_f16 v255.l, v127.l
+// GFX11: v_rcp_f16_e64 v255.l, v127.l ; encoding: [0xff,0x00,0xd4,0xd5,0x7f,0x01,0x00,0x00]
+
+v_rcp_f16 v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_rcp_f16_e64_dpp v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd4,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+
+v_rcp_f16 v255.l, v127.l quad_perm:[3,2,1,0]
+// GFX11: v_rcp_f16_e64_dpp v255.l, v127.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+
+v_rcp_f16 v255.l, vcc_hi
+// GFX11: v_rcp_f16_e64 v255.l, vcc_hi ; encoding: [0xff,0x00,0xd4,0xd5,0x6b,0x00,0x00,0x00]
+
+v_rcp_f16 v255.l, vcc_lo
+// GFX11: v_rcp_f16_e64 v255.l, vcc_lo ; encoding: [0xff,0x00,0xd4,0xd5,0x6a,0x00,0x00,0x00]
+
+v_rcp_f16 v5.h, v199.h
+// GFX11: v_rcp_f16_e64 v5.h, v199.h op_sel:[1,1] ; encoding: [0x05,0x48,0xd4,0xd5,0xc7,0x01,0x00,0x00]
+
+v_rcp_f16 v5.h, v199.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_rcp_f16_e64_dpp v5.h, v199.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xd4,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+
+v_rcp_f16 v5.h, v199.h quad_perm:[3,2,1,0]
+// GFX11: v_rcp_f16_e64_dpp v5.h, v199.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x48,0xd4,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+
+v_rcp_f16 v5.l, v199.l
+// GFX11: v_rcp_f16_e64 v5.l, v199.l ; encoding: [0x05,0x00,0xd4,0xd5,0xc7,0x01,0x00,0x00]
+
+v_rcp_f16 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_rcp_f16_e64_dpp v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd4,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+
+v_rcp_f16 v5.l, v199.l quad_perm:[3,2,1,0]
+// GFX11: v_rcp_f16_e64_dpp v5.l, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
v_rndne_f16 v128.h, 0xfe0b
// GFX11: v_rndne_f16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xde,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
@@ -2084,71 +2414,137 @@ v_rndne_f16 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
v_rndne_f16 v5.l, v199.l quad_perm:[3,2,1,0]
// GFX11: v_rndne_f16_e64_dpp v5.l, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xde,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
-v_rsq_f16 v128, 0xfe0b
-// GFX11: v_rsq_f16_e64 v128, 0xfe0b ; encoding: [0x80,0x00,0xd6,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+v_rsq_f16 v128.h, 0xfe0b
+// GFX11: v_rsq_f16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xd6,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+
+v_rsq_f16 v128.l, 0xfe0b
+// GFX11: v_rsq_f16_e64 v128.l, 0xfe0b ; encoding: [0x80,0x00,0xd6,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+
+v_rsq_f16 v255.h, -1
+// GFX11: v_rsq_f16_e64 v255.h, -1 op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0xc1,0x00,0x00,0x00]
+
+v_rsq_f16 v255.h, 0.5
+// GFX11: v_rsq_f16_e64 v255.h, 0.5 op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0xf0,0x00,0x00,0x00]
+
+v_rsq_f16 v255.h, exec_hi
+// GFX11: v_rsq_f16_e64 v255.h, exec_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0x7f,0x00,0x00,0x00]
+
+v_rsq_f16 v255.h, exec_lo
+// GFX11: v_rsq_f16_e64 v255.h, exec_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0x7e,0x00,0x00,0x00]
+
+v_rsq_f16 v255.h, m0
+// GFX11: v_rsq_f16_e64 v255.h, m0 op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0x7d,0x00,0x00,0x00]
-v_rsq_f16 v255, -1
-// GFX11: v_rsq_f16_e64 v255, -1 ; encoding: [0xff,0x00,0xd6,0xd5,0xc1,0x00,0x00,0x00]
+v_rsq_f16 v255.h, null
+// GFX11: v_rsq_f16_e64 v255.h, null op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0x7c,0x00,0x00,0x00]
-v_rsq_f16 v255, 0.5
-// GFX11: v_rsq_f16_e64 v255, 0.5 ; encoding: [0xff,0x00,0xd6,0xd5,0xf0,0x00,0x00,0x00]
+v_rsq_f16 v255.h, s1
+// GFX11: v_rsq_f16_e64 v255.h, s1 op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0x01,0x00,0x00,0x00]
-v_rsq_f16 v255, exec_hi
-// GFX11: v_rsq_f16_e64 v255, exec_hi ; encoding: [0xff,0x00,0xd6,0xd5,0x7f,0x00,0x00,0x00]
+v_rsq_f16 v255.h, s105
+// GFX11: v_rsq_f16_e64 v255.h, s105 op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0x69,0x00,0x00,0x00]
-v_rsq_f16 v255, exec_lo
-// GFX11: v_rsq_f16_e64 v255, exec_lo ; encoding: [0xff,0x00,0xd6,0xd5,0x7e,0x00,0x00,0x00]
+v_rsq_f16 v255.h, src_scc
+// GFX11: v_rsq_f16_e64 v255.h, src_scc op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0xfd,0x00,0x00,0x00]
-v_rsq_f16 v255, m0
-// GFX11: v_rsq_f16_e64 v255, m0 ; encoding: [0xff,0x00,0xd6,0xd5,0x7d,0x00,0x00,0x00]
+v_rsq_f16 v255.h, ttmp15
+// GFX11: v_rsq_f16_e64 v255.h, ttmp15 op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0x7b,0x00,0x00,0x00]
-v_rsq_f16 v255, null
-// GFX11: v_rsq_f16_e64 v255, null ; encoding: [0xff,0x00,0xd6,0xd5,0x7c,0x00,0x00,0x00]
+v_rsq_f16 v255.h, v1.h
+// GFX11: v_rsq_f16_e64 v255.h, v1.h op_sel:[1,1] ; encoding: [0xff,0x48,0xd6,0xd5,0x01,0x01,0x00,0x00]
-v_rsq_f16 v255, s1
-// GFX11: v_rsq_f16_e64 v255, s1 ; encoding: [0xff,0x00,0xd6,0xd5,0x01,0x00,0x00,0x00]
+v_rsq_f16 v255.h, v1.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_rsq_f16_e64_dpp v255.h, v1.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xd6,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
-v_rsq_f16 v255, s105
-// GFX11: v_rsq_f16_e64 v255, s105 ; encoding: [0xff,0x00,0xd6,0xd5,0x69,0x00,0x00,0x00]
+v_rsq_f16 v255.h, v1.h quad_perm:[3,2,1,0]
+// GFX11: v_rsq_f16_e64_dpp v255.h, v1.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
-v_rsq_f16 v255, src_scc
-// GFX11: v_rsq_f16_e64 v255, src_scc ; encoding: [0xff,0x00,0xd6,0xd5,0xfd,0x00,0x00,0x00]
+v_rsq_f16 v255.h, v127.h
+// GFX11: v_rsq_f16_e64 v255.h, v127.h op_sel:[1,1] ; encoding: [0xff,0x48,0xd6,0xd5,0x7f,0x01,0x00,0x00]
-v_rsq_f16 v255, ttmp15
-// GFX11: v_rsq_f16_e64 v255, ttmp15 ; encoding: [0xff,0x00,0xd6,0xd5,0x7b,0x00,0x00,0x00]
+v_rsq_f16 v255.h, v127.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_rsq_f16_e64_dpp v255.h, v127.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xd6,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
-v_rsq_f16 v255, v1
-// GFX11: v_rsq_f16_e64 v255, v1 ; encoding: [0xff,0x00,0xd6,0xd5,0x01,0x01,0x00,0x00]
+v_rsq_f16 v255.h, v127.h quad_perm:[3,2,1,0]
+// GFX11: v_rsq_f16_e64_dpp v255.h, v127.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
-v_rsq_f16 v255, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_rsq_f16_e64_dpp v255, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd6,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+v_rsq_f16 v255.h, vcc_hi
+// GFX11: v_rsq_f16_e64 v255.h, vcc_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0x6b,0x00,0x00,0x00]
-v_rsq_f16 v255, v1 quad_perm:[3,2,1,0]
-// GFX11: v_rsq_f16_e64_dpp v255, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+v_rsq_f16 v255.h, vcc_lo
+// GFX11: v_rsq_f16_e64 v255.h, vcc_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0x6a,0x00,0x00,0x00]
-v_rsq_f16 v255, v127
-// GFX11: v_rsq_f16_e64 v255, v127 ; encoding: [0xff,0x00,0xd6,0xd5,0x7f,0x01,0x00,0x00]
+v_rsq_f16 v255.l, -1
+// GFX11: v_rsq_f16_e64 v255.l, -1 ; encoding: [0xff,0x00,0xd6,0xd5,0xc1,0x00,0x00,0x00]
-v_rsq_f16 v255, v127 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_rsq_f16_e64_dpp v255, v127 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd6,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+v_rsq_f16 v255.l, 0.5
+// GFX11: v_rsq_f16_e64 v255.l, 0.5 ; encoding: [0xff,0x00,0xd6,0xd5,0xf0,0x00,0x00,0x00]
-v_rsq_f16 v255, v127 quad_perm:[3,2,1,0]
-// GFX11: v_rsq_f16_e64_dpp v255, v127 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+v_rsq_f16 v255.l, exec_hi
+// GFX11: v_rsq_f16_e64 v255.l, exec_hi ; encoding: [0xff,0x00,0xd6,0xd5,0x7f,0x00,0x00,0x00]
-v_rsq_f16 v255, vcc_hi
-// GFX11: v_rsq_f16_e64 v255, vcc_hi ; encoding: [0xff,0x00,0xd6,0xd5,0x6b,0x00,0x00,0x00]
+v_rsq_f16 v255.l, exec_lo
+// GFX11: v_rsq_f16_e64 v255.l, exec_lo ; encoding: [0xff,0x00,0xd6,0xd5,0x7e,0x00,0x00,0x00]
-v_rsq_f16 v255, vcc_lo
-// GFX11: v_rsq_f16_e64 v255, vcc_lo ; encoding: [0xff,0x00,0xd6,0xd5,0x6a,0x00,0x00,0x00]
+v_rsq_f16 v255.l, m0
+// GFX11: v_rsq_f16_e64 v255.l, m0 ; encoding: [0xff,0x00,0xd6,0xd5,0x7d,0x00,0x00,0x00]
-v_rsq_f16 v5, v199
-// GFX11: v_rsq_f16_e64 v5, v199 ; encoding: [0x05,0x00,0xd6,0xd5,0xc7,0x01,0x00,0x00]
+v_rsq_f16 v255.l, null
+// GFX11: v_rsq_f16_e64 v255.l, null ; encoding: [0xff,0x00,0xd6,0xd5,0x7c,0x00,0x00,0x00]
-v_rsq_f16 v5, v199 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_rsq_f16_e64_dpp v5, v199 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd6,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+v_rsq_f16 v255.l, s1
+// GFX11: v_rsq_f16_e64 v255.l, s1 ; encoding: [0xff,0x00,0xd6,0xd5,0x01,0x00,0x00,0x00]
-v_rsq_f16 v5, v199 quad_perm:[3,2,1,0]
-// GFX11: v_rsq_f16_e64_dpp v5, v199 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+v_rsq_f16 v255.l, s105
+// GFX11: v_rsq_f16_e64 v255.l, s105 ; encoding: [0xff,0x00,0xd6,0xd5,0x69,0x00,0x00,0x00]
+
+v_rsq_f16 v255.l, src_scc
+// GFX11: v_rsq_f16_e64 v255.l, src_scc ; encoding: [0xff,0x00,0xd6,0xd5,0xfd,0x00,0x00,0x00]
+
+v_rsq_f16 v255.l, ttmp15
+// GFX11: v_rsq_f16_e64 v255.l, ttmp15 ; encoding: [0xff,0x00,0xd6,0xd5,0x7b,0x00,0x00,0x00]
+
+v_rsq_f16 v255.l, v1.l
+// GFX11: v_rsq_f16_e64 v255.l, v1.l ; encoding: [0xff,0x00,0xd6,0xd5,0x01,0x01,0x00,0x00]
+
+v_rsq_f16 v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_rsq_f16_e64_dpp v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd6,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+
+v_rsq_f16 v255.l, v1.l quad_perm:[3,2,1,0]
+// GFX11: v_rsq_f16_e64_dpp v255.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+
+v_rsq_f16 v255.l, v127.l
+// GFX11: v_rsq_f16_e64 v255.l, v127.l ; encoding: [0xff,0x00,0xd6,0xd5,0x7f,0x01,0x00,0x00]
+
+v_rsq_f16 v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_rsq_f16_e64_dpp v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd6,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+
+v_rsq_f16 v255.l, v127.l quad_perm:[3,2,1,0]
+// GFX11: v_rsq_f16_e64_dpp v255.l, v127.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+
+v_rsq_f16 v255.l, vcc_hi
+// GFX11: v_rsq_f16_e64 v255.l, vcc_hi ; encoding: [0xff,0x00,0xd6,0xd5,0x6b,0x00,0x00,0x00]
+
+v_rsq_f16 v255.l, vcc_lo
+// GFX11: v_rsq_f16_e64 v255.l, vcc_lo ; encoding: [0xff,0x00,0xd6,0xd5,0x6a,0x00,0x00,0x00]
+
+v_rsq_f16 v5.h, v199.h
+// GFX11: v_rsq_f16_e64 v5.h, v199.h op_sel:[1,1] ; encoding: [0x05,0x48,0xd6,0xd5,0xc7,0x01,0x00,0x00]
+
+v_rsq_f16 v5.h, v199.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_rsq_f16_e64_dpp v5.h, v199.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xd6,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+
+v_rsq_f16 v5.h, v199.h quad_perm:[3,2,1,0]
+// GFX11: v_rsq_f16_e64_dpp v5.h, v199.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x48,0xd6,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+
+v_rsq_f16 v5.l, v199.l
+// GFX11: v_rsq_f16_e64 v5.l, v199.l ; encoding: [0x05,0x00,0xd6,0xd5,0xc7,0x01,0x00,0x00]
+
+v_rsq_f16 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_rsq_f16_e64_dpp v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd6,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+
+v_rsq_f16 v5.l, v199.l quad_perm:[3,2,1,0]
+// GFX11: v_rsq_f16_e64_dpp v5.l, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
v_sat_pk_u8_i16 v199.h, v5
// GFX11: v_sat_pk_u8_i16_e64 v199.h, v5 op_sel:[0,1] ; encoding: [0xc7,0x40,0xe2,0xd5,0x05,0x01,0x00,0x00]
@@ -2300,71 +2696,137 @@ v_sin_f16 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
v_sin_f16 v5.l, v199.l quad_perm:[3,2,1,0]
// GFX11: v_sin_f16_e64_dpp v5.l, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xe0,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
-v_sqrt_f16 v128, 0xfe0b
-// GFX11: v_sqrt_f16_e64 v128, 0xfe0b ; encoding: [0x80,0x00,0xd5,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+v_sqrt_f16 v128.h, 0xfe0b
+// GFX11: v_sqrt_f16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xd5,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+
+v_sqrt_f16 v128.l, 0xfe0b
+// GFX11: v_sqrt_f16_e64 v128.l, 0xfe0b ; encoding: [0x80,0x00,0xd5,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+
+v_sqrt_f16 v255.h, -1
+// GFX11: v_sqrt_f16_e64 v255.h, -1 op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0xc1,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.h, 0.5
+// GFX11: v_sqrt_f16_e64 v255.h, 0.5 op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0xf0,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.h, exec_hi
+// GFX11: v_sqrt_f16_e64 v255.h, exec_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0x7f,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.h, exec_lo
+// GFX11: v_sqrt_f16_e64 v255.h, exec_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0x7e,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.h, m0
+// GFX11: v_sqrt_f16_e64 v255.h, m0 op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0x7d,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.h, null
+// GFX11: v_sqrt_f16_e64 v255.h, null op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0x7c,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.h, s1
+// GFX11: v_sqrt_f16_e64 v255.h, s1 op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0x01,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.h, s105
+// GFX11: v_sqrt_f16_e64 v255.h, s105 op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0x69,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.h, src_scc
+// GFX11: v_sqrt_f16_e64 v255.h, src_scc op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0xfd,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.h, ttmp15
+// GFX11: v_sqrt_f16_e64 v255.h, ttmp15 op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0x7b,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.h, v1.h
+// GFX11: v_sqrt_f16_e64 v255.h, v1.h op_sel:[1,1] ; encoding: [0xff,0x48,0xd5,0xd5,0x01,0x01,0x00,0x00]
+
+v_sqrt_f16 v255.h, v1.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_sqrt_f16_e64_dpp v255.h, v1.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xd5,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+
+v_sqrt_f16 v255.h, v1.h quad_perm:[3,2,1,0]
+// GFX11: v_sqrt_f16_e64_dpp v255.h, v1.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+
+v_sqrt_f16 v255.h, v127.h
+// GFX11: v_sqrt_f16_e64 v255.h, v127.h op_sel:[1,1] ; encoding: [0xff,0x48,0xd5,0xd5,0x7f,0x01,0x00,0x00]
+
+v_sqrt_f16 v255.h, v127.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_sqrt_f16_e64_dpp v255.h, v127.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xd5,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+
+v_sqrt_f16 v255.h, v127.h quad_perm:[3,2,1,0]
+// GFX11: v_sqrt_f16_e64_dpp v255.h, v127.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+
+v_sqrt_f16 v255.h, vcc_hi
+// GFX11: v_sqrt_f16_e64 v255.h, vcc_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0x6b,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.h, vcc_lo
+// GFX11: v_sqrt_f16_e64 v255.h, vcc_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0x6a,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.l, -1
+// GFX11: v_sqrt_f16_e64 v255.l, -1 ; encoding: [0xff,0x00,0xd5,0xd5,0xc1,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.l, 0.5
+// GFX11: v_sqrt_f16_e64 v255.l, 0.5 ; encoding: [0xff,0x00,0xd5,0xd5,0xf0,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.l, exec_hi
+// GFX11: v_sqrt_f16_e64 v255.l, exec_hi ; encoding: [0xff,0x00,0xd5,0xd5,0x7f,0x00,0x00,0x00]
-v_sqrt_f16 v255, -1
-// GFX11: v_sqrt_f16_e64 v255, -1 ; encoding: [0xff,0x00,0xd5,0xd5,0xc1,0x00,0x00,0x00]
+v_sqrt_f16 v255.l, exec_lo
+// GFX11: v_sqrt_f16_e64 v255.l, exec_lo ; encoding: [0xff,0x00,0xd5,0xd5,0x7e,0x00,0x00,0x00]
-v_sqrt_f16 v255, 0.5
-// GFX11: v_sqrt_f16_e64 v255, 0.5 ; encoding: [0xff,0x00,0xd5,0xd5,0xf0,0x00,0x00,0x00]
+v_sqrt_f16 v255.l, m0
+// GFX11: v_sqrt_f16_e64 v255.l, m0 ; encoding: [0xff,0x00,0xd5,0xd5,0x7d,0x00,0x00,0x00]
-v_sqrt_f16 v255, exec_hi
-// GFX11: v_sqrt_f16_e64 v255, exec_hi ; encoding: [0xff,0x00,0xd5,0xd5,0x7f,0x00,0x00,0x00]
+v_sqrt_f16 v255.l, null
+// GFX11: v_sqrt_f16_e64 v255.l, null ; encoding: [0xff,0x00,0xd5,0xd5,0x7c,0x00,0x00,0x00]
-v_sqrt_f16 v255, exec_lo
-// GFX11: v_sqrt_f16_e64 v255, exec_lo ; encoding: [0xff,0x00,0xd5,0xd5,0x7e,0x00,0x00,0x00]
+v_sqrt_f16 v255.l, s1
+// GFX11: v_sqrt_f16_e64 v255.l, s1 ; encoding: [0xff,0x00,0xd5,0xd5,0x01,0x00,0x00,0x00]
-v_sqrt_f16 v255, m0
-// GFX11: v_sqrt_f16_e64 v255, m0 ; encoding: [0xff,0x00,0xd5,0xd5,0x7d,0x00,0x00,0x00]
+v_sqrt_f16 v255.l, s105
+// GFX11: v_sqrt_f16_e64 v255.l, s105 ; encoding: [0xff,0x00,0xd5,0xd5,0x69,0x00,0x00,0x00]
-v_sqrt_f16 v255, null
-// GFX11: v_sqrt_f16_e64 v255, null ; encoding: [0xff,0x00,0xd5,0xd5,0x7c,0x00,0x00,0x00]
+v_sqrt_f16 v255.l, src_scc
+// GFX11: v_sqrt_f16_e64 v255.l, src_scc ; encoding: [0xff,0x00,0xd5,0xd5,0xfd,0x00,0x00,0x00]
-v_sqrt_f16 v255, s1
-// GFX11: v_sqrt_f16_e64 v255, s1 ; encoding: [0xff,0x00,0xd5,0xd5,0x01,0x00,0x00,0x00]
+v_sqrt_f16 v255.l, ttmp15
+// GFX11: v_sqrt_f16_e64 v255.l, ttmp15 ; encoding: [0xff,0x00,0xd5,0xd5,0x7b,0x00,0x00,0x00]
-v_sqrt_f16 v255, s105
-// GFX11: v_sqrt_f16_e64 v255, s105 ; encoding: [0xff,0x00,0xd5,0xd5,0x69,0x00,0x00,0x00]
+v_sqrt_f16 v255.l, v1.l
+// GFX11: v_sqrt_f16_e64 v255.l, v1.l ; encoding: [0xff,0x00,0xd5,0xd5,0x01,0x01,0x00,0x00]
-v_sqrt_f16 v255, src_scc
-// GFX11: v_sqrt_f16_e64 v255, src_scc ; encoding: [0xff,0x00,0xd5,0xd5,0xfd,0x00,0x00,0x00]
+v_sqrt_f16 v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_sqrt_f16_e64_dpp v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd5,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
-v_sqrt_f16 v255, ttmp15
-// GFX11: v_sqrt_f16_e64 v255, ttmp15 ; encoding: [0xff,0x00,0xd5,0xd5,0x7b,0x00,0x00,0x00]
+v_sqrt_f16 v255.l, v1.l quad_perm:[3,2,1,0]
+// GFX11: v_sqrt_f16_e64_dpp v255.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
-v_sqrt_f16 v255, v1
-// GFX11: v_sqrt_f16_e64 v255, v1 ; encoding: [0xff,0x00,0xd5,0xd5,0x01,0x01,0x00,0x00]
+v_sqrt_f16 v255.l, v127.l
+// GFX11: v_sqrt_f16_e64 v255.l, v127.l ; encoding: [0xff,0x00,0xd5,0xd5,0x7f,0x01,0x00,0x00]
-v_sqrt_f16 v255, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_sqrt_f16_e64_dpp v255, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd5,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+v_sqrt_f16 v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_sqrt_f16_e64_dpp v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd5,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
-v_sqrt_f16 v255, v1 quad_perm:[3,2,1,0]
-// GFX11: v_sqrt_f16_e64_dpp v255, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+v_sqrt_f16 v255.l, v127.l quad_perm:[3,2,1,0]
+// GFX11: v_sqrt_f16_e64_dpp v255.l, v127.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
-v_sqrt_f16 v255, v127
-// GFX11: v_sqrt_f16_e64 v255, v127 ; encoding: [0xff,0x00,0xd5,0xd5,0x7f,0x01,0x00,0x00]
+v_sqrt_f16 v255.l, vcc_hi
+// GFX11: v_sqrt_f16_e64 v255.l, vcc_hi ; encoding: [0xff,0x00,0xd5,0xd5,0x6b,0x00,0x00,0x00]
-v_sqrt_f16 v255, v127 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_sqrt_f16_e64_dpp v255, v127 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd5,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+v_sqrt_f16 v255.l, vcc_lo
+// GFX11: v_sqrt_f16_e64 v255.l, vcc_lo ; encoding: [0xff,0x00,0xd5,0xd5,0x6a,0x00,0x00,0x00]
-v_sqrt_f16 v255, v127 quad_perm:[3,2,1,0]
-// GFX11: v_sqrt_f16_e64_dpp v255, v127 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+v_sqrt_f16 v5.h, v199.h
+// GFX11: v_sqrt_f16_e64 v5.h, v199.h op_sel:[1,1] ; encoding: [0x05,0x48,0xd5,0xd5,0xc7,0x01,0x00,0x00]
-v_sqrt_f16 v255, vcc_hi
-// GFX11: v_sqrt_f16_e64 v255, vcc_hi ; encoding: [0xff,0x00,0xd5,0xd5,0x6b,0x00,0x00,0x00]
+v_sqrt_f16 v5.h, v199.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_sqrt_f16_e64_dpp v5.h, v199.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xd5,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
-v_sqrt_f16 v255, vcc_lo
-// GFX11: v_sqrt_f16_e64 v255, vcc_lo ; encoding: [0xff,0x00,0xd5,0xd5,0x6a,0x00,0x00,0x00]
+v_sqrt_f16 v5.h, v199.h quad_perm:[3,2,1,0]
+// GFX11: v_sqrt_f16_e64_dpp v5.h, v199.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x48,0xd5,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
-v_sqrt_f16 v5, v199
-// GFX11: v_sqrt_f16_e64 v5, v199 ; encoding: [0x05,0x00,0xd5,0xd5,0xc7,0x01,0x00,0x00]
+v_sqrt_f16 v5.l, v199.l
+// GFX11: v_sqrt_f16_e64 v5.l, v199.l ; encoding: [0x05,0x00,0xd5,0xd5,0xc7,0x01,0x00,0x00]
-v_sqrt_f16 v5, v199 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_sqrt_f16_e64_dpp v5, v199 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd5,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+v_sqrt_f16 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX11: v_sqrt_f16_e64_dpp v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd5,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
-v_sqrt_f16 v5, v199 quad_perm:[3,2,1,0]
-// GFX11: v_sqrt_f16_e64_dpp v5, v199 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+v_sqrt_f16 v5.l, v199.l quad_perm:[3,2,1,0]
+// GFX11: v_sqrt_f16_e64_dpp v5.l, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
v_trunc_f16 v128.h, 0xfe0b
// GFX11: v_trunc_f16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xdd,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
diff --git a/llvm/test/MC/AMDGPU/gfx11_asm_vop3_from_vop1.s b/llvm/test/MC/AMDGPU/gfx11_asm_vop3_from_vop1.s
index f730487157ebc..ab3287fdb03d1 100644
--- a/llvm/test/MC/AMDGPU/gfx11_asm_vop3_from_vop1.s
+++ b/llvm/test/MC/AMDGPU/gfx11_asm_vop3_from_vop1.s
@@ -46,50 +46,59 @@ v_bfrev_b32_e64 v5, src_scc
v_bfrev_b32_e64 v255, 0xaf123456
// GFX11: v_bfrev_b32_e64 v255, 0xaf123456 ; encoding: [0xff,0x00,0xb8,0xd5,0xff,0x00,0x00,0x00,0x56,0x34,0x12,0xaf]
-v_ceil_f16_e64 v5, v1
-// GFX11: v_ceil_f16_e64 v5, v1 ; encoding: [0x05,0x00,0xdc,0xd5,0x01,0x01,0x00,0x00]
+v_ceil_f16_e64 v5.l, v1.l
+// GFX11: v_ceil_f16_e64 v5.l, v1.l ; encoding: [0x05,0x00,0xdc,0xd5,0x01,0x01,0x00,0x00]
-v_ceil_f16_e64 v5, v255
-// GFX11: v_ceil_f16_e64 v5, v255 ; encoding: [0x05,0x00,0xdc,0xd5,0xff,0x01,0x00,0x00]
+v_ceil_f16_e64 v5.l, v255.l
+// GFX11: v_ceil_f16_e64 v5.l, v255.l ; encoding: [0x05,0x00,0xdc,0xd5,0xff,0x01,0x00,0x00]
-v_ceil_f16_e64 v5, s1
-// GFX11: v_ceil_f16_e64 v5, s1 ; encoding: [0x05,0x00,0xdc,0xd5,0x01,0x00,0x00,0x00]
+v_ceil_f16_e64 v5.l, s1
+// GFX11: v_ceil_f16_e64 v5.l, s1 ; encoding: [0x05,0x00,0xdc,0xd5,0x01,0x00,0x00,0x00]
-v_ceil_f16_e64 v5, s105
-// GFX11: v_ceil_f16_e64 v5, s105 ; encoding: [0x05,0x00,0xdc,0xd5,0x69,0x00,0x00,0x00]
+v_ceil_f16_e64 v5.l, s105
+// GFX11: v_ceil_f16_e64 v5.l, s105 ; encoding: [0x05,0x00,0xdc,0xd5,0x69,0x00,0x00,0x00]
-v_ceil_f16_e64 v5, vcc_lo
-// GFX11: v_ceil_f16_e64 v5, vcc_lo ; encoding: [0x05,0x00,0xdc,0xd5,0x6a,0x00,0x00,0x00]
+v_ceil_f16_e64 v5.l, vcc_lo
+// GFX11: v_ceil_f16_e64 v5.l, vcc_lo ; encoding: [0x05,0x00,0xdc,0xd5,0x6a,0x00,0x00,0x00]
-v_ceil_f16_e64 v5, vcc_hi
-// GFX11: v_ceil_f16_e64 v5, vcc_hi ; encoding: [0x05,0x00,0xdc,0xd5,0x6b,0x00,0x00,0x00]
+v_ceil_f16_e64 v5.l, vcc_hi
+// GFX11: v_ceil_f16_e64 v5.l, vcc_hi ; encoding: [0x05,0x00,0xdc,0xd5,0x6b,0x00,0x00,0x00]
-v_ceil_f16_e64 v5, ttmp15
-// GFX11: v_ceil_f16_e64 v5, ttmp15 ; encoding: [0x05,0x00,0xdc,0xd5,0x7b,0x00,0x00,0x00]
+v_ceil_f16_e64 v5.l, ttmp15
+// GFX11: v_ceil_f16_e64 v5.l, ttmp15 ; encoding: [0x05,0x00,0xdc,0xd5,0x7b,0x00,0x00,0x00]
-v_ceil_f16_e64 v5, m0
-// GFX11: v_ceil_f16_e64 v5, m0 ; encoding: [0x05,0x00,0xdc,0xd5,0x7d,0x00,0x00,0x00]
+v_ceil_f16_e64 v5.l, m0
+// GFX11: v_ceil_f16_e64 v5.l, m0 ; encoding: [0x05,0x00,0xdc,0xd5,0x7d,0x00,0x00,0x00]
-v_ceil_f16_e64 v5, exec_lo
-// GFX11: v_ceil_f16_e64 v5, exec_lo ; encoding: [0x05,0x00,0xdc,0xd5,0x7e,0x00,0x00,0x00]
+v_ceil_f16_e64 v5.l, exec_lo
+// GFX11: v_ceil_f16_e64 v5.l, exec_lo ; encoding: [0x05,0x00,0xdc,0xd5,0x7e,0x00,0x00,0x00]
-v_ceil_f16_e64 v5, exec_hi
-// GFX11: v_ceil_f16_e64 v5, exec_hi ; encoding: [0x05,0x00,0xdc,0xd5,0x7f,0x00,0x00,0x00]
+v_ceil_f16_e64 v5.l, exec_hi
+// GFX11: v_ceil_f16_e64 v5.l, exec_hi ; encoding: [0x05,0x00,0xdc,0xd5,0x7f,0x00,0x00,0x00]
-v_ceil_f16_e64 v5, null
-// GFX11: v_ceil_f16_e64 v5, null ; encoding: [0x05,0x00,0xdc,0xd5,0x7c,0x00,0x00,0x00]
+v_ceil_f16_e64 v5.l, null
+// GFX11: v_ceil_f16_e64 v5.l, null ; encoding: [0x05,0x00,0xdc,0xd5,0x7c,0x00,0x00,0x00]
-v_ceil_f16_e64 v5, -1
-// GFX11: v_ceil_f16_e64 v5, -1 ; encoding: [0x05,0x00,0xdc,0xd5,0xc1,0x00,0x00,0x00]
+v_ceil_f16_e64 v5.l, -1
+// GFX11: v_ceil_f16_e64 v5.l, -1 ; encoding: [0x05,0x00,0xdc,0xd5,0xc1,0x00,0x00,0x00]
-v_ceil_f16_e64 v5, 0.5 mul:2
-// GFX11: v_ceil_f16_e64 v5, 0.5 mul:2 ; encoding: [0x05,0x00,0xdc,0xd5,0xf0,0x00,0x00,0x08]
+v_ceil_f16_e64 v5.l, 0.5 mul:2
+// GFX11: v_ceil_f16_e64 v5.l, 0.5 mul:2 ; encoding: [0x05,0x00,0xdc,0xd5,0xf0,0x00,0x00,0x08]
-v_ceil_f16_e64 v5, src_scc mul:4
-// GFX11: v_ceil_f16_e64 v5, src_scc mul:4 ; encoding: [0x05,0x00,0xdc,0xd5,0xfd,0x00,0x00,0x10]
+v_ceil_f16_e64 v5.l, src_scc mul:4
+// GFX11: v_ceil_f16_e64 v5.l, src_scc mul:4 ; encoding: [0x05,0x00,0xdc,0xd5,0xfd,0x00,0x00,0x10]
-v_ceil_f16_e64 v255, -|0xfe0b| clamp div:2
-// GFX11: v_ceil_f16_e64 v255, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xdc,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
+v_ceil_f16_e64 v255.l, -|0xfe0b| clamp div:2
+// GFX11: v_ceil_f16_e64 v255.l, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xdc,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
+
+v_ceil_f16_e64 v5.h, v1.h
+// GFX11: v_ceil_f16_e64 v5.h, v1.h op_sel:[1,1] ; encoding: [0x05,0x48,0xdc,0xd5,0x01,0x01,0x00,0x00]
+
+v_ceil_f16_e64 v5.l, v255.h
+// GFX11: v_ceil_f16_e64 v5.l, v255.h op_sel:[1,0] ; encoding: [0x05,0x08,0xdc,0xd5,0xff,0x01,0x00,0x00]
+
+v_ceil_f16_e64 v255.h, -|0xfe0b| clamp div:2
+// GFX11: v_ceil_f16_e64 v255.h, -|0xfe0b| op_sel:[0,1] clamp div:2 ; encoding: [0xff,0xc1,0xdc,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
v_ceil_f32_e64 v5, v1
// GFX11: v_ceil_f32_e64 v5, v1 ; encoding: [0x05,0x00,0xa2,0xd5,0x01,0x01,0x00,0x00]
@@ -1750,50 +1759,59 @@ v_cvt_u32_u16_e64 v255, 0xfe0b
v_cvt_u32_u16_e64 v5, v255.h
// GFX11: v_cvt_u32_u16_e64 v5, v255.h op_sel:[1,0] ; encoding: [0x05,0x08,0xeb,0xd5,0xff,0x01,0x00,0x00]
-v_exp_f16_e64 v5, v1
-// GFX11: v_exp_f16_e64 v5, v1 ; encoding: [0x05,0x00,0xd8,0xd5,0x01,0x01,0x00,0x00]
+v_exp_f16_e64 v5.l, v1.l
+// GFX11: v_exp_f16_e64 v5.l, v1.l ; encoding: [0x05,0x00,0xd8,0xd5,0x01,0x01,0x00,0x00]
+
+v_exp_f16_e64 v5.l, v255.l
+// GFX11: v_exp_f16_e64 v5.l, v255.l ; encoding: [0x05,0x00,0xd8,0xd5,0xff,0x01,0x00,0x00]
+
+v_exp_f16_e64 v5.l, s1
+// GFX11: v_exp_f16_e64 v5.l, s1 ; encoding: [0x05,0x00,0xd8,0xd5,0x01,0x00,0x00,0x00]
+
+v_exp_f16_e64 v5.l, s105
+// GFX11: v_exp_f16_e64 v5.l, s105 ; encoding: [0x05,0x00,0xd8,0xd5,0x69,0x00,0x00,0x00]
-v_exp_f16_e64 v5, v255
-// GFX11: v_exp_f16_e64 v5, v255 ; encoding: [0x05,0x00,0xd8,0xd5,0xff,0x01,0x00,0x00]
+v_exp_f16_e64 v5.l, vcc_lo
+// GFX11: v_exp_f16_e64 v5.l, vcc_lo ; encoding: [0x05,0x00,0xd8,0xd5,0x6a,0x00,0x00,0x00]
-v_exp_f16_e64 v5, s1
-// GFX11: v_exp_f16_e64 v5, s1 ; encoding: [0x05,0x00,0xd8,0xd5,0x01,0x00,0x00,0x00]
+v_exp_f16_e64 v5.l, vcc_hi
+// GFX11: v_exp_f16_e64 v5.l, vcc_hi ; encoding: [0x05,0x00,0xd8,0xd5,0x6b,0x00,0x00,0x00]
-v_exp_f16_e64 v5, s105
-// GFX11: v_exp_f16_e64 v5, s105 ; encoding: [0x05,0x00,0xd8,0xd5,0x69,0x00,0x00,0x00]
+v_exp_f16_e64 v5.l, ttmp15
+// GFX11: v_exp_f16_e64 v5.l, ttmp15 ; encoding: [0x05,0x00,0xd8,0xd5,0x7b,0x00,0x00,0x00]
-v_exp_f16_e64 v5, vcc_lo
-// GFX11: v_exp_f16_e64 v5, vcc_lo ; encoding: [0x05,0x00,0xd8,0xd5,0x6a,0x00,0x00,0x00]
+v_exp_f16_e64 v5.l, m0
+// GFX11: v_exp_f16_e64 v5.l, m0 ; encoding: [0x05,0x00,0xd8,0xd5,0x7d,0x00,0x00,0x00]
-v_exp_f16_e64 v5, vcc_hi
-// GFX11: v_exp_f16_e64 v5, vcc_hi ; encoding: [0x05,0x00,0xd8,0xd5,0x6b,0x00,0x00,0x00]
+v_exp_f16_e64 v5.l, exec_lo
+// GFX11: v_exp_f16_e64 v5.l, exec_lo ; encoding: [0x05,0x00,0xd8,0xd5,0x7e,0x00,0x00,0x00]
-v_exp_f16_e64 v5, ttmp15
-// GFX11: v_exp_f16_e64 v5, ttmp15 ; encoding: [0x05,0x00,0xd8,0xd5,0x7b,0x00,0x00,0x00]
+v_exp_f16_e64 v5.l, exec_hi
+// GFX11: v_exp_f16_e64 v5.l, exec_hi ; encoding: [0x05,0x00,0xd8,0xd5,0x7f,0x00,0x00,0x00]
-v_exp_f16_e64 v5, m0
-// GFX11: v_exp_f16_e64 v5, m0 ; encoding: [0x05,0x00,0xd8,0xd5,0x7d,0x00,0x00,0x00]
+v_exp_f16_e64 v5.l, null
+// GFX11: v_exp_f16_e64 v5.l, null ; encoding: [0x05,0x00,0xd8,0xd5,0x7c,0x00,0x00,0x00]
-v_exp_f16_e64 v5, exec_lo
-// GFX11: v_exp_f16_e64 v5, exec_lo ; encoding: [0x05,0x00,0xd8,0xd5,0x7e,0x00,0x00,0x00]
+v_exp_f16_e64 v5.l, -1
+// GFX11: v_exp_f16_e64 v5.l, -1 ; encoding: [0x05,0x00,0xd8,0xd5,0xc1,0x00,0x00,0x00]
-v_exp_f16_e64 v5, exec_hi
-// GFX11: v_exp_f16_e64 v5, exec_hi ; encoding: [0x05,0x00,0xd8,0xd5,0x7f,0x00,0x00,0x00]
+v_exp_f16_e64 v5.l, 0.5 mul:2
+// GFX11: v_exp_f16_e64 v5.l, 0.5 mul:2 ; encoding: [0x05,0x00,0xd8,0xd5,0xf0,0x00,0x00,0x08]
-v_exp_f16_e64 v5, null
-// GFX11: v_exp_f16_e64 v5, null ; encoding: [0x05,0x00,0xd8,0xd5,0x7c,0x00,0x00,0x00]
+v_exp_f16_e64 v5.l, src_scc mul:4
+// GFX11: v_exp_f16_e64 v5.l, src_scc mul:4 ; encoding: [0x05,0x00,0xd8,0xd5,0xfd,0x00,0x00,0x10]
-v_exp_f16_e64 v5, -1
-// GFX11: v_exp_f16_e64 v5, -1 ; encoding: [0x05,0x00,0xd8,0xd5,0xc1,0x00,0x00,0x00]
+v_exp_f16_e64 v255.l, -|0xfe0b| clamp div:2
+// GFX11: v_exp_f16_e64 v255.l, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xd8,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
-v_exp_f16_e64 v5, 0.5 mul:2
-// GFX11: v_exp_f16_e64 v5, 0.5 mul:2 ; encoding: [0x05,0x00,0xd8,0xd5,0xf0,0x00,0x00,0x08]
+v_exp_f16_e64 v5.h, v1.h
+// GFX11: v_exp_f16_e64 v5.h, v1.h op_sel:[1,1] ; encoding: [0x05,0x48,0xd8,0xd5,0x01,0x01,0x00,0x00]
-v_exp_f16_e64 v5, src_scc mul:4
-// GFX11: v_exp_f16_e64 v5, src_scc mul:4 ; encoding: [0x05,0x00,0xd8,0xd5,0xfd,0x00,0x00,0x10]
+v_exp_f16_e64 v5.l, v255.h
+// GFX11: v_exp_f16_e64 v5.l, v255.h op_sel:[1,0] ; encoding: [0x05,0x08,0xd8,0xd5,0xff,0x01,0x00,0x00]
-v_exp_f16_e64 v255, -|0xfe0b| clamp div:2
-// GFX11: v_exp_f16_e64 v255, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xd8,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
+v_exp_f16_e64 v255.h, -|0xfe0b| clamp div:2
+// GFX11: v_exp_f16_e64 v255.h, -|0xfe0b| op_sel:[0,1] clamp div:2 ; encoding: [0xff,0xc1,0xd8,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
v_exp_f32_e64 v5, v1
// GFX11: v_exp_f32_e64 v5, v1 ; encoding: [0x05,0x00,0xa5,0xd5,0x01,0x01,0x00,0x00]
@@ -1975,50 +1993,59 @@ v_ffbl_b32_e64 v5, src_scc
v_ffbl_b32_e64 v255, 0xaf123456
// GFX11: v_ctz_i32_b32_e64 v255, 0xaf123456 ; encoding: [0xff,0x00,0xba,0xd5,0xff,0x00,0x00,0x00,0x56,0x34,0x12,0xaf]
-v_floor_f16_e64 v5, v1
-// GFX11: v_floor_f16_e64 v5, v1 ; encoding: [0x05,0x00,0xdb,0xd5,0x01,0x01,0x00,0x00]
+v_floor_f16_e64 v5.l, v1.l
+// GFX11: v_floor_f16_e64 v5.l, v1.l ; encoding: [0x05,0x00,0xdb,0xd5,0x01,0x01,0x00,0x00]
-v_floor_f16_e64 v5, v255
-// GFX11: v_floor_f16_e64 v5, v255 ; encoding: [0x05,0x00,0xdb,0xd5,0xff,0x01,0x00,0x00]
+v_floor_f16_e64 v5.l, v255.l
+// GFX11: v_floor_f16_e64 v5.l, v255.l ; encoding: [0x05,0x00,0xdb,0xd5,0xff,0x01,0x00,0x00]
-v_floor_f16_e64 v5, s1
-// GFX11: v_floor_f16_e64 v5, s1 ; encoding: [0x05,0x00,0xdb,0xd5,0x01,0x00,0x00,0x00]
+v_floor_f16_e64 v5.l, s1
+// GFX11: v_floor_f16_e64 v5.l, s1 ; encoding: [0x05,0x00,0xdb,0xd5,0x01,0x00,0x00,0x00]
-v_floor_f16_e64 v5, s105
-// GFX11: v_floor_f16_e64 v5, s105 ; encoding: [0x05,0x00,0xdb,0xd5,0x69,0x00,0x00,0x00]
+v_floor_f16_e64 v5.l, s105
+// GFX11: v_floor_f16_e64 v5.l, s105 ; encoding: [0x05,0x00,0xdb,0xd5,0x69,0x00,0x00,0x00]
-v_floor_f16_e64 v5, vcc_lo
-// GFX11: v_floor_f16_e64 v5, vcc_lo ; encoding: [0x05,0x00,0xdb,0xd5,0x6a,0x00,0x00,0x00]
+v_floor_f16_e64 v5.l, vcc_lo
+// GFX11: v_floor_f16_e64 v5.l, vcc_lo ; encoding: [0x05,0x00,0xdb,0xd5,0x6a,0x00,0x00,0x00]
-v_floor_f16_e64 v5, vcc_hi
-// GFX11: v_floor_f16_e64 v5, vcc_hi ; encoding: [0x05,0x00,0xdb,0xd5,0x6b,0x00,0x00,0x00]
+v_floor_f16_e64 v5.l, vcc_hi
+// GFX11: v_floor_f16_e64 v5.l, vcc_hi ; encoding: [0x05,0x00,0xdb,0xd5,0x6b,0x00,0x00,0x00]
-v_floor_f16_e64 v5, ttmp15
-// GFX11: v_floor_f16_e64 v5, ttmp15 ; encoding: [0x05,0x00,0xdb,0xd5,0x7b,0x00,0x00,0x00]
+v_floor_f16_e64 v5.l, ttmp15
+// GFX11: v_floor_f16_e64 v5.l, ttmp15 ; encoding: [0x05,0x00,0xdb,0xd5,0x7b,0x00,0x00,0x00]
-v_floor_f16_e64 v5, m0
-// GFX11: v_floor_f16_e64 v5, m0 ; encoding: [0x05,0x00,0xdb,0xd5,0x7d,0x00,0x00,0x00]
+v_floor_f16_e64 v5.l, m0
+// GFX11: v_floor_f16_e64 v5.l, m0 ; encoding: [0x05,0x00,0xdb,0xd5,0x7d,0x00,0x00,0x00]
-v_floor_f16_e64 v5, exec_lo
-// GFX11: v_floor_f16_e64 v5, exec_lo ; encoding: [0x05,0x00,0xdb,0xd5,0x7e,0x00,0x00,0x00]
+v_floor_f16_e64 v5.l, exec_lo
+// GFX11: v_floor_f16_e64 v5.l, exec_lo ; encoding: [0x05,0x00,0xdb,0xd5,0x7e,0x00,0x00,0x00]
-v_floor_f16_e64 v5, exec_hi
-// GFX11: v_floor_f16_e64 v5, exec_hi ; encoding: [0x05,0x00,0xdb,0xd5,0x7f,0x00,0x00,0x00]
+v_floor_f16_e64 v5.l, exec_hi
+// GFX11: v_floor_f16_e64 v5.l, exec_hi ; encoding: [0x05,0x00,0xdb,0xd5,0x7f,0x00,0x00,0x00]
-v_floor_f16_e64 v5, null
-// GFX11: v_floor_f16_e64 v5, null ; encoding: [0x05,0x00,0xdb,0xd5,0x7c,0x00,0x00,0x00]
+v_floor_f16_e64 v5.l, null
+// GFX11: v_floor_f16_e64 v5.l, null ; encoding: [0x05,0x00,0xdb,0xd5,0x7c,0x00,0x00,0x00]
-v_floor_f16_e64 v5, -1
-// GFX11: v_floor_f16_e64 v5, -1 ; encoding: [0x05,0x00,0xdb,0xd5,0xc1,0x00,0x00,0x00]
+v_floor_f16_e64 v5.l, -1
+// GFX11: v_floor_f16_e64 v5.l, -1 ; encoding: [0x05,0x00,0xdb,0xd5,0xc1,0x00,0x00,0x00]
-v_floor_f16_e64 v5, 0.5 mul:2
-// GFX11: v_floor_f16_e64 v5, 0.5 mul:2 ; encoding: [0x05,0x00,0xdb,0xd5,0xf0,0x00,0x00,0x08]
+v_floor_f16_e64 v5.l, 0.5 mul:2
+// GFX11: v_floor_f16_e64 v5.l, 0.5 mul:2 ; encoding: [0x05,0x00,0xdb,0xd5,0xf0,0x00,0x00,0x08]
-v_floor_f16_e64 v5, src_scc mul:4
-// GFX11: v_floor_f16_e64 v5, src_scc mul:4 ; encoding: [0x05,0x00,0xdb,0xd5,0xfd,0x00,0x00,0x10]
+v_floor_f16_e64 v5.l, src_scc mul:4
+// GFX11: v_floor_f16_e64 v5.l, src_scc mul:4 ; encoding: [0x05,0x00,0xdb,0xd5,0xfd,0x00,0x00,0x10]
-v_floor_f16_e64 v255, -|0xfe0b| clamp div:2
-// GFX11: v_floor_f16_e64 v255, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xdb,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
+v_floor_f16_e64 v255.l, -|0xfe0b| clamp div:2
+// GFX11: v_floor_f16_e64 v255.l, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xdb,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
+
+v_floor_f16_e64 v5.h, v1.h
+// GFX11: v_floor_f16_e64 v5.h, v1.h op_sel:[1,1] ; encoding: [0x05,0x48,0xdb,0xd5,0x01,0x01,0x00,0x00]
+
+v_floor_f16_e64 v5.l, v255.h
+// GFX11: v_floor_f16_e64 v5.l, v255.h op_sel:[1,0] ; encoding: [0x05,0x08,0xdb,0xd5,0xff,0x01,0x00,0x00]
+
+v_floor_f16_e64 v255.h, -|0xfe0b| clamp div:2
+// GFX11: v_floor_f16_e64 v255.h, -|0xfe0b| op_sel:[0,1] clamp div:2 ; encoding: [0xff,0xc1,0xdb,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
v_floor_f32_e64 v5, v1
// GFX11: v_floor_f32_e64 v5, v1 ; encoding: [0x05,0x00,0xa4,0xd5,0x01,0x01,0x00,0x00]
@@ -2506,50 +2533,59 @@ v_frexp_mant_f64_e64 v[5:6], -|src_scc| mul:4
v_frexp_mant_f64_e64 v[254:255], 0xaf123456 clamp div:2
// GFX11: v_frexp_mant_f64_e64 v[254:255], 0xaf123456 clamp div:2 ; encoding: [0xfe,0x80,0xbd,0xd5,0xff,0x00,0x00,0x18,0x56,0x34,0x12,0xaf]
-v_log_f16_e64 v5, v1
-// GFX11: v_log_f16_e64 v5, v1 ; encoding: [0x05,0x00,0xd7,0xd5,0x01,0x01,0x00,0x00]
+v_log_f16_e64 v5.l, v1.l
+// GFX11: v_log_f16_e64 v5.l, v1.l ; encoding: [0x05,0x00,0xd7,0xd5,0x01,0x01,0x00,0x00]
+
+v_log_f16_e64 v5.l, v255.l
+// GFX11: v_log_f16_e64 v5.l, v255.l ; encoding: [0x05,0x00,0xd7,0xd5,0xff,0x01,0x00,0x00]
+
+v_log_f16_e64 v5.l, s1
+// GFX11: v_log_f16_e64 v5.l, s1 ; encoding: [0x05,0x00,0xd7,0xd5,0x01,0x00,0x00,0x00]
-v_log_f16_e64 v5, v255
-// GFX11: v_log_f16_e64 v5, v255 ; encoding: [0x05,0x00,0xd7,0xd5,0xff,0x01,0x00,0x00]
+v_log_f16_e64 v5.l, s105
+// GFX11: v_log_f16_e64 v5.l, s105 ; encoding: [0x05,0x00,0xd7,0xd5,0x69,0x00,0x00,0x00]
-v_log_f16_e64 v5, s1
-// GFX11: v_log_f16_e64 v5, s1 ; encoding: [0x05,0x00,0xd7,0xd5,0x01,0x00,0x00,0x00]
+v_log_f16_e64 v5.l, vcc_lo
+// GFX11: v_log_f16_e64 v5.l, vcc_lo ; encoding: [0x05,0x00,0xd7,0xd5,0x6a,0x00,0x00,0x00]
-v_log_f16_e64 v5, s105
-// GFX11: v_log_f16_e64 v5, s105 ; encoding: [0x05,0x00,0xd7,0xd5,0x69,0x00,0x00,0x00]
+v_log_f16_e64 v5.l, vcc_hi
+// GFX11: v_log_f16_e64 v5.l, vcc_hi ; encoding: [0x05,0x00,0xd7,0xd5,0x6b,0x00,0x00,0x00]
-v_log_f16_e64 v5, vcc_lo
-// GFX11: v_log_f16_e64 v5, vcc_lo ; encoding: [0x05,0x00,0xd7,0xd5,0x6a,0x00,0x00,0x00]
+v_log_f16_e64 v5.l, ttmp15
+// GFX11: v_log_f16_e64 v5.l, ttmp15 ; encoding: [0x05,0x00,0xd7,0xd5,0x7b,0x00,0x00,0x00]
-v_log_f16_e64 v5, vcc_hi
-// GFX11: v_log_f16_e64 v5, vcc_hi ; encoding: [0x05,0x00,0xd7,0xd5,0x6b,0x00,0x00,0x00]
+v_log_f16_e64 v5.l, m0
+// GFX11: v_log_f16_e64 v5.l, m0 ; encoding: [0x05,0x00,0xd7,0xd5,0x7d,0x00,0x00,0x00]
-v_log_f16_e64 v5, ttmp15
-// GFX11: v_log_f16_e64 v5, ttmp15 ; encoding: [0x05,0x00,0xd7,0xd5,0x7b,0x00,0x00,0x00]
+v_log_f16_e64 v5.l, exec_lo
+// GFX11: v_log_f16_e64 v5.l, exec_lo ; encoding: [0x05,0x00,0xd7,0xd5,0x7e,0x00,0x00,0x00]
-v_log_f16_e64 v5, m0
-// GFX11: v_log_f16_e64 v5, m0 ; encoding: [0x05,0x00,0xd7,0xd5,0x7d,0x00,0x00,0x00]
+v_log_f16_e64 v5.l, exec_hi
+// GFX11: v_log_f16_e64 v5.l, exec_hi ; encoding: [0x05,0x00,0xd7,0xd5,0x7f,0x00,0x00,0x00]
-v_log_f16_e64 v5, exec_lo
-// GFX11: v_log_f16_e64 v5, exec_lo ; encoding: [0x05,0x00,0xd7,0xd5,0x7e,0x00,0x00,0x00]
+v_log_f16_e64 v5.l, null
+// GFX11: v_log_f16_e64 v5.l, null ; encoding: [0x05,0x00,0xd7,0xd5,0x7c,0x00,0x00,0x00]
-v_log_f16_e64 v5, exec_hi
-// GFX11: v_log_f16_e64 v5, exec_hi ; encoding: [0x05,0x00,0xd7,0xd5,0x7f,0x00,0x00,0x00]
+v_log_f16_e64 v5.l, -1
+// GFX11: v_log_f16_e64 v5.l, -1 ; encoding: [0x05,0x00,0xd7,0xd5,0xc1,0x00,0x00,0x00]
-v_log_f16_e64 v5, null
-// GFX11: v_log_f16_e64 v5, null ; encoding: [0x05,0x00,0xd7,0xd5,0x7c,0x00,0x00,0x00]
+v_log_f16_e64 v5.l, 0.5 mul:2
+// GFX11: v_log_f16_e64 v5.l, 0.5 mul:2 ; encoding: [0x05,0x00,0xd7,0xd5,0xf0,0x00,0x00,0x08]
-v_log_f16_e64 v5, -1
-// GFX11: v_log_f16_e64 v5, -1 ; encoding: [0x05,0x00,0xd7,0xd5,0xc1,0x00,0x00,0x00]
+v_log_f16_e64 v5.l, src_scc mul:4
+// GFX11: v_log_f16_e64 v5.l, src_scc mul:4 ; encoding: [0x05,0x00,0xd7,0xd5,0xfd,0x00,0x00,0x10]
-v_log_f16_e64 v5, 0.5 mul:2
-// GFX11: v_log_f16_e64 v5, 0.5 mul:2 ; encoding: [0x05,0x00,0xd7,0xd5,0xf0,0x00,0x00,0x08]
+v_log_f16_e64 v255.l, -|0xfe0b| clamp div:2
+// GFX11: v_log_f16_e64 v255.l, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xd7,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
-v_log_f16_e64 v5, src_scc mul:4
-// GFX11: v_log_f16_e64 v5, src_scc mul:4 ; encoding: [0x05,0x00,0xd7,0xd5,0xfd,0x00,0x00,0x10]
+v_log_f16_e64 v5.h, v1.h
+// GFX11: v_log_f16_e64 v5.h, v1.h op_sel:[1,1] ; encoding: [0x05,0x48,0xd7,0xd5,0x01,0x01,0x00,0x00]
-v_log_f16_e64 v255, -|0xfe0b| clamp div:2
-// GFX11: v_log_f16_e64 v255, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xd7,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
+v_log_f16_e64 v5.l, v255.h
+// GFX11: v_log_f16_e64 v5.l, v255.h op_sel:[1,0] ; encoding: [0x05,0x08,0xd7,0xd5,0xff,0x01,0x00,0x00]
+
+v_log_f16_e64 v255.h, -|0xfe0b| clamp div:2
+// GFX11: v_log_f16_e64 v255.h, -|0xfe0b| op_sel:[0,1] clamp div:2 ; encoding: [0xff,0xc1,0xd7,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
v_log_f32_e64 v5, v1
// GFX11: v_log_f32_e64 v5, v1 ; encoding: [0x05,0x00,0xa7,0xd5,0x01,0x01,0x00,0x00]
@@ -2809,50 +2845,59 @@ v_not_b32_e64 v255, 0xaf123456
v_pipeflush_e64
// GFX11: v_pipeflush ; encoding: [0x00,0x00,0x9b,0xd5,0x00,0x00,0x00,0x00]
-v_rcp_f16_e64 v5, v1
-// GFX11: v_rcp_f16_e64 v5, v1 ; encoding: [0x05,0x00,0xd4,0xd5,0x01,0x01,0x00,0x00]
+v_rcp_f16_e64 v5.l, v1.l
+// GFX11: v_rcp_f16_e64 v5.l, v1.l ; encoding: [0x05,0x00,0xd4,0xd5,0x01,0x01,0x00,0x00]
+
+v_rcp_f16_e64 v5.l, v255.l
+// GFX11: v_rcp_f16_e64 v5.l, v255.l ; encoding: [0x05,0x00,0xd4,0xd5,0xff,0x01,0x00,0x00]
+
+v_rcp_f16_e64 v5.l, s1
+// GFX11: v_rcp_f16_e64 v5.l, s1 ; encoding: [0x05,0x00,0xd4,0xd5,0x01,0x00,0x00,0x00]
-v_rcp_f16_e64 v5, v255
-// GFX11: v_rcp_f16_e64 v5, v255 ; encoding: [0x05,0x00,0xd4,0xd5,0xff,0x01,0x00,0x00]
+v_rcp_f16_e64 v5.l, s105
+// GFX11: v_rcp_f16_e64 v5.l, s105 ; encoding: [0x05,0x00,0xd4,0xd5,0x69,0x00,0x00,0x00]
-v_rcp_f16_e64 v5, s1
-// GFX11: v_rcp_f16_e64 v5, s1 ; encoding: [0x05,0x00,0xd4,0xd5,0x01,0x00,0x00,0x00]
+v_rcp_f16_e64 v5.l, vcc_lo
+// GFX11: v_rcp_f16_e64 v5.l, vcc_lo ; encoding: [0x05,0x00,0xd4,0xd5,0x6a,0x00,0x00,0x00]
-v_rcp_f16_e64 v5, s105
-// GFX11: v_rcp_f16_e64 v5, s105 ; encoding: [0x05,0x00,0xd4,0xd5,0x69,0x00,0x00,0x00]
+v_rcp_f16_e64 v5.l, vcc_hi
+// GFX11: v_rcp_f16_e64 v5.l, vcc_hi ; encoding: [0x05,0x00,0xd4,0xd5,0x6b,0x00,0x00,0x00]
-v_rcp_f16_e64 v5, vcc_lo
-// GFX11: v_rcp_f16_e64 v5, vcc_lo ; encoding: [0x05,0x00,0xd4,0xd5,0x6a,0x00,0x00,0x00]
+v_rcp_f16_e64 v5.l, ttmp15
+// GFX11: v_rcp_f16_e64 v5.l, ttmp15 ; encoding: [0x05,0x00,0xd4,0xd5,0x7b,0x00,0x00,0x00]
-v_rcp_f16_e64 v5, vcc_hi
-// GFX11: v_rcp_f16_e64 v5, vcc_hi ; encoding: [0x05,0x00,0xd4,0xd5,0x6b,0x00,0x00,0x00]
+v_rcp_f16_e64 v5.l, m0
+// GFX11: v_rcp_f16_e64 v5.l, m0 ; encoding: [0x05,0x00,0xd4,0xd5,0x7d,0x00,0x00,0x00]
-v_rcp_f16_e64 v5, ttmp15
-// GFX11: v_rcp_f16_e64 v5, ttmp15 ; encoding: [0x05,0x00,0xd4,0xd5,0x7b,0x00,0x00,0x00]
+v_rcp_f16_e64 v5.l, exec_lo
+// GFX11: v_rcp_f16_e64 v5.l, exec_lo ; encoding: [0x05,0x00,0xd4,0xd5,0x7e,0x00,0x00,0x00]
-v_rcp_f16_e64 v5, m0
-// GFX11: v_rcp_f16_e64 v5, m0 ; encoding: [0x05,0x00,0xd4,0xd5,0x7d,0x00,0x00,0x00]
+v_rcp_f16_e64 v5.l, exec_hi
+// GFX11: v_rcp_f16_e64 v5.l, exec_hi ; encoding: [0x05,0x00,0xd4,0xd5,0x7f,0x00,0x00,0x00]
-v_rcp_f16_e64 v5, exec_lo
-// GFX11: v_rcp_f16_e64 v5, exec_lo ; encoding: [0x05,0x00,0xd4,0xd5,0x7e,0x00,0x00,0x00]
+v_rcp_f16_e64 v5.l, null
+// GFX11: v_rcp_f16_e64 v5.l, null ; encoding: [0x05,0x00,0xd4,0xd5,0x7c,0x00,0x00,0x00]
-v_rcp_f16_e64 v5, exec_hi
-// GFX11: v_rcp_f16_e64 v5, exec_hi ; encoding: [0x05,0x00,0xd4,0xd5,0x7f,0x00,0x00,0x00]
+v_rcp_f16_e64 v5.l, -1
+// GFX11: v_rcp_f16_e64 v5.l, -1 ; encoding: [0x05,0x00,0xd4,0xd5,0xc1,0x00,0x00,0x00]
-v_rcp_f16_e64 v5, null
-// GFX11: v_rcp_f16_e64 v5, null ; encoding: [0x05,0x00,0xd4,0xd5,0x7c,0x00,0x00,0x00]
+v_rcp_f16_e64 v5.l, 0.5 mul:2
+// GFX11: v_rcp_f16_e64 v5.l, 0.5 mul:2 ; encoding: [0x05,0x00,0xd4,0xd5,0xf0,0x00,0x00,0x08]
-v_rcp_f16_e64 v5, -1
-// GFX11: v_rcp_f16_e64 v5, -1 ; encoding: [0x05,0x00,0xd4,0xd5,0xc1,0x00,0x00,0x00]
+v_rcp_f16_e64 v5.l, src_scc mul:4
+// GFX11: v_rcp_f16_e64 v5.l, src_scc mul:4 ; encoding: [0x05,0x00,0xd4,0xd5,0xfd,0x00,0x00,0x10]
-v_rcp_f16_e64 v5, 0.5 mul:2
-// GFX11: v_rcp_f16_e64 v5, 0.5 mul:2 ; encoding: [0x05,0x00,0xd4,0xd5,0xf0,0x00,0x00,0x08]
+v_rcp_f16_e64 v255.l, -|0xfe0b| clamp div:2
+// GFX11: v_rcp_f16_e64 v255.l, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xd4,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
-v_rcp_f16_e64 v5, src_scc mul:4
-// GFX11: v_rcp_f16_e64 v5, src_scc mul:4 ; encoding: [0x05,0x00,0xd4,0xd5,0xfd,0x00,0x00,0x10]
+v_rcp_f16_e64 v5.h, v1.h
+// GFX11: v_rcp_f16_e64 v5.h, v1.h op_sel:[1,1] ; encoding: [0x05,0x48,0xd4,0xd5,0x01,0x01,0x00,0x00]
-v_rcp_f16_e64 v255, -|0xfe0b| clamp div:2
-// GFX11: v_rcp_f16_e64 v255, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xd4,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
+v_rcp_f16_e64 v5.l, v255.h
+// GFX11: v_rcp_f16_e64 v5.l, v255.h op_sel:[1,0] ; encoding: [0x05,0x08,0xd4,0xd5,0xff,0x01,0x00,0x00]
+
+v_rcp_f16_e64 v255.h, -|0xfe0b| clamp div:2
+// GFX11: v_rcp_f16_e64 v255.h, -|0xfe0b| op_sel:[0,1] clamp div:2 ; encoding: [0xff,0xc1,0xd4,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
v_rcp_f32_e64 v5, v1
// GFX11: v_rcp_f32_e64 v5, v1 ; encoding: [0x05,0x00,0xaa,0xd5,0x01,0x01,0x00,0x00]
@@ -3115,50 +3160,59 @@ v_rndne_f64_e64 v[5:6], -|src_scc| mul:4
v_rndne_f64_e64 v[254:255], 0xaf123456 clamp div:2
// GFX11: v_rndne_f64_e64 v[254:255], 0xaf123456 clamp div:2 ; encoding: [0xfe,0x80,0x99,0xd5,0xff,0x00,0x00,0x18,0x56,0x34,0x12,0xaf]
-v_rsq_f16_e64 v5, v1
-// GFX11: v_rsq_f16_e64 v5, v1 ; encoding: [0x05,0x00,0xd6,0xd5,0x01,0x01,0x00,0x00]
+v_rsq_f16_e64 v5.l, v1.l
+// GFX11: v_rsq_f16_e64 v5.l, v1.l ; encoding: [0x05,0x00,0xd6,0xd5,0x01,0x01,0x00,0x00]
+
+v_rsq_f16_e64 v5.l, v255.l
+// GFX11: v_rsq_f16_e64 v5.l, v255.l ; encoding: [0x05,0x00,0xd6,0xd5,0xff,0x01,0x00,0x00]
-v_rsq_f16_e64 v5, v255
-// GFX11: v_rsq_f16_e64 v5, v255 ; encoding: [0x05,0x00,0xd6,0xd5,0xff,0x01,0x00,0x00]
+v_rsq_f16_e64 v5.l, s1
+// GFX11: v_rsq_f16_e64 v5.l, s1 ; encoding: [0x05,0x00,0xd6,0xd5,0x01,0x00,0x00,0x00]
-v_rsq_f16_e64 v5, s1
-// GFX11: v_rsq_f16_e64 v5, s1 ; encoding: [0x05,0x00,0xd6,0xd5,0x01,0x00,0x00,0x00]
+v_rsq_f16_e64 v5.l, s105
+// GFX11: v_rsq_f16_e64 v5.l, s105 ; encoding: [0x05,0x00,0xd6,0xd5,0x69,0x00,0x00,0x00]
-v_rsq_f16_e64 v5, s105
-// GFX11: v_rsq_f16_e64 v5, s105 ; encoding: [0x05,0x00,0xd6,0xd5,0x69,0x00,0x00,0x00]
+v_rsq_f16_e64 v5.l, vcc_lo
+// GFX11: v_rsq_f16_e64 v5.l, vcc_lo ; encoding: [0x05,0x00,0xd6,0xd5,0x6a,0x00,0x00,0x00]
-v_rsq_f16_e64 v5, vcc_lo
-// GFX11: v_rsq_f16_e64 v5, vcc_lo ; encoding: [0x05,0x00,0xd6,0xd5,0x6a,0x00,0x00,0x00]
+v_rsq_f16_e64 v5.l, vcc_hi
+// GFX11: v_rsq_f16_e64 v5.l, vcc_hi ; encoding: [0x05,0x00,0xd6,0xd5,0x6b,0x00,0x00,0x00]
-v_rsq_f16_e64 v5, vcc_hi
-// GFX11: v_rsq_f16_e64 v5, vcc_hi ; encoding: [0x05,0x00,0xd6,0xd5,0x6b,0x00,0x00,0x00]
+v_rsq_f16_e64 v5.l, ttmp15
+// GFX11: v_rsq_f16_e64 v5.l, ttmp15 ; encoding: [0x05,0x00,0xd6,0xd5,0x7b,0x00,0x00,0x00]
-v_rsq_f16_e64 v5, ttmp15
-// GFX11: v_rsq_f16_e64 v5, ttmp15 ; encoding: [0x05,0x00,0xd6,0xd5,0x7b,0x00,0x00,0x00]
+v_rsq_f16_e64 v5.l, m0
+// GFX11: v_rsq_f16_e64 v5.l, m0 ; encoding: [0x05,0x00,0xd6,0xd5,0x7d,0x00,0x00,0x00]
-v_rsq_f16_e64 v5, m0
-// GFX11: v_rsq_f16_e64 v5, m0 ; encoding: [0x05,0x00,0xd6,0xd5,0x7d,0x00,0x00,0x00]
+v_rsq_f16_e64 v5.l, exec_lo
+// GFX11: v_rsq_f16_e64 v5.l, exec_lo ; encoding: [0x05,0x00,0xd6,0xd5,0x7e,0x00,0x00,0x00]
-v_rsq_f16_e64 v5, exec_lo
-// GFX11: v_rsq_f16_e64 v5, exec_lo ; encoding: [0x05,0x00,0xd6,0xd5,0x7e,0x00,0x00,0x00]
+v_rsq_f16_e64 v5.l, exec_hi
+// GFX11: v_rsq_f16_e64 v5.l, exec_hi ; encoding: [0x05,0x00,0xd6,0xd5,0x7f,0x00,0x00,0x00]
-v_rsq_f16_e64 v5, exec_hi
-// GFX11: v_rsq_f16_e64 v5, exec_hi ; encoding: [0x05,0x00,0xd6,0xd5,0x7f,0x00,0x00,0x00]
+v_rsq_f16_e64 v5.l, null
+// GFX11: v_rsq_f16_e64 v5.l, null ; encoding: [0x05,0x00,0xd6,0xd5,0x7c,0x00,0x00,0x00]
-v_rsq_f16_e64 v5, null
-// GFX11: v_rsq_f16_e64 v5, null ; encoding: [0x05,0x00,0xd6,0xd5,0x7c,0x00,0x00,0x00]
+v_rsq_f16_e64 v5.l, -1
+// GFX11: v_rsq_f16_e64 v5.l, -1 ; encoding: [0x05,0x00,0xd6,0xd5,0xc1,0x00,0x00,0x00]
-v_rsq_f16_e64 v5, -1
-// GFX11: v_rsq_f16_e64 v5, -1 ; encoding: [0x05,0x00,0xd6,0xd5,0xc1,0x00,0x00,0x00]
+v_rsq_f16_e64 v5.l, 0.5 mul:2
+// GFX11: v_rsq_f16_e64 v5.l, 0.5 mul:2 ; encoding: [0x05,0x00,0xd6,0xd5,0xf0,0x00,0x00,0x08]
-v_rsq_f16_e64 v5, 0.5 mul:2
-// GFX11: v_rsq_f16_e64 v5, 0.5 mul:2 ; encoding: [0x05,0x00,0xd6,0xd5,0xf0,0x00,0x00,0x08]
+v_rsq_f16_e64 v5.l, src_scc mul:4
+// GFX11: v_rsq_f16_e64 v5.l, src_scc mul:4 ; encoding: [0x05,0x00,0xd6,0xd5,0xfd,0x00,0x00,0x10]
-v_rsq_f16_e64 v5, src_scc mul:4
-// GFX11: v_rsq_f16_e64 v5, src_scc mul:4 ; encoding: [0x05,0x00,0xd6,0xd5,0xfd,0x00,0x00,0x10]
+v_rsq_f16_e64 v255.l, -|0xfe0b| clamp div:2
+// GFX11: v_rsq_f16_e64 v255.l, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xd6,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
-v_rsq_f16_e64 v255, -|0xfe0b| clamp div:2
-// GFX11: v_rsq_f16_e64 v255, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xd6,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
+v_rsq_f16_e64 v5.h, v1.h
+// GFX11: v_rsq_f16_e64 v5.h, v1.h op_sel:[1,1] ; encoding: [0x05,0x48,0xd6,0xd5,0x01,0x01,0x00,0x00]
+
+v_rsq_f16_e64 v5.l, v255.h
+// GFX11: v_rsq_f16_e64 v5.l, v255.h op_sel:[1,0] ; encoding: [0x05,0x08,0xd6,0xd5,0xff,0x01,0x00,0x00]
+
+v_rsq_f16_e64 v255.h, -|0xfe0b| clamp div:2
+// GFX11: v_rsq_f16_e64 v255.h, -|0xfe0b| op_sel:[0,1] clamp div:2 ; encoding: [0xff,0xc1,0xd6,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
v_rsq_f32_e64 v5, v1
// GFX11: v_rsq_f32_e64 v5, v1 ; encoding: [0x05,0x00,0xae,0xd5,0x01,0x01,0x00,0x00]
@@ -3388,50 +3442,59 @@ v_sin_f32_e64 v5, src_scc mul:4
v_sin_f32_e64 v255, -|0xaf123456| clamp div:2
// GFX11: v_sin_f32_e64 v255, -|0xaf123456| clamp div:2 ; encoding: [0xff,0x81,0xb5,0xd5,0xff,0x00,0x00,0x38,0x56,0x34,0x12,0xaf]
-v_sqrt_f16_e64 v5, v1
-// GFX11: v_sqrt_f16_e64 v5, v1 ; encoding: [0x05,0x00,0xd5,0xd5,0x01,0x01,0x00,0x00]
+v_sqrt_f16_e64 v5.l, v1.l
+// GFX11: v_sqrt_f16_e64 v5.l, v1.l ; encoding: [0x05,0x00,0xd5,0xd5,0x01,0x01,0x00,0x00]
+
+v_sqrt_f16_e64 v5.l, v255.l
+// GFX11: v_sqrt_f16_e64 v5.l, v255.l ; encoding: [0x05,0x00,0xd5,0xd5,0xff,0x01,0x00,0x00]
+
+v_sqrt_f16_e64 v5.l, s1
+// GFX11: v_sqrt_f16_e64 v5.l, s1 ; encoding: [0x05,0x00,0xd5,0xd5,0x01,0x00,0x00,0x00]
+
+v_sqrt_f16_e64 v5.l, s105
+// GFX11: v_sqrt_f16_e64 v5.l, s105 ; encoding: [0x05,0x00,0xd5,0xd5,0x69,0x00,0x00,0x00]
-v_sqrt_f16_e64 v5, v255
-// GFX11: v_sqrt_f16_e64 v5, v255 ; encoding: [0x05,0x00,0xd5,0xd5,0xff,0x01,0x00,0x00]
+v_sqrt_f16_e64 v5.l, vcc_lo
+// GFX11: v_sqrt_f16_e64 v5.l, vcc_lo ; encoding: [0x05,0x00,0xd5,0xd5,0x6a,0x00,0x00,0x00]
-v_sqrt_f16_e64 v5, s1
-// GFX11: v_sqrt_f16_e64 v5, s1 ; encoding: [0x05,0x00,0xd5,0xd5,0x01,0x00,0x00,0x00]
+v_sqrt_f16_e64 v5.l, vcc_hi
+// GFX11: v_sqrt_f16_e64 v5.l, vcc_hi ; encoding: [0x05,0x00,0xd5,0xd5,0x6b,0x00,0x00,0x00]
-v_sqrt_f16_e64 v5, s105
-// GFX11: v_sqrt_f16_e64 v5, s105 ; encoding: [0x05,0x00,0xd5,0xd5,0x69,0x00,0x00,0x00]
+v_sqrt_f16_e64 v5.l, ttmp15
+// GFX11: v_sqrt_f16_e64 v5.l, ttmp15 ; encoding: [0x05,0x00,0xd5,0xd5,0x7b,0x00,0x00,0x00]
-v_sqrt_f16_e64 v5, vcc_lo
-// GFX11: v_sqrt_f16_e64 v5, vcc_lo ; encoding: [0x05,0x00,0xd5,0xd5,0x6a,0x00,0x00,0x00]
+v_sqrt_f16_e64 v5.l, m0
+// GFX11: v_sqrt_f16_e64 v5.l, m0 ; encoding: [0x05,0x00,0xd5,0xd5,0x7d,0x00,0x00,0x00]
-v_sqrt_f16_e64 v5, vcc_hi
-// GFX11: v_sqrt_f16_e64 v5, vcc_hi ; encoding: [0x05,0x00,0xd5,0xd5,0x6b,0x00,0x00,0x00]
+v_sqrt_f16_e64 v5.l, exec_lo
+// GFX11: v_sqrt_f16_e64 v5.l, exec_lo ; encoding: [0x05,0x00,0xd5,0xd5,0x7e,0x00,0x00,0x00]
-v_sqrt_f16_e64 v5, ttmp15
-// GFX11: v_sqrt_f16_e64 v5, ttmp15 ; encoding: [0x05,0x00,0xd5,0xd5,0x7b,0x00,0x00,0x00]
+v_sqrt_f16_e64 v5.l, exec_hi
+// GFX11: v_sqrt_f16_e64 v5.l, exec_hi ; encoding: [0x05,0x00,0xd5,0xd5,0x7f,0x00,0x00,0x00]
-v_sqrt_f16_e64 v5, m0
-// GFX11: v_sqrt_f16_e64 v5, m0 ; encoding: [0x05,0x00,0xd5,0xd5,0x7d,0x00,0x00,0x00]
+v_sqrt_f16_e64 v5.l, null
+// GFX11: v_sqrt_f16_e64 v5.l, null ; encoding: [0x05,0x00,0xd5,0xd5,0x7c,0x00,0x00,0x00]
-v_sqrt_f16_e64 v5, exec_lo
-// GFX11: v_sqrt_f16_e64 v5, exec_lo ; encoding: [0x05,0x00,0xd5,0xd5,0x7e,0x00,0x00,0x00]
+v_sqrt_f16_e64 v5.l, -1
+// GFX11: v_sqrt_f16_e64 v5.l, -1 ; encoding: [0x05,0x00,0xd5,0xd5,0xc1,0x00,0x00,0x00]
-v_sqrt_f16_e64 v5, exec_hi
-// GFX11: v_sqrt_f16_e64 v5, exec_hi ; encoding: [0x05,0x00,0xd5,0xd5,0x7f,0x00,0x00,0x00]
+v_sqrt_f16_e64 v5.l, 0.5 mul:2
+// GFX11: v_sqrt_f16_e64 v5.l, 0.5 mul:2 ; encoding: [0x05,0x00,0xd5,0xd5,0xf0,0x00,0x00,0x08]
-v_sqrt_f16_e64 v5, null
-// GFX11: v_sqrt_f16_e64 v5, null ; encoding: [0x05,0x00,0xd5,0xd5,0x7c,0x00,0x00,0x00]
+v_sqrt_f16_e64 v5.l, src_scc mul:4
+// GFX11: v_sqrt_f16_e64 v5.l, src_scc mul:4 ; encoding: [0x05,0x00,0xd5,0xd5,0xfd,0x00,0x00,0x10]
-v_sqrt_f16_e64 v5, -1
-// GFX11: v_sqrt_f16_e64 v5, -1 ; encoding: [0x05,0x00,0xd5,0xd5,0xc1,0x00,0x00,0x00]
+v_sqrt_f16_e64 v255.l, -|0xfe0b| clamp div:2
+// GFX11: v_sqrt_f16_e64 v255.l, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xd5,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
-v_sqrt_f16_e64 v5, 0.5 mul:2
-// GFX11: v_sqrt_f16_e64 v5, 0.5 mul:2 ; encoding: [0x05,0x00,0xd5,0xd5,0xf0,0x00,0x00,0x08]
+v_sqrt_f16_e64 v5.h, v1.h
+// GFX11: v_sqrt_f16_e64 v5.h, v1.h op_sel:[1,1] ; encoding: [0x05,0x48,0xd5,0xd5,0x01,0x01,0x00,0x00]
-v_sqrt_f16_e64 v5, src_scc mul:4
-// GFX11: v_sqrt_f16_e64 v5, src_scc mul:4 ; encoding: [0x05,0x00,0xd5,0xd5,0xfd,0x00,0x00,0x10]
+v_sqrt_f16_e64 v5.l, v255.h
+// GFX11: v_sqrt_f16_e64 v5.l, v255.h op_sel:[1,0] ; encoding: [0x05,0x08,0xd5,0xd5,0xff,0x01,0x00,0x00]
-v_sqrt_f16_e64 v255, -|0xfe0b| clamp div:2
-// GFX11: v_sqrt_f16_e64 v255, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xd5,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
+v_sqrt_f16_e64 v255.h, -|0xfe0b| clamp div:2
+// GFX11: v_sqrt_f16_e64 v255.h, -|0xfe0b| op_sel:[0,1] clamp div:2 ; encoding: [0xff,0xc1,0xd5,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
v_sqrt_f32_e64 v5, v1
// GFX11: v_sqrt_f32_e64 v5, v1 ; encoding: [0x05,0x00,0xb3,0xd5,0x01,0x01,0x00,0x00]
diff --git a/llvm/test/MC/AMDGPU/gfx12_asm_vop1.s b/llvm/test/MC/AMDGPU/gfx12_asm_vop1.s
index e21e5bf827ed1..6f47e6c03af1e 100644
--- a/llvm/test/MC/AMDGPU/gfx12_asm_vop1.s
+++ b/llvm/test/MC/AMDGPU/gfx12_asm_vop1.s
@@ -94,6 +94,18 @@ v_ceil_f16 v5.l, src_scc
v_ceil_f16 v127.l, 0xfe0b
// GFX12: v_ceil_f16_e32 v127.l, 0xfe0b ; encoding: [0xff,0xb8,0xfe,0x7e,0x0b,0xfe,0x00,0x00]
+v_ceil_f16 v5.l, v1.h
+// GFX12: v_ceil_f16_e32 v5.l, v1.h ; encoding: [0x81,0xb9,0x0a,0x7e]
+
+v_ceil_f16 v5.l, v127.h
+// GFX12: v_ceil_f16_e32 v5.l, v127.h ; encoding: [0xff,0xb9,0x0a,0x7e]
+
+v_ceil_f16 v5.h, src_scc
+// GFX12: v_ceil_f16_e32 v5.h, src_scc ; encoding: [0xfd,0xb8,0x0a,0x7f]
+
+v_ceil_f16 v127.h, 0xfe0b
+// GFX12: v_ceil_f16_e32 v127.h, 0xfe0b ; encoding: [0xff,0xb8,0xfe,0x7f,0x0b,0xfe,0x00,0x00]
+
v_ceil_f32 v5, v1
// GFX12: v_ceil_f32_e32 v5, v1 ; encoding: [0x01,0x45,0x0a,0x7e]
@@ -1886,6 +1898,18 @@ v_exp_f16 v5.l, src_scc
v_exp_f16 v127.l, 0xfe0b
// GFX12: v_exp_f16_e32 v127.l, 0xfe0b ; encoding: [0xff,0xb0,0xfe,0x7e,0x0b,0xfe,0x00,0x00]
+v_exp_f16 v5.l, v1.h
+// GFX12: v_exp_f16_e32 v5.l, v1.h ; encoding: [0x81,0xb1,0x0a,0x7e]
+
+v_exp_f16 v5.l, v127.h
+// GFX12: v_exp_f16_e32 v5.l, v127.h ; encoding: [0xff,0xb1,0x0a,0x7e]
+
+v_exp_f16 v5.h, src_scc
+// GFX12: v_exp_f16_e32 v5.h, src_scc ; encoding: [0xfd,0xb0,0x0a,0x7f]
+
+v_exp_f16 v127.h, 0xfe0b
+// GFX12: v_exp_f16_e32 v127.h, 0xfe0b ; encoding: [0xff,0xb0,0xfe,0x7f,0x0b,0xfe,0x00,0x00]
+
v_exp_f32 v5, v1
// GFX12: v_exp_f32_e32 v5, v1 ; encoding: [0x01,0x4b,0x0a,0x7e]
@@ -2111,6 +2135,18 @@ v_floor_f16 v5.l, src_scc
v_floor_f16 v127.l, 0xfe0b
// GFX12: v_floor_f16_e32 v127.l, 0xfe0b ; encoding: [0xff,0xb6,0xfe,0x7e,0x0b,0xfe,0x00,0x00]
+v_floor_f16 v5.l, v1.h
+// GFX12: v_floor_f16_e32 v5.l, v1.h ; encoding: [0x81,0xb7,0x0a,0x7e]
+
+v_floor_f16 v5.l, v127.h
+// GFX12: v_floor_f16_e32 v5.l, v127.h ; encoding: [0xff,0xb7,0x0a,0x7e]
+
+v_floor_f16 v5.h, src_scc
+// GFX12: v_floor_f16_e32 v5.h, src_scc ; encoding: [0xfd,0xb6,0x0a,0x7f]
+
+v_floor_f16 v127.h, 0xfe0b
+// GFX12: v_floor_f16_e32 v127.h, 0xfe0b ; encoding: [0xff,0xb6,0xfe,0x7f,0x0b,0xfe,0x00,0x00]
+
v_floor_f32 v5, v1
// GFX12: v_floor_f32_e32 v5, v1 ; encoding: [0x01,0x49,0x0a,0x7e]
@@ -2651,6 +2687,18 @@ v_log_f16 v5.l, src_scc
v_log_f16 v127.l, 0xfe0b
// GFX12: v_log_f16_e32 v127.l, 0xfe0b ; encoding: [0xff,0xae,0xfe,0x7e,0x0b,0xfe,0x00,0x00]
+v_log_f16 v5.l, v1.h
+// GFX12: v_log_f16_e32 v5.l, v1.h ; encoding: [0x81,0xaf,0x0a,0x7e]
+
+v_log_f16 v5.l, v127.h
+// GFX12: v_log_f16_e32 v5.l, v127.h ; encoding: [0xff,0xaf,0x0a,0x7e]
+
+v_log_f16 v5.h, src_scc
+// GFX12: v_log_f16_e32 v5.h, src_scc ; encoding: [0xfd,0xae,0x0a,0x7f]
+
+v_log_f16 v127.h, 0xfe0b
+// GFX12: v_log_f16_e32 v127.h, 0xfe0b ; encoding: [0xff,0xae,0xfe,0x7f,0x0b,0xfe,0x00,0x00]
+
v_log_f32 v5, v1
// GFX12: v_log_f32_e32 v5, v1 ; encoding: [0x01,0x4f,0x0a,0x7e]
@@ -2964,6 +3012,18 @@ v_rcp_f16 v5.l, src_scc
v_rcp_f16 v127.l, 0xfe0b
// GFX12: v_rcp_f16_e32 v127.l, 0xfe0b ; encoding: [0xff,0xa8,0xfe,0x7e,0x0b,0xfe,0x00,0x00]
+v_rcp_f16 v5.l, v1.h
+// GFX12: v_rcp_f16_e32 v5.l, v1.h ; encoding: [0x81,0xa9,0x0a,0x7e]
+
+v_rcp_f16 v5.l, v127.h
+// GFX12: v_rcp_f16_e32 v5.l, v127.h ; encoding: [0xff,0xa9,0x0a,0x7e]
+
+v_rcp_f16 v5.h, src_scc
+// GFX12: v_rcp_f16_e32 v5.h, src_scc ; encoding: [0xfd,0xa8,0x0a,0x7f]
+
+v_rcp_f16 v127.h, 0xfe0b
+// GFX12: v_rcp_f16_e32 v127.h, 0xfe0b ; encoding: [0xff,0xa8,0xfe,0x7f,0x0b,0xfe,0x00,0x00]
+
v_rcp_f32 v5, v1
// GFX12: v_rcp_f32_e32 v5, v1 ; encoding: [0x01,0x55,0x0a,0x7e]
@@ -3291,6 +3351,18 @@ v_rsq_f16 v5.l, src_scc
v_rsq_f16 v127.l, 0xfe0b
// GFX12: v_rsq_f16_e32 v127.l, 0xfe0b ; encoding: [0xff,0xac,0xfe,0x7e,0x0b,0xfe,0x00,0x00]
+v_rsq_f16 v5.l, v1.h
+// GFX12: v_rsq_f16_e32 v5.l, v1.h ; encoding: [0x81,0xad,0x0a,0x7e]
+
+v_rsq_f16 v5.l, v127.h
+// GFX12: v_rsq_f16_e32 v5.l, v127.h ; encoding: [0xff,0xad,0x0a,0x7e]
+
+v_rsq_f16 v5.h, src_scc
+// GFX12: v_rsq_f16_e32 v5.h, src_scc ; encoding: [0xfd,0xac,0x0a,0x7f]
+
+v_rsq_f16 v127.h, 0xfe0b
+// GFX12: v_rsq_f16_e32 v127.h, 0xfe0b ; encoding: [0xff,0xac,0xfe,0x7f,0x0b,0xfe,0x00,0x00]
+
v_rsq_f32 v5, v1
// GFX12: v_rsq_f32_e32 v5, v1 ; encoding: [0x01,0x5d,0x0a,0x7e]
@@ -3585,6 +3657,18 @@ v_sqrt_f16 v5.l, src_scc
v_sqrt_f16 v127.l, 0xfe0b
// GFX12: v_sqrt_f16_e32 v127.l, 0xfe0b ; encoding: [0xff,0xaa,0xfe,0x7e,0x0b,0xfe,0x00,0x00]
+v_sqrt_f16 v5.l, v1.h
+// GFX12: v_sqrt_f16_e32 v5.l, v1.h ; encoding: [0x81,0xab,0x0a,0x7e]
+
+v_sqrt_f16 v5.l, v127.h
+// GFX12: v_sqrt_f16_e32 v5.l, v127.h ; encoding: [0xff,0xab,0x0a,0x7e]
+
+v_sqrt_f16 v5.h, src_scc
+// GFX12: v_sqrt_f16_e32 v5.h, src_scc ; encoding: [0xfd,0xaa,0x0a,0x7f]
+
+v_sqrt_f16 v127.h, 0xfe0b
+// GFX12: v_sqrt_f16_e32 v127.h, 0xfe0b ; encoding: [0xff,0xaa,0xfe,0x7f,0x0b,0xfe,0x00,0x00]
+
v_sqrt_f32 v5, v1
// GFX12: v_sqrt_f32_e32 v5, v1 ; encoding: [0x01,0x67,0x0a,0x7e]
diff --git a/llvm/test/MC/AMDGPU/gfx12_asm_vop1_dpp16.s b/llvm/test/MC/AMDGPU/gfx12_asm_vop1_dpp16.s
index e821fb30edac1..3cd040a253bca 100644
--- a/llvm/test/MC/AMDGPU/gfx12_asm_vop1_dpp16.s
+++ b/llvm/test/MC/AMDGPU/gfx12_asm_vop1_dpp16.s
@@ -46,47 +46,53 @@ v_bfrev_b32 v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
v_bfrev_b32 v255, v255 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
// GFX12: v_bfrev_b32_dpp v255, v255 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0x70,0xfe,0x7f,0xff,0x6f,0x05,0x30]
-v_ceil_f16 v5, v1 quad_perm:[3,2,1,0]
-// GFX12: v_ceil_f16_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x1b,0x00,0xff]
+v_ceil_f16 v5.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: v_ceil_f16_dpp v5.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x1b,0x00,0xff]
-v_ceil_f16 v5, v1 quad_perm:[0,1,2,3]
-// GFX12: v_ceil_f16_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0xe4,0x00,0xff]
+v_ceil_f16 v5.l, v1.l quad_perm:[0,1,2,3]
+// GFX12: v_ceil_f16_dpp v5.l, v1.l quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0xe4,0x00,0xff]
-v_ceil_f16 v5, v1 row_mirror
-// GFX12: v_ceil_f16_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x40,0x01,0xff]
+v_ceil_f16 v5.l, v1.l row_mirror
+// GFX12: v_ceil_f16_dpp v5.l, v1.l row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x40,0x01,0xff]
-v_ceil_f16 v5, v1 row_half_mirror
-// GFX12: v_ceil_f16_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x41,0x01,0xff]
+v_ceil_f16 v5.l, v1.l row_half_mirror
+// GFX12: v_ceil_f16_dpp v5.l, v1.l row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x41,0x01,0xff]
-v_ceil_f16 v5, v1 row_shl:1
-// GFX12: v_ceil_f16_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x01,0x01,0xff]
+v_ceil_f16 v5.l, v1.l row_shl:1
+// GFX12: v_ceil_f16_dpp v5.l, v1.l row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x01,0x01,0xff]
-v_ceil_f16 v5, v1 row_shl:15
-// GFX12: v_ceil_f16_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x0f,0x01,0xff]
+v_ceil_f16 v5.l, v1.l row_shl:15
+// GFX12: v_ceil_f16_dpp v5.l, v1.l row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x0f,0x01,0xff]
-v_ceil_f16 v5, v1 row_shr:1
-// GFX12: v_ceil_f16_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x11,0x01,0xff]
+v_ceil_f16 v5.l, v1.l row_shr:1
+// GFX12: v_ceil_f16_dpp v5.l, v1.l row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x11,0x01,0xff]
-v_ceil_f16 v5, v1 row_shr:15
-// GFX12: v_ceil_f16_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x1f,0x01,0xff]
+v_ceil_f16 v5.l, v1.l row_shr:15
+// GFX12: v_ceil_f16_dpp v5.l, v1.l row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x1f,0x01,0xff]
-v_ceil_f16 v5, v1 row_ror:1
-// GFX12: v_ceil_f16_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x21,0x01,0xff]
+v_ceil_f16 v5.l, v1.l row_ror:1
+// GFX12: v_ceil_f16_dpp v5.l, v1.l row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x21,0x01,0xff]
-v_ceil_f16 v5, v1 row_ror:15
-// GFX12: v_ceil_f16_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x2f,0x01,0xff]
+v_ceil_f16 v5.l, v1.l row_ror:15
+// GFX12: v_ceil_f16_dpp v5.l, v1.l row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x2f,0x01,0xff]
-v_ceil_f16 v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
-// GFX12: v_ceil_f16_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x50,0x01,0xff]
+v_ceil_f16 v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf
+// GFX12: v_ceil_f16_dpp v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x50,0x01,0xff]
-v_ceil_f16 v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1
-// GFX12: v_ceil_f16_dpp v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x5f,0x01,0x01]
+v_ceil_f16 v5.l, v1.l row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX12: v_ceil_f16_dpp v5.l, v1.l row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x5f,0x01,0x01]
-v_ceil_f16 v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
-// GFX12: v_ceil_f16_dpp v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x60,0x09,0x13]
+v_ceil_f16 v5.l, v1.l row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_ceil_f16_dpp v5.l, v1.l row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xb8,0x0a,0x7e,0x01,0x60,0x09,0x13]
-v_ceil_f16 v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
-// GFX12: v_ceil_f16_dpp v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xb8,0xfe,0x7e,0x7f,0x6f,0x35,0x30]
+v_ceil_f16 v127.l, -|v127.l| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_ceil_f16_dpp v127.l, -|v127.l| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xb8,0xfe,0x7e,0x7f,0x6f,0x35,0x30]
+
+v_ceil_f16 v5.h, v1.h row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_ceil_f16_dpp v5.h, v1.h row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xb8,0x0a,0x7f,0x81,0x60,0x09,0x13]
+
+v_ceil_f16 v127.h, -|v127.h| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_ceil_f16_dpp v127.h, -|v127.h| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xb8,0xfe,0x7f,0xff,0x6f,0x35,0x30]
v_ceil_f32 v5, v1 quad_perm:[3,2,1,0]
// GFX12: v_ceil_f32_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x44,0x0a,0x7e,0x01,0x1b,0x00,0xff]
@@ -1384,47 +1390,53 @@ v_cvt_u32_u16 v5, v1.h row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
v_cvt_u32_u16 v255, v127.h row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
// GFX12: v_cvt_u32_u16_dpp v255, v127.h row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xd6,0xfe,0x7f,0xff,0x6f,0x05,0x30]
-v_exp_f16 v5, v1 quad_perm:[3,2,1,0]
-// GFX12: v_exp_f16_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x1b,0x00,0xff]
+v_exp_f16 v5.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: v_exp_f16_dpp v5.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x1b,0x00,0xff]
+
+v_exp_f16 v5.l, v1.l quad_perm:[0,1,2,3]
+// GFX12: v_exp_f16_dpp v5.l, v1.l quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0xe4,0x00,0xff]
+
+v_exp_f16 v5.l, v1.l row_mirror
+// GFX12: v_exp_f16_dpp v5.l, v1.l row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x40,0x01,0xff]
-v_exp_f16 v5, v1 quad_perm:[0,1,2,3]
-// GFX12: v_exp_f16_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0xe4,0x00,0xff]
+v_exp_f16 v5.l, v1.l row_half_mirror
+// GFX12: v_exp_f16_dpp v5.l, v1.l row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x41,0x01,0xff]
-v_exp_f16 v5, v1 row_mirror
-// GFX12: v_exp_f16_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x40,0x01,0xff]
+v_exp_f16 v5.l, v1.l row_shl:1
+// GFX12: v_exp_f16_dpp v5.l, v1.l row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x01,0x01,0xff]
-v_exp_f16 v5, v1 row_half_mirror
-// GFX12: v_exp_f16_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x41,0x01,0xff]
+v_exp_f16 v5.l, v1.l row_shl:15
+// GFX12: v_exp_f16_dpp v5.l, v1.l row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x0f,0x01,0xff]
-v_exp_f16 v5, v1 row_shl:1
-// GFX12: v_exp_f16_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x01,0x01,0xff]
+v_exp_f16 v5.l, v1.l row_shr:1
+// GFX12: v_exp_f16_dpp v5.l, v1.l row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x11,0x01,0xff]
-v_exp_f16 v5, v1 row_shl:15
-// GFX12: v_exp_f16_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x0f,0x01,0xff]
+v_exp_f16 v5.l, v1.l row_shr:15
+// GFX12: v_exp_f16_dpp v5.l, v1.l row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x1f,0x01,0xff]
-v_exp_f16 v5, v1 row_shr:1
-// GFX12: v_exp_f16_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x11,0x01,0xff]
+v_exp_f16 v5.l, v1.l row_ror:1
+// GFX12: v_exp_f16_dpp v5.l, v1.l row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x21,0x01,0xff]
-v_exp_f16 v5, v1 row_shr:15
-// GFX12: v_exp_f16_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x1f,0x01,0xff]
+v_exp_f16 v5.l, v1.l row_ror:15
+// GFX12: v_exp_f16_dpp v5.l, v1.l row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x2f,0x01,0xff]
-v_exp_f16 v5, v1 row_ror:1
-// GFX12: v_exp_f16_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x21,0x01,0xff]
+v_exp_f16 v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf
+// GFX12: v_exp_f16_dpp v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x50,0x01,0xff]
-v_exp_f16 v5, v1 row_ror:15
-// GFX12: v_exp_f16_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x2f,0x01,0xff]
+v_exp_f16 v5.l, v1.l row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX12: v_exp_f16_dpp v5.l, v1.l row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x5f,0x01,0x01]
-v_exp_f16 v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
-// GFX12: v_exp_f16_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x50,0x01,0xff]
+v_exp_f16 v5.l, v1.l row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_exp_f16_dpp v5.l, v1.l row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x60,0x09,0x13]
-v_exp_f16 v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1
-// GFX12: v_exp_f16_dpp v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x5f,0x01,0x01]
+v_exp_f16 v127.l, -|v127.l| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_exp_f16_dpp v127.l, -|v127.l| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xb0,0xfe,0x7e,0x7f,0x6f,0x35,0x30]
-v_exp_f16 v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
-// GFX12: v_exp_f16_dpp v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xb0,0x0a,0x7e,0x01,0x60,0x09,0x13]
+v_exp_f16 v5.h, v1.h row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_exp_f16_dpp v5.h, v1.h row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xb0,0x0a,0x7f,0x81,0x60,0x09,0x13]
-v_exp_f16 v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
-// GFX12: v_exp_f16_dpp v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xb0,0xfe,0x7e,0x7f,0x6f,0x35,0x30]
+v_exp_f16 v127.h, -|v127.h| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_exp_f16_dpp v127.h, -|v127.h| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xb0,0xfe,0x7f,0xff,0x6f,0x35,0x30]
v_exp_f32 v5, v1 quad_perm:[3,2,1,0]
// GFX12: v_exp_f32_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x4a,0x0a,0x7e,0x01,0x1b,0x00,0xff]
@@ -1594,47 +1606,53 @@ v_ffbl_b32 v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
v_ffbl_b32 v255, v255 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
// GFX12: v_ctz_i32_b32_dpp v255, v255 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0x74,0xfe,0x7f,0xff,0x6f,0x05,0x30]
-v_floor_f16 v5, v1 quad_perm:[3,2,1,0]
-// GFX12: v_floor_f16_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x1b,0x00,0xff]
+v_floor_f16 v5.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: v_floor_f16_dpp v5.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x1b,0x00,0xff]
-v_floor_f16 v5, v1 quad_perm:[0,1,2,3]
-// GFX12: v_floor_f16_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0xe4,0x00,0xff]
+v_floor_f16 v5.l, v1.l quad_perm:[0,1,2,3]
+// GFX12: v_floor_f16_dpp v5.l, v1.l quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0xe4,0x00,0xff]
-v_floor_f16 v5, v1 row_mirror
-// GFX12: v_floor_f16_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x40,0x01,0xff]
+v_floor_f16 v5.l, v1.l row_mirror
+// GFX12: v_floor_f16_dpp v5.l, v1.l row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x40,0x01,0xff]
-v_floor_f16 v5, v1 row_half_mirror
-// GFX12: v_floor_f16_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x41,0x01,0xff]
+v_floor_f16 v5.l, v1.l row_half_mirror
+// GFX12: v_floor_f16_dpp v5.l, v1.l row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x41,0x01,0xff]
-v_floor_f16 v5, v1 row_shl:1
-// GFX12: v_floor_f16_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x01,0x01,0xff]
+v_floor_f16 v5.l, v1.l row_shl:1
+// GFX12: v_floor_f16_dpp v5.l, v1.l row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x01,0x01,0xff]
-v_floor_f16 v5, v1 row_shl:15
-// GFX12: v_floor_f16_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x0f,0x01,0xff]
+v_floor_f16 v5.l, v1.l row_shl:15
+// GFX12: v_floor_f16_dpp v5.l, v1.l row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x0f,0x01,0xff]
-v_floor_f16 v5, v1 row_shr:1
-// GFX12: v_floor_f16_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x11,0x01,0xff]
+v_floor_f16 v5.l, v1.l row_shr:1
+// GFX12: v_floor_f16_dpp v5.l, v1.l row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x11,0x01,0xff]
-v_floor_f16 v5, v1 row_shr:15
-// GFX12: v_floor_f16_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x1f,0x01,0xff]
+v_floor_f16 v5.l, v1.l row_shr:15
+// GFX12: v_floor_f16_dpp v5.l, v1.l row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x1f,0x01,0xff]
-v_floor_f16 v5, v1 row_ror:1
-// GFX12: v_floor_f16_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x21,0x01,0xff]
+v_floor_f16 v5.l, v1.l row_ror:1
+// GFX12: v_floor_f16_dpp v5.l, v1.l row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x21,0x01,0xff]
-v_floor_f16 v5, v1 row_ror:15
-// GFX12: v_floor_f16_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x2f,0x01,0xff]
+v_floor_f16 v5.l, v1.l row_ror:15
+// GFX12: v_floor_f16_dpp v5.l, v1.l row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x2f,0x01,0xff]
-v_floor_f16 v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
-// GFX12: v_floor_f16_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x50,0x01,0xff]
+v_floor_f16 v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf
+// GFX12: v_floor_f16_dpp v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x50,0x01,0xff]
-v_floor_f16 v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1
-// GFX12: v_floor_f16_dpp v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x5f,0x01,0x01]
+v_floor_f16 v5.l, v1.l row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX12: v_floor_f16_dpp v5.l, v1.l row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x5f,0x01,0x01]
-v_floor_f16 v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
-// GFX12: v_floor_f16_dpp v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x60,0x09,0x13]
+v_floor_f16 v5.l, v1.l row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_floor_f16_dpp v5.l, v1.l row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xb6,0x0a,0x7e,0x01,0x60,0x09,0x13]
-v_floor_f16 v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
-// GFX12: v_floor_f16_dpp v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xb6,0xfe,0x7e,0x7f,0x6f,0x35,0x30]
+v_floor_f16 v127.l, -|v127.l| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_floor_f16_dpp v127.l, -|v127.l| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xb6,0xfe,0x7e,0x7f,0x6f,0x35,0x30]
+
+v_floor_f16 v5.h, v1.h row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_floor_f16_dpp v5.h, v1.h row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xb6,0x0a,0x7f,0x81,0x60,0x09,0x13]
+
+v_floor_f16 v127.h, -|v127.h| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_floor_f16_dpp v127.h, -|v127.h| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xb6,0xfe,0x7f,0xff,0x6f,0x35,0x30]
v_floor_f32 v5, v1 quad_perm:[3,2,1,0]
// GFX12: v_floor_f32_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x48,0x0a,0x7e,0x01,0x1b,0x00,0xff]
@@ -1948,47 +1966,53 @@ v_frexp_mant_f32 v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
v_frexp_mant_f32 v255, -|v255| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
// GFX12: v_frexp_mant_f32_dpp v255, -|v255| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0x80,0xfe,0x7f,0xff,0x6f,0x35,0x30]
-v_log_f16 v5, v1 quad_perm:[3,2,1,0]
-// GFX12: v_log_f16_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x1b,0x00,0xff]
+v_log_f16 v5.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: v_log_f16_dpp v5.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x1b,0x00,0xff]
+
+v_log_f16 v5.l, v1.l quad_perm:[0,1,2,3]
+// GFX12: v_log_f16_dpp v5.l, v1.l quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0xe4,0x00,0xff]
-v_log_f16 v5, v1 quad_perm:[0,1,2,3]
-// GFX12: v_log_f16_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0xe4,0x00,0xff]
+v_log_f16 v5.l, v1.l row_mirror
+// GFX12: v_log_f16_dpp v5.l, v1.l row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x40,0x01,0xff]
-v_log_f16 v5, v1 row_mirror
-// GFX12: v_log_f16_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x40,0x01,0xff]
+v_log_f16 v5.l, v1.l row_half_mirror
+// GFX12: v_log_f16_dpp v5.l, v1.l row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x41,0x01,0xff]
-v_log_f16 v5, v1 row_half_mirror
-// GFX12: v_log_f16_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x41,0x01,0xff]
+v_log_f16 v5.l, v1.l row_shl:1
+// GFX12: v_log_f16_dpp v5.l, v1.l row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x01,0x01,0xff]
-v_log_f16 v5, v1 row_shl:1
-// GFX12: v_log_f16_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x01,0x01,0xff]
+v_log_f16 v5.l, v1.l row_shl:15
+// GFX12: v_log_f16_dpp v5.l, v1.l row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x0f,0x01,0xff]
-v_log_f16 v5, v1 row_shl:15
-// GFX12: v_log_f16_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x0f,0x01,0xff]
+v_log_f16 v5.l, v1.l row_shr:1
+// GFX12: v_log_f16_dpp v5.l, v1.l row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x11,0x01,0xff]
-v_log_f16 v5, v1 row_shr:1
-// GFX12: v_log_f16_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x11,0x01,0xff]
+v_log_f16 v5.l, v1.l row_shr:15
+// GFX12: v_log_f16_dpp v5.l, v1.l row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x1f,0x01,0xff]
-v_log_f16 v5, v1 row_shr:15
-// GFX12: v_log_f16_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x1f,0x01,0xff]
+v_log_f16 v5.l, v1.l row_ror:1
+// GFX12: v_log_f16_dpp v5.l, v1.l row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x21,0x01,0xff]
-v_log_f16 v5, v1 row_ror:1
-// GFX12: v_log_f16_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x21,0x01,0xff]
+v_log_f16 v5.l, v1.l row_ror:15
+// GFX12: v_log_f16_dpp v5.l, v1.l row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x2f,0x01,0xff]
-v_log_f16 v5, v1 row_ror:15
-// GFX12: v_log_f16_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x2f,0x01,0xff]
+v_log_f16 v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf
+// GFX12: v_log_f16_dpp v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x50,0x01,0xff]
-v_log_f16 v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
-// GFX12: v_log_f16_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x50,0x01,0xff]
+v_log_f16 v5.l, v1.l row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX12: v_log_f16_dpp v5.l, v1.l row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x5f,0x01,0x01]
-v_log_f16 v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1
-// GFX12: v_log_f16_dpp v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x5f,0x01,0x01]
+v_log_f16 v5.l, v1.l row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_log_f16_dpp v5.l, v1.l row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x60,0x09,0x13]
-v_log_f16 v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
-// GFX12: v_log_f16_dpp v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xae,0x0a,0x7e,0x01,0x60,0x09,0x13]
+v_log_f16 v127.l, -|v127.l| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_log_f16_dpp v127.l, -|v127.l| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xae,0xfe,0x7e,0x7f,0x6f,0x35,0x30]
-v_log_f16 v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
-// GFX12: v_log_f16_dpp v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xae,0xfe,0x7e,0x7f,0x6f,0x35,0x30]
+v_log_f16 v5.h, v1.h row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_log_f16_dpp v5.h, v1.h row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xae,0x0a,0x7f,0x81,0x60,0x09,0x13]
+
+v_log_f16 v127.h, -|v127.h| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_log_f16_dpp v127.h, -|v127.h| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xae,0xfe,0x7f,0xff,0x6f,0x35,0x30]
v_log_f32 v5, v1 quad_perm:[3,2,1,0]
// GFX12: v_log_f32_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x4e,0x0a,0x7e,0x01,0x1b,0x00,0xff]
@@ -2332,47 +2356,53 @@ v_not_b32 v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
v_not_b32 v255, v255 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
// GFX12: v_not_b32_dpp v255, v255 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0x6e,0xfe,0x7f,0xff,0x6f,0x05,0x30]
-v_rcp_f16 v5, v1 quad_perm:[3,2,1,0]
-// GFX12: v_rcp_f16_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x1b,0x00,0xff]
+v_rcp_f16 v5.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: v_rcp_f16_dpp v5.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x1b,0x00,0xff]
+
+v_rcp_f16 v5.l, v1.l quad_perm:[0,1,2,3]
+// GFX12: v_rcp_f16_dpp v5.l, v1.l quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0xe4,0x00,0xff]
-v_rcp_f16 v5, v1 quad_perm:[0,1,2,3]
-// GFX12: v_rcp_f16_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0xe4,0x00,0xff]
+v_rcp_f16 v5.l, v1.l row_mirror
+// GFX12: v_rcp_f16_dpp v5.l, v1.l row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x40,0x01,0xff]
-v_rcp_f16 v5, v1 row_mirror
-// GFX12: v_rcp_f16_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x40,0x01,0xff]
+v_rcp_f16 v5.l, v1.l row_half_mirror
+// GFX12: v_rcp_f16_dpp v5.l, v1.l row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x41,0x01,0xff]
-v_rcp_f16 v5, v1 row_half_mirror
-// GFX12: v_rcp_f16_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x41,0x01,0xff]
+v_rcp_f16 v5.l, v1.l row_shl:1
+// GFX12: v_rcp_f16_dpp v5.l, v1.l row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x01,0x01,0xff]
-v_rcp_f16 v5, v1 row_shl:1
-// GFX12: v_rcp_f16_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x01,0x01,0xff]
+v_rcp_f16 v5.l, v1.l row_shl:15
+// GFX12: v_rcp_f16_dpp v5.l, v1.l row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x0f,0x01,0xff]
-v_rcp_f16 v5, v1 row_shl:15
-// GFX12: v_rcp_f16_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x0f,0x01,0xff]
+v_rcp_f16 v5.l, v1.l row_shr:1
+// GFX12: v_rcp_f16_dpp v5.l, v1.l row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x11,0x01,0xff]
-v_rcp_f16 v5, v1 row_shr:1
-// GFX12: v_rcp_f16_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x11,0x01,0xff]
+v_rcp_f16 v5.l, v1.l row_shr:15
+// GFX12: v_rcp_f16_dpp v5.l, v1.l row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x1f,0x01,0xff]
-v_rcp_f16 v5, v1 row_shr:15
-// GFX12: v_rcp_f16_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x1f,0x01,0xff]
+v_rcp_f16 v5.l, v1.l row_ror:1
+// GFX12: v_rcp_f16_dpp v5.l, v1.l row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x21,0x01,0xff]
-v_rcp_f16 v5, v1 row_ror:1
-// GFX12: v_rcp_f16_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x21,0x01,0xff]
+v_rcp_f16 v5.l, v1.l row_ror:15
+// GFX12: v_rcp_f16_dpp v5.l, v1.l row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x2f,0x01,0xff]
-v_rcp_f16 v5, v1 row_ror:15
-// GFX12: v_rcp_f16_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x2f,0x01,0xff]
+v_rcp_f16 v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf
+// GFX12: v_rcp_f16_dpp v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x50,0x01,0xff]
-v_rcp_f16 v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
-// GFX12: v_rcp_f16_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x50,0x01,0xff]
+v_rcp_f16 v5.l, v1.l row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX12: v_rcp_f16_dpp v5.l, v1.l row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x5f,0x01,0x01]
-v_rcp_f16 v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1
-// GFX12: v_rcp_f16_dpp v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x5f,0x01,0x01]
+v_rcp_f16 v5.l, v1.l row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_rcp_f16_dpp v5.l, v1.l row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x60,0x09,0x13]
-v_rcp_f16 v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
-// GFX12: v_rcp_f16_dpp v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xa8,0x0a,0x7e,0x01,0x60,0x09,0x13]
+v_rcp_f16 v127.l, -|v127.l| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_rcp_f16_dpp v127.l, -|v127.l| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xa8,0xfe,0x7e,0x7f,0x6f,0x35,0x30]
-v_rcp_f16 v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
-// GFX12: v_rcp_f16_dpp v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xa8,0xfe,0x7e,0x7f,0x6f,0x35,0x30]
+v_rcp_f16 v5.h, v1.h row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_rcp_f16_dpp v5.h, v1.h row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xa8,0x0a,0x7f,0x81,0x60,0x09,0x13]
+
+v_rcp_f16 v127.h, -|v127.h| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_rcp_f16_dpp v127.h, -|v127.h| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xa8,0xfe,0x7f,0xff,0x6f,0x35,0x30]
v_rcp_f32 v5, v1 quad_perm:[3,2,1,0]
// GFX12: v_rcp_f32_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x54,0x0a,0x7e,0x01,0x1b,0x00,0xff]
@@ -2548,47 +2578,53 @@ v_rndne_f32 v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
v_rndne_f32 v255, -|v255| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
// GFX12: v_rndne_f32_dpp v255, -|v255| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0x46,0xfe,0x7f,0xff,0x6f,0x35,0x30]
-v_rsq_f16 v5, v1 quad_perm:[3,2,1,0]
-// GFX12: v_rsq_f16_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x1b,0x00,0xff]
+v_rsq_f16 v5.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: v_rsq_f16_dpp v5.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x1b,0x00,0xff]
+
+v_rsq_f16 v5.l, v1.l quad_perm:[0,1,2,3]
+// GFX12: v_rsq_f16_dpp v5.l, v1.l quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0xe4,0x00,0xff]
-v_rsq_f16 v5, v1 quad_perm:[0,1,2,3]
-// GFX12: v_rsq_f16_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0xe4,0x00,0xff]
+v_rsq_f16 v5.l, v1.l row_mirror
+// GFX12: v_rsq_f16_dpp v5.l, v1.l row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x40,0x01,0xff]
-v_rsq_f16 v5, v1 row_mirror
-// GFX12: v_rsq_f16_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x40,0x01,0xff]
+v_rsq_f16 v5.l, v1.l row_half_mirror
+// GFX12: v_rsq_f16_dpp v5.l, v1.l row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x41,0x01,0xff]
-v_rsq_f16 v5, v1 row_half_mirror
-// GFX12: v_rsq_f16_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x41,0x01,0xff]
+v_rsq_f16 v5.l, v1.l row_shl:1
+// GFX12: v_rsq_f16_dpp v5.l, v1.l row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x01,0x01,0xff]
-v_rsq_f16 v5, v1 row_shl:1
-// GFX12: v_rsq_f16_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x01,0x01,0xff]
+v_rsq_f16 v5.l, v1.l row_shl:15
+// GFX12: v_rsq_f16_dpp v5.l, v1.l row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x0f,0x01,0xff]
-v_rsq_f16 v5, v1 row_shl:15
-// GFX12: v_rsq_f16_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x0f,0x01,0xff]
+v_rsq_f16 v5.l, v1.l row_shr:1
+// GFX12: v_rsq_f16_dpp v5.l, v1.l row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x11,0x01,0xff]
-v_rsq_f16 v5, v1 row_shr:1
-// GFX12: v_rsq_f16_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x11,0x01,0xff]
+v_rsq_f16 v5.l, v1.l row_shr:15
+// GFX12: v_rsq_f16_dpp v5.l, v1.l row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x1f,0x01,0xff]
-v_rsq_f16 v5, v1 row_shr:15
-// GFX12: v_rsq_f16_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x1f,0x01,0xff]
+v_rsq_f16 v5.l, v1.l row_ror:1
+// GFX12: v_rsq_f16_dpp v5.l, v1.l row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x21,0x01,0xff]
-v_rsq_f16 v5, v1 row_ror:1
-// GFX12: v_rsq_f16_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x21,0x01,0xff]
+v_rsq_f16 v5.l, v1.l row_ror:15
+// GFX12: v_rsq_f16_dpp v5.l, v1.l row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x2f,0x01,0xff]
-v_rsq_f16 v5, v1 row_ror:15
-// GFX12: v_rsq_f16_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x2f,0x01,0xff]
+v_rsq_f16 v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf
+// GFX12: v_rsq_f16_dpp v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x50,0x01,0xff]
-v_rsq_f16 v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
-// GFX12: v_rsq_f16_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x50,0x01,0xff]
+v_rsq_f16 v5.l, v1.l row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX12: v_rsq_f16_dpp v5.l, v1.l row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x5f,0x01,0x01]
-v_rsq_f16 v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1
-// GFX12: v_rsq_f16_dpp v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x5f,0x01,0x01]
+v_rsq_f16 v5.l, v1.l row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_rsq_f16_dpp v5.l, v1.l row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x60,0x09,0x13]
-v_rsq_f16 v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
-// GFX12: v_rsq_f16_dpp v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xac,0x0a,0x7e,0x01,0x60,0x09,0x13]
+v_rsq_f16 v127.l, -|v127.l| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_rsq_f16_dpp v127.l, -|v127.l| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xac,0xfe,0x7e,0x7f,0x6f,0x35,0x30]
-v_rsq_f16 v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
-// GFX12: v_rsq_f16_dpp v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xac,0xfe,0x7e,0x7f,0x6f,0x35,0x30]
+v_rsq_f16 v5.h, v1.h row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_rsq_f16_dpp v5.h, v1.h row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xac,0x0a,0x7f,0x81,0x60,0x09,0x13]
+
+v_rsq_f16 v127.h, -|v127.h| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_rsq_f16_dpp v127.h, -|v127.h| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xac,0xfe,0x7f,0xff,0x6f,0x35,0x30]
v_rsq_f32 v5, v1 quad_perm:[3,2,1,0]
// GFX12: v_rsq_f32_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x5c,0x0a,0x7e,0x01,0x1b,0x00,0xff]
@@ -2770,47 +2806,53 @@ v_sin_f32 v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
v_sin_f32 v255, -|v255| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
// GFX12: v_sin_f32_dpp v255, -|v255| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0x6a,0xfe,0x7f,0xff,0x6f,0x35,0x30]
-v_sqrt_f16 v5, v1 quad_perm:[3,2,1,0]
-// GFX12: v_sqrt_f16_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x1b,0x00,0xff]
+v_sqrt_f16 v5.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: v_sqrt_f16_dpp v5.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x1b,0x00,0xff]
+
+v_sqrt_f16 v5.l, v1.l quad_perm:[0,1,2,3]
+// GFX12: v_sqrt_f16_dpp v5.l, v1.l quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0xe4,0x00,0xff]
+
+v_sqrt_f16 v5.l, v1.l row_mirror
+// GFX12: v_sqrt_f16_dpp v5.l, v1.l row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x40,0x01,0xff]
-v_sqrt_f16 v5, v1 quad_perm:[0,1,2,3]
-// GFX12: v_sqrt_f16_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0xe4,0x00,0xff]
+v_sqrt_f16 v5.l, v1.l row_half_mirror
+// GFX12: v_sqrt_f16_dpp v5.l, v1.l row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x41,0x01,0xff]
-v_sqrt_f16 v5, v1 row_mirror
-// GFX12: v_sqrt_f16_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x40,0x01,0xff]
+v_sqrt_f16 v5.l, v1.l row_shl:1
+// GFX12: v_sqrt_f16_dpp v5.l, v1.l row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x01,0x01,0xff]
-v_sqrt_f16 v5, v1 row_half_mirror
-// GFX12: v_sqrt_f16_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x41,0x01,0xff]
+v_sqrt_f16 v5.l, v1.l row_shl:15
+// GFX12: v_sqrt_f16_dpp v5.l, v1.l row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x0f,0x01,0xff]
-v_sqrt_f16 v5, v1 row_shl:1
-// GFX12: v_sqrt_f16_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x01,0x01,0xff]
+v_sqrt_f16 v5.l, v1.l row_shr:1
+// GFX12: v_sqrt_f16_dpp v5.l, v1.l row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x11,0x01,0xff]
-v_sqrt_f16 v5, v1 row_shl:15
-// GFX12: v_sqrt_f16_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x0f,0x01,0xff]
+v_sqrt_f16 v5.l, v1.l row_shr:15
+// GFX12: v_sqrt_f16_dpp v5.l, v1.l row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x1f,0x01,0xff]
-v_sqrt_f16 v5, v1 row_shr:1
-// GFX12: v_sqrt_f16_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x11,0x01,0xff]
+v_sqrt_f16 v5.l, v1.l row_ror:1
+// GFX12: v_sqrt_f16_dpp v5.l, v1.l row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x21,0x01,0xff]
-v_sqrt_f16 v5, v1 row_shr:15
-// GFX12: v_sqrt_f16_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x1f,0x01,0xff]
+v_sqrt_f16 v5.l, v1.l row_ror:15
+// GFX12: v_sqrt_f16_dpp v5.l, v1.l row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x2f,0x01,0xff]
-v_sqrt_f16 v5, v1 row_ror:1
-// GFX12: v_sqrt_f16_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x21,0x01,0xff]
+v_sqrt_f16 v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf
+// GFX12: v_sqrt_f16_dpp v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x50,0x01,0xff]
-v_sqrt_f16 v5, v1 row_ror:15
-// GFX12: v_sqrt_f16_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x2f,0x01,0xff]
+v_sqrt_f16 v5.l, v1.l row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX12: v_sqrt_f16_dpp v5.l, v1.l row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x5f,0x01,0x01]
-v_sqrt_f16 v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
-// GFX12: v_sqrt_f16_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x50,0x01,0xff]
+v_sqrt_f16 v5.l, v1.l row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_sqrt_f16_dpp v5.l, v1.l row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x60,0x09,0x13]
-v_sqrt_f16 v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1
-// GFX12: v_sqrt_f16_dpp v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x5f,0x01,0x01]
+v_sqrt_f16 v127.l, -|v127.l| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_sqrt_f16_dpp v127.l, -|v127.l| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xaa,0xfe,0x7e,0x7f,0x6f,0x35,0x30]
-v_sqrt_f16 v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
-// GFX12: v_sqrt_f16_dpp v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xaa,0x0a,0x7e,0x01,0x60,0x09,0x13]
+v_sqrt_f16 v5.h, v1.h row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_sqrt_f16_dpp v5.h, v1.h row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xaa,0x0a,0x7f,0x81,0x60,0x09,0x13]
-v_sqrt_f16 v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
-// GFX12: v_sqrt_f16_dpp v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xaa,0xfe,0x7e,0x7f,0x6f,0x35,0x30]
+v_sqrt_f16 v127.h, -|v127.h| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_sqrt_f16_dpp v127.h, -|v127.h| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xaa,0xfe,0x7f,0xff,0x6f,0x35,0x30]
v_sqrt_f32 v5, v1 quad_perm:[3,2,1,0]
// GFX12: v_sqrt_f32_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x66,0x0a,0x7e,0x01,0x1b,0x00,0xff]
diff --git a/llvm/test/MC/AMDGPU/gfx12_asm_vop1_dpp8.s b/llvm/test/MC/AMDGPU/gfx12_asm_vop1_dpp8.s
index ecf408ee85445..fe56179e77775 100644
--- a/llvm/test/MC/AMDGPU/gfx12_asm_vop1_dpp8.s
+++ b/llvm/test/MC/AMDGPU/gfx12_asm_vop1_dpp8.s
@@ -13,14 +13,20 @@ v_bfrev_b32 v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1
v_bfrev_b32 v255, v255 dpp8:[0,0,0,0,0,0,0,0] fi:0
// GFX12: v_bfrev_b32_dpp v255, v255 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0x70,0xfe,0x7f,0xff,0x00,0x00,0x00]
-v_ceil_f16 v5, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_ceil_f16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xb8,0x0a,0x7e,0x01,0x77,0x39,0x05]
+v_ceil_f16 v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_ceil_f16_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xb8,0x0a,0x7e,0x01,0x77,0x39,0x05]
-v_ceil_f16 v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1
-// GFX12: v_ceil_f16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xb8,0x0a,0x7e,0x01,0x77,0x39,0x05]
+v_ceil_f16 v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_ceil_f16_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xb8,0x0a,0x7e,0x01,0x77,0x39,0x05]
-v_ceil_f16 v127, v127 dpp8:[0,0,0,0,0,0,0,0] fi:0
-// GFX12: v_ceil_f16_dpp v127, v127 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xb8,0xfe,0x7e,0x7f,0x00,0x00,0x00]
+v_ceil_f16 v127.l, v127.l dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_ceil_f16_dpp v127.l, v127.l dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xb8,0xfe,0x7e,0x7f,0x00,0x00,0x00]
+
+v_ceil_f16 v5.h, v1.h dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_ceil_f16_dpp v5.h, v1.h dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xb8,0x0a,0x7f,0x81,0x77,0x39,0x05]
+
+v_ceil_f16 v127.h, v127.h dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_ceil_f16_dpp v127.h, v127.h dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xb8,0xfe,0x7f,0xff,0x00,0x00,0x00]
v_ceil_f32 v5, v1 dpp8:[7,6,5,4,3,2,1,0]
// GFX12: v_ceil_f32_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0x44,0x0a,0x7e,0x01,0x77,0x39,0x05]
@@ -361,14 +367,20 @@ v_cvt_u32_u16 v5, v1.h dpp8:[7,6,5,4,3,2,1,0] fi:1
v_cvt_u32_u16 v255, v127.h dpp8:[0,0,0,0,0,0,0,0] fi:0
// GFX12: v_cvt_u32_u16_dpp v255, v127.h dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xd6,0xfe,0x7f,0xff,0x00,0x00,0x00]
-v_exp_f16 v5, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_exp_f16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xb0,0x0a,0x7e,0x01,0x77,0x39,0x05]
+v_exp_f16 v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_exp_f16_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xb0,0x0a,0x7e,0x01,0x77,0x39,0x05]
+
+v_exp_f16 v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_exp_f16_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xb0,0x0a,0x7e,0x01,0x77,0x39,0x05]
+
+v_exp_f16 v127.l, v127.l dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_exp_f16_dpp v127.l, v127.l dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xb0,0xfe,0x7e,0x7f,0x00,0x00,0x00]
-v_exp_f16 v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1
-// GFX12: v_exp_f16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xb0,0x0a,0x7e,0x01,0x77,0x39,0x05]
+v_exp_f16 v5.h, v1.h dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_exp_f16_dpp v5.h, v1.h dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xb0,0x0a,0x7f,0x81,0x77,0x39,0x05]
-v_exp_f16 v127, v127 dpp8:[0,0,0,0,0,0,0,0] fi:0
-// GFX12: v_exp_f16_dpp v127, v127 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xb0,0xfe,0x7e,0x7f,0x00,0x00,0x00]
+v_exp_f16 v127.h, v127.h dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_exp_f16_dpp v127.h, v127.h dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xb0,0xfe,0x7f,0xff,0x00,0x00,0x00]
v_exp_f32 v5, v1 dpp8:[7,6,5,4,3,2,1,0]
// GFX12: v_exp_f32_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0x4a,0x0a,0x7e,0x01,0x77,0x39,0x05]
@@ -406,14 +418,20 @@ v_ffbl_b32 v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1
v_ffbl_b32 v255, v255 dpp8:[0,0,0,0,0,0,0,0] fi:0
// GFX12: v_ctz_i32_b32_dpp v255, v255 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0x74,0xfe,0x7f,0xff,0x00,0x00,0x00]
-v_floor_f16 v5, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_floor_f16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xb6,0x0a,0x7e,0x01,0x77,0x39,0x05]
+v_floor_f16 v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_floor_f16_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xb6,0x0a,0x7e,0x01,0x77,0x39,0x05]
-v_floor_f16 v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1
-// GFX12: v_floor_f16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xb6,0x0a,0x7e,0x01,0x77,0x39,0x05]
+v_floor_f16 v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_floor_f16_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xb6,0x0a,0x7e,0x01,0x77,0x39,0x05]
-v_floor_f16 v127, v127 dpp8:[0,0,0,0,0,0,0,0] fi:0
-// GFX12: v_floor_f16_dpp v127, v127 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xb6,0xfe,0x7e,0x7f,0x00,0x00,0x00]
+v_floor_f16 v127.l, v127.l dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_floor_f16_dpp v127.l, v127.l dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xb6,0xfe,0x7e,0x7f,0x00,0x00,0x00]
+
+v_floor_f16 v5.h, v1.h dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_floor_f16_dpp v5.h, v1.h dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xb6,0x0a,0x7f,0x81,0x77,0x39,0x05]
+
+v_floor_f16 v127.h, v127.h dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_floor_f16_dpp v127.h, v127.h dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xb6,0xfe,0x7f,0xff,0x00,0x00,0x00]
v_floor_f32 v5, v1 dpp8:[7,6,5,4,3,2,1,0]
// GFX12: v_floor_f32_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0x48,0x0a,0x7e,0x01,0x77,0x39,0x05]
@@ -496,14 +514,20 @@ v_frexp_mant_f32 v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1
v_frexp_mant_f32 v255, v255 dpp8:[0,0,0,0,0,0,0,0] fi:0
// GFX12: v_frexp_mant_f32_dpp v255, v255 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0x80,0xfe,0x7f,0xff,0x00,0x00,0x00]
-v_log_f16 v5, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_log_f16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xae,0x0a,0x7e,0x01,0x77,0x39,0x05]
+v_log_f16 v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_log_f16_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xae,0x0a,0x7e,0x01,0x77,0x39,0x05]
+
+v_log_f16 v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_log_f16_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xae,0x0a,0x7e,0x01,0x77,0x39,0x05]
-v_log_f16 v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1
-// GFX12: v_log_f16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xae,0x0a,0x7e,0x01,0x77,0x39,0x05]
+v_log_f16 v127.l, v127.l dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_log_f16_dpp v127.l, v127.l dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xae,0xfe,0x7e,0x7f,0x00,0x00,0x00]
-v_log_f16 v127, v127 dpp8:[0,0,0,0,0,0,0,0] fi:0
-// GFX12: v_log_f16_dpp v127, v127 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xae,0xfe,0x7e,0x7f,0x00,0x00,0x00]
+v_log_f16 v5.h, v1.h dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_log_f16_dpp v5.h, v1.h dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xae,0x0a,0x7f,0x81,0x77,0x39,0x05]
+
+v_log_f16 v127.h, v127.h dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_log_f16_dpp v127.h, v127.h dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xae,0xfe,0x7f,0xff,0x00,0x00,0x00]
v_log_f32 v5, v1 dpp8:[7,6,5,4,3,2,1,0]
// GFX12: v_log_f32_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0x4e,0x0a,0x7e,0x01,0x77,0x39,0x05]
@@ -583,14 +607,20 @@ v_not_b32 v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1
v_not_b32 v255, v255 dpp8:[0,0,0,0,0,0,0,0] fi:0
// GFX12: v_not_b32_dpp v255, v255 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0x6e,0xfe,0x7f,0xff,0x00,0x00,0x00]
-v_rcp_f16 v5, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_rcp_f16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xa8,0x0a,0x7e,0x01,0x77,0x39,0x05]
+v_rcp_f16 v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_rcp_f16_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xa8,0x0a,0x7e,0x01,0x77,0x39,0x05]
+
+v_rcp_f16 v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_rcp_f16_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xa8,0x0a,0x7e,0x01,0x77,0x39,0x05]
-v_rcp_f16 v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1
-// GFX12: v_rcp_f16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xa8,0x0a,0x7e,0x01,0x77,0x39,0x05]
+v_rcp_f16 v127.l, v127.l dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_rcp_f16_dpp v127.l, v127.l dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xa8,0xfe,0x7e,0x7f,0x00,0x00,0x00]
-v_rcp_f16 v127, v127 dpp8:[0,0,0,0,0,0,0,0] fi:0
-// GFX12: v_rcp_f16_dpp v127, v127 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xa8,0xfe,0x7e,0x7f,0x00,0x00,0x00]
+v_rcp_f16 v5.h, v1.h dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_rcp_f16_dpp v5.h, v1.h dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xa8,0x0a,0x7f,0x81,0x77,0x39,0x05]
+
+v_rcp_f16 v127.h, v127.h dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_rcp_f16_dpp v127.h, v127.h dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xa8,0xfe,0x7f,0xff,0x00,0x00,0x00]
v_rcp_f32 v5, v1 dpp8:[7,6,5,4,3,2,1,0]
// GFX12: v_rcp_f32_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0x54,0x0a,0x7e,0x01,0x77,0x39,0x05]
@@ -634,14 +664,20 @@ v_rndne_f32 v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1
v_rndne_f32 v255, v255 dpp8:[0,0,0,0,0,0,0,0] fi:0
// GFX12: v_rndne_f32_dpp v255, v255 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0x46,0xfe,0x7f,0xff,0x00,0x00,0x00]
-v_rsq_f16 v5, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_rsq_f16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xac,0x0a,0x7e,0x01,0x77,0x39,0x05]
+v_rsq_f16 v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_rsq_f16_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xac,0x0a,0x7e,0x01,0x77,0x39,0x05]
+
+v_rsq_f16 v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_rsq_f16_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xac,0x0a,0x7e,0x01,0x77,0x39,0x05]
-v_rsq_f16 v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1
-// GFX12: v_rsq_f16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xac,0x0a,0x7e,0x01,0x77,0x39,0x05]
+v_rsq_f16 v127.l, v127.l dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_rsq_f16_dpp v127.l, v127.l dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xac,0xfe,0x7e,0x7f,0x00,0x00,0x00]
-v_rsq_f16 v127, v127 dpp8:[0,0,0,0,0,0,0,0] fi:0
-// GFX12: v_rsq_f16_dpp v127, v127 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xac,0xfe,0x7e,0x7f,0x00,0x00,0x00]
+v_rsq_f16 v5.h, v1.h dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_rsq_f16_dpp v5.h, v1.h dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xac,0x0a,0x7f,0x81,0x77,0x39,0x05]
+
+v_rsq_f16 v127.h, v127.h dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_rsq_f16_dpp v127.h, v127.h dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xac,0xfe,0x7f,0xff,0x00,0x00,0x00]
v_rsq_f32 v5, v1 dpp8:[7,6,5,4,3,2,1,0]
// GFX12: v_rsq_f32_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0x5c,0x0a,0x7e,0x01,0x77,0x39,0x05]
@@ -691,14 +727,20 @@ v_sin_f32 v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1
v_sin_f32 v255, v255 dpp8:[0,0,0,0,0,0,0,0] fi:0
// GFX12: v_sin_f32_dpp v255, v255 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0x6a,0xfe,0x7f,0xff,0x00,0x00,0x00]
-v_sqrt_f16 v5, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_sqrt_f16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xaa,0x0a,0x7e,0x01,0x77,0x39,0x05]
+v_sqrt_f16 v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_sqrt_f16_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xaa,0x0a,0x7e,0x01,0x77,0x39,0x05]
+
+v_sqrt_f16 v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_sqrt_f16_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xaa,0x0a,0x7e,0x01,0x77,0x39,0x05]
+
+v_sqrt_f16 v127.l, v127.l dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_sqrt_f16_dpp v127.l, v127.l dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xaa,0xfe,0x7e,0x7f,0x00,0x00,0x00]
-v_sqrt_f16 v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1
-// GFX12: v_sqrt_f16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xaa,0x0a,0x7e,0x01,0x77,0x39,0x05]
+v_sqrt_f16 v5.h, v1.h dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_sqrt_f16_dpp v5.h, v1.h dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xaa,0x0a,0x7f,0x81,0x77,0x39,0x05]
-v_sqrt_f16 v127, v127 dpp8:[0,0,0,0,0,0,0,0] fi:0
-// GFX12: v_sqrt_f16_dpp v127, v127 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xaa,0xfe,0x7e,0x7f,0x00,0x00,0x00]
+v_sqrt_f16 v127.h, v127.h dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_sqrt_f16_dpp v127.h, v127.h dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xaa,0xfe,0x7f,0xff,0x00,0x00,0x00]
v_sqrt_f32 v5, v1 dpp8:[7,6,5,4,3,2,1,0]
// GFX12: v_sqrt_f32_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0x66,0x0a,0x7e,0x01,0x77,0x39,0x05]
diff --git a/llvm/test/MC/AMDGPU/gfx12_asm_vop1_t16_err.s b/llvm/test/MC/AMDGPU/gfx12_asm_vop1_t16_err.s
index 158684b85b098..4ad22f8732e91 100644
--- a/llvm/test/MC/AMDGPU/gfx12_asm_vop1_t16_err.s
+++ b/llvm/test/MC/AMDGPU/gfx12_asm_vop1_t16_err.s
@@ -5,6 +5,12 @@
v_ceil_f16_e32 v128, 0xfe0b
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
+v_ceil_f16_e32 v128.h, 0xfe0b
+// GFX12: :[[@LINE-1]]:16: error: invalid operand for instruction
+
+v_ceil_f16_e32 v128.l, 0xfe0b
+// GFX12: :[[@LINE-1]]:16: error: invalid operand for instruction
+
v_ceil_f16_e32 v255, v1
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
@@ -14,6 +20,24 @@ v_ceil_f16_e32 v255, v1 dpp8:[7,6,5,4,3,2,1,0]
v_ceil_f16_e32 v255, v1 quad_perm:[3,2,1,0]
// GFX12: :[[@LINE-1]]:25: error: invalid operand for instruction
+v_ceil_f16_e32 v255.h, v1.h
+// GFX12: :[[@LINE-1]]:16: error: invalid operand for instruction
+
+v_ceil_f16_e32 v255.h, v1.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:16: error: invalid operand for instruction
+
+v_ceil_f16_e32 v255.h, v1.h quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:16: error: invalid operand for instruction
+
+v_ceil_f16_e32 v255.l, v1.l
+// GFX12: :[[@LINE-1]]:16: error: invalid operand for instruction
+
+v_ceil_f16_e32 v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:16: error: invalid operand for instruction
+
+v_ceil_f16_e32 v255.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:16: error: invalid operand for instruction
+
v_ceil_f16_e32 v5, v199
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
@@ -23,6 +47,24 @@ v_ceil_f16_e32 v5, v199 dpp8:[7,6,5,4,3,2,1,0]
v_ceil_f16_e32 v5, v199 quad_perm:[3,2,1,0]
// GFX12: :[[@LINE-1]]:25: error: invalid operand for instruction
+v_ceil_f16_e32 v5.h, v199.h
+// GFX12: :[[@LINE-1]]:22: error: invalid operand for instruction
+
+v_ceil_f16_e32 v5.h, v199.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:22: error: invalid operand for instruction
+
+v_ceil_f16_e32 v5.h, v199.h quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:22: error: invalid operand for instruction
+
+v_ceil_f16_e32 v5.l, v199.l
+// GFX12: :[[@LINE-1]]:22: error: invalid operand for instruction
+
+v_ceil_f16_e32 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:22: error: invalid operand for instruction
+
+v_ceil_f16_e32 v5.l, v199.l quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:22: error: invalid operand for instruction
+
v_cos_f16_e32 v128, 0xfe0b
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
@@ -467,6 +509,12 @@ v_cvt_u32_u16_e32 v5, v199.l quad_perm:[3,2,1,0]
v_exp_f16_e32 v128, 0xfe0b
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
+v_exp_f16_e32 v128.h, 0xfe0b
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_exp_f16_e32 v128.l, 0xfe0b
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
v_exp_f16_e32 v255, v1
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
@@ -476,6 +524,24 @@ v_exp_f16_e32 v255, v1 dpp8:[7,6,5,4,3,2,1,0]
v_exp_f16_e32 v255, v1 quad_perm:[3,2,1,0]
// GFX12: :[[@LINE-1]]:24: error: invalid operand for instruction
+v_exp_f16_e32 v255.h, v1.h
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_exp_f16_e32 v255.h, v1.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_exp_f16_e32 v255.h, v1.h quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_exp_f16_e32 v255.l, v1.l
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_exp_f16_e32 v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_exp_f16_e32 v255.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
v_exp_f16_e32 v5, v199
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
@@ -485,9 +551,33 @@ v_exp_f16_e32 v5, v199 dpp8:[7,6,5,4,3,2,1,0]
v_exp_f16_e32 v5, v199 quad_perm:[3,2,1,0]
// GFX12: :[[@LINE-1]]:24: error: invalid operand for instruction
+v_exp_f16_e32 v5.h, v199.h
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
+v_exp_f16_e32 v5.h, v199.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
+v_exp_f16_e32 v5.h, v199.h quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
+v_exp_f16_e32 v5.l, v199.l
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
+v_exp_f16_e32 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
+v_exp_f16_e32 v5.l, v199.l quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
v_floor_f16_e32 v128, 0xfe0b
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
+v_floor_f16_e32 v128.h, 0xfe0b
+// GFX12: :[[@LINE-1]]:17: error: invalid operand for instruction
+
+v_floor_f16_e32 v128.l, 0xfe0b
+// GFX12: :[[@LINE-1]]:17: error: invalid operand for instruction
+
v_floor_f16_e32 v255, v1
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
@@ -497,6 +587,24 @@ v_floor_f16_e32 v255, v1 dpp8:[7,6,5,4,3,2,1,0]
v_floor_f16_e32 v255, v1 quad_perm:[3,2,1,0]
// GFX12: :[[@LINE-1]]:26: error: invalid operand for instruction
+v_floor_f16_e32 v255.h, v1.h
+// GFX12: :[[@LINE-1]]:17: error: invalid operand for instruction
+
+v_floor_f16_e32 v255.h, v1.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:17: error: invalid operand for instruction
+
+v_floor_f16_e32 v255.h, v1.h quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:17: error: invalid operand for instruction
+
+v_floor_f16_e32 v255.l, v1.l
+// GFX12: :[[@LINE-1]]:17: error: invalid operand for instruction
+
+v_floor_f16_e32 v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:17: error: invalid operand for instruction
+
+v_floor_f16_e32 v255.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:17: error: invalid operand for instruction
+
v_floor_f16_e32 v5, v199
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
@@ -506,6 +614,24 @@ v_floor_f16_e32 v5, v199 dpp8:[7,6,5,4,3,2,1,0]
v_floor_f16_e32 v5, v199 quad_perm:[3,2,1,0]
// GFX12: :[[@LINE-1]]:26: error: invalid operand for instruction
+v_floor_f16_e32 v5.h, v199.h
+// GFX12: :[[@LINE-1]]:23: error: invalid operand for instruction
+
+v_floor_f16_e32 v5.h, v199.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:23: error: invalid operand for instruction
+
+v_floor_f16_e32 v5.h, v199.h quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:23: error: invalid operand for instruction
+
+v_floor_f16_e32 v5.l, v199.l
+// GFX12: :[[@LINE-1]]:23: error: invalid operand for instruction
+
+v_floor_f16_e32 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:23: error: invalid operand for instruction
+
+v_floor_f16_e32 v5.l, v199.l quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:23: error: invalid operand for instruction
+
v_fract_f16_e32 v128, 0xfe0b
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
@@ -677,6 +803,12 @@ v_frexp_mant_f16_e32 v5.l, v199.l quad_perm:[3,2,1,0]
v_log_f16_e32 v128, 0xfe0b
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
+v_log_f16_e32 v128.h, 0xfe0b
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_log_f16_e32 v128.l, 0xfe0b
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
v_log_f16_e32 v255, v1
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
@@ -686,6 +818,24 @@ v_log_f16_e32 v255, v1 dpp8:[7,6,5,4,3,2,1,0]
v_log_f16_e32 v255, v1 quad_perm:[3,2,1,0]
// GFX12: :[[@LINE-1]]:24: error: invalid operand for instruction
+v_log_f16_e32 v255.h, v1.h
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_log_f16_e32 v255.h, v1.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_log_f16_e32 v255.h, v1.h quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_log_f16_e32 v255.l, v1.l
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_log_f16_e32 v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_log_f16_e32 v255.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
v_log_f16_e32 v5, v199
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
@@ -695,6 +845,24 @@ v_log_f16_e32 v5, v199 dpp8:[7,6,5,4,3,2,1,0]
v_log_f16_e32 v5, v199 quad_perm:[3,2,1,0]
// GFX12: :[[@LINE-1]]:24: error: invalid operand for instruction
+v_log_f16_e32 v5.h, v199.h
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
+v_log_f16_e32 v5.h, v199.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
+v_log_f16_e32 v5.h, v199.h quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
+v_log_f16_e32 v5.l, v199.l
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
+v_log_f16_e32 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
+v_log_f16_e32 v5.l, v199.l quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
v_not_b16_e32 v128, 0xfe0b
// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
@@ -761,6 +929,12 @@ v_not_b16_e32 v5.l, v199.l quad_perm:[3,2,1,0]
v_rcp_f16_e32 v128, 0xfe0b
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
+v_rcp_f16_e32 v128.h, 0xfe0b
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_rcp_f16_e32 v128.l, 0xfe0b
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
v_rcp_f16_e32 v255, v1
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
@@ -770,6 +944,24 @@ v_rcp_f16_e32 v255, v1 dpp8:[7,6,5,4,3,2,1,0]
v_rcp_f16_e32 v255, v1 quad_perm:[3,2,1,0]
// GFX12: :[[@LINE-1]]:24: error: invalid operand for instruction
+v_rcp_f16_e32 v255.h, v1.h
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_rcp_f16_e32 v255.h, v1.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_rcp_f16_e32 v255.h, v1.h quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_rcp_f16_e32 v255.l, v1.l
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_rcp_f16_e32 v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_rcp_f16_e32 v255.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
v_rcp_f16_e32 v5, v199
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
@@ -779,6 +971,24 @@ v_rcp_f16_e32 v5, v199 dpp8:[7,6,5,4,3,2,1,0]
v_rcp_f16_e32 v5, v199 quad_perm:[3,2,1,0]
// GFX12: :[[@LINE-1]]:24: error: invalid operand for instruction
+v_rcp_f16_e32 v5.h, v199.h
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
+v_rcp_f16_e32 v5.h, v199.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
+v_rcp_f16_e32 v5.h, v199.h quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
+v_rcp_f16_e32 v5.l, v199.l
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
+v_rcp_f16_e32 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
+v_rcp_f16_e32 v5.l, v199.l quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
v_rndne_f16_e32 v128, 0xfe0b
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
@@ -845,6 +1055,12 @@ v_rndne_f16_e32 v5.l, v199.l quad_perm:[3,2,1,0]
v_rsq_f16_e32 v128, 0xfe0b
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
+v_rsq_f16_e32 v128.h, 0xfe0b
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_rsq_f16_e32 v128.l, 0xfe0b
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
v_rsq_f16_e32 v255, v1
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
@@ -854,6 +1070,24 @@ v_rsq_f16_e32 v255, v1 dpp8:[7,6,5,4,3,2,1,0]
v_rsq_f16_e32 v255, v1 quad_perm:[3,2,1,0]
// GFX12: :[[@LINE-1]]:24: error: invalid operand for instruction
+v_rsq_f16_e32 v255.h, v1.h
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_rsq_f16_e32 v255.h, v1.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_rsq_f16_e32 v255.h, v1.h quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_rsq_f16_e32 v255.l, v1.l
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_rsq_f16_e32 v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
+v_rsq_f16_e32 v255.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:15: error: invalid operand for instruction
+
v_rsq_f16_e32 v5, v199
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
@@ -863,6 +1097,24 @@ v_rsq_f16_e32 v5, v199 dpp8:[7,6,5,4,3,2,1,0]
v_rsq_f16_e32 v5, v199 quad_perm:[3,2,1,0]
// GFX12: :[[@LINE-1]]:24: error: invalid operand for instruction
+v_rsq_f16_e32 v5.h, v199.h
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
+v_rsq_f16_e32 v5.h, v199.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
+v_rsq_f16_e32 v5.h, v199.h quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
+v_rsq_f16_e32 v5.l, v199.l
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
+v_rsq_f16_e32 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
+v_rsq_f16_e32 v5.l, v199.l quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:21: error: invalid operand for instruction
+
v_sat_pk_u8_i16_e32 v199, v5
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
@@ -956,6 +1208,12 @@ v_sin_f16_e32 v5.l, v199.l quad_perm:[3,2,1,0]
v_sqrt_f16_e32 v128, 0xfe0b
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
+v_sqrt_f16_e32 v128.h, 0xfe0b
+// GFX12: :[[@LINE-1]]:16: error: invalid operand for instruction
+
+v_sqrt_f16_e32 v128.l, 0xfe0b
+// GFX12: :[[@LINE-1]]:16: error: invalid operand for instruction
+
v_sqrt_f16_e32 v255, v1
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
@@ -965,6 +1223,24 @@ v_sqrt_f16_e32 v255, v1 dpp8:[7,6,5,4,3,2,1,0]
v_sqrt_f16_e32 v255, v1 quad_perm:[3,2,1,0]
// GFX12: :[[@LINE-1]]:25: error: invalid operand for instruction
+v_sqrt_f16_e32 v255.h, v1.h
+// GFX12: :[[@LINE-1]]:16: error: invalid operand for instruction
+
+v_sqrt_f16_e32 v255.h, v1.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:16: error: invalid operand for instruction
+
+v_sqrt_f16_e32 v255.h, v1.h quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:16: error: invalid operand for instruction
+
+v_sqrt_f16_e32 v255.l, v1.l
+// GFX12: :[[@LINE-1]]:16: error: invalid operand for instruction
+
+v_sqrt_f16_e32 v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:16: error: invalid operand for instruction
+
+v_sqrt_f16_e32 v255.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:16: error: invalid operand for instruction
+
v_sqrt_f16_e32 v5, v199
// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode
@@ -974,6 +1250,24 @@ v_sqrt_f16_e32 v5, v199 dpp8:[7,6,5,4,3,2,1,0]
v_sqrt_f16_e32 v5, v199 quad_perm:[3,2,1,0]
// GFX12: :[[@LINE-1]]:25: error: invalid operand for instruction
+v_sqrt_f16_e32 v5.h, v199.h
+// GFX12: :[[@LINE-1]]:22: error: invalid operand for instruction
+
+v_sqrt_f16_e32 v5.h, v199.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:22: error: invalid operand for instruction
+
+v_sqrt_f16_e32 v5.h, v199.h quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:22: error: invalid operand for instruction
+
+v_sqrt_f16_e32 v5.l, v199.l
+// GFX12: :[[@LINE-1]]:22: error: invalid operand for instruction
+
+v_sqrt_f16_e32 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: :[[@LINE-1]]:22: error: invalid operand for instruction
+
+v_sqrt_f16_e32 v5.l, v199.l quad_perm:[3,2,1,0]
+// GFX12: :[[@LINE-1]]:22: error: invalid operand for instruction
+
v_swap_b16_e32 v0.l, 0
// GFX12: :[[@LINE-1]]:22: error: invalid operand for instruction
diff --git a/llvm/test/MC/AMDGPU/gfx12_asm_vop1_t16_promote.s b/llvm/test/MC/AMDGPU/gfx12_asm_vop1_t16_promote.s
index e5c8d761710b0..8a8b16fb7afaa 100644
--- a/llvm/test/MC/AMDGPU/gfx12_asm_vop1_t16_promote.s
+++ b/llvm/test/MC/AMDGPU/gfx12_asm_vop1_t16_promote.s
@@ -1,71 +1,137 @@
// NOTE: Assertions have been autogenerated by utils/update_mc_test_checks.py UTC_ARGS: --unique --version 5
// RUN: llvm-mc -triple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 -show-encoding %s | FileCheck --check-prefix=GFX12 --implicit-check-not=_e32 %s
-v_ceil_f16 v128, 0xfe0b
-// GFX12: v_ceil_f16_e64 v128, 0xfe0b ; encoding: [0x80,0x00,0xdc,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+v_ceil_f16 v128.h, 0xfe0b
+// GFX12: v_ceil_f16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xdc,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
-v_ceil_f16 v255, -1
-// GFX12: v_ceil_f16_e64 v255, -1 ; encoding: [0xff,0x00,0xdc,0xd5,0xc1,0x00,0x00,0x00]
+v_ceil_f16 v128.l, 0xfe0b
+// GFX12: v_ceil_f16_e64 v128.l, 0xfe0b ; encoding: [0x80,0x00,0xdc,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
-v_ceil_f16 v255, 0.5
-// GFX12: v_ceil_f16_e64 v255, 0.5 ; encoding: [0xff,0x00,0xdc,0xd5,0xf0,0x00,0x00,0x00]
+v_ceil_f16 v255.h, -1
+// GFX12: v_ceil_f16_e64 v255.h, -1 op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0xc1,0x00,0x00,0x00]
-v_ceil_f16 v255, exec_hi
-// GFX12: v_ceil_f16_e64 v255, exec_hi ; encoding: [0xff,0x00,0xdc,0xd5,0x7f,0x00,0x00,0x00]
+v_ceil_f16 v255.h, 0.5
+// GFX12: v_ceil_f16_e64 v255.h, 0.5 op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0xf0,0x00,0x00,0x00]
-v_ceil_f16 v255, exec_lo
-// GFX12: v_ceil_f16_e64 v255, exec_lo ; encoding: [0xff,0x00,0xdc,0xd5,0x7e,0x00,0x00,0x00]
+v_ceil_f16 v255.h, exec_hi
+// GFX12: v_ceil_f16_e64 v255.h, exec_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0x7f,0x00,0x00,0x00]
-v_ceil_f16 v255, m0
-// GFX12: v_ceil_f16_e64 v255, m0 ; encoding: [0xff,0x00,0xdc,0xd5,0x7d,0x00,0x00,0x00]
+v_ceil_f16 v255.h, exec_lo
+// GFX12: v_ceil_f16_e64 v255.h, exec_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0x7e,0x00,0x00,0x00]
-v_ceil_f16 v255, null
-// GFX12: v_ceil_f16_e64 v255, null ; encoding: [0xff,0x00,0xdc,0xd5,0x7c,0x00,0x00,0x00]
+v_ceil_f16 v255.h, m0
+// GFX12: v_ceil_f16_e64 v255.h, m0 op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0x7d,0x00,0x00,0x00]
-v_ceil_f16 v255, s1
-// GFX12: v_ceil_f16_e64 v255, s1 ; encoding: [0xff,0x00,0xdc,0xd5,0x01,0x00,0x00,0x00]
+v_ceil_f16 v255.h, null
+// GFX12: v_ceil_f16_e64 v255.h, null op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0x7c,0x00,0x00,0x00]
-v_ceil_f16 v255, s105
-// GFX12: v_ceil_f16_e64 v255, s105 ; encoding: [0xff,0x00,0xdc,0xd5,0x69,0x00,0x00,0x00]
+v_ceil_f16 v255.h, s1
+// GFX12: v_ceil_f16_e64 v255.h, s1 op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0x01,0x00,0x00,0x00]
-v_ceil_f16 v255, src_scc
-// GFX12: v_ceil_f16_e64 v255, src_scc ; encoding: [0xff,0x00,0xdc,0xd5,0xfd,0x00,0x00,0x00]
+v_ceil_f16 v255.h, s105
+// GFX12: v_ceil_f16_e64 v255.h, s105 op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0x69,0x00,0x00,0x00]
-v_ceil_f16 v255, ttmp15
-// GFX12: v_ceil_f16_e64 v255, ttmp15 ; encoding: [0xff,0x00,0xdc,0xd5,0x7b,0x00,0x00,0x00]
+v_ceil_f16 v255.h, src_scc
+// GFX12: v_ceil_f16_e64 v255.h, src_scc op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0xfd,0x00,0x00,0x00]
-v_ceil_f16 v255, v1
-// GFX12: v_ceil_f16_e64 v255, v1 ; encoding: [0xff,0x00,0xdc,0xd5,0x01,0x01,0x00,0x00]
+v_ceil_f16 v255.h, ttmp15
+// GFX12: v_ceil_f16_e64 v255.h, ttmp15 op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0x7b,0x00,0x00,0x00]
-v_ceil_f16 v255, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_ceil_f16_e64_dpp v255, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xdc,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+v_ceil_f16 v255.h, v1.h
+// GFX12: v_ceil_f16_e64 v255.h, v1.h op_sel:[1,1] ; encoding: [0xff,0x48,0xdc,0xd5,0x01,0x01,0x00,0x00]
-v_ceil_f16 v255, v1 quad_perm:[3,2,1,0]
-// GFX12: v_ceil_f16_e64_dpp v255, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+v_ceil_f16 v255.h, v1.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_ceil_f16_e64_dpp v255.h, v1.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xdc,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
-v_ceil_f16 v255, v127
-// GFX12: v_ceil_f16_e64 v255, v127 ; encoding: [0xff,0x00,0xdc,0xd5,0x7f,0x01,0x00,0x00]
+v_ceil_f16 v255.h, v1.h quad_perm:[3,2,1,0]
+// GFX12: v_ceil_f16_e64_dpp v255.h, v1.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
-v_ceil_f16 v255, v127 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_ceil_f16_e64_dpp v255, v127 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xdc,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+v_ceil_f16 v255.h, v127.h
+// GFX12: v_ceil_f16_e64 v255.h, v127.h op_sel:[1,1] ; encoding: [0xff,0x48,0xdc,0xd5,0x7f,0x01,0x00,0x00]
-v_ceil_f16 v255, v127 quad_perm:[3,2,1,0]
-// GFX12: v_ceil_f16_e64_dpp v255, v127 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+v_ceil_f16 v255.h, v127.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_ceil_f16_e64_dpp v255.h, v127.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xdc,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
-v_ceil_f16 v255, vcc_hi
-// GFX12: v_ceil_f16_e64 v255, vcc_hi ; encoding: [0xff,0x00,0xdc,0xd5,0x6b,0x00,0x00,0x00]
+v_ceil_f16 v255.h, v127.h quad_perm:[3,2,1,0]
+// GFX12: v_ceil_f16_e64_dpp v255.h, v127.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
-v_ceil_f16 v255, vcc_lo
-// GFX12: v_ceil_f16_e64 v255, vcc_lo ; encoding: [0xff,0x00,0xdc,0xd5,0x6a,0x00,0x00,0x00]
+v_ceil_f16 v255.h, vcc_hi
+// GFX12: v_ceil_f16_e64 v255.h, vcc_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0x6b,0x00,0x00,0x00]
-v_ceil_f16 v5, v199
-// GFX12: v_ceil_f16_e64 v5, v199 ; encoding: [0x05,0x00,0xdc,0xd5,0xc7,0x01,0x00,0x00]
+v_ceil_f16 v255.h, vcc_lo
+// GFX12: v_ceil_f16_e64 v255.h, vcc_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xdc,0xd5,0x6a,0x00,0x00,0x00]
-v_ceil_f16 v5, v199 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_ceil_f16_e64_dpp v5, v199 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xdc,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+v_ceil_f16 v255.l, -1
+// GFX12: v_ceil_f16_e64 v255.l, -1 ; encoding: [0xff,0x00,0xdc,0xd5,0xc1,0x00,0x00,0x00]
-v_ceil_f16 v5, v199 quad_perm:[3,2,1,0]
-// GFX12: v_ceil_f16_e64_dpp v5, v199 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+v_ceil_f16 v255.l, 0.5
+// GFX12: v_ceil_f16_e64 v255.l, 0.5 ; encoding: [0xff,0x00,0xdc,0xd5,0xf0,0x00,0x00,0x00]
+
+v_ceil_f16 v255.l, exec_hi
+// GFX12: v_ceil_f16_e64 v255.l, exec_hi ; encoding: [0xff,0x00,0xdc,0xd5,0x7f,0x00,0x00,0x00]
+
+v_ceil_f16 v255.l, exec_lo
+// GFX12: v_ceil_f16_e64 v255.l, exec_lo ; encoding: [0xff,0x00,0xdc,0xd5,0x7e,0x00,0x00,0x00]
+
+v_ceil_f16 v255.l, m0
+// GFX12: v_ceil_f16_e64 v255.l, m0 ; encoding: [0xff,0x00,0xdc,0xd5,0x7d,0x00,0x00,0x00]
+
+v_ceil_f16 v255.l, null
+// GFX12: v_ceil_f16_e64 v255.l, null ; encoding: [0xff,0x00,0xdc,0xd5,0x7c,0x00,0x00,0x00]
+
+v_ceil_f16 v255.l, s1
+// GFX12: v_ceil_f16_e64 v255.l, s1 ; encoding: [0xff,0x00,0xdc,0xd5,0x01,0x00,0x00,0x00]
+
+v_ceil_f16 v255.l, s105
+// GFX12: v_ceil_f16_e64 v255.l, s105 ; encoding: [0xff,0x00,0xdc,0xd5,0x69,0x00,0x00,0x00]
+
+v_ceil_f16 v255.l, src_scc
+// GFX12: v_ceil_f16_e64 v255.l, src_scc ; encoding: [0xff,0x00,0xdc,0xd5,0xfd,0x00,0x00,0x00]
+
+v_ceil_f16 v255.l, ttmp15
+// GFX12: v_ceil_f16_e64 v255.l, ttmp15 ; encoding: [0xff,0x00,0xdc,0xd5,0x7b,0x00,0x00,0x00]
+
+v_ceil_f16 v255.l, v1.l
+// GFX12: v_ceil_f16_e64 v255.l, v1.l ; encoding: [0xff,0x00,0xdc,0xd5,0x01,0x01,0x00,0x00]
+
+v_ceil_f16 v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_ceil_f16_e64_dpp v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xdc,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+
+v_ceil_f16 v255.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: v_ceil_f16_e64_dpp v255.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+
+v_ceil_f16 v255.l, v127.l
+// GFX12: v_ceil_f16_e64 v255.l, v127.l ; encoding: [0xff,0x00,0xdc,0xd5,0x7f,0x01,0x00,0x00]
+
+v_ceil_f16 v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_ceil_f16_e64_dpp v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xdc,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+
+v_ceil_f16 v255.l, v127.l quad_perm:[3,2,1,0]
+// GFX12: v_ceil_f16_e64_dpp v255.l, v127.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+
+v_ceil_f16 v255.l, vcc_hi
+// GFX12: v_ceil_f16_e64 v255.l, vcc_hi ; encoding: [0xff,0x00,0xdc,0xd5,0x6b,0x00,0x00,0x00]
+
+v_ceil_f16 v255.l, vcc_lo
+// GFX12: v_ceil_f16_e64 v255.l, vcc_lo ; encoding: [0xff,0x00,0xdc,0xd5,0x6a,0x00,0x00,0x00]
+
+v_ceil_f16 v5.h, v199.h
+// GFX12: v_ceil_f16_e64 v5.h, v199.h op_sel:[1,1] ; encoding: [0x05,0x48,0xdc,0xd5,0xc7,0x01,0x00,0x00]
+
+v_ceil_f16 v5.h, v199.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_ceil_f16_e64_dpp v5.h, v199.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xdc,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+
+v_ceil_f16 v5.h, v199.h quad_perm:[3,2,1,0]
+// GFX12: v_ceil_f16_e64_dpp v5.h, v199.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x48,0xdc,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+
+v_ceil_f16 v5.l, v199.l
+// GFX12: v_ceil_f16_e64 v5.l, v199.l ; encoding: [0x05,0x00,0xdc,0xd5,0xc7,0x01,0x00,0x00]
+
+v_ceil_f16 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_ceil_f16_e64_dpp v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xdc,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+
+v_ceil_f16 v5.l, v199.l quad_perm:[3,2,1,0]
+// GFX12: v_ceil_f16_e64_dpp v5.l, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
v_cos_f16 v128.h, 0xfe0b
// GFX12: v_cos_f16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xe1,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
@@ -1120,137 +1186,269 @@ v_cvt_u32_u16 v5, v199.l dpp8:[7,6,5,4,3,2,1,0]
v_cvt_u32_u16 v5, v199.l quad_perm:[3,2,1,0]
// GFX12: v_cvt_u32_u16_e64_dpp v5, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xeb,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
-v_exp_f16 v128, 0xfe0b
-// GFX12: v_exp_f16_e64 v128, 0xfe0b ; encoding: [0x80,0x00,0xd8,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+v_exp_f16 v128.h, 0xfe0b
+// GFX12: v_exp_f16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xd8,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+
+v_exp_f16 v128.l, 0xfe0b
+// GFX12: v_exp_f16_e64 v128.l, 0xfe0b ; encoding: [0x80,0x00,0xd8,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+
+v_exp_f16 v255.h, -1
+// GFX12: v_exp_f16_e64 v255.h, -1 op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0xc1,0x00,0x00,0x00]
+
+v_exp_f16 v255.h, 0.5
+// GFX12: v_exp_f16_e64 v255.h, 0.5 op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0xf0,0x00,0x00,0x00]
+
+v_exp_f16 v255.h, exec_hi
+// GFX12: v_exp_f16_e64 v255.h, exec_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0x7f,0x00,0x00,0x00]
+
+v_exp_f16 v255.h, exec_lo
+// GFX12: v_exp_f16_e64 v255.h, exec_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0x7e,0x00,0x00,0x00]
+
+v_exp_f16 v255.h, m0
+// GFX12: v_exp_f16_e64 v255.h, m0 op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0x7d,0x00,0x00,0x00]
+
+v_exp_f16 v255.h, null
+// GFX12: v_exp_f16_e64 v255.h, null op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0x7c,0x00,0x00,0x00]
+
+v_exp_f16 v255.h, s1
+// GFX12: v_exp_f16_e64 v255.h, s1 op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0x01,0x00,0x00,0x00]
+
+v_exp_f16 v255.h, s105
+// GFX12: v_exp_f16_e64 v255.h, s105 op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0x69,0x00,0x00,0x00]
+
+v_exp_f16 v255.h, src_scc
+// GFX12: v_exp_f16_e64 v255.h, src_scc op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0xfd,0x00,0x00,0x00]
+
+v_exp_f16 v255.h, ttmp15
+// GFX12: v_exp_f16_e64 v255.h, ttmp15 op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0x7b,0x00,0x00,0x00]
+
+v_exp_f16 v255.h, v1.h
+// GFX12: v_exp_f16_e64 v255.h, v1.h op_sel:[1,1] ; encoding: [0xff,0x48,0xd8,0xd5,0x01,0x01,0x00,0x00]
+
+v_exp_f16 v255.h, v1.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_exp_f16_e64_dpp v255.h, v1.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xd8,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+
+v_exp_f16 v255.h, v1.h quad_perm:[3,2,1,0]
+// GFX12: v_exp_f16_e64_dpp v255.h, v1.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+
+v_exp_f16 v255.h, v127.h
+// GFX12: v_exp_f16_e64 v255.h, v127.h op_sel:[1,1] ; encoding: [0xff,0x48,0xd8,0xd5,0x7f,0x01,0x00,0x00]
+
+v_exp_f16 v255.h, v127.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_exp_f16_e64_dpp v255.h, v127.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xd8,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+
+v_exp_f16 v255.h, v127.h quad_perm:[3,2,1,0]
+// GFX12: v_exp_f16_e64_dpp v255.h, v127.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
-v_exp_f16 v255, -1
-// GFX12: v_exp_f16_e64 v255, -1 ; encoding: [0xff,0x00,0xd8,0xd5,0xc1,0x00,0x00,0x00]
+v_exp_f16 v255.h, vcc_hi
+// GFX12: v_exp_f16_e64 v255.h, vcc_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0x6b,0x00,0x00,0x00]
-v_exp_f16 v255, 0.5
-// GFX12: v_exp_f16_e64 v255, 0.5 ; encoding: [0xff,0x00,0xd8,0xd5,0xf0,0x00,0x00,0x00]
+v_exp_f16 v255.h, vcc_lo
+// GFX12: v_exp_f16_e64 v255.h, vcc_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xd8,0xd5,0x6a,0x00,0x00,0x00]
-v_exp_f16 v255, exec_hi
-// GFX12: v_exp_f16_e64 v255, exec_hi ; encoding: [0xff,0x00,0xd8,0xd5,0x7f,0x00,0x00,0x00]
+v_exp_f16 v255.l, -1
+// GFX12: v_exp_f16_e64 v255.l, -1 ; encoding: [0xff,0x00,0xd8,0xd5,0xc1,0x00,0x00,0x00]
-v_exp_f16 v255, exec_lo
-// GFX12: v_exp_f16_e64 v255, exec_lo ; encoding: [0xff,0x00,0xd8,0xd5,0x7e,0x00,0x00,0x00]
+v_exp_f16 v255.l, 0.5
+// GFX12: v_exp_f16_e64 v255.l, 0.5 ; encoding: [0xff,0x00,0xd8,0xd5,0xf0,0x00,0x00,0x00]
-v_exp_f16 v255, m0
-// GFX12: v_exp_f16_e64 v255, m0 ; encoding: [0xff,0x00,0xd8,0xd5,0x7d,0x00,0x00,0x00]
+v_exp_f16 v255.l, exec_hi
+// GFX12: v_exp_f16_e64 v255.l, exec_hi ; encoding: [0xff,0x00,0xd8,0xd5,0x7f,0x00,0x00,0x00]
-v_exp_f16 v255, null
-// GFX12: v_exp_f16_e64 v255, null ; encoding: [0xff,0x00,0xd8,0xd5,0x7c,0x00,0x00,0x00]
+v_exp_f16 v255.l, exec_lo
+// GFX12: v_exp_f16_e64 v255.l, exec_lo ; encoding: [0xff,0x00,0xd8,0xd5,0x7e,0x00,0x00,0x00]
-v_exp_f16 v255, s1
-// GFX12: v_exp_f16_e64 v255, s1 ; encoding: [0xff,0x00,0xd8,0xd5,0x01,0x00,0x00,0x00]
+v_exp_f16 v255.l, m0
+// GFX12: v_exp_f16_e64 v255.l, m0 ; encoding: [0xff,0x00,0xd8,0xd5,0x7d,0x00,0x00,0x00]
-v_exp_f16 v255, s105
-// GFX12: v_exp_f16_e64 v255, s105 ; encoding: [0xff,0x00,0xd8,0xd5,0x69,0x00,0x00,0x00]
+v_exp_f16 v255.l, null
+// GFX12: v_exp_f16_e64 v255.l, null ; encoding: [0xff,0x00,0xd8,0xd5,0x7c,0x00,0x00,0x00]
-v_exp_f16 v255, src_scc
-// GFX12: v_exp_f16_e64 v255, src_scc ; encoding: [0xff,0x00,0xd8,0xd5,0xfd,0x00,0x00,0x00]
+v_exp_f16 v255.l, s1
+// GFX12: v_exp_f16_e64 v255.l, s1 ; encoding: [0xff,0x00,0xd8,0xd5,0x01,0x00,0x00,0x00]
-v_exp_f16 v255, ttmp15
-// GFX12: v_exp_f16_e64 v255, ttmp15 ; encoding: [0xff,0x00,0xd8,0xd5,0x7b,0x00,0x00,0x00]
+v_exp_f16 v255.l, s105
+// GFX12: v_exp_f16_e64 v255.l, s105 ; encoding: [0xff,0x00,0xd8,0xd5,0x69,0x00,0x00,0x00]
-v_exp_f16 v255, v1
-// GFX12: v_exp_f16_e64 v255, v1 ; encoding: [0xff,0x00,0xd8,0xd5,0x01,0x01,0x00,0x00]
+v_exp_f16 v255.l, src_scc
+// GFX12: v_exp_f16_e64 v255.l, src_scc ; encoding: [0xff,0x00,0xd8,0xd5,0xfd,0x00,0x00,0x00]
-v_exp_f16 v255, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_exp_f16_e64_dpp v255, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd8,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+v_exp_f16 v255.l, ttmp15
+// GFX12: v_exp_f16_e64 v255.l, ttmp15 ; encoding: [0xff,0x00,0xd8,0xd5,0x7b,0x00,0x00,0x00]
-v_exp_f16 v255, v1 quad_perm:[3,2,1,0]
-// GFX12: v_exp_f16_e64_dpp v255, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+v_exp_f16 v255.l, v1.l
+// GFX12: v_exp_f16_e64 v255.l, v1.l ; encoding: [0xff,0x00,0xd8,0xd5,0x01,0x01,0x00,0x00]
-v_exp_f16 v255, v127
-// GFX12: v_exp_f16_e64 v255, v127 ; encoding: [0xff,0x00,0xd8,0xd5,0x7f,0x01,0x00,0x00]
+v_exp_f16 v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_exp_f16_e64_dpp v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd8,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
-v_exp_f16 v255, v127 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_exp_f16_e64_dpp v255, v127 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd8,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+v_exp_f16 v255.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: v_exp_f16_e64_dpp v255.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
-v_exp_f16 v255, v127 quad_perm:[3,2,1,0]
-// GFX12: v_exp_f16_e64_dpp v255, v127 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+v_exp_f16 v255.l, v127.l
+// GFX12: v_exp_f16_e64 v255.l, v127.l ; encoding: [0xff,0x00,0xd8,0xd5,0x7f,0x01,0x00,0x00]
-v_exp_f16 v255, vcc_hi
-// GFX12: v_exp_f16_e64 v255, vcc_hi ; encoding: [0xff,0x00,0xd8,0xd5,0x6b,0x00,0x00,0x00]
+v_exp_f16 v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_exp_f16_e64_dpp v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd8,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
-v_exp_f16 v255, vcc_lo
-// GFX12: v_exp_f16_e64 v255, vcc_lo ; encoding: [0xff,0x00,0xd8,0xd5,0x6a,0x00,0x00,0x00]
+v_exp_f16 v255.l, v127.l quad_perm:[3,2,1,0]
+// GFX12: v_exp_f16_e64_dpp v255.l, v127.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
-v_exp_f16 v5, v199
-// GFX12: v_exp_f16_e64 v5, v199 ; encoding: [0x05,0x00,0xd8,0xd5,0xc7,0x01,0x00,0x00]
+v_exp_f16 v255.l, vcc_hi
+// GFX12: v_exp_f16_e64 v255.l, vcc_hi ; encoding: [0xff,0x00,0xd8,0xd5,0x6b,0x00,0x00,0x00]
-v_exp_f16 v5, v199 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_exp_f16_e64_dpp v5, v199 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd8,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+v_exp_f16 v255.l, vcc_lo
+// GFX12: v_exp_f16_e64 v255.l, vcc_lo ; encoding: [0xff,0x00,0xd8,0xd5,0x6a,0x00,0x00,0x00]
-v_exp_f16 v5, v199 quad_perm:[3,2,1,0]
-// GFX12: v_exp_f16_e64_dpp v5, v199 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+v_exp_f16 v5.h, v199.h
+// GFX12: v_exp_f16_e64 v5.h, v199.h op_sel:[1,1] ; encoding: [0x05,0x48,0xd8,0xd5,0xc7,0x01,0x00,0x00]
-v_floor_f16 v128, 0xfe0b
-// GFX12: v_floor_f16_e64 v128, 0xfe0b ; encoding: [0x80,0x00,0xdb,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+v_exp_f16 v5.h, v199.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_exp_f16_e64_dpp v5.h, v199.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xd8,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
-v_floor_f16 v255, -1
-// GFX12: v_floor_f16_e64 v255, -1 ; encoding: [0xff,0x00,0xdb,0xd5,0xc1,0x00,0x00,0x00]
+v_exp_f16 v5.h, v199.h quad_perm:[3,2,1,0]
+// GFX12: v_exp_f16_e64_dpp v5.h, v199.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x48,0xd8,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
-v_floor_f16 v255, 0.5
-// GFX12: v_floor_f16_e64 v255, 0.5 ; encoding: [0xff,0x00,0xdb,0xd5,0xf0,0x00,0x00,0x00]
+v_exp_f16 v5.l, v199.l
+// GFX12: v_exp_f16_e64 v5.l, v199.l ; encoding: [0x05,0x00,0xd8,0xd5,0xc7,0x01,0x00,0x00]
-v_floor_f16 v255, exec_hi
-// GFX12: v_floor_f16_e64 v255, exec_hi ; encoding: [0xff,0x00,0xdb,0xd5,0x7f,0x00,0x00,0x00]
+v_exp_f16 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_exp_f16_e64_dpp v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd8,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
-v_floor_f16 v255, exec_lo
-// GFX12: v_floor_f16_e64 v255, exec_lo ; encoding: [0xff,0x00,0xdb,0xd5,0x7e,0x00,0x00,0x00]
+v_exp_f16 v5.l, v199.l quad_perm:[3,2,1,0]
+// GFX12: v_exp_f16_e64_dpp v5.l, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
-v_floor_f16 v255, m0
-// GFX12: v_floor_f16_e64 v255, m0 ; encoding: [0xff,0x00,0xdb,0xd5,0x7d,0x00,0x00,0x00]
+v_floor_f16 v128.h, 0xfe0b
+// GFX12: v_floor_f16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xdb,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
-v_floor_f16 v255, null
-// GFX12: v_floor_f16_e64 v255, null ; encoding: [0xff,0x00,0xdb,0xd5,0x7c,0x00,0x00,0x00]
+v_floor_f16 v128.l, 0xfe0b
+// GFX12: v_floor_f16_e64 v128.l, 0xfe0b ; encoding: [0x80,0x00,0xdb,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
-v_floor_f16 v255, s1
-// GFX12: v_floor_f16_e64 v255, s1 ; encoding: [0xff,0x00,0xdb,0xd5,0x01,0x00,0x00,0x00]
+v_floor_f16 v255.h, -1
+// GFX12: v_floor_f16_e64 v255.h, -1 op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0xc1,0x00,0x00,0x00]
-v_floor_f16 v255, s105
-// GFX12: v_floor_f16_e64 v255, s105 ; encoding: [0xff,0x00,0xdb,0xd5,0x69,0x00,0x00,0x00]
+v_floor_f16 v255.h, 0.5
+// GFX12: v_floor_f16_e64 v255.h, 0.5 op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0xf0,0x00,0x00,0x00]
-v_floor_f16 v255, src_scc
-// GFX12: v_floor_f16_e64 v255, src_scc ; encoding: [0xff,0x00,0xdb,0xd5,0xfd,0x00,0x00,0x00]
+v_floor_f16 v255.h, exec_hi
+// GFX12: v_floor_f16_e64 v255.h, exec_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0x7f,0x00,0x00,0x00]
-v_floor_f16 v255, ttmp15
-// GFX12: v_floor_f16_e64 v255, ttmp15 ; encoding: [0xff,0x00,0xdb,0xd5,0x7b,0x00,0x00,0x00]
+v_floor_f16 v255.h, exec_lo
+// GFX12: v_floor_f16_e64 v255.h, exec_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0x7e,0x00,0x00,0x00]
-v_floor_f16 v255, v1
-// GFX12: v_floor_f16_e64 v255, v1 ; encoding: [0xff,0x00,0xdb,0xd5,0x01,0x01,0x00,0x00]
+v_floor_f16 v255.h, m0
+// GFX12: v_floor_f16_e64 v255.h, m0 op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0x7d,0x00,0x00,0x00]
-v_floor_f16 v255, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_floor_f16_e64_dpp v255, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xdb,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+v_floor_f16 v255.h, null
+// GFX12: v_floor_f16_e64 v255.h, null op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0x7c,0x00,0x00,0x00]
-v_floor_f16 v255, v1 quad_perm:[3,2,1,0]
-// GFX12: v_floor_f16_e64_dpp v255, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+v_floor_f16 v255.h, s1
+// GFX12: v_floor_f16_e64 v255.h, s1 op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0x01,0x00,0x00,0x00]
-v_floor_f16 v255, v127
-// GFX12: v_floor_f16_e64 v255, v127 ; encoding: [0xff,0x00,0xdb,0xd5,0x7f,0x01,0x00,0x00]
+v_floor_f16 v255.h, s105
+// GFX12: v_floor_f16_e64 v255.h, s105 op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0x69,0x00,0x00,0x00]
-v_floor_f16 v255, v127 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_floor_f16_e64_dpp v255, v127 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xdb,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+v_floor_f16 v255.h, src_scc
+// GFX12: v_floor_f16_e64 v255.h, src_scc op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0xfd,0x00,0x00,0x00]
-v_floor_f16 v255, v127 quad_perm:[3,2,1,0]
-// GFX12: v_floor_f16_e64_dpp v255, v127 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+v_floor_f16 v255.h, ttmp15
+// GFX12: v_floor_f16_e64 v255.h, ttmp15 op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0x7b,0x00,0x00,0x00]
-v_floor_f16 v255, vcc_hi
-// GFX12: v_floor_f16_e64 v255, vcc_hi ; encoding: [0xff,0x00,0xdb,0xd5,0x6b,0x00,0x00,0x00]
+v_floor_f16 v255.h, v1.h
+// GFX12: v_floor_f16_e64 v255.h, v1.h op_sel:[1,1] ; encoding: [0xff,0x48,0xdb,0xd5,0x01,0x01,0x00,0x00]
-v_floor_f16 v255, vcc_lo
-// GFX12: v_floor_f16_e64 v255, vcc_lo ; encoding: [0xff,0x00,0xdb,0xd5,0x6a,0x00,0x00,0x00]
+v_floor_f16 v255.h, v1.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_floor_f16_e64_dpp v255.h, v1.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xdb,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
-v_floor_f16 v5, v199
-// GFX12: v_floor_f16_e64 v5, v199 ; encoding: [0x05,0x00,0xdb,0xd5,0xc7,0x01,0x00,0x00]
+v_floor_f16 v255.h, v1.h quad_perm:[3,2,1,0]
+// GFX12: v_floor_f16_e64_dpp v255.h, v1.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
-v_floor_f16 v5, v199 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_floor_f16_e64_dpp v5, v199 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xdb,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+v_floor_f16 v255.h, v127.h
+// GFX12: v_floor_f16_e64 v255.h, v127.h op_sel:[1,1] ; encoding: [0xff,0x48,0xdb,0xd5,0x7f,0x01,0x00,0x00]
-v_floor_f16 v5, v199 quad_perm:[3,2,1,0]
-// GFX12: v_floor_f16_e64_dpp v5, v199 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+v_floor_f16 v255.h, v127.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_floor_f16_e64_dpp v255.h, v127.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xdb,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+
+v_floor_f16 v255.h, v127.h quad_perm:[3,2,1,0]
+// GFX12: v_floor_f16_e64_dpp v255.h, v127.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+
+v_floor_f16 v255.h, vcc_hi
+// GFX12: v_floor_f16_e64 v255.h, vcc_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0x6b,0x00,0x00,0x00]
+
+v_floor_f16 v255.h, vcc_lo
+// GFX12: v_floor_f16_e64 v255.h, vcc_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xdb,0xd5,0x6a,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, -1
+// GFX12: v_floor_f16_e64 v255.l, -1 ; encoding: [0xff,0x00,0xdb,0xd5,0xc1,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, 0.5
+// GFX12: v_floor_f16_e64 v255.l, 0.5 ; encoding: [0xff,0x00,0xdb,0xd5,0xf0,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, exec_hi
+// GFX12: v_floor_f16_e64 v255.l, exec_hi ; encoding: [0xff,0x00,0xdb,0xd5,0x7f,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, exec_lo
+// GFX12: v_floor_f16_e64 v255.l, exec_lo ; encoding: [0xff,0x00,0xdb,0xd5,0x7e,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, m0
+// GFX12: v_floor_f16_e64 v255.l, m0 ; encoding: [0xff,0x00,0xdb,0xd5,0x7d,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, null
+// GFX12: v_floor_f16_e64 v255.l, null ; encoding: [0xff,0x00,0xdb,0xd5,0x7c,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, s1
+// GFX12: v_floor_f16_e64 v255.l, s1 ; encoding: [0xff,0x00,0xdb,0xd5,0x01,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, s105
+// GFX12: v_floor_f16_e64 v255.l, s105 ; encoding: [0xff,0x00,0xdb,0xd5,0x69,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, src_scc
+// GFX12: v_floor_f16_e64 v255.l, src_scc ; encoding: [0xff,0x00,0xdb,0xd5,0xfd,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, ttmp15
+// GFX12: v_floor_f16_e64 v255.l, ttmp15 ; encoding: [0xff,0x00,0xdb,0xd5,0x7b,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, v1.l
+// GFX12: v_floor_f16_e64 v255.l, v1.l ; encoding: [0xff,0x00,0xdb,0xd5,0x01,0x01,0x00,0x00]
+
+v_floor_f16 v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_floor_f16_e64_dpp v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xdb,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+
+v_floor_f16 v255.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: v_floor_f16_e64_dpp v255.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+
+v_floor_f16 v255.l, v127.l
+// GFX12: v_floor_f16_e64 v255.l, v127.l ; encoding: [0xff,0x00,0xdb,0xd5,0x7f,0x01,0x00,0x00]
+
+v_floor_f16 v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_floor_f16_e64_dpp v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xdb,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+
+v_floor_f16 v255.l, v127.l quad_perm:[3,2,1,0]
+// GFX12: v_floor_f16_e64_dpp v255.l, v127.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+
+v_floor_f16 v255.l, vcc_hi
+// GFX12: v_floor_f16_e64 v255.l, vcc_hi ; encoding: [0xff,0x00,0xdb,0xd5,0x6b,0x00,0x00,0x00]
+
+v_floor_f16 v255.l, vcc_lo
+// GFX12: v_floor_f16_e64 v255.l, vcc_lo ; encoding: [0xff,0x00,0xdb,0xd5,0x6a,0x00,0x00,0x00]
+
+v_floor_f16 v5.h, v199.h
+// GFX12: v_floor_f16_e64 v5.h, v199.h op_sel:[1,1] ; encoding: [0x05,0x48,0xdb,0xd5,0xc7,0x01,0x00,0x00]
+
+v_floor_f16 v5.h, v199.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_floor_f16_e64_dpp v5.h, v199.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xdb,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+
+v_floor_f16 v5.h, v199.h quad_perm:[3,2,1,0]
+// GFX12: v_floor_f16_e64_dpp v5.h, v199.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x48,0xdb,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+
+v_floor_f16 v5.l, v199.l
+// GFX12: v_floor_f16_e64 v5.l, v199.l ; encoding: [0x05,0x00,0xdb,0xd5,0xc7,0x01,0x00,0x00]
+
+v_floor_f16 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_floor_f16_e64_dpp v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xdb,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+
+v_floor_f16 v5.l, v199.l quad_perm:[3,2,1,0]
+// GFX12: v_floor_f16_e64_dpp v5.l, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
v_fract_f16 v128.h, 0xfe0b
// GFX12: v_fract_f16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xdf,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
@@ -1648,71 +1846,137 @@ v_frexp_mant_f16 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
v_frexp_mant_f16 v5.l, v199.l quad_perm:[3,2,1,0]
// GFX12: v_frexp_mant_f16_e64_dpp v5.l, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd9,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
-v_log_f16 v128, 0xfe0b
-// GFX12: v_log_f16_e64 v128, 0xfe0b ; encoding: [0x80,0x00,0xd7,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+v_log_f16 v128.h, 0xfe0b
+// GFX12: v_log_f16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xd7,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+
+v_log_f16 v128.l, 0xfe0b
+// GFX12: v_log_f16_e64 v128.l, 0xfe0b ; encoding: [0x80,0x00,0xd7,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+
+v_log_f16 v255.h, -1
+// GFX12: v_log_f16_e64 v255.h, -1 op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0xc1,0x00,0x00,0x00]
+
+v_log_f16 v255.h, 0.5
+// GFX12: v_log_f16_e64 v255.h, 0.5 op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0xf0,0x00,0x00,0x00]
+
+v_log_f16 v255.h, exec_hi
+// GFX12: v_log_f16_e64 v255.h, exec_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0x7f,0x00,0x00,0x00]
+
+v_log_f16 v255.h, exec_lo
+// GFX12: v_log_f16_e64 v255.h, exec_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0x7e,0x00,0x00,0x00]
+
+v_log_f16 v255.h, m0
+// GFX12: v_log_f16_e64 v255.h, m0 op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0x7d,0x00,0x00,0x00]
+
+v_log_f16 v255.h, null
+// GFX12: v_log_f16_e64 v255.h, null op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0x7c,0x00,0x00,0x00]
+
+v_log_f16 v255.h, s1
+// GFX12: v_log_f16_e64 v255.h, s1 op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0x01,0x00,0x00,0x00]
+
+v_log_f16 v255.h, s105
+// GFX12: v_log_f16_e64 v255.h, s105 op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0x69,0x00,0x00,0x00]
+
+v_log_f16 v255.h, src_scc
+// GFX12: v_log_f16_e64 v255.h, src_scc op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0xfd,0x00,0x00,0x00]
+
+v_log_f16 v255.h, ttmp15
+// GFX12: v_log_f16_e64 v255.h, ttmp15 op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0x7b,0x00,0x00,0x00]
-v_log_f16 v255, -1
-// GFX12: v_log_f16_e64 v255, -1 ; encoding: [0xff,0x00,0xd7,0xd5,0xc1,0x00,0x00,0x00]
+v_log_f16 v255.h, v1.h
+// GFX12: v_log_f16_e64 v255.h, v1.h op_sel:[1,1] ; encoding: [0xff,0x48,0xd7,0xd5,0x01,0x01,0x00,0x00]
-v_log_f16 v255, 0.5
-// GFX12: v_log_f16_e64 v255, 0.5 ; encoding: [0xff,0x00,0xd7,0xd5,0xf0,0x00,0x00,0x00]
+v_log_f16 v255.h, v1.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_log_f16_e64_dpp v255.h, v1.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xd7,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
-v_log_f16 v255, exec_hi
-// GFX12: v_log_f16_e64 v255, exec_hi ; encoding: [0xff,0x00,0xd7,0xd5,0x7f,0x00,0x00,0x00]
+v_log_f16 v255.h, v1.h quad_perm:[3,2,1,0]
+// GFX12: v_log_f16_e64_dpp v255.h, v1.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
-v_log_f16 v255, exec_lo
-// GFX12: v_log_f16_e64 v255, exec_lo ; encoding: [0xff,0x00,0xd7,0xd5,0x7e,0x00,0x00,0x00]
+v_log_f16 v255.h, v127.h
+// GFX12: v_log_f16_e64 v255.h, v127.h op_sel:[1,1] ; encoding: [0xff,0x48,0xd7,0xd5,0x7f,0x01,0x00,0x00]
-v_log_f16 v255, m0
-// GFX12: v_log_f16_e64 v255, m0 ; encoding: [0xff,0x00,0xd7,0xd5,0x7d,0x00,0x00,0x00]
+v_log_f16 v255.h, v127.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_log_f16_e64_dpp v255.h, v127.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xd7,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
-v_log_f16 v255, null
-// GFX12: v_log_f16_e64 v255, null ; encoding: [0xff,0x00,0xd7,0xd5,0x7c,0x00,0x00,0x00]
+v_log_f16 v255.h, v127.h quad_perm:[3,2,1,0]
+// GFX12: v_log_f16_e64_dpp v255.h, v127.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
-v_log_f16 v255, s1
-// GFX12: v_log_f16_e64 v255, s1 ; encoding: [0xff,0x00,0xd7,0xd5,0x01,0x00,0x00,0x00]
+v_log_f16 v255.h, vcc_hi
+// GFX12: v_log_f16_e64 v255.h, vcc_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0x6b,0x00,0x00,0x00]
-v_log_f16 v255, s105
-// GFX12: v_log_f16_e64 v255, s105 ; encoding: [0xff,0x00,0xd7,0xd5,0x69,0x00,0x00,0x00]
+v_log_f16 v255.h, vcc_lo
+// GFX12: v_log_f16_e64 v255.h, vcc_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xd7,0xd5,0x6a,0x00,0x00,0x00]
-v_log_f16 v255, src_scc
-// GFX12: v_log_f16_e64 v255, src_scc ; encoding: [0xff,0x00,0xd7,0xd5,0xfd,0x00,0x00,0x00]
+v_log_f16 v255.l, -1
+// GFX12: v_log_f16_e64 v255.l, -1 ; encoding: [0xff,0x00,0xd7,0xd5,0xc1,0x00,0x00,0x00]
-v_log_f16 v255, ttmp15
-// GFX12: v_log_f16_e64 v255, ttmp15 ; encoding: [0xff,0x00,0xd7,0xd5,0x7b,0x00,0x00,0x00]
+v_log_f16 v255.l, 0.5
+// GFX12: v_log_f16_e64 v255.l, 0.5 ; encoding: [0xff,0x00,0xd7,0xd5,0xf0,0x00,0x00,0x00]
-v_log_f16 v255, v1
-// GFX12: v_log_f16_e64 v255, v1 ; encoding: [0xff,0x00,0xd7,0xd5,0x01,0x01,0x00,0x00]
+v_log_f16 v255.l, exec_hi
+// GFX12: v_log_f16_e64 v255.l, exec_hi ; encoding: [0xff,0x00,0xd7,0xd5,0x7f,0x00,0x00,0x00]
-v_log_f16 v255, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_log_f16_e64_dpp v255, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd7,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+v_log_f16 v255.l, exec_lo
+// GFX12: v_log_f16_e64 v255.l, exec_lo ; encoding: [0xff,0x00,0xd7,0xd5,0x7e,0x00,0x00,0x00]
-v_log_f16 v255, v1 quad_perm:[3,2,1,0]
-// GFX12: v_log_f16_e64_dpp v255, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+v_log_f16 v255.l, m0
+// GFX12: v_log_f16_e64 v255.l, m0 ; encoding: [0xff,0x00,0xd7,0xd5,0x7d,0x00,0x00,0x00]
-v_log_f16 v255, v127
-// GFX12: v_log_f16_e64 v255, v127 ; encoding: [0xff,0x00,0xd7,0xd5,0x7f,0x01,0x00,0x00]
+v_log_f16 v255.l, null
+// GFX12: v_log_f16_e64 v255.l, null ; encoding: [0xff,0x00,0xd7,0xd5,0x7c,0x00,0x00,0x00]
-v_log_f16 v255, v127 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_log_f16_e64_dpp v255, v127 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd7,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+v_log_f16 v255.l, s1
+// GFX12: v_log_f16_e64 v255.l, s1 ; encoding: [0xff,0x00,0xd7,0xd5,0x01,0x00,0x00,0x00]
-v_log_f16 v255, v127 quad_perm:[3,2,1,0]
-// GFX12: v_log_f16_e64_dpp v255, v127 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+v_log_f16 v255.l, s105
+// GFX12: v_log_f16_e64 v255.l, s105 ; encoding: [0xff,0x00,0xd7,0xd5,0x69,0x00,0x00,0x00]
-v_log_f16 v255, vcc_hi
-// GFX12: v_log_f16_e64 v255, vcc_hi ; encoding: [0xff,0x00,0xd7,0xd5,0x6b,0x00,0x00,0x00]
+v_log_f16 v255.l, src_scc
+// GFX12: v_log_f16_e64 v255.l, src_scc ; encoding: [0xff,0x00,0xd7,0xd5,0xfd,0x00,0x00,0x00]
-v_log_f16 v255, vcc_lo
-// GFX12: v_log_f16_e64 v255, vcc_lo ; encoding: [0xff,0x00,0xd7,0xd5,0x6a,0x00,0x00,0x00]
+v_log_f16 v255.l, ttmp15
+// GFX12: v_log_f16_e64 v255.l, ttmp15 ; encoding: [0xff,0x00,0xd7,0xd5,0x7b,0x00,0x00,0x00]
-v_log_f16 v5, v199
-// GFX12: v_log_f16_e64 v5, v199 ; encoding: [0x05,0x00,0xd7,0xd5,0xc7,0x01,0x00,0x00]
+v_log_f16 v255.l, v1.l
+// GFX12: v_log_f16_e64 v255.l, v1.l ; encoding: [0xff,0x00,0xd7,0xd5,0x01,0x01,0x00,0x00]
-v_log_f16 v5, v199 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_log_f16_e64_dpp v5, v199 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd7,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+v_log_f16 v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_log_f16_e64_dpp v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd7,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
-v_log_f16 v5, v199 quad_perm:[3,2,1,0]
-// GFX12: v_log_f16_e64_dpp v5, v199 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+v_log_f16 v255.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: v_log_f16_e64_dpp v255.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+
+v_log_f16 v255.l, v127.l
+// GFX12: v_log_f16_e64 v255.l, v127.l ; encoding: [0xff,0x00,0xd7,0xd5,0x7f,0x01,0x00,0x00]
+
+v_log_f16 v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_log_f16_e64_dpp v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd7,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+
+v_log_f16 v255.l, v127.l quad_perm:[3,2,1,0]
+// GFX12: v_log_f16_e64_dpp v255.l, v127.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+
+v_log_f16 v255.l, vcc_hi
+// GFX12: v_log_f16_e64 v255.l, vcc_hi ; encoding: [0xff,0x00,0xd7,0xd5,0x6b,0x00,0x00,0x00]
+
+v_log_f16 v255.l, vcc_lo
+// GFX12: v_log_f16_e64 v255.l, vcc_lo ; encoding: [0xff,0x00,0xd7,0xd5,0x6a,0x00,0x00,0x00]
+
+v_log_f16 v5.h, v199.h
+// GFX12: v_log_f16_e64 v5.h, v199.h op_sel:[1,1] ; encoding: [0x05,0x48,0xd7,0xd5,0xc7,0x01,0x00,0x00]
+
+v_log_f16 v5.h, v199.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_log_f16_e64_dpp v5.h, v199.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xd7,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+
+v_log_f16 v5.h, v199.h quad_perm:[3,2,1,0]
+// GFX12: v_log_f16_e64_dpp v5.h, v199.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x48,0xd7,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+
+v_log_f16 v5.l, v199.l
+// GFX12: v_log_f16_e64 v5.l, v199.l ; encoding: [0x05,0x00,0xd7,0xd5,0xc7,0x01,0x00,0x00]
+
+v_log_f16 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_log_f16_e64_dpp v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd7,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+
+v_log_f16 v5.l, v199.l quad_perm:[3,2,1,0]
+// GFX12: v_log_f16_e64_dpp v5.l, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
v_not_b16 v128.h, 0xfe0b
// GFX12: v_not_b16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xe9,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
@@ -1846,71 +2110,137 @@ v_not_b16 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
v_not_b16 v5.l, v199.l quad_perm:[3,2,1,0]
// GFX12: v_not_b16_e64_dpp v5.l, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xe9,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
-v_rcp_f16 v128, 0xfe0b
-// GFX12: v_rcp_f16_e64 v128, 0xfe0b ; encoding: [0x80,0x00,0xd4,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+v_rcp_f16 v128.h, 0xfe0b
+// GFX12: v_rcp_f16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xd4,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+
+v_rcp_f16 v128.l, 0xfe0b
+// GFX12: v_rcp_f16_e64 v128.l, 0xfe0b ; encoding: [0x80,0x00,0xd4,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+
+v_rcp_f16 v255.h, -1
+// GFX12: v_rcp_f16_e64 v255.h, -1 op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0xc1,0x00,0x00,0x00]
+
+v_rcp_f16 v255.h, 0.5
+// GFX12: v_rcp_f16_e64 v255.h, 0.5 op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0xf0,0x00,0x00,0x00]
+
+v_rcp_f16 v255.h, exec_hi
+// GFX12: v_rcp_f16_e64 v255.h, exec_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0x7f,0x00,0x00,0x00]
+
+v_rcp_f16 v255.h, exec_lo
+// GFX12: v_rcp_f16_e64 v255.h, exec_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0x7e,0x00,0x00,0x00]
+
+v_rcp_f16 v255.h, m0
+// GFX12: v_rcp_f16_e64 v255.h, m0 op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0x7d,0x00,0x00,0x00]
+
+v_rcp_f16 v255.h, null
+// GFX12: v_rcp_f16_e64 v255.h, null op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0x7c,0x00,0x00,0x00]
+
+v_rcp_f16 v255.h, s1
+// GFX12: v_rcp_f16_e64 v255.h, s1 op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0x01,0x00,0x00,0x00]
+
+v_rcp_f16 v255.h, s105
+// GFX12: v_rcp_f16_e64 v255.h, s105 op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0x69,0x00,0x00,0x00]
-v_rcp_f16 v255, -1
-// GFX12: v_rcp_f16_e64 v255, -1 ; encoding: [0xff,0x00,0xd4,0xd5,0xc1,0x00,0x00,0x00]
+v_rcp_f16 v255.h, src_scc
+// GFX12: v_rcp_f16_e64 v255.h, src_scc op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0xfd,0x00,0x00,0x00]
-v_rcp_f16 v255, 0.5
-// GFX12: v_rcp_f16_e64 v255, 0.5 ; encoding: [0xff,0x00,0xd4,0xd5,0xf0,0x00,0x00,0x00]
+v_rcp_f16 v255.h, ttmp15
+// GFX12: v_rcp_f16_e64 v255.h, ttmp15 op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0x7b,0x00,0x00,0x00]
-v_rcp_f16 v255, exec_hi
-// GFX12: v_rcp_f16_e64 v255, exec_hi ; encoding: [0xff,0x00,0xd4,0xd5,0x7f,0x00,0x00,0x00]
+v_rcp_f16 v255.h, v1.h
+// GFX12: v_rcp_f16_e64 v255.h, v1.h op_sel:[1,1] ; encoding: [0xff,0x48,0xd4,0xd5,0x01,0x01,0x00,0x00]
-v_rcp_f16 v255, exec_lo
-// GFX12: v_rcp_f16_e64 v255, exec_lo ; encoding: [0xff,0x00,0xd4,0xd5,0x7e,0x00,0x00,0x00]
+v_rcp_f16 v255.h, v1.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_rcp_f16_e64_dpp v255.h, v1.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xd4,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
-v_rcp_f16 v255, m0
-// GFX12: v_rcp_f16_e64 v255, m0 ; encoding: [0xff,0x00,0xd4,0xd5,0x7d,0x00,0x00,0x00]
+v_rcp_f16 v255.h, v1.h quad_perm:[3,2,1,0]
+// GFX12: v_rcp_f16_e64_dpp v255.h, v1.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
-v_rcp_f16 v255, null
-// GFX12: v_rcp_f16_e64 v255, null ; encoding: [0xff,0x00,0xd4,0xd5,0x7c,0x00,0x00,0x00]
+v_rcp_f16 v255.h, v127.h
+// GFX12: v_rcp_f16_e64 v255.h, v127.h op_sel:[1,1] ; encoding: [0xff,0x48,0xd4,0xd5,0x7f,0x01,0x00,0x00]
-v_rcp_f16 v255, s1
-// GFX12: v_rcp_f16_e64 v255, s1 ; encoding: [0xff,0x00,0xd4,0xd5,0x01,0x00,0x00,0x00]
+v_rcp_f16 v255.h, v127.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_rcp_f16_e64_dpp v255.h, v127.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xd4,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
-v_rcp_f16 v255, s105
-// GFX12: v_rcp_f16_e64 v255, s105 ; encoding: [0xff,0x00,0xd4,0xd5,0x69,0x00,0x00,0x00]
+v_rcp_f16 v255.h, v127.h quad_perm:[3,2,1,0]
+// GFX12: v_rcp_f16_e64_dpp v255.h, v127.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
-v_rcp_f16 v255, src_scc
-// GFX12: v_rcp_f16_e64 v255, src_scc ; encoding: [0xff,0x00,0xd4,0xd5,0xfd,0x00,0x00,0x00]
+v_rcp_f16 v255.h, vcc_hi
+// GFX12: v_rcp_f16_e64 v255.h, vcc_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0x6b,0x00,0x00,0x00]
-v_rcp_f16 v255, ttmp15
-// GFX12: v_rcp_f16_e64 v255, ttmp15 ; encoding: [0xff,0x00,0xd4,0xd5,0x7b,0x00,0x00,0x00]
+v_rcp_f16 v255.h, vcc_lo
+// GFX12: v_rcp_f16_e64 v255.h, vcc_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xd4,0xd5,0x6a,0x00,0x00,0x00]
-v_rcp_f16 v255, v1
-// GFX12: v_rcp_f16_e64 v255, v1 ; encoding: [0xff,0x00,0xd4,0xd5,0x01,0x01,0x00,0x00]
+v_rcp_f16 v255.l, -1
+// GFX12: v_rcp_f16_e64 v255.l, -1 ; encoding: [0xff,0x00,0xd4,0xd5,0xc1,0x00,0x00,0x00]
-v_rcp_f16 v255, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_rcp_f16_e64_dpp v255, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd4,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+v_rcp_f16 v255.l, 0.5
+// GFX12: v_rcp_f16_e64 v255.l, 0.5 ; encoding: [0xff,0x00,0xd4,0xd5,0xf0,0x00,0x00,0x00]
-v_rcp_f16 v255, v1 quad_perm:[3,2,1,0]
-// GFX12: v_rcp_f16_e64_dpp v255, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+v_rcp_f16 v255.l, exec_hi
+// GFX12: v_rcp_f16_e64 v255.l, exec_hi ; encoding: [0xff,0x00,0xd4,0xd5,0x7f,0x00,0x00,0x00]
-v_rcp_f16 v255, v127
-// GFX12: v_rcp_f16_e64 v255, v127 ; encoding: [0xff,0x00,0xd4,0xd5,0x7f,0x01,0x00,0x00]
+v_rcp_f16 v255.l, exec_lo
+// GFX12: v_rcp_f16_e64 v255.l, exec_lo ; encoding: [0xff,0x00,0xd4,0xd5,0x7e,0x00,0x00,0x00]
-v_rcp_f16 v255, v127 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_rcp_f16_e64_dpp v255, v127 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd4,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+v_rcp_f16 v255.l, m0
+// GFX12: v_rcp_f16_e64 v255.l, m0 ; encoding: [0xff,0x00,0xd4,0xd5,0x7d,0x00,0x00,0x00]
-v_rcp_f16 v255, v127 quad_perm:[3,2,1,0]
-// GFX12: v_rcp_f16_e64_dpp v255, v127 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+v_rcp_f16 v255.l, null
+// GFX12: v_rcp_f16_e64 v255.l, null ; encoding: [0xff,0x00,0xd4,0xd5,0x7c,0x00,0x00,0x00]
-v_rcp_f16 v255, vcc_hi
-// GFX12: v_rcp_f16_e64 v255, vcc_hi ; encoding: [0xff,0x00,0xd4,0xd5,0x6b,0x00,0x00,0x00]
+v_rcp_f16 v255.l, s1
+// GFX12: v_rcp_f16_e64 v255.l, s1 ; encoding: [0xff,0x00,0xd4,0xd5,0x01,0x00,0x00,0x00]
-v_rcp_f16 v255, vcc_lo
-// GFX12: v_rcp_f16_e64 v255, vcc_lo ; encoding: [0xff,0x00,0xd4,0xd5,0x6a,0x00,0x00,0x00]
+v_rcp_f16 v255.l, s105
+// GFX12: v_rcp_f16_e64 v255.l, s105 ; encoding: [0xff,0x00,0xd4,0xd5,0x69,0x00,0x00,0x00]
-v_rcp_f16 v5, v199
-// GFX12: v_rcp_f16_e64 v5, v199 ; encoding: [0x05,0x00,0xd4,0xd5,0xc7,0x01,0x00,0x00]
+v_rcp_f16 v255.l, src_scc
+// GFX12: v_rcp_f16_e64 v255.l, src_scc ; encoding: [0xff,0x00,0xd4,0xd5,0xfd,0x00,0x00,0x00]
-v_rcp_f16 v5, v199 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_rcp_f16_e64_dpp v5, v199 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd4,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+v_rcp_f16 v255.l, ttmp15
+// GFX12: v_rcp_f16_e64 v255.l, ttmp15 ; encoding: [0xff,0x00,0xd4,0xd5,0x7b,0x00,0x00,0x00]
-v_rcp_f16 v5, v199 quad_perm:[3,2,1,0]
-// GFX12: v_rcp_f16_e64_dpp v5, v199 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+v_rcp_f16 v255.l, v1.l
+// GFX12: v_rcp_f16_e64 v255.l, v1.l ; encoding: [0xff,0x00,0xd4,0xd5,0x01,0x01,0x00,0x00]
+
+v_rcp_f16 v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_rcp_f16_e64_dpp v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd4,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+
+v_rcp_f16 v255.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: v_rcp_f16_e64_dpp v255.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+
+v_rcp_f16 v255.l, v127.l
+// GFX12: v_rcp_f16_e64 v255.l, v127.l ; encoding: [0xff,0x00,0xd4,0xd5,0x7f,0x01,0x00,0x00]
+
+v_rcp_f16 v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_rcp_f16_e64_dpp v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd4,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+
+v_rcp_f16 v255.l, v127.l quad_perm:[3,2,1,0]
+// GFX12: v_rcp_f16_e64_dpp v255.l, v127.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+
+v_rcp_f16 v255.l, vcc_hi
+// GFX12: v_rcp_f16_e64 v255.l, vcc_hi ; encoding: [0xff,0x00,0xd4,0xd5,0x6b,0x00,0x00,0x00]
+
+v_rcp_f16 v255.l, vcc_lo
+// GFX12: v_rcp_f16_e64 v255.l, vcc_lo ; encoding: [0xff,0x00,0xd4,0xd5,0x6a,0x00,0x00,0x00]
+
+v_rcp_f16 v5.h, v199.h
+// GFX12: v_rcp_f16_e64 v5.h, v199.h op_sel:[1,1] ; encoding: [0x05,0x48,0xd4,0xd5,0xc7,0x01,0x00,0x00]
+
+v_rcp_f16 v5.h, v199.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_rcp_f16_e64_dpp v5.h, v199.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xd4,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+
+v_rcp_f16 v5.h, v199.h quad_perm:[3,2,1,0]
+// GFX12: v_rcp_f16_e64_dpp v5.h, v199.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x48,0xd4,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+
+v_rcp_f16 v5.l, v199.l
+// GFX12: v_rcp_f16_e64 v5.l, v199.l ; encoding: [0x05,0x00,0xd4,0xd5,0xc7,0x01,0x00,0x00]
+
+v_rcp_f16 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_rcp_f16_e64_dpp v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd4,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+
+v_rcp_f16 v5.l, v199.l quad_perm:[3,2,1,0]
+// GFX12: v_rcp_f16_e64_dpp v5.l, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
v_rndne_f16 v128.h, 0xfe0b
// GFX12: v_rndne_f16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xde,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
@@ -2044,71 +2374,137 @@ v_rndne_f16 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
v_rndne_f16 v5.l, v199.l quad_perm:[3,2,1,0]
// GFX12: v_rndne_f16_e64_dpp v5.l, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xde,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
-v_rsq_f16 v128, 0xfe0b
-// GFX12: v_rsq_f16_e64 v128, 0xfe0b ; encoding: [0x80,0x00,0xd6,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+v_rsq_f16 v128.h, 0xfe0b
+// GFX12: v_rsq_f16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xd6,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+
+v_rsq_f16 v128.l, 0xfe0b
+// GFX12: v_rsq_f16_e64 v128.l, 0xfe0b ; encoding: [0x80,0x00,0xd6,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+
+v_rsq_f16 v255.h, -1
+// GFX12: v_rsq_f16_e64 v255.h, -1 op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0xc1,0x00,0x00,0x00]
+
+v_rsq_f16 v255.h, 0.5
+// GFX12: v_rsq_f16_e64 v255.h, 0.5 op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0xf0,0x00,0x00,0x00]
+
+v_rsq_f16 v255.h, exec_hi
+// GFX12: v_rsq_f16_e64 v255.h, exec_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0x7f,0x00,0x00,0x00]
+
+v_rsq_f16 v255.h, exec_lo
+// GFX12: v_rsq_f16_e64 v255.h, exec_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0x7e,0x00,0x00,0x00]
+
+v_rsq_f16 v255.h, m0
+// GFX12: v_rsq_f16_e64 v255.h, m0 op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0x7d,0x00,0x00,0x00]
-v_rsq_f16 v255, -1
-// GFX12: v_rsq_f16_e64 v255, -1 ; encoding: [0xff,0x00,0xd6,0xd5,0xc1,0x00,0x00,0x00]
+v_rsq_f16 v255.h, null
+// GFX12: v_rsq_f16_e64 v255.h, null op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0x7c,0x00,0x00,0x00]
-v_rsq_f16 v255, 0.5
-// GFX12: v_rsq_f16_e64 v255, 0.5 ; encoding: [0xff,0x00,0xd6,0xd5,0xf0,0x00,0x00,0x00]
+v_rsq_f16 v255.h, s1
+// GFX12: v_rsq_f16_e64 v255.h, s1 op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0x01,0x00,0x00,0x00]
-v_rsq_f16 v255, exec_hi
-// GFX12: v_rsq_f16_e64 v255, exec_hi ; encoding: [0xff,0x00,0xd6,0xd5,0x7f,0x00,0x00,0x00]
+v_rsq_f16 v255.h, s105
+// GFX12: v_rsq_f16_e64 v255.h, s105 op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0x69,0x00,0x00,0x00]
-v_rsq_f16 v255, exec_lo
-// GFX12: v_rsq_f16_e64 v255, exec_lo ; encoding: [0xff,0x00,0xd6,0xd5,0x7e,0x00,0x00,0x00]
+v_rsq_f16 v255.h, src_scc
+// GFX12: v_rsq_f16_e64 v255.h, src_scc op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0xfd,0x00,0x00,0x00]
-v_rsq_f16 v255, m0
-// GFX12: v_rsq_f16_e64 v255, m0 ; encoding: [0xff,0x00,0xd6,0xd5,0x7d,0x00,0x00,0x00]
+v_rsq_f16 v255.h, ttmp15
+// GFX12: v_rsq_f16_e64 v255.h, ttmp15 op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0x7b,0x00,0x00,0x00]
-v_rsq_f16 v255, null
-// GFX12: v_rsq_f16_e64 v255, null ; encoding: [0xff,0x00,0xd6,0xd5,0x7c,0x00,0x00,0x00]
+v_rsq_f16 v255.h, v1.h
+// GFX12: v_rsq_f16_e64 v255.h, v1.h op_sel:[1,1] ; encoding: [0xff,0x48,0xd6,0xd5,0x01,0x01,0x00,0x00]
-v_rsq_f16 v255, s1
-// GFX12: v_rsq_f16_e64 v255, s1 ; encoding: [0xff,0x00,0xd6,0xd5,0x01,0x00,0x00,0x00]
+v_rsq_f16 v255.h, v1.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_rsq_f16_e64_dpp v255.h, v1.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xd6,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
-v_rsq_f16 v255, s105
-// GFX12: v_rsq_f16_e64 v255, s105 ; encoding: [0xff,0x00,0xd6,0xd5,0x69,0x00,0x00,0x00]
+v_rsq_f16 v255.h, v1.h quad_perm:[3,2,1,0]
+// GFX12: v_rsq_f16_e64_dpp v255.h, v1.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
-v_rsq_f16 v255, src_scc
-// GFX12: v_rsq_f16_e64 v255, src_scc ; encoding: [0xff,0x00,0xd6,0xd5,0xfd,0x00,0x00,0x00]
+v_rsq_f16 v255.h, v127.h
+// GFX12: v_rsq_f16_e64 v255.h, v127.h op_sel:[1,1] ; encoding: [0xff,0x48,0xd6,0xd5,0x7f,0x01,0x00,0x00]
-v_rsq_f16 v255, ttmp15
-// GFX12: v_rsq_f16_e64 v255, ttmp15 ; encoding: [0xff,0x00,0xd6,0xd5,0x7b,0x00,0x00,0x00]
+v_rsq_f16 v255.h, v127.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_rsq_f16_e64_dpp v255.h, v127.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xd6,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
-v_rsq_f16 v255, v1
-// GFX12: v_rsq_f16_e64 v255, v1 ; encoding: [0xff,0x00,0xd6,0xd5,0x01,0x01,0x00,0x00]
+v_rsq_f16 v255.h, v127.h quad_perm:[3,2,1,0]
+// GFX12: v_rsq_f16_e64_dpp v255.h, v127.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
-v_rsq_f16 v255, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_rsq_f16_e64_dpp v255, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd6,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+v_rsq_f16 v255.h, vcc_hi
+// GFX12: v_rsq_f16_e64 v255.h, vcc_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0x6b,0x00,0x00,0x00]
-v_rsq_f16 v255, v1 quad_perm:[3,2,1,0]
-// GFX12: v_rsq_f16_e64_dpp v255, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+v_rsq_f16 v255.h, vcc_lo
+// GFX12: v_rsq_f16_e64 v255.h, vcc_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xd6,0xd5,0x6a,0x00,0x00,0x00]
-v_rsq_f16 v255, v127
-// GFX12: v_rsq_f16_e64 v255, v127 ; encoding: [0xff,0x00,0xd6,0xd5,0x7f,0x01,0x00,0x00]
+v_rsq_f16 v255.l, -1
+// GFX12: v_rsq_f16_e64 v255.l, -1 ; encoding: [0xff,0x00,0xd6,0xd5,0xc1,0x00,0x00,0x00]
-v_rsq_f16 v255, v127 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_rsq_f16_e64_dpp v255, v127 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd6,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+v_rsq_f16 v255.l, 0.5
+// GFX12: v_rsq_f16_e64 v255.l, 0.5 ; encoding: [0xff,0x00,0xd6,0xd5,0xf0,0x00,0x00,0x00]
-v_rsq_f16 v255, v127 quad_perm:[3,2,1,0]
-// GFX12: v_rsq_f16_e64_dpp v255, v127 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+v_rsq_f16 v255.l, exec_hi
+// GFX12: v_rsq_f16_e64 v255.l, exec_hi ; encoding: [0xff,0x00,0xd6,0xd5,0x7f,0x00,0x00,0x00]
-v_rsq_f16 v255, vcc_hi
-// GFX12: v_rsq_f16_e64 v255, vcc_hi ; encoding: [0xff,0x00,0xd6,0xd5,0x6b,0x00,0x00,0x00]
+v_rsq_f16 v255.l, exec_lo
+// GFX12: v_rsq_f16_e64 v255.l, exec_lo ; encoding: [0xff,0x00,0xd6,0xd5,0x7e,0x00,0x00,0x00]
-v_rsq_f16 v255, vcc_lo
-// GFX12: v_rsq_f16_e64 v255, vcc_lo ; encoding: [0xff,0x00,0xd6,0xd5,0x6a,0x00,0x00,0x00]
+v_rsq_f16 v255.l, m0
+// GFX12: v_rsq_f16_e64 v255.l, m0 ; encoding: [0xff,0x00,0xd6,0xd5,0x7d,0x00,0x00,0x00]
-v_rsq_f16 v5, v199
-// GFX12: v_rsq_f16_e64 v5, v199 ; encoding: [0x05,0x00,0xd6,0xd5,0xc7,0x01,0x00,0x00]
+v_rsq_f16 v255.l, null
+// GFX12: v_rsq_f16_e64 v255.l, null ; encoding: [0xff,0x00,0xd6,0xd5,0x7c,0x00,0x00,0x00]
-v_rsq_f16 v5, v199 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_rsq_f16_e64_dpp v5, v199 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd6,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+v_rsq_f16 v255.l, s1
+// GFX12: v_rsq_f16_e64 v255.l, s1 ; encoding: [0xff,0x00,0xd6,0xd5,0x01,0x00,0x00,0x00]
-v_rsq_f16 v5, v199 quad_perm:[3,2,1,0]
-// GFX12: v_rsq_f16_e64_dpp v5, v199 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+v_rsq_f16 v255.l, s105
+// GFX12: v_rsq_f16_e64 v255.l, s105 ; encoding: [0xff,0x00,0xd6,0xd5,0x69,0x00,0x00,0x00]
+
+v_rsq_f16 v255.l, src_scc
+// GFX12: v_rsq_f16_e64 v255.l, src_scc ; encoding: [0xff,0x00,0xd6,0xd5,0xfd,0x00,0x00,0x00]
+
+v_rsq_f16 v255.l, ttmp15
+// GFX12: v_rsq_f16_e64 v255.l, ttmp15 ; encoding: [0xff,0x00,0xd6,0xd5,0x7b,0x00,0x00,0x00]
+
+v_rsq_f16 v255.l, v1.l
+// GFX12: v_rsq_f16_e64 v255.l, v1.l ; encoding: [0xff,0x00,0xd6,0xd5,0x01,0x01,0x00,0x00]
+
+v_rsq_f16 v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_rsq_f16_e64_dpp v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd6,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+
+v_rsq_f16 v255.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: v_rsq_f16_e64_dpp v255.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+
+v_rsq_f16 v255.l, v127.l
+// GFX12: v_rsq_f16_e64 v255.l, v127.l ; encoding: [0xff,0x00,0xd6,0xd5,0x7f,0x01,0x00,0x00]
+
+v_rsq_f16 v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_rsq_f16_e64_dpp v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd6,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+
+v_rsq_f16 v255.l, v127.l quad_perm:[3,2,1,0]
+// GFX12: v_rsq_f16_e64_dpp v255.l, v127.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+
+v_rsq_f16 v255.l, vcc_hi
+// GFX12: v_rsq_f16_e64 v255.l, vcc_hi ; encoding: [0xff,0x00,0xd6,0xd5,0x6b,0x00,0x00,0x00]
+
+v_rsq_f16 v255.l, vcc_lo
+// GFX12: v_rsq_f16_e64 v255.l, vcc_lo ; encoding: [0xff,0x00,0xd6,0xd5,0x6a,0x00,0x00,0x00]
+
+v_rsq_f16 v5.h, v199.h
+// GFX12: v_rsq_f16_e64 v5.h, v199.h op_sel:[1,1] ; encoding: [0x05,0x48,0xd6,0xd5,0xc7,0x01,0x00,0x00]
+
+v_rsq_f16 v5.h, v199.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_rsq_f16_e64_dpp v5.h, v199.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xd6,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+
+v_rsq_f16 v5.h, v199.h quad_perm:[3,2,1,0]
+// GFX12: v_rsq_f16_e64_dpp v5.h, v199.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x48,0xd6,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+
+v_rsq_f16 v5.l, v199.l
+// GFX12: v_rsq_f16_e64 v5.l, v199.l ; encoding: [0x05,0x00,0xd6,0xd5,0xc7,0x01,0x00,0x00]
+
+v_rsq_f16 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_rsq_f16_e64_dpp v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd6,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+
+v_rsq_f16 v5.l, v199.l quad_perm:[3,2,1,0]
+// GFX12: v_rsq_f16_e64_dpp v5.l, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
v_sat_pk_u8_i16 v199, v5
// GFX12: v_sat_pk_u8_i16_e64 v199, v5 ; encoding: [0xc7,0x00,0xe2,0xd5,0x05,0x01,0x00,0x00]
@@ -2260,71 +2656,137 @@ v_sin_f16 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
v_sin_f16 v5.l, v199.l quad_perm:[3,2,1,0]
// GFX12: v_sin_f16_e64_dpp v5.l, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xe0,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
-v_sqrt_f16 v128, 0xfe0b
-// GFX12: v_sqrt_f16_e64 v128, 0xfe0b ; encoding: [0x80,0x00,0xd5,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+v_sqrt_f16 v128.h, 0xfe0b
+// GFX12: v_sqrt_f16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xd5,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+
+v_sqrt_f16 v128.l, 0xfe0b
+// GFX12: v_sqrt_f16_e64 v128.l, 0xfe0b ; encoding: [0x80,0x00,0xd5,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
+
+v_sqrt_f16 v255.h, -1
+// GFX12: v_sqrt_f16_e64 v255.h, -1 op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0xc1,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.h, 0.5
+// GFX12: v_sqrt_f16_e64 v255.h, 0.5 op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0xf0,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.h, exec_hi
+// GFX12: v_sqrt_f16_e64 v255.h, exec_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0x7f,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.h, exec_lo
+// GFX12: v_sqrt_f16_e64 v255.h, exec_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0x7e,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.h, m0
+// GFX12: v_sqrt_f16_e64 v255.h, m0 op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0x7d,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.h, null
+// GFX12: v_sqrt_f16_e64 v255.h, null op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0x7c,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.h, s1
+// GFX12: v_sqrt_f16_e64 v255.h, s1 op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0x01,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.h, s105
+// GFX12: v_sqrt_f16_e64 v255.h, s105 op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0x69,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.h, src_scc
+// GFX12: v_sqrt_f16_e64 v255.h, src_scc op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0xfd,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.h, ttmp15
+// GFX12: v_sqrt_f16_e64 v255.h, ttmp15 op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0x7b,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.h, v1.h
+// GFX12: v_sqrt_f16_e64 v255.h, v1.h op_sel:[1,1] ; encoding: [0xff,0x48,0xd5,0xd5,0x01,0x01,0x00,0x00]
+
+v_sqrt_f16 v255.h, v1.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_sqrt_f16_e64_dpp v255.h, v1.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xd5,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+
+v_sqrt_f16 v255.h, v1.h quad_perm:[3,2,1,0]
+// GFX12: v_sqrt_f16_e64_dpp v255.h, v1.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+
+v_sqrt_f16 v255.h, v127.h
+// GFX12: v_sqrt_f16_e64 v255.h, v127.h op_sel:[1,1] ; encoding: [0xff,0x48,0xd5,0xd5,0x7f,0x01,0x00,0x00]
+
+v_sqrt_f16 v255.h, v127.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_sqrt_f16_e64_dpp v255.h, v127.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x48,0xd5,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+
+v_sqrt_f16 v255.h, v127.h quad_perm:[3,2,1,0]
+// GFX12: v_sqrt_f16_e64_dpp v255.h, v127.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x48,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+
+v_sqrt_f16 v255.h, vcc_hi
+// GFX12: v_sqrt_f16_e64 v255.h, vcc_hi op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0x6b,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.h, vcc_lo
+// GFX12: v_sqrt_f16_e64 v255.h, vcc_lo op_sel:[0,1] ; encoding: [0xff,0x40,0xd5,0xd5,0x6a,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.l, -1
+// GFX12: v_sqrt_f16_e64 v255.l, -1 ; encoding: [0xff,0x00,0xd5,0xd5,0xc1,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.l, 0.5
+// GFX12: v_sqrt_f16_e64 v255.l, 0.5 ; encoding: [0xff,0x00,0xd5,0xd5,0xf0,0x00,0x00,0x00]
+
+v_sqrt_f16 v255.l, exec_hi
+// GFX12: v_sqrt_f16_e64 v255.l, exec_hi ; encoding: [0xff,0x00,0xd5,0xd5,0x7f,0x00,0x00,0x00]
-v_sqrt_f16 v255, -1
-// GFX12: v_sqrt_f16_e64 v255, -1 ; encoding: [0xff,0x00,0xd5,0xd5,0xc1,0x00,0x00,0x00]
+v_sqrt_f16 v255.l, exec_lo
+// GFX12: v_sqrt_f16_e64 v255.l, exec_lo ; encoding: [0xff,0x00,0xd5,0xd5,0x7e,0x00,0x00,0x00]
-v_sqrt_f16 v255, 0.5
-// GFX12: v_sqrt_f16_e64 v255, 0.5 ; encoding: [0xff,0x00,0xd5,0xd5,0xf0,0x00,0x00,0x00]
+v_sqrt_f16 v255.l, m0
+// GFX12: v_sqrt_f16_e64 v255.l, m0 ; encoding: [0xff,0x00,0xd5,0xd5,0x7d,0x00,0x00,0x00]
-v_sqrt_f16 v255, exec_hi
-// GFX12: v_sqrt_f16_e64 v255, exec_hi ; encoding: [0xff,0x00,0xd5,0xd5,0x7f,0x00,0x00,0x00]
+v_sqrt_f16 v255.l, null
+// GFX12: v_sqrt_f16_e64 v255.l, null ; encoding: [0xff,0x00,0xd5,0xd5,0x7c,0x00,0x00,0x00]
-v_sqrt_f16 v255, exec_lo
-// GFX12: v_sqrt_f16_e64 v255, exec_lo ; encoding: [0xff,0x00,0xd5,0xd5,0x7e,0x00,0x00,0x00]
+v_sqrt_f16 v255.l, s1
+// GFX12: v_sqrt_f16_e64 v255.l, s1 ; encoding: [0xff,0x00,0xd5,0xd5,0x01,0x00,0x00,0x00]
-v_sqrt_f16 v255, m0
-// GFX12: v_sqrt_f16_e64 v255, m0 ; encoding: [0xff,0x00,0xd5,0xd5,0x7d,0x00,0x00,0x00]
+v_sqrt_f16 v255.l, s105
+// GFX12: v_sqrt_f16_e64 v255.l, s105 ; encoding: [0xff,0x00,0xd5,0xd5,0x69,0x00,0x00,0x00]
-v_sqrt_f16 v255, null
-// GFX12: v_sqrt_f16_e64 v255, null ; encoding: [0xff,0x00,0xd5,0xd5,0x7c,0x00,0x00,0x00]
+v_sqrt_f16 v255.l, src_scc
+// GFX12: v_sqrt_f16_e64 v255.l, src_scc ; encoding: [0xff,0x00,0xd5,0xd5,0xfd,0x00,0x00,0x00]
-v_sqrt_f16 v255, s1
-// GFX12: v_sqrt_f16_e64 v255, s1 ; encoding: [0xff,0x00,0xd5,0xd5,0x01,0x00,0x00,0x00]
+v_sqrt_f16 v255.l, ttmp15
+// GFX12: v_sqrt_f16_e64 v255.l, ttmp15 ; encoding: [0xff,0x00,0xd5,0xd5,0x7b,0x00,0x00,0x00]
-v_sqrt_f16 v255, s105
-// GFX12: v_sqrt_f16_e64 v255, s105 ; encoding: [0xff,0x00,0xd5,0xd5,0x69,0x00,0x00,0x00]
+v_sqrt_f16 v255.l, v1.l
+// GFX12: v_sqrt_f16_e64 v255.l, v1.l ; encoding: [0xff,0x00,0xd5,0xd5,0x01,0x01,0x00,0x00]
-v_sqrt_f16 v255, src_scc
-// GFX12: v_sqrt_f16_e64 v255, src_scc ; encoding: [0xff,0x00,0xd5,0xd5,0xfd,0x00,0x00,0x00]
+v_sqrt_f16 v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_sqrt_f16_e64_dpp v255.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd5,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
-v_sqrt_f16 v255, ttmp15
-// GFX12: v_sqrt_f16_e64 v255, ttmp15 ; encoding: [0xff,0x00,0xd5,0xd5,0x7b,0x00,0x00,0x00]
+v_sqrt_f16 v255.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: v_sqrt_f16_e64_dpp v255.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
-v_sqrt_f16 v255, v1
-// GFX12: v_sqrt_f16_e64 v255, v1 ; encoding: [0xff,0x00,0xd5,0xd5,0x01,0x01,0x00,0x00]
+v_sqrt_f16 v255.l, v127.l
+// GFX12: v_sqrt_f16_e64 v255.l, v127.l ; encoding: [0xff,0x00,0xd5,0xd5,0x7f,0x01,0x00,0x00]
-v_sqrt_f16 v255, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_sqrt_f16_e64_dpp v255, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd5,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+v_sqrt_f16 v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_sqrt_f16_e64_dpp v255.l, v127.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd5,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
-v_sqrt_f16 v255, v1 quad_perm:[3,2,1,0]
-// GFX12: v_sqrt_f16_e64_dpp v255, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+v_sqrt_f16 v255.l, v127.l quad_perm:[3,2,1,0]
+// GFX12: v_sqrt_f16_e64_dpp v255.l, v127.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
-v_sqrt_f16 v255, v127
-// GFX12: v_sqrt_f16_e64 v255, v127 ; encoding: [0xff,0x00,0xd5,0xd5,0x7f,0x01,0x00,0x00]
+v_sqrt_f16 v255.l, vcc_hi
+// GFX12: v_sqrt_f16_e64 v255.l, vcc_hi ; encoding: [0xff,0x00,0xd5,0xd5,0x6b,0x00,0x00,0x00]
-v_sqrt_f16 v255, v127 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_sqrt_f16_e64_dpp v255, v127 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xff,0x00,0xd5,0xd5,0xe9,0x00,0x00,0x00,0x7f,0x77,0x39,0x05]
+v_sqrt_f16 v255.l, vcc_lo
+// GFX12: v_sqrt_f16_e64 v255.l, vcc_lo ; encoding: [0xff,0x00,0xd5,0xd5,0x6a,0x00,0x00,0x00]
-v_sqrt_f16 v255, v127 quad_perm:[3,2,1,0]
-// GFX12: v_sqrt_f16_e64_dpp v255, v127 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x7f,0x1b,0x00,0xff]
+v_sqrt_f16 v5.h, v199.h
+// GFX12: v_sqrt_f16_e64 v5.h, v199.h op_sel:[1,1] ; encoding: [0x05,0x48,0xd5,0xd5,0xc7,0x01,0x00,0x00]
-v_sqrt_f16 v255, vcc_hi
-// GFX12: v_sqrt_f16_e64 v255, vcc_hi ; encoding: [0xff,0x00,0xd5,0xd5,0x6b,0x00,0x00,0x00]
+v_sqrt_f16 v5.h, v199.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_sqrt_f16_e64_dpp v5.h, v199.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xd5,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
-v_sqrt_f16 v255, vcc_lo
-// GFX12: v_sqrt_f16_e64 v255, vcc_lo ; encoding: [0xff,0x00,0xd5,0xd5,0x6a,0x00,0x00,0x00]
+v_sqrt_f16 v5.h, v199.h quad_perm:[3,2,1,0]
+// GFX12: v_sqrt_f16_e64_dpp v5.h, v199.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x48,0xd5,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
-v_sqrt_f16 v5, v199
-// GFX12: v_sqrt_f16_e64 v5, v199 ; encoding: [0x05,0x00,0xd5,0xd5,0xc7,0x01,0x00,0x00]
+v_sqrt_f16 v5.l, v199.l
+// GFX12: v_sqrt_f16_e64 v5.l, v199.l ; encoding: [0x05,0x00,0xd5,0xd5,0xc7,0x01,0x00,0x00]
-v_sqrt_f16 v5, v199 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_sqrt_f16_e64_dpp v5, v199 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd5,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
+v_sqrt_f16 v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_sqrt_f16_e64_dpp v5.l, v199.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd5,0xd5,0xe9,0x00,0x00,0x00,0xc7,0x77,0x39,0x05]
-v_sqrt_f16 v5, v199 quad_perm:[3,2,1,0]
-// GFX12: v_sqrt_f16_e64_dpp v5, v199 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
+v_sqrt_f16 v5.l, v199.l quad_perm:[3,2,1,0]
+// GFX12: v_sqrt_f16_e64_dpp v5.l, v199.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0xc7,0x1b,0x00,0xff]
v_trunc_f16 v128.h, 0xfe0b
// GFX12: v_trunc_f16_e64 v128.h, 0xfe0b op_sel:[0,1] ; encoding: [0x80,0x40,0xdd,0xd5,0xff,0x00,0x00,0x00,0x0b,0xfe,0x00,0x00]
diff --git a/llvm/test/MC/AMDGPU/gfx12_asm_vop3_from_vop1.s b/llvm/test/MC/AMDGPU/gfx12_asm_vop3_from_vop1.s
index d49a7085ba497..fc9cd06de0245 100644
--- a/llvm/test/MC/AMDGPU/gfx12_asm_vop3_from_vop1.s
+++ b/llvm/test/MC/AMDGPU/gfx12_asm_vop3_from_vop1.s
@@ -46,50 +46,59 @@ v_bfrev_b32_e64 v5, src_scc
v_bfrev_b32_e64 v255, 0xaf123456
// GFX12: v_bfrev_b32_e64 v255, 0xaf123456 ; encoding: [0xff,0x00,0xb8,0xd5,0xff,0x00,0x00,0x00,0x56,0x34,0x12,0xaf]
-v_ceil_f16_e64 v5, v1
-// GFX12: v_ceil_f16_e64 v5, v1 ; encoding: [0x05,0x00,0xdc,0xd5,0x01,0x01,0x00,0x00]
+v_ceil_f16_e64 v5.l, v1.l
+// GFX12: v_ceil_f16_e64 v5.l, v1.l ; encoding: [0x05,0x00,0xdc,0xd5,0x01,0x01,0x00,0x00]
-v_ceil_f16_e64 v5, v255
-// GFX12: v_ceil_f16_e64 v5, v255 ; encoding: [0x05,0x00,0xdc,0xd5,0xff,0x01,0x00,0x00]
+v_ceil_f16_e64 v5.l, v255.l
+// GFX12: v_ceil_f16_e64 v5.l, v255.l ; encoding: [0x05,0x00,0xdc,0xd5,0xff,0x01,0x00,0x00]
-v_ceil_f16_e64 v5, s1
-// GFX12: v_ceil_f16_e64 v5, s1 ; encoding: [0x05,0x00,0xdc,0xd5,0x01,0x00,0x00,0x00]
+v_ceil_f16_e64 v5.l, s1
+// GFX12: v_ceil_f16_e64 v5.l, s1 ; encoding: [0x05,0x00,0xdc,0xd5,0x01,0x00,0x00,0x00]
-v_ceil_f16_e64 v5, s105
-// GFX12: v_ceil_f16_e64 v5, s105 ; encoding: [0x05,0x00,0xdc,0xd5,0x69,0x00,0x00,0x00]
+v_ceil_f16_e64 v5.l, s105
+// GFX12: v_ceil_f16_e64 v5.l, s105 ; encoding: [0x05,0x00,0xdc,0xd5,0x69,0x00,0x00,0x00]
-v_ceil_f16_e64 v5, vcc_lo
-// GFX12: v_ceil_f16_e64 v5, vcc_lo ; encoding: [0x05,0x00,0xdc,0xd5,0x6a,0x00,0x00,0x00]
+v_ceil_f16_e64 v5.l, vcc_lo
+// GFX12: v_ceil_f16_e64 v5.l, vcc_lo ; encoding: [0x05,0x00,0xdc,0xd5,0x6a,0x00,0x00,0x00]
-v_ceil_f16_e64 v5, vcc_hi
-// GFX12: v_ceil_f16_e64 v5, vcc_hi ; encoding: [0x05,0x00,0xdc,0xd5,0x6b,0x00,0x00,0x00]
+v_ceil_f16_e64 v5.l, vcc_hi
+// GFX12: v_ceil_f16_e64 v5.l, vcc_hi ; encoding: [0x05,0x00,0xdc,0xd5,0x6b,0x00,0x00,0x00]
-v_ceil_f16_e64 v5, ttmp15
-// GFX12: v_ceil_f16_e64 v5, ttmp15 ; encoding: [0x05,0x00,0xdc,0xd5,0x7b,0x00,0x00,0x00]
+v_ceil_f16_e64 v5.l, ttmp15
+// GFX12: v_ceil_f16_e64 v5.l, ttmp15 ; encoding: [0x05,0x00,0xdc,0xd5,0x7b,0x00,0x00,0x00]
-v_ceil_f16_e64 v5, m0
-// GFX12: v_ceil_f16_e64 v5, m0 ; encoding: [0x05,0x00,0xdc,0xd5,0x7d,0x00,0x00,0x00]
+v_ceil_f16_e64 v5.l, m0
+// GFX12: v_ceil_f16_e64 v5.l, m0 ; encoding: [0x05,0x00,0xdc,0xd5,0x7d,0x00,0x00,0x00]
-v_ceil_f16_e64 v5, exec_lo
-// GFX12: v_ceil_f16_e64 v5, exec_lo ; encoding: [0x05,0x00,0xdc,0xd5,0x7e,0x00,0x00,0x00]
+v_ceil_f16_e64 v5.l, exec_lo
+// GFX12: v_ceil_f16_e64 v5.l, exec_lo ; encoding: [0x05,0x00,0xdc,0xd5,0x7e,0x00,0x00,0x00]
-v_ceil_f16_e64 v5, exec_hi
-// GFX12: v_ceil_f16_e64 v5, exec_hi ; encoding: [0x05,0x00,0xdc,0xd5,0x7f,0x00,0x00,0x00]
+v_ceil_f16_e64 v5.l, exec_hi
+// GFX12: v_ceil_f16_e64 v5.l, exec_hi ; encoding: [0x05,0x00,0xdc,0xd5,0x7f,0x00,0x00,0x00]
-v_ceil_f16_e64 v5, null
-// GFX12: v_ceil_f16_e64 v5, null ; encoding: [0x05,0x00,0xdc,0xd5,0x7c,0x00,0x00,0x00]
+v_ceil_f16_e64 v5.l, null
+// GFX12: v_ceil_f16_e64 v5.l, null ; encoding: [0x05,0x00,0xdc,0xd5,0x7c,0x00,0x00,0x00]
-v_ceil_f16_e64 v5, -1
-// GFX12: v_ceil_f16_e64 v5, -1 ; encoding: [0x05,0x00,0xdc,0xd5,0xc1,0x00,0x00,0x00]
+v_ceil_f16_e64 v5.l, -1
+// GFX12: v_ceil_f16_e64 v5.l, -1 ; encoding: [0x05,0x00,0xdc,0xd5,0xc1,0x00,0x00,0x00]
-v_ceil_f16_e64 v5, 0.5 mul:2
-// GFX12: v_ceil_f16_e64 v5, 0.5 mul:2 ; encoding: [0x05,0x00,0xdc,0xd5,0xf0,0x00,0x00,0x08]
+v_ceil_f16_e64 v5.l, 0.5 mul:2
+// GFX12: v_ceil_f16_e64 v5.l, 0.5 mul:2 ; encoding: [0x05,0x00,0xdc,0xd5,0xf0,0x00,0x00,0x08]
-v_ceil_f16_e64 v5, src_scc mul:4
-// GFX12: v_ceil_f16_e64 v5, src_scc mul:4 ; encoding: [0x05,0x00,0xdc,0xd5,0xfd,0x00,0x00,0x10]
+v_ceil_f16_e64 v5.l, src_scc mul:4
+// GFX12: v_ceil_f16_e64 v5.l, src_scc mul:4 ; encoding: [0x05,0x00,0xdc,0xd5,0xfd,0x00,0x00,0x10]
-v_ceil_f16_e64 v255, -|0xfe0b| clamp div:2
-// GFX12: v_ceil_f16_e64 v255, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xdc,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
+v_ceil_f16_e64 v255.l, -|0xfe0b| clamp div:2
+// GFX12: v_ceil_f16_e64 v255.l, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xdc,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
+
+v_ceil_f16_e64 v5.h, v1.h
+// GFX12: v_ceil_f16_e64 v5.h, v1.h op_sel:[1,1] ; encoding: [0x05,0x48,0xdc,0xd5,0x01,0x01,0x00,0x00]
+
+v_ceil_f16_e64 v5.l, v255.h
+// GFX12: v_ceil_f16_e64 v5.l, v255.h op_sel:[1,0] ; encoding: [0x05,0x08,0xdc,0xd5,0xff,0x01,0x00,0x00]
+
+v_ceil_f16_e64 v255.h, -|0xfe0b| clamp div:2
+// GFX12: v_ceil_f16_e64 v255.h, -|0xfe0b| op_sel:[0,1] clamp div:2 ; encoding: [0xff,0xc1,0xdc,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
v_ceil_f32_e64 v5, v1
// GFX12: v_ceil_f32_e64 v5, v1 ; encoding: [0x05,0x00,0xa2,0xd5,0x01,0x01,0x00,0x00]
@@ -1903,47 +1912,59 @@ v_cvt_u32_u16_e64 v5, v255.h
v_exp_f16_e64 v5, v1
// GFX12: v_exp_f16_e64 v5, v1 ; encoding: [0x05,0x00,0xd8,0xd5,0x01,0x01,0x00,0x00]
-v_exp_f16_e64 v5, v255
-// GFX12: v_exp_f16_e64 v5, v255 ; encoding: [0x05,0x00,0xd8,0xd5,0xff,0x01,0x00,0x00]
+v_exp_f16_e64 v5.l, v1.l
+// GFX12: v_exp_f16_e64 v5.l, v1.l ; encoding: [0x05,0x00,0xd8,0xd5,0x01,0x01,0x00,0x00]
+
+v_exp_f16_e64 v5.l, v255.l
+// GFX12: v_exp_f16_e64 v5.l, v255.l ; encoding: [0x05,0x00,0xd8,0xd5,0xff,0x01,0x00,0x00]
+
+v_exp_f16_e64 v5.l, s1
+// GFX12: v_exp_f16_e64 v5.l, s1 ; encoding: [0x05,0x00,0xd8,0xd5,0x01,0x00,0x00,0x00]
+
+v_exp_f16_e64 v5.l, s105
+// GFX12: v_exp_f16_e64 v5.l, s105 ; encoding: [0x05,0x00,0xd8,0xd5,0x69,0x00,0x00,0x00]
-v_exp_f16_e64 v5, s1
-// GFX12: v_exp_f16_e64 v5, s1 ; encoding: [0x05,0x00,0xd8,0xd5,0x01,0x00,0x00,0x00]
+v_exp_f16_e64 v5.l, vcc_lo
+// GFX12: v_exp_f16_e64 v5.l, vcc_lo ; encoding: [0x05,0x00,0xd8,0xd5,0x6a,0x00,0x00,0x00]
-v_exp_f16_e64 v5, s105
-// GFX12: v_exp_f16_e64 v5, s105 ; encoding: [0x05,0x00,0xd8,0xd5,0x69,0x00,0x00,0x00]
+v_exp_f16_e64 v5.l, vcc_hi
+// GFX12: v_exp_f16_e64 v5.l, vcc_hi ; encoding: [0x05,0x00,0xd8,0xd5,0x6b,0x00,0x00,0x00]
-v_exp_f16_e64 v5, vcc_lo
-// GFX12: v_exp_f16_e64 v5, vcc_lo ; encoding: [0x05,0x00,0xd8,0xd5,0x6a,0x00,0x00,0x00]
+v_exp_f16_e64 v5.l, ttmp15
+// GFX12: v_exp_f16_e64 v5.l, ttmp15 ; encoding: [0x05,0x00,0xd8,0xd5,0x7b,0x00,0x00,0x00]
-v_exp_f16_e64 v5, vcc_hi
-// GFX12: v_exp_f16_e64 v5, vcc_hi ; encoding: [0x05,0x00,0xd8,0xd5,0x6b,0x00,0x00,0x00]
+v_exp_f16_e64 v5.l, m0
+// GFX12: v_exp_f16_e64 v5.l, m0 ; encoding: [0x05,0x00,0xd8,0xd5,0x7d,0x00,0x00,0x00]
-v_exp_f16_e64 v5, ttmp15
-// GFX12: v_exp_f16_e64 v5, ttmp15 ; encoding: [0x05,0x00,0xd8,0xd5,0x7b,0x00,0x00,0x00]
+v_exp_f16_e64 v5.l, exec_lo
+// GFX12: v_exp_f16_e64 v5.l, exec_lo ; encoding: [0x05,0x00,0xd8,0xd5,0x7e,0x00,0x00,0x00]
-v_exp_f16_e64 v5, m0
-// GFX12: v_exp_f16_e64 v5, m0 ; encoding: [0x05,0x00,0xd8,0xd5,0x7d,0x00,0x00,0x00]
+v_exp_f16_e64 v5.l, exec_hi
+// GFX12: v_exp_f16_e64 v5.l, exec_hi ; encoding: [0x05,0x00,0xd8,0xd5,0x7f,0x00,0x00,0x00]
-v_exp_f16_e64 v5, exec_lo
-// GFX12: v_exp_f16_e64 v5, exec_lo ; encoding: [0x05,0x00,0xd8,0xd5,0x7e,0x00,0x00,0x00]
+v_exp_f16_e64 v5.l, null
+// GFX12: v_exp_f16_e64 v5.l, null ; encoding: [0x05,0x00,0xd8,0xd5,0x7c,0x00,0x00,0x00]
-v_exp_f16_e64 v5, exec_hi
-// GFX12: v_exp_f16_e64 v5, exec_hi ; encoding: [0x05,0x00,0xd8,0xd5,0x7f,0x00,0x00,0x00]
+v_exp_f16_e64 v5.l, -1
+// GFX12: v_exp_f16_e64 v5.l, -1 ; encoding: [0x05,0x00,0xd8,0xd5,0xc1,0x00,0x00,0x00]
-v_exp_f16_e64 v5, null
-// GFX12: v_exp_f16_e64 v5, null ; encoding: [0x05,0x00,0xd8,0xd5,0x7c,0x00,0x00,0x00]
+v_exp_f16_e64 v5.l, 0.5 mul:2
+// GFX12: v_exp_f16_e64 v5.l, 0.5 mul:2 ; encoding: [0x05,0x00,0xd8,0xd5,0xf0,0x00,0x00,0x08]
-v_exp_f16_e64 v5, -1
-// GFX12: v_exp_f16_e64 v5, -1 ; encoding: [0x05,0x00,0xd8,0xd5,0xc1,0x00,0x00,0x00]
+v_exp_f16_e64 v5.l, src_scc mul:4
+// GFX12: v_exp_f16_e64 v5.l, src_scc mul:4 ; encoding: [0x05,0x00,0xd8,0xd5,0xfd,0x00,0x00,0x10]
-v_exp_f16_e64 v5, 0.5 mul:2
-// GFX12: v_exp_f16_e64 v5, 0.5 mul:2 ; encoding: [0x05,0x00,0xd8,0xd5,0xf0,0x00,0x00,0x08]
+v_exp_f16_e64 v255.l, -|0xfe0b| clamp div:2
+// GFX12: v_exp_f16_e64 v255.l, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xd8,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
-v_exp_f16_e64 v5, src_scc mul:4
-// GFX12: v_exp_f16_e64 v5, src_scc mul:4 ; encoding: [0x05,0x00,0xd8,0xd5,0xfd,0x00,0x00,0x10]
+v_exp_f16_e64 v5.h, v1.h
+// GFX12: v_exp_f16_e64 v5.h, v1.h op_sel:[1,1] ; encoding: [0x05,0x48,0xd8,0xd5,0x01,0x01,0x00,0x00]
-v_exp_f16_e64 v255, -|0xfe0b| clamp div:2
-// GFX12: v_exp_f16_e64 v255, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xd8,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
+v_exp_f16_e64 v5.l, v255.h
+// GFX12: v_exp_f16_e64 v5.l, v255.h op_sel:[1,0] ; encoding: [0x05,0x08,0xd8,0xd5,0xff,0x01,0x00,0x00]
+
+v_exp_f16_e64 v255.h, -|0xfe0b| clamp div:2
+// GFX12: v_exp_f16_e64 v255.h, -|0xfe0b| op_sel:[0,1] clamp div:2 ; encoding: [0xff,0xc1,0xd8,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
v_exp_f32_e64 v5, v1
// GFX12: v_exp_f32_e64 v5, v1 ; encoding: [0x05,0x00,0xa5,0xd5,0x01,0x01,0x00,0x00]
@@ -2125,50 +2146,59 @@ v_ffbl_b32_e64 v5, src_scc
v_ffbl_b32_e64 v255, 0xaf123456
// GFX12: v_ctz_i32_b32_e64 v255, 0xaf123456 ; encoding: [0xff,0x00,0xba,0xd5,0xff,0x00,0x00,0x00,0x56,0x34,0x12,0xaf]
-v_floor_f16_e64 v5, v1
-// GFX12: v_floor_f16_e64 v5, v1 ; encoding: [0x05,0x00,0xdb,0xd5,0x01,0x01,0x00,0x00]
+v_floor_f16_e64 v5.l, v1.l
+// GFX12: v_floor_f16_e64 v5.l, v1.l ; encoding: [0x05,0x00,0xdb,0xd5,0x01,0x01,0x00,0x00]
+
+v_floor_f16_e64 v5.l, v255.l
+// GFX12: v_floor_f16_e64 v5.l, v255.l ; encoding: [0x05,0x00,0xdb,0xd5,0xff,0x01,0x00,0x00]
+
+v_floor_f16_e64 v5.l, s1
+// GFX12: v_floor_f16_e64 v5.l, s1 ; encoding: [0x05,0x00,0xdb,0xd5,0x01,0x00,0x00,0x00]
-v_floor_f16_e64 v5, v255
-// GFX12: v_floor_f16_e64 v5, v255 ; encoding: [0x05,0x00,0xdb,0xd5,0xff,0x01,0x00,0x00]
+v_floor_f16_e64 v5.l, s105
+// GFX12: v_floor_f16_e64 v5.l, s105 ; encoding: [0x05,0x00,0xdb,0xd5,0x69,0x00,0x00,0x00]
-v_floor_f16_e64 v5, s1
-// GFX12: v_floor_f16_e64 v5, s1 ; encoding: [0x05,0x00,0xdb,0xd5,0x01,0x00,0x00,0x00]
+v_floor_f16_e64 v5.l, vcc_lo
+// GFX12: v_floor_f16_e64 v5.l, vcc_lo ; encoding: [0x05,0x00,0xdb,0xd5,0x6a,0x00,0x00,0x00]
-v_floor_f16_e64 v5, s105
-// GFX12: v_floor_f16_e64 v5, s105 ; encoding: [0x05,0x00,0xdb,0xd5,0x69,0x00,0x00,0x00]
+v_floor_f16_e64 v5.l, vcc_hi
+// GFX12: v_floor_f16_e64 v5.l, vcc_hi ; encoding: [0x05,0x00,0xdb,0xd5,0x6b,0x00,0x00,0x00]
-v_floor_f16_e64 v5, vcc_lo
-// GFX12: v_floor_f16_e64 v5, vcc_lo ; encoding: [0x05,0x00,0xdb,0xd5,0x6a,0x00,0x00,0x00]
+v_floor_f16_e64 v5.l, ttmp15
+// GFX12: v_floor_f16_e64 v5.l, ttmp15 ; encoding: [0x05,0x00,0xdb,0xd5,0x7b,0x00,0x00,0x00]
-v_floor_f16_e64 v5, vcc_hi
-// GFX12: v_floor_f16_e64 v5, vcc_hi ; encoding: [0x05,0x00,0xdb,0xd5,0x6b,0x00,0x00,0x00]
+v_floor_f16_e64 v5.l, m0
+// GFX12: v_floor_f16_e64 v5.l, m0 ; encoding: [0x05,0x00,0xdb,0xd5,0x7d,0x00,0x00,0x00]
-v_floor_f16_e64 v5, ttmp15
-// GFX12: v_floor_f16_e64 v5, ttmp15 ; encoding: [0x05,0x00,0xdb,0xd5,0x7b,0x00,0x00,0x00]
+v_floor_f16_e64 v5.l, exec_lo
+// GFX12: v_floor_f16_e64 v5.l, exec_lo ; encoding: [0x05,0x00,0xdb,0xd5,0x7e,0x00,0x00,0x00]
-v_floor_f16_e64 v5, m0
-// GFX12: v_floor_f16_e64 v5, m0 ; encoding: [0x05,0x00,0xdb,0xd5,0x7d,0x00,0x00,0x00]
+v_floor_f16_e64 v5.l, exec_hi
+// GFX12: v_floor_f16_e64 v5.l, exec_hi ; encoding: [0x05,0x00,0xdb,0xd5,0x7f,0x00,0x00,0x00]
-v_floor_f16_e64 v5, exec_lo
-// GFX12: v_floor_f16_e64 v5, exec_lo ; encoding: [0x05,0x00,0xdb,0xd5,0x7e,0x00,0x00,0x00]
+v_floor_f16_e64 v5.l, null
+// GFX12: v_floor_f16_e64 v5.l, null ; encoding: [0x05,0x00,0xdb,0xd5,0x7c,0x00,0x00,0x00]
-v_floor_f16_e64 v5, exec_hi
-// GFX12: v_floor_f16_e64 v5, exec_hi ; encoding: [0x05,0x00,0xdb,0xd5,0x7f,0x00,0x00,0x00]
+v_floor_f16_e64 v5.l, -1
+// GFX12: v_floor_f16_e64 v5.l, -1 ; encoding: [0x05,0x00,0xdb,0xd5,0xc1,0x00,0x00,0x00]
-v_floor_f16_e64 v5, null
-// GFX12: v_floor_f16_e64 v5, null ; encoding: [0x05,0x00,0xdb,0xd5,0x7c,0x00,0x00,0x00]
+v_floor_f16_e64 v5.l, 0.5 mul:2
+// GFX12: v_floor_f16_e64 v5.l, 0.5 mul:2 ; encoding: [0x05,0x00,0xdb,0xd5,0xf0,0x00,0x00,0x08]
-v_floor_f16_e64 v5, -1
-// GFX12: v_floor_f16_e64 v5, -1 ; encoding: [0x05,0x00,0xdb,0xd5,0xc1,0x00,0x00,0x00]
+v_floor_f16_e64 v5.l, src_scc mul:4
+// GFX12: v_floor_f16_e64 v5.l, src_scc mul:4 ; encoding: [0x05,0x00,0xdb,0xd5,0xfd,0x00,0x00,0x10]
-v_floor_f16_e64 v5, 0.5 mul:2
-// GFX12: v_floor_f16_e64 v5, 0.5 mul:2 ; encoding: [0x05,0x00,0xdb,0xd5,0xf0,0x00,0x00,0x08]
+v_floor_f16_e64 v255.l, -|0xfe0b| clamp div:2
+// GFX12: v_floor_f16_e64 v255.l, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xdb,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
-v_floor_f16_e64 v5, src_scc mul:4
-// GFX12: v_floor_f16_e64 v5, src_scc mul:4 ; encoding: [0x05,0x00,0xdb,0xd5,0xfd,0x00,0x00,0x10]
+v_floor_f16_e64 v5.h, v1.h
+// GFX12: v_floor_f16_e64 v5.h, v1.h op_sel:[1,1] ; encoding: [0x05,0x48,0xdb,0xd5,0x01,0x01,0x00,0x00]
-v_floor_f16_e64 v255, -|0xfe0b| clamp div:2
-// GFX12: v_floor_f16_e64 v255, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xdb,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
+v_floor_f16_e64 v5.l, v255.h
+// GFX12: v_floor_f16_e64 v5.l, v255.h op_sel:[1,0] ; encoding: [0x05,0x08,0xdb,0xd5,0xff,0x01,0x00,0x00]
+
+v_floor_f16_e64 v255.h, -|0xfe0b| clamp div:2
+// GFX12: v_floor_f16_e64 v255.h, -|0xfe0b| op_sel:[0,1] clamp div:2 ; encoding: [0xff,0xc1,0xdb,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
v_floor_f32_e64 v5, v1
// GFX12: v_floor_f32_e64 v5, v1 ; encoding: [0x05,0x00,0xa4,0xd5,0x01,0x01,0x00,0x00]
@@ -2656,50 +2686,59 @@ v_frexp_mant_f64_e64 v[5:6], -|src_scc| mul:4
v_frexp_mant_f64_e64 v[254:255], 0xaf123456 clamp div:2
// GFX12: v_frexp_mant_f64_e64 v[254:255], 0xaf123456 clamp div:2 ; encoding: [0xfe,0x80,0xbd,0xd5,0xff,0x00,0x00,0x18,0x56,0x34,0x12,0xaf]
-v_log_f16_e64 v5, v1
-// GFX12: v_log_f16_e64 v5, v1 ; encoding: [0x05,0x00,0xd7,0xd5,0x01,0x01,0x00,0x00]
+v_log_f16_e64 v5.l, v1.l
+// GFX12: v_log_f16_e64 v5.l, v1.l ; encoding: [0x05,0x00,0xd7,0xd5,0x01,0x01,0x00,0x00]
+
+v_log_f16_e64 v5.l, v255.l
+// GFX12: v_log_f16_e64 v5.l, v255.l ; encoding: [0x05,0x00,0xd7,0xd5,0xff,0x01,0x00,0x00]
-v_log_f16_e64 v5, v255
-// GFX12: v_log_f16_e64 v5, v255 ; encoding: [0x05,0x00,0xd7,0xd5,0xff,0x01,0x00,0x00]
+v_log_f16_e64 v5.l, s1
+// GFX12: v_log_f16_e64 v5.l, s1 ; encoding: [0x05,0x00,0xd7,0xd5,0x01,0x00,0x00,0x00]
-v_log_f16_e64 v5, s1
-// GFX12: v_log_f16_e64 v5, s1 ; encoding: [0x05,0x00,0xd7,0xd5,0x01,0x00,0x00,0x00]
+v_log_f16_e64 v5.l, s105
+// GFX12: v_log_f16_e64 v5.l, s105 ; encoding: [0x05,0x00,0xd7,0xd5,0x69,0x00,0x00,0x00]
-v_log_f16_e64 v5, s105
-// GFX12: v_log_f16_e64 v5, s105 ; encoding: [0x05,0x00,0xd7,0xd5,0x69,0x00,0x00,0x00]
+v_log_f16_e64 v5.l, vcc_lo
+// GFX12: v_log_f16_e64 v5.l, vcc_lo ; encoding: [0x05,0x00,0xd7,0xd5,0x6a,0x00,0x00,0x00]
-v_log_f16_e64 v5, vcc_lo
-// GFX12: v_log_f16_e64 v5, vcc_lo ; encoding: [0x05,0x00,0xd7,0xd5,0x6a,0x00,0x00,0x00]
+v_log_f16_e64 v5.l, vcc_hi
+// GFX12: v_log_f16_e64 v5.l, vcc_hi ; encoding: [0x05,0x00,0xd7,0xd5,0x6b,0x00,0x00,0x00]
-v_log_f16_e64 v5, vcc_hi
-// GFX12: v_log_f16_e64 v5, vcc_hi ; encoding: [0x05,0x00,0xd7,0xd5,0x6b,0x00,0x00,0x00]
+v_log_f16_e64 v5.l, ttmp15
+// GFX12: v_log_f16_e64 v5.l, ttmp15 ; encoding: [0x05,0x00,0xd7,0xd5,0x7b,0x00,0x00,0x00]
-v_log_f16_e64 v5, ttmp15
-// GFX12: v_log_f16_e64 v5, ttmp15 ; encoding: [0x05,0x00,0xd7,0xd5,0x7b,0x00,0x00,0x00]
+v_log_f16_e64 v5.l, m0
+// GFX12: v_log_f16_e64 v5.l, m0 ; encoding: [0x05,0x00,0xd7,0xd5,0x7d,0x00,0x00,0x00]
-v_log_f16_e64 v5, m0
-// GFX12: v_log_f16_e64 v5, m0 ; encoding: [0x05,0x00,0xd7,0xd5,0x7d,0x00,0x00,0x00]
+v_log_f16_e64 v5.l, exec_lo
+// GFX12: v_log_f16_e64 v5.l, exec_lo ; encoding: [0x05,0x00,0xd7,0xd5,0x7e,0x00,0x00,0x00]
-v_log_f16_e64 v5, exec_lo
-// GFX12: v_log_f16_e64 v5, exec_lo ; encoding: [0x05,0x00,0xd7,0xd5,0x7e,0x00,0x00,0x00]
+v_log_f16_e64 v5.l, exec_hi
+// GFX12: v_log_f16_e64 v5.l, exec_hi ; encoding: [0x05,0x00,0xd7,0xd5,0x7f,0x00,0x00,0x00]
-v_log_f16_e64 v5, exec_hi
-// GFX12: v_log_f16_e64 v5, exec_hi ; encoding: [0x05,0x00,0xd7,0xd5,0x7f,0x00,0x00,0x00]
+v_log_f16_e64 v5.l, null
+// GFX12: v_log_f16_e64 v5.l, null ; encoding: [0x05,0x00,0xd7,0xd5,0x7c,0x00,0x00,0x00]
-v_log_f16_e64 v5, null
-// GFX12: v_log_f16_e64 v5, null ; encoding: [0x05,0x00,0xd7,0xd5,0x7c,0x00,0x00,0x00]
+v_log_f16_e64 v5.l, -1
+// GFX12: v_log_f16_e64 v5.l, -1 ; encoding: [0x05,0x00,0xd7,0xd5,0xc1,0x00,0x00,0x00]
-v_log_f16_e64 v5, -1
-// GFX12: v_log_f16_e64 v5, -1 ; encoding: [0x05,0x00,0xd7,0xd5,0xc1,0x00,0x00,0x00]
+v_log_f16_e64 v5.l, 0.5 mul:2
+// GFX12: v_log_f16_e64 v5.l, 0.5 mul:2 ; encoding: [0x05,0x00,0xd7,0xd5,0xf0,0x00,0x00,0x08]
-v_log_f16_e64 v5, 0.5 mul:2
-// GFX12: v_log_f16_e64 v5, 0.5 mul:2 ; encoding: [0x05,0x00,0xd7,0xd5,0xf0,0x00,0x00,0x08]
+v_log_f16_e64 v5.l, src_scc mul:4
+// GFX12: v_log_f16_e64 v5.l, src_scc mul:4 ; encoding: [0x05,0x00,0xd7,0xd5,0xfd,0x00,0x00,0x10]
-v_log_f16_e64 v5, src_scc mul:4
-// GFX12: v_log_f16_e64 v5, src_scc mul:4 ; encoding: [0x05,0x00,0xd7,0xd5,0xfd,0x00,0x00,0x10]
+v_log_f16_e64 v255.l, -|0xfe0b| clamp div:2
+// GFX12: v_log_f16_e64 v255.l, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xd7,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
-v_log_f16_e64 v255, -|0xfe0b| clamp div:2
-// GFX12: v_log_f16_e64 v255, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xd7,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
+v_log_f16_e64 v5.h, v1.h
+// GFX12: v_log_f16_e64 v5.h, v1.h op_sel:[1,1] ; encoding: [0x05,0x48,0xd7,0xd5,0x01,0x01,0x00,0x00]
+
+v_log_f16_e64 v5.l, v255.h
+// GFX12: v_log_f16_e64 v5.l, v255.h op_sel:[1,0] ; encoding: [0x05,0x08,0xd7,0xd5,0xff,0x01,0x00,0x00]
+
+v_log_f16_e64 v255.h, -|0xfe0b| clamp div:2
+// GFX12: v_log_f16_e64 v255.h, -|0xfe0b| op_sel:[0,1] clamp div:2 ; encoding: [0xff,0xc1,0xd7,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
v_log_f32_e64 v5, v1
// GFX12: v_log_f32_e64 v5, v1 ; encoding: [0x05,0x00,0xa7,0xd5,0x01,0x01,0x00,0x00]
@@ -2959,50 +2998,59 @@ v_not_b32_e64 v255, 0xaf123456
v_pipeflush_e64
// GFX12: v_pipeflush ; encoding: [0x00,0x00,0x9b,0xd5,0x00,0x00,0x00,0x00]
-v_rcp_f16_e64 v5, v1
-// GFX12: v_rcp_f16_e64 v5, v1 ; encoding: [0x05,0x00,0xd4,0xd5,0x01,0x01,0x00,0x00]
+v_rcp_f16_e64 v5.l, v1.l
+// GFX12: v_rcp_f16_e64 v5.l, v1.l ; encoding: [0x05,0x00,0xd4,0xd5,0x01,0x01,0x00,0x00]
+
+v_rcp_f16_e64 v5.l, v255.l
+// GFX12: v_rcp_f16_e64 v5.l, v255.l ; encoding: [0x05,0x00,0xd4,0xd5,0xff,0x01,0x00,0x00]
-v_rcp_f16_e64 v5, v255
-// GFX12: v_rcp_f16_e64 v5, v255 ; encoding: [0x05,0x00,0xd4,0xd5,0xff,0x01,0x00,0x00]
+v_rcp_f16_e64 v5.l, s1
+// GFX12: v_rcp_f16_e64 v5.l, s1 ; encoding: [0x05,0x00,0xd4,0xd5,0x01,0x00,0x00,0x00]
-v_rcp_f16_e64 v5, s1
-// GFX12: v_rcp_f16_e64 v5, s1 ; encoding: [0x05,0x00,0xd4,0xd5,0x01,0x00,0x00,0x00]
+v_rcp_f16_e64 v5.l, s105
+// GFX12: v_rcp_f16_e64 v5.l, s105 ; encoding: [0x05,0x00,0xd4,0xd5,0x69,0x00,0x00,0x00]
-v_rcp_f16_e64 v5, s105
-// GFX12: v_rcp_f16_e64 v5, s105 ; encoding: [0x05,0x00,0xd4,0xd5,0x69,0x00,0x00,0x00]
+v_rcp_f16_e64 v5.l, vcc_lo
+// GFX12: v_rcp_f16_e64 v5.l, vcc_lo ; encoding: [0x05,0x00,0xd4,0xd5,0x6a,0x00,0x00,0x00]
-v_rcp_f16_e64 v5, vcc_lo
-// GFX12: v_rcp_f16_e64 v5, vcc_lo ; encoding: [0x05,0x00,0xd4,0xd5,0x6a,0x00,0x00,0x00]
+v_rcp_f16_e64 v5.l, vcc_hi
+// GFX12: v_rcp_f16_e64 v5.l, vcc_hi ; encoding: [0x05,0x00,0xd4,0xd5,0x6b,0x00,0x00,0x00]
-v_rcp_f16_e64 v5, vcc_hi
-// GFX12: v_rcp_f16_e64 v5, vcc_hi ; encoding: [0x05,0x00,0xd4,0xd5,0x6b,0x00,0x00,0x00]
+v_rcp_f16_e64 v5.l, ttmp15
+// GFX12: v_rcp_f16_e64 v5.l, ttmp15 ; encoding: [0x05,0x00,0xd4,0xd5,0x7b,0x00,0x00,0x00]
-v_rcp_f16_e64 v5, ttmp15
-// GFX12: v_rcp_f16_e64 v5, ttmp15 ; encoding: [0x05,0x00,0xd4,0xd5,0x7b,0x00,0x00,0x00]
+v_rcp_f16_e64 v5.l, m0
+// GFX12: v_rcp_f16_e64 v5.l, m0 ; encoding: [0x05,0x00,0xd4,0xd5,0x7d,0x00,0x00,0x00]
-v_rcp_f16_e64 v5, m0
-// GFX12: v_rcp_f16_e64 v5, m0 ; encoding: [0x05,0x00,0xd4,0xd5,0x7d,0x00,0x00,0x00]
+v_rcp_f16_e64 v5.l, exec_lo
+// GFX12: v_rcp_f16_e64 v5.l, exec_lo ; encoding: [0x05,0x00,0xd4,0xd5,0x7e,0x00,0x00,0x00]
-v_rcp_f16_e64 v5, exec_lo
-// GFX12: v_rcp_f16_e64 v5, exec_lo ; encoding: [0x05,0x00,0xd4,0xd5,0x7e,0x00,0x00,0x00]
+v_rcp_f16_e64 v5.l, exec_hi
+// GFX12: v_rcp_f16_e64 v5.l, exec_hi ; encoding: [0x05,0x00,0xd4,0xd5,0x7f,0x00,0x00,0x00]
-v_rcp_f16_e64 v5, exec_hi
-// GFX12: v_rcp_f16_e64 v5, exec_hi ; encoding: [0x05,0x00,0xd4,0xd5,0x7f,0x00,0x00,0x00]
+v_rcp_f16_e64 v5.l, null
+// GFX12: v_rcp_f16_e64 v5.l, null ; encoding: [0x05,0x00,0xd4,0xd5,0x7c,0x00,0x00,0x00]
-v_rcp_f16_e64 v5, null
-// GFX12: v_rcp_f16_e64 v5, null ; encoding: [0x05,0x00,0xd4,0xd5,0x7c,0x00,0x00,0x00]
+v_rcp_f16_e64 v5.l, -1
+// GFX12: v_rcp_f16_e64 v5.l, -1 ; encoding: [0x05,0x00,0xd4,0xd5,0xc1,0x00,0x00,0x00]
-v_rcp_f16_e64 v5, -1
-// GFX12: v_rcp_f16_e64 v5, -1 ; encoding: [0x05,0x00,0xd4,0xd5,0xc1,0x00,0x00,0x00]
+v_rcp_f16_e64 v5.l, 0.5 mul:2
+// GFX12: v_rcp_f16_e64 v5.l, 0.5 mul:2 ; encoding: [0x05,0x00,0xd4,0xd5,0xf0,0x00,0x00,0x08]
-v_rcp_f16_e64 v5, 0.5 mul:2
-// GFX12: v_rcp_f16_e64 v5, 0.5 mul:2 ; encoding: [0x05,0x00,0xd4,0xd5,0xf0,0x00,0x00,0x08]
+v_rcp_f16_e64 v5.l, src_scc mul:4
+// GFX12: v_rcp_f16_e64 v5.l, src_scc mul:4 ; encoding: [0x05,0x00,0xd4,0xd5,0xfd,0x00,0x00,0x10]
-v_rcp_f16_e64 v5, src_scc mul:4
-// GFX12: v_rcp_f16_e64 v5, src_scc mul:4 ; encoding: [0x05,0x00,0xd4,0xd5,0xfd,0x00,0x00,0x10]
+v_rcp_f16_e64 v255.l, -|0xfe0b| clamp div:2
+// GFX12: v_rcp_f16_e64 v255.l, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xd4,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
-v_rcp_f16_e64 v255, -|0xfe0b| clamp div:2
-// GFX12: v_rcp_f16_e64 v255, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xd4,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
+v_rcp_f16_e64 v5.h, v1.h
+// GFX12: v_rcp_f16_e64 v5.h, v1.h op_sel:[1,1] ; encoding: [0x05,0x48,0xd4,0xd5,0x01,0x01,0x00,0x00]
+
+v_rcp_f16_e64 v5.l, v255.h
+// GFX12: v_rcp_f16_e64 v5.l, v255.h op_sel:[1,0] ; encoding: [0x05,0x08,0xd4,0xd5,0xff,0x01,0x00,0x00]
+
+v_rcp_f16_e64 v255.h, -|0xfe0b| clamp div:2
+// GFX12: v_rcp_f16_e64 v255.h, -|0xfe0b| op_sel:[0,1] clamp div:2 ; encoding: [0xff,0xc1,0xd4,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
v_rcp_f32_e64 v5, v1
// GFX12: v_rcp_f32_e64 v5, v1 ; encoding: [0x05,0x00,0xaa,0xd5,0x01,0x01,0x00,0x00]
@@ -3265,50 +3313,59 @@ v_rndne_f64_e64 v[5:6], -|src_scc| mul:4
v_rndne_f64_e64 v[254:255], 0xaf123456 clamp div:2
// GFX12: v_rndne_f64_e64 v[254:255], 0xaf123456 clamp div:2 ; encoding: [0xfe,0x80,0x99,0xd5,0xff,0x00,0x00,0x18,0x56,0x34,0x12,0xaf]
-v_rsq_f16_e64 v5, v1
-// GFX12: v_rsq_f16_e64 v5, v1 ; encoding: [0x05,0x00,0xd6,0xd5,0x01,0x01,0x00,0x00]
+v_rsq_f16_e64 v5.l, v1.l
+// GFX12: v_rsq_f16_e64 v5.l, v1.l ; encoding: [0x05,0x00,0xd6,0xd5,0x01,0x01,0x00,0x00]
+
+v_rsq_f16_e64 v5.l, v255.l
+// GFX12: v_rsq_f16_e64 v5.l, v255.l ; encoding: [0x05,0x00,0xd6,0xd5,0xff,0x01,0x00,0x00]
-v_rsq_f16_e64 v5, v255
-// GFX12: v_rsq_f16_e64 v5, v255 ; encoding: [0x05,0x00,0xd6,0xd5,0xff,0x01,0x00,0x00]
+v_rsq_f16_e64 v5.l, s1
+// GFX12: v_rsq_f16_e64 v5.l, s1 ; encoding: [0x05,0x00,0xd6,0xd5,0x01,0x00,0x00,0x00]
-v_rsq_f16_e64 v5, s1
-// GFX12: v_rsq_f16_e64 v5, s1 ; encoding: [0x05,0x00,0xd6,0xd5,0x01,0x00,0x00,0x00]
+v_rsq_f16_e64 v5.l, s105
+// GFX12: v_rsq_f16_e64 v5.l, s105 ; encoding: [0x05,0x00,0xd6,0xd5,0x69,0x00,0x00,0x00]
-v_rsq_f16_e64 v5, s105
-// GFX12: v_rsq_f16_e64 v5, s105 ; encoding: [0x05,0x00,0xd6,0xd5,0x69,0x00,0x00,0x00]
+v_rsq_f16_e64 v5.l, vcc_lo
+// GFX12: v_rsq_f16_e64 v5.l, vcc_lo ; encoding: [0x05,0x00,0xd6,0xd5,0x6a,0x00,0x00,0x00]
-v_rsq_f16_e64 v5, vcc_lo
-// GFX12: v_rsq_f16_e64 v5, vcc_lo ; encoding: [0x05,0x00,0xd6,0xd5,0x6a,0x00,0x00,0x00]
+v_rsq_f16_e64 v5.l, vcc_hi
+// GFX12: v_rsq_f16_e64 v5.l, vcc_hi ; encoding: [0x05,0x00,0xd6,0xd5,0x6b,0x00,0x00,0x00]
-v_rsq_f16_e64 v5, vcc_hi
-// GFX12: v_rsq_f16_e64 v5, vcc_hi ; encoding: [0x05,0x00,0xd6,0xd5,0x6b,0x00,0x00,0x00]
+v_rsq_f16_e64 v5.l, ttmp15
+// GFX12: v_rsq_f16_e64 v5.l, ttmp15 ; encoding: [0x05,0x00,0xd6,0xd5,0x7b,0x00,0x00,0x00]
-v_rsq_f16_e64 v5, ttmp15
-// GFX12: v_rsq_f16_e64 v5, ttmp15 ; encoding: [0x05,0x00,0xd6,0xd5,0x7b,0x00,0x00,0x00]
+v_rsq_f16_e64 v5.l, m0
+// GFX12: v_rsq_f16_e64 v5.l, m0 ; encoding: [0x05,0x00,0xd6,0xd5,0x7d,0x00,0x00,0x00]
-v_rsq_f16_e64 v5, m0
-// GFX12: v_rsq_f16_e64 v5, m0 ; encoding: [0x05,0x00,0xd6,0xd5,0x7d,0x00,0x00,0x00]
+v_rsq_f16_e64 v5.l, exec_lo
+// GFX12: v_rsq_f16_e64 v5.l, exec_lo ; encoding: [0x05,0x00,0xd6,0xd5,0x7e,0x00,0x00,0x00]
-v_rsq_f16_e64 v5, exec_lo
-// GFX12: v_rsq_f16_e64 v5, exec_lo ; encoding: [0x05,0x00,0xd6,0xd5,0x7e,0x00,0x00,0x00]
+v_rsq_f16_e64 v5.l, exec_hi
+// GFX12: v_rsq_f16_e64 v5.l, exec_hi ; encoding: [0x05,0x00,0xd6,0xd5,0x7f,0x00,0x00,0x00]
-v_rsq_f16_e64 v5, exec_hi
-// GFX12: v_rsq_f16_e64 v5, exec_hi ; encoding: [0x05,0x00,0xd6,0xd5,0x7f,0x00,0x00,0x00]
+v_rsq_f16_e64 v5.l, null
+// GFX12: v_rsq_f16_e64 v5.l, null ; encoding: [0x05,0x00,0xd6,0xd5,0x7c,0x00,0x00,0x00]
-v_rsq_f16_e64 v5, null
-// GFX12: v_rsq_f16_e64 v5, null ; encoding: [0x05,0x00,0xd6,0xd5,0x7c,0x00,0x00,0x00]
+v_rsq_f16_e64 v5.l, -1
+// GFX12: v_rsq_f16_e64 v5.l, -1 ; encoding: [0x05,0x00,0xd6,0xd5,0xc1,0x00,0x00,0x00]
-v_rsq_f16_e64 v5, -1
-// GFX12: v_rsq_f16_e64 v5, -1 ; encoding: [0x05,0x00,0xd6,0xd5,0xc1,0x00,0x00,0x00]
+v_rsq_f16_e64 v5.l, 0.5 mul:2
+// GFX12: v_rsq_f16_e64 v5.l, 0.5 mul:2 ; encoding: [0x05,0x00,0xd6,0xd5,0xf0,0x00,0x00,0x08]
-v_rsq_f16_e64 v5, 0.5 mul:2
-// GFX12: v_rsq_f16_e64 v5, 0.5 mul:2 ; encoding: [0x05,0x00,0xd6,0xd5,0xf0,0x00,0x00,0x08]
+v_rsq_f16_e64 v5.l, src_scc mul:4
+// GFX12: v_rsq_f16_e64 v5.l, src_scc mul:4 ; encoding: [0x05,0x00,0xd6,0xd5,0xfd,0x00,0x00,0x10]
-v_rsq_f16_e64 v5, src_scc mul:4
-// GFX12: v_rsq_f16_e64 v5, src_scc mul:4 ; encoding: [0x05,0x00,0xd6,0xd5,0xfd,0x00,0x00,0x10]
+v_rsq_f16_e64 v255.l, -|0xfe0b| clamp div:2
+// GFX12: v_rsq_f16_e64 v255.l, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xd6,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
-v_rsq_f16_e64 v255, -|0xfe0b| clamp div:2
-// GFX12: v_rsq_f16_e64 v255, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xd6,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
+v_rsq_f16_e64 v5.h, v1.h
+// GFX12: v_rsq_f16_e64 v5.h, v1.h op_sel:[1,1] ; encoding: [0x05,0x48,0xd6,0xd5,0x01,0x01,0x00,0x00]
+
+v_rsq_f16_e64 v5.l, v255.h
+// GFX12: v_rsq_f16_e64 v5.l, v255.h op_sel:[1,0] ; encoding: [0x05,0x08,0xd6,0xd5,0xff,0x01,0x00,0x00]
+
+v_rsq_f16_e64 v255.h, -|0xfe0b| clamp div:2
+// GFX12: v_rsq_f16_e64 v255.h, -|0xfe0b| op_sel:[0,1] clamp div:2 ; encoding: [0xff,0xc1,0xd6,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
v_rsq_f32_e64 v5, v1
// GFX12: v_rsq_f32_e64 v5, v1 ; encoding: [0x05,0x00,0xae,0xd5,0x01,0x01,0x00,0x00]
@@ -3538,50 +3595,59 @@ v_sin_f32_e64 v5, src_scc mul:4
v_sin_f32_e64 v255, -|0xaf123456| clamp div:2
// GFX12: v_sin_f32_e64 v255, -|0xaf123456| clamp div:2 ; encoding: [0xff,0x81,0xb5,0xd5,0xff,0x00,0x00,0x38,0x56,0x34,0x12,0xaf]
-v_sqrt_f16_e64 v5, v1
-// GFX12: v_sqrt_f16_e64 v5, v1 ; encoding: [0x05,0x00,0xd5,0xd5,0x01,0x01,0x00,0x00]
+v_sqrt_f16_e64 v5.l, v1.l
+// GFX12: v_sqrt_f16_e64 v5.l, v1.l ; encoding: [0x05,0x00,0xd5,0xd5,0x01,0x01,0x00,0x00]
+
+v_sqrt_f16_e64 v5.l, v255.l
+// GFX12: v_sqrt_f16_e64 v5.l, v255.l ; encoding: [0x05,0x00,0xd5,0xd5,0xff,0x01,0x00,0x00]
+
+v_sqrt_f16_e64 v5.l, s1
+// GFX12: v_sqrt_f16_e64 v5.l, s1 ; encoding: [0x05,0x00,0xd5,0xd5,0x01,0x00,0x00,0x00]
+
+v_sqrt_f16_e64 v5.l, s105
+// GFX12: v_sqrt_f16_e64 v5.l, s105 ; encoding: [0x05,0x00,0xd5,0xd5,0x69,0x00,0x00,0x00]
-v_sqrt_f16_e64 v5, v255
-// GFX12: v_sqrt_f16_e64 v5, v255 ; encoding: [0x05,0x00,0xd5,0xd5,0xff,0x01,0x00,0x00]
+v_sqrt_f16_e64 v5.l, vcc_lo
+// GFX12: v_sqrt_f16_e64 v5.l, vcc_lo ; encoding: [0x05,0x00,0xd5,0xd5,0x6a,0x00,0x00,0x00]
-v_sqrt_f16_e64 v5, s1
-// GFX12: v_sqrt_f16_e64 v5, s1 ; encoding: [0x05,0x00,0xd5,0xd5,0x01,0x00,0x00,0x00]
+v_sqrt_f16_e64 v5.l, vcc_hi
+// GFX12: v_sqrt_f16_e64 v5.l, vcc_hi ; encoding: [0x05,0x00,0xd5,0xd5,0x6b,0x00,0x00,0x00]
-v_sqrt_f16_e64 v5, s105
-// GFX12: v_sqrt_f16_e64 v5, s105 ; encoding: [0x05,0x00,0xd5,0xd5,0x69,0x00,0x00,0x00]
+v_sqrt_f16_e64 v5.l, ttmp15
+// GFX12: v_sqrt_f16_e64 v5.l, ttmp15 ; encoding: [0x05,0x00,0xd5,0xd5,0x7b,0x00,0x00,0x00]
-v_sqrt_f16_e64 v5, vcc_lo
-// GFX12: v_sqrt_f16_e64 v5, vcc_lo ; encoding: [0x05,0x00,0xd5,0xd5,0x6a,0x00,0x00,0x00]
+v_sqrt_f16_e64 v5.l, m0
+// GFX12: v_sqrt_f16_e64 v5.l, m0 ; encoding: [0x05,0x00,0xd5,0xd5,0x7d,0x00,0x00,0x00]
-v_sqrt_f16_e64 v5, vcc_hi
-// GFX12: v_sqrt_f16_e64 v5, vcc_hi ; encoding: [0x05,0x00,0xd5,0xd5,0x6b,0x00,0x00,0x00]
+v_sqrt_f16_e64 v5.l, exec_lo
+// GFX12: v_sqrt_f16_e64 v5.l, exec_lo ; encoding: [0x05,0x00,0xd5,0xd5,0x7e,0x00,0x00,0x00]
-v_sqrt_f16_e64 v5, ttmp15
-// GFX12: v_sqrt_f16_e64 v5, ttmp15 ; encoding: [0x05,0x00,0xd5,0xd5,0x7b,0x00,0x00,0x00]
+v_sqrt_f16_e64 v5.l, exec_hi
+// GFX12: v_sqrt_f16_e64 v5.l, exec_hi ; encoding: [0x05,0x00,0xd5,0xd5,0x7f,0x00,0x00,0x00]
-v_sqrt_f16_e64 v5, m0
-// GFX12: v_sqrt_f16_e64 v5, m0 ; encoding: [0x05,0x00,0xd5,0xd5,0x7d,0x00,0x00,0x00]
+v_sqrt_f16_e64 v5.l, null
+// GFX12: v_sqrt_f16_e64 v5.l, null ; encoding: [0x05,0x00,0xd5,0xd5,0x7c,0x00,0x00,0x00]
-v_sqrt_f16_e64 v5, exec_lo
-// GFX12: v_sqrt_f16_e64 v5, exec_lo ; encoding: [0x05,0x00,0xd5,0xd5,0x7e,0x00,0x00,0x00]
+v_sqrt_f16_e64 v5.l, -1
+// GFX12: v_sqrt_f16_e64 v5.l, -1 ; encoding: [0x05,0x00,0xd5,0xd5,0xc1,0x00,0x00,0x00]
-v_sqrt_f16_e64 v5, exec_hi
-// GFX12: v_sqrt_f16_e64 v5, exec_hi ; encoding: [0x05,0x00,0xd5,0xd5,0x7f,0x00,0x00,0x00]
+v_sqrt_f16_e64 v5.l, 0.5 mul:2
+// GFX12: v_sqrt_f16_e64 v5.l, 0.5 mul:2 ; encoding: [0x05,0x00,0xd5,0xd5,0xf0,0x00,0x00,0x08]
-v_sqrt_f16_e64 v5, null
-// GFX12: v_sqrt_f16_e64 v5, null ; encoding: [0x05,0x00,0xd5,0xd5,0x7c,0x00,0x00,0x00]
+v_sqrt_f16_e64 v5.l, src_scc mul:4
+// GFX12: v_sqrt_f16_e64 v5.l, src_scc mul:4 ; encoding: [0x05,0x00,0xd5,0xd5,0xfd,0x00,0x00,0x10]
-v_sqrt_f16_e64 v5, -1
-// GFX12: v_sqrt_f16_e64 v5, -1 ; encoding: [0x05,0x00,0xd5,0xd5,0xc1,0x00,0x00,0x00]
+v_sqrt_f16_e64 v255.l, -|0xfe0b| clamp div:2
+// GFX12: v_sqrt_f16_e64 v255.l, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xd5,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
-v_sqrt_f16_e64 v5, 0.5 mul:2
-// GFX12: v_sqrt_f16_e64 v5, 0.5 mul:2 ; encoding: [0x05,0x00,0xd5,0xd5,0xf0,0x00,0x00,0x08]
+v_sqrt_f16_e64 v5.h, v1.h
+// GFX12: v_sqrt_f16_e64 v5.h, v1.h op_sel:[1,1] ; encoding: [0x05,0x48,0xd5,0xd5,0x01,0x01,0x00,0x00]
-v_sqrt_f16_e64 v5, src_scc mul:4
-// GFX12: v_sqrt_f16_e64 v5, src_scc mul:4 ; encoding: [0x05,0x00,0xd5,0xd5,0xfd,0x00,0x00,0x10]
+v_sqrt_f16_e64 v5.l, v255.h
+// GFX12: v_sqrt_f16_e64 v5.l, v255.h op_sel:[1,0] ; encoding: [0x05,0x08,0xd5,0xd5,0xff,0x01,0x00,0x00]
-v_sqrt_f16_e64 v255, -|0xfe0b| clamp div:2
-// GFX12: v_sqrt_f16_e64 v255, -|0xfe0b| clamp div:2 ; encoding: [0xff,0x81,0xd5,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
+v_sqrt_f16_e64 v255.h, -|0xfe0b| clamp div:2
+// GFX12: v_sqrt_f16_e64 v255.h, -|0xfe0b| op_sel:[0,1] clamp div:2 ; encoding: [0xff,0xc1,0xd5,0xd5,0xff,0x00,0x00,0x38,0x0b,0xfe,0x00,0x00]
v_sqrt_f32_e64 v5, v1
// GFX12: v_sqrt_f32_e64 v5, v1 ; encoding: [0x05,0x00,0xb3,0xd5,0x01,0x01,0x00,0x00]
diff --git a/llvm/test/MC/AMDGPU/gfx12_asm_vop3_from_vop1_dpp16.s b/llvm/test/MC/AMDGPU/gfx12_asm_vop3_from_vop1_dpp16.s
index 89102ae780f16..6d07f299d1d62 100644
--- a/llvm/test/MC/AMDGPU/gfx12_asm_vop3_from_vop1_dpp16.s
+++ b/llvm/test/MC/AMDGPU/gfx12_asm_vop3_from_vop1_dpp16.s
@@ -43,47 +43,56 @@ v_bfrev_b32_e64_dpp v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 f
v_bfrev_b32_e64_dpp v255, v255 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
// GFX12: v_bfrev_b32_e64_dpp v255, v255 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x00,0xb8,0xd5,0xfa,0x00,0x00,0x00,0xff,0x6f,0x05,0x30]
-v_ceil_f16_e64_dpp v5, v1 quad_perm:[3,2,1,0]
-// GFX12: v_ceil_f16_e64_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+v_ceil_f16_e64_dpp v5.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: v_ceil_f16_e64_dpp v5.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
-v_ceil_f16_e64_dpp v5, v1 quad_perm:[0,1,2,3]
-// GFX12: v_ceil_f16_e64_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]
+v_ceil_f16_e64_dpp v5.l, v1.l quad_perm:[0,1,2,3]
+// GFX12: v_ceil_f16_e64_dpp v5.l, v1.l quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]
-v_ceil_f16_e64_dpp v5, v1 row_mirror
-// GFX12: v_ceil_f16_e64_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]
+v_ceil_f16_e64_dpp v5.l, v1.l row_mirror
+// GFX12: v_ceil_f16_e64_dpp v5.l, v1.l row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]
-v_ceil_f16_e64_dpp v5, v1 row_half_mirror
-// GFX12: v_ceil_f16_e64_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]
+v_ceil_f16_e64_dpp v5.l, v1.l row_half_mirror
+// GFX12: v_ceil_f16_e64_dpp v5.l, v1.l row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]
-v_ceil_f16_e64_dpp v5, v1 row_shl:1
-// GFX12: v_ceil_f16_e64_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]
+v_ceil_f16_e64_dpp v5.l, v1.l row_shl:1
+// GFX12: v_ceil_f16_e64_dpp v5.l, v1.l row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]
-v_ceil_f16_e64_dpp v5, v1 row_shl:15
-// GFX12: v_ceil_f16_e64_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]
+v_ceil_f16_e64_dpp v5.l, v1.l row_shl:15
+// GFX12: v_ceil_f16_e64_dpp v5.l, v1.l row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]
-v_ceil_f16_e64_dpp v5, v1 row_shr:1
-// GFX12: v_ceil_f16_e64_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]
+v_ceil_f16_e64_dpp v5.l, v1.l row_shr:1
+// GFX12: v_ceil_f16_e64_dpp v5.l, v1.l row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]
-v_ceil_f16_e64_dpp v5, v1 row_shr:15
-// GFX12: v_ceil_f16_e64_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]
+v_ceil_f16_e64_dpp v5.l, v1.l row_shr:15
+// GFX12: v_ceil_f16_e64_dpp v5.l, v1.l row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]
-v_ceil_f16_e64_dpp v5, v1 row_ror:1
-// GFX12: v_ceil_f16_e64_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]
+v_ceil_f16_e64_dpp v5.l, v1.l row_ror:1
+// GFX12: v_ceil_f16_e64_dpp v5.l, v1.l row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]
-v_ceil_f16_e64_dpp v5, v1 row_ror:15
-// GFX12: v_ceil_f16_e64_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]
+v_ceil_f16_e64_dpp v5.l, v1.l row_ror:15
+// GFX12: v_ceil_f16_e64_dpp v5.l, v1.l row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]
-v_ceil_f16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
-// GFX12: v_ceil_f16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]
+v_ceil_f16_e64_dpp v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf
+// GFX12: v_ceil_f16_e64_dpp v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]
-v_ceil_f16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
-// GFX12: v_ceil_f16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
+v_ceil_f16_e64_dpp v5.l, v1.l mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX12: v_ceil_f16_e64_dpp v5.l, v1.l mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
-v_ceil_f16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
-// GFX12: v_ceil_f16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
+v_ceil_f16_e64_dpp v5.l, v1.l mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_ceil_f16_e64_dpp v5.l, v1.l mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
-v_ceil_f16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
-// GFX12: v_ceil_f16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xdc,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
+v_ceil_f16_e64_dpp v255.l, -|v255.l| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_ceil_f16_e64_dpp v255.l, -|v255.l| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xdc,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
+
+v_ceil_f16_e64_dpp v5.h, v1.h mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX12: v_ceil_f16_e64_dpp v5.h, v1.h op_sel:[1,1] mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x48,0xdc,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
+
+v_ceil_f16_e64_dpp v5.l, v1.h mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_ceil_f16_e64_dpp v5.l, v1.h op_sel:[1,0] mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x08,0xdc,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
+
+v_ceil_f16_e64_dpp v255.h, -|v255.l| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_ceil_f16_e64_dpp v255.h, -|v255.l| op_sel:[0,1] clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0xc1,0xdc,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
v_ceil_f32_e64_dpp v5, v1 quad_perm:[3,2,1,0]
// GFX12: v_ceil_f32_e64_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xa2,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
@@ -1420,47 +1429,56 @@ v_cvt_u32_u16_e64_dpp v255, v255.l row_xmask:15 row_mask:0x3 bank_mask:0x0 bound
v_cvt_u32_u16_e64_dpp v255, v255.h row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
// GFX12: v_cvt_u32_u16_e64_dpp v255, v255.h op_sel:[1,0] row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x08,0xeb,0xd5,0xfa,0x00,0x00,0x00,0xff,0x6f,0x05,0x30]
-v_exp_f16_e64_dpp v5, v1 quad_perm:[3,2,1,0]
-// GFX12: v_exp_f16_e64_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+v_exp_f16_e64_dpp v5.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: v_exp_f16_e64_dpp v5.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+
+v_exp_f16_e64_dpp v5.l, v1.l quad_perm:[0,1,2,3]
+// GFX12: v_exp_f16_e64_dpp v5.l, v1.l quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]
+
+v_exp_f16_e64_dpp v5.l, v1.l row_mirror
+// GFX12: v_exp_f16_e64_dpp v5.l, v1.l row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]
+
+v_exp_f16_e64_dpp v5.l, v1.l row_half_mirror
+// GFX12: v_exp_f16_e64_dpp v5.l, v1.l row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]
-v_exp_f16_e64_dpp v5, v1 quad_perm:[0,1,2,3]
-// GFX12: v_exp_f16_e64_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]
+v_exp_f16_e64_dpp v5.l, v1.l row_shl:1
+// GFX12: v_exp_f16_e64_dpp v5.l, v1.l row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]
-v_exp_f16_e64_dpp v5, v1 row_mirror
-// GFX12: v_exp_f16_e64_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]
+v_exp_f16_e64_dpp v5.l, v1.l row_shl:15
+// GFX12: v_exp_f16_e64_dpp v5.l, v1.l row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]
-v_exp_f16_e64_dpp v5, v1 row_half_mirror
-// GFX12: v_exp_f16_e64_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]
+v_exp_f16_e64_dpp v5.l, v1.l row_shr:1
+// GFX12: v_exp_f16_e64_dpp v5.l, v1.l row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]
-v_exp_f16_e64_dpp v5, v1 row_shl:1
-// GFX12: v_exp_f16_e64_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]
+v_exp_f16_e64_dpp v5.l, v1.l row_shr:15
+// GFX12: v_exp_f16_e64_dpp v5.l, v1.l row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]
-v_exp_f16_e64_dpp v5, v1 row_shl:15
-// GFX12: v_exp_f16_e64_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]
+v_exp_f16_e64_dpp v5.l, v1.l row_ror:1
+// GFX12: v_exp_f16_e64_dpp v5.l, v1.l row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]
-v_exp_f16_e64_dpp v5, v1 row_shr:1
-// GFX12: v_exp_f16_e64_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]
+v_exp_f16_e64_dpp v5.l, v1.l row_ror:15
+// GFX12: v_exp_f16_e64_dpp v5.l, v1.l row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]
-v_exp_f16_e64_dpp v5, v1 row_shr:15
-// GFX12: v_exp_f16_e64_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]
+v_exp_f16_e64_dpp v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf
+// GFX12: v_exp_f16_e64_dpp v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]
-v_exp_f16_e64_dpp v5, v1 row_ror:1
-// GFX12: v_exp_f16_e64_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]
+v_exp_f16_e64_dpp v5.l, v1.l mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX12: v_exp_f16_e64_dpp v5.l, v1.l mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
-v_exp_f16_e64_dpp v5, v1 row_ror:15
-// GFX12: v_exp_f16_e64_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]
+v_exp_f16_e64_dpp v5.l, v1.l mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_exp_f16_e64_dpp v5.l, v1.l mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
-v_exp_f16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
-// GFX12: v_exp_f16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]
+v_exp_f16_e64_dpp v255.l, -|v255.l| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_exp_f16_e64_dpp v255.l, -|v255.l| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xd8,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
-v_exp_f16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
-// GFX12: v_exp_f16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
+v_exp_f16_e64_dpp v5.h, v1.h mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX12: v_exp_f16_e64_dpp v5.h, v1.h op_sel:[1,1] mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x48,0xd8,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
-v_exp_f16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
-// GFX12: v_exp_f16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
+v_exp_f16_e64_dpp v5.l, v1.h mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_exp_f16_e64_dpp v5.l, v1.h op_sel:[1,0] mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x08,0xd8,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
-v_exp_f16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
-// GFX12: v_exp_f16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xd8,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
+v_exp_f16_e64_dpp v255.h, -|v255.l| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_exp_f16_e64_dpp v255.h, -|v255.l| op_sel:[0,1] clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0xc1,0xd8,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
v_exp_f32_e64_dpp v5, v1 quad_perm:[3,2,1,0]
// GFX12: v_exp_f32_e64_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xa5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
@@ -1630,47 +1648,56 @@ v_ffbl_b32_e64_dpp v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi
v_ffbl_b32_e64_dpp v255, v255 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
// GFX12: v_ctz_i32_b32_e64_dpp v255, v255 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x00,0xba,0xd5,0xfa,0x00,0x00,0x00,0xff,0x6f,0x05,0x30]
-v_floor_f16_e64_dpp v5, v1 quad_perm:[3,2,1,0]
-// GFX12: v_floor_f16_e64_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+v_floor_f16_e64_dpp v5.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: v_floor_f16_e64_dpp v5.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
-v_floor_f16_e64_dpp v5, v1 quad_perm:[0,1,2,3]
-// GFX12: v_floor_f16_e64_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]
+v_floor_f16_e64_dpp v5.l, v1.l quad_perm:[0,1,2,3]
+// GFX12: v_floor_f16_e64_dpp v5.l, v1.l quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]
-v_floor_f16_e64_dpp v5, v1 row_mirror
-// GFX12: v_floor_f16_e64_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]
+v_floor_f16_e64_dpp v5.l, v1.l row_mirror
+// GFX12: v_floor_f16_e64_dpp v5.l, v1.l row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]
-v_floor_f16_e64_dpp v5, v1 row_half_mirror
-// GFX12: v_floor_f16_e64_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]
+v_floor_f16_e64_dpp v5.l, v1.l row_half_mirror
+// GFX12: v_floor_f16_e64_dpp v5.l, v1.l row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]
-v_floor_f16_e64_dpp v5, v1 row_shl:1
-// GFX12: v_floor_f16_e64_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]
+v_floor_f16_e64_dpp v5.l, v1.l row_shl:1
+// GFX12: v_floor_f16_e64_dpp v5.l, v1.l row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]
-v_floor_f16_e64_dpp v5, v1 row_shl:15
-// GFX12: v_floor_f16_e64_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]
+v_floor_f16_e64_dpp v5.l, v1.l row_shl:15
+// GFX12: v_floor_f16_e64_dpp v5.l, v1.l row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]
-v_floor_f16_e64_dpp v5, v1 row_shr:1
-// GFX12: v_floor_f16_e64_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]
+v_floor_f16_e64_dpp v5.l, v1.l row_shr:1
+// GFX12: v_floor_f16_e64_dpp v5.l, v1.l row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]
-v_floor_f16_e64_dpp v5, v1 row_shr:15
-// GFX12: v_floor_f16_e64_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]
+v_floor_f16_e64_dpp v5.l, v1.l row_shr:15
+// GFX12: v_floor_f16_e64_dpp v5.l, v1.l row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]
-v_floor_f16_e64_dpp v5, v1 row_ror:1
-// GFX12: v_floor_f16_e64_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]
+v_floor_f16_e64_dpp v5.l, v1.l row_ror:1
+// GFX12: v_floor_f16_e64_dpp v5.l, v1.l row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]
-v_floor_f16_e64_dpp v5, v1 row_ror:15
-// GFX12: v_floor_f16_e64_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]
+v_floor_f16_e64_dpp v5.l, v1.l row_ror:15
+// GFX12: v_floor_f16_e64_dpp v5.l, v1.l row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]
-v_floor_f16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
-// GFX12: v_floor_f16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]
+v_floor_f16_e64_dpp v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf
+// GFX12: v_floor_f16_e64_dpp v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]
-v_floor_f16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
-// GFX12: v_floor_f16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
+v_floor_f16_e64_dpp v5.l, v1.l mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX12: v_floor_f16_e64_dpp v5.l, v1.l mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
-v_floor_f16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
-// GFX12: v_floor_f16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
+v_floor_f16_e64_dpp v5.l, v1.l mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_floor_f16_e64_dpp v5.l, v1.l mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
-v_floor_f16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
-// GFX12: v_floor_f16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xdb,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
+v_floor_f16_e64_dpp v255.l, -|v255.l| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_floor_f16_e64_dpp v255.l, -|v255.l| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xdb,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
+
+v_floor_f16_e64_dpp v5.h, v1.h mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX12: v_floor_f16_e64_dpp v5.h, v1.h op_sel:[1,1] mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x48,0xdb,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
+
+v_floor_f16_e64_dpp v5.l, v1.h mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_floor_f16_e64_dpp v5.l, v1.h op_sel:[1,0] mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x08,0xdb,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
+
+v_floor_f16_e64_dpp v255.h, -|v255.l| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_floor_f16_e64_dpp v255.h, -|v255.l| op_sel:[0,1] clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0xc1,0xdb,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
v_floor_f32_e64_dpp v5, v1 quad_perm:[3,2,1,0]
// GFX12: v_floor_f32_e64_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xa4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
@@ -1993,47 +2020,56 @@ v_frexp_mant_f32_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bou
v_frexp_mant_f32_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
// GFX12: v_frexp_mant_f32_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xc0,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
-v_log_f16_e64_dpp v5, v1 quad_perm:[3,2,1,0]
-// GFX12: v_log_f16_e64_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+v_log_f16_e64_dpp v5.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: v_log_f16_e64_dpp v5.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+
+v_log_f16_e64_dpp v5.l, v1.l quad_perm:[0,1,2,3]
+// GFX12: v_log_f16_e64_dpp v5.l, v1.l quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]
+
+v_log_f16_e64_dpp v5.l, v1.l row_mirror
+// GFX12: v_log_f16_e64_dpp v5.l, v1.l row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]
-v_log_f16_e64_dpp v5, v1 quad_perm:[0,1,2,3]
-// GFX12: v_log_f16_e64_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]
+v_log_f16_e64_dpp v5.l, v1.l row_half_mirror
+// GFX12: v_log_f16_e64_dpp v5.l, v1.l row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]
-v_log_f16_e64_dpp v5, v1 row_mirror
-// GFX12: v_log_f16_e64_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]
+v_log_f16_e64_dpp v5.l, v1.l row_shl:1
+// GFX12: v_log_f16_e64_dpp v5.l, v1.l row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]
-v_log_f16_e64_dpp v5, v1 row_half_mirror
-// GFX12: v_log_f16_e64_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]
+v_log_f16_e64_dpp v5.l, v1.l row_shl:15
+// GFX12: v_log_f16_e64_dpp v5.l, v1.l row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]
-v_log_f16_e64_dpp v5, v1 row_shl:1
-// GFX12: v_log_f16_e64_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]
+v_log_f16_e64_dpp v5.l, v1.l row_shr:1
+// GFX12: v_log_f16_e64_dpp v5.l, v1.l row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]
-v_log_f16_e64_dpp v5, v1 row_shl:15
-// GFX12: v_log_f16_e64_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]
+v_log_f16_e64_dpp v5.l, v1.l row_shr:15
+// GFX12: v_log_f16_e64_dpp v5.l, v1.l row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]
-v_log_f16_e64_dpp v5, v1 row_shr:1
-// GFX12: v_log_f16_e64_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]
+v_log_f16_e64_dpp v5.l, v1.l row_ror:1
+// GFX12: v_log_f16_e64_dpp v5.l, v1.l row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]
-v_log_f16_e64_dpp v5, v1 row_shr:15
-// GFX12: v_log_f16_e64_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]
+v_log_f16_e64_dpp v5.l, v1.l row_ror:15
+// GFX12: v_log_f16_e64_dpp v5.l, v1.l row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]
-v_log_f16_e64_dpp v5, v1 row_ror:1
-// GFX12: v_log_f16_e64_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]
+v_log_f16_e64_dpp v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf
+// GFX12: v_log_f16_e64_dpp v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]
-v_log_f16_e64_dpp v5, v1 row_ror:15
-// GFX12: v_log_f16_e64_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]
+v_log_f16_e64_dpp v5.l, v1.l mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX12: v_log_f16_e64_dpp v5.l, v1.l mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
-v_log_f16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
-// GFX12: v_log_f16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]
+v_log_f16_e64_dpp v5.l, v1.l mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_log_f16_e64_dpp v5.l, v1.l mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
-v_log_f16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
-// GFX12: v_log_f16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
+v_log_f16_e64_dpp v255.l, -|v255.l| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_log_f16_e64_dpp v255.l, -|v255.l| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xd7,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
-v_log_f16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
-// GFX12: v_log_f16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
+v_log_f16_e64_dpp v5.h, v1.h mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX12: v_log_f16_e64_dpp v5.h, v1.h op_sel:[1,1] mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x48,0xd7,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
-v_log_f16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
-// GFX12: v_log_f16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xd7,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
+v_log_f16_e64_dpp v5.l, v1.h mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_log_f16_e64_dpp v5.l, v1.h op_sel:[1,0] mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x08,0xd7,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
+
+v_log_f16_e64_dpp v255.h, -|v255.l| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_log_f16_e64_dpp v255.h, -|v255.l| op_sel:[0,1] clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0xc1,0xd7,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
v_log_f32_e64_dpp v5, v1 quad_perm:[3,2,1,0]
// GFX12: v_log_f32_e64_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xa7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
@@ -2254,47 +2290,56 @@ v_not_b32_e64_dpp v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:
v_not_b32_e64_dpp v255, v255 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
// GFX12: v_not_b32_e64_dpp v255, v255 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x00,0xb7,0xd5,0xfa,0x00,0x00,0x00,0xff,0x6f,0x05,0x30]
-v_rcp_f16_e64_dpp v5, v1 quad_perm:[3,2,1,0]
-// GFX12: v_rcp_f16_e64_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+v_rcp_f16_e64_dpp v5.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: v_rcp_f16_e64_dpp v5.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+
+v_rcp_f16_e64_dpp v5.l, v1.l quad_perm:[0,1,2,3]
+// GFX12: v_rcp_f16_e64_dpp v5.l, v1.l quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]
+
+v_rcp_f16_e64_dpp v5.l, v1.l row_mirror
+// GFX12: v_rcp_f16_e64_dpp v5.l, v1.l row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]
-v_rcp_f16_e64_dpp v5, v1 quad_perm:[0,1,2,3]
-// GFX12: v_rcp_f16_e64_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]
+v_rcp_f16_e64_dpp v5.l, v1.l row_half_mirror
+// GFX12: v_rcp_f16_e64_dpp v5.l, v1.l row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]
-v_rcp_f16_e64_dpp v5, v1 row_mirror
-// GFX12: v_rcp_f16_e64_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]
+v_rcp_f16_e64_dpp v5.l, v1.l row_shl:1
+// GFX12: v_rcp_f16_e64_dpp v5.l, v1.l row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]
-v_rcp_f16_e64_dpp v5, v1 row_half_mirror
-// GFX12: v_rcp_f16_e64_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]
+v_rcp_f16_e64_dpp v5.l, v1.l row_shl:15
+// GFX12: v_rcp_f16_e64_dpp v5.l, v1.l row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]
-v_rcp_f16_e64_dpp v5, v1 row_shl:1
-// GFX12: v_rcp_f16_e64_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]
+v_rcp_f16_e64_dpp v5.l, v1.l row_shr:1
+// GFX12: v_rcp_f16_e64_dpp v5.l, v1.l row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]
-v_rcp_f16_e64_dpp v5, v1 row_shl:15
-// GFX12: v_rcp_f16_e64_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]
+v_rcp_f16_e64_dpp v5.l, v1.l row_shr:15
+// GFX12: v_rcp_f16_e64_dpp v5.l, v1.l row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]
-v_rcp_f16_e64_dpp v5, v1 row_shr:1
-// GFX12: v_rcp_f16_e64_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]
+v_rcp_f16_e64_dpp v5.l, v1.l row_ror:1
+// GFX12: v_rcp_f16_e64_dpp v5.l, v1.l row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]
-v_rcp_f16_e64_dpp v5, v1 row_shr:15
-// GFX12: v_rcp_f16_e64_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]
+v_rcp_f16_e64_dpp v5.l, v1.l row_ror:15
+// GFX12: v_rcp_f16_e64_dpp v5.l, v1.l row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]
-v_rcp_f16_e64_dpp v5, v1 row_ror:1
-// GFX12: v_rcp_f16_e64_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]
+v_rcp_f16_e64_dpp v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf
+// GFX12: v_rcp_f16_e64_dpp v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]
-v_rcp_f16_e64_dpp v5, v1 row_ror:15
-// GFX12: v_rcp_f16_e64_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]
+v_rcp_f16_e64_dpp v5.l, v1.l mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX12: v_rcp_f16_e64_dpp v5.l, v1.l mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
-v_rcp_f16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
-// GFX12: v_rcp_f16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]
+v_rcp_f16_e64_dpp v5.l, v1.l mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_rcp_f16_e64_dpp v5.l, v1.l mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
-v_rcp_f16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
-// GFX12: v_rcp_f16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
+v_rcp_f16_e64_dpp v255.l, -|v255.l| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_rcp_f16_e64_dpp v255.l, -|v255.l| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xd4,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
-v_rcp_f16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
-// GFX12: v_rcp_f16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
+v_rcp_f16_e64_dpp v5.h, v1.h mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX12: v_rcp_f16_e64_dpp v5.h, v1.h op_sel:[1,1] mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x48,0xd4,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
-v_rcp_f16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
-// GFX12: v_rcp_f16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xd4,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
+v_rcp_f16_e64_dpp v5.l, v1.h mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_rcp_f16_e64_dpp v5.l, v1.h op_sel:[1,0] mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x08,0xd4,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
+
+v_rcp_f16_e64_dpp v255.h, -|v255.l| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_rcp_f16_e64_dpp v255.h, -|v255.l| op_sel:[0,1] clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0xc1,0xd4,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
v_rcp_f32_e64_dpp v5, v1 quad_perm:[3,2,1,0]
// GFX12: v_rcp_f32_e64_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xaa,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
@@ -2473,47 +2518,56 @@ v_rndne_f32_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ct
v_rndne_f32_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
// GFX12: v_rndne_f32_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xa3,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
-v_rsq_f16_e64_dpp v5, v1 quad_perm:[3,2,1,0]
-// GFX12: v_rsq_f16_e64_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+v_rsq_f16_e64_dpp v5.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: v_rsq_f16_e64_dpp v5.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+
+v_rsq_f16_e64_dpp v5.l, v1.l quad_perm:[0,1,2,3]
+// GFX12: v_rsq_f16_e64_dpp v5.l, v1.l quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]
-v_rsq_f16_e64_dpp v5, v1 quad_perm:[0,1,2,3]
-// GFX12: v_rsq_f16_e64_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]
+v_rsq_f16_e64_dpp v5.l, v1.l row_mirror
+// GFX12: v_rsq_f16_e64_dpp v5.l, v1.l row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]
-v_rsq_f16_e64_dpp v5, v1 row_mirror
-// GFX12: v_rsq_f16_e64_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]
+v_rsq_f16_e64_dpp v5.l, v1.l row_half_mirror
+// GFX12: v_rsq_f16_e64_dpp v5.l, v1.l row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]
-v_rsq_f16_e64_dpp v5, v1 row_half_mirror
-// GFX12: v_rsq_f16_e64_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]
+v_rsq_f16_e64_dpp v5.l, v1.l row_shl:1
+// GFX12: v_rsq_f16_e64_dpp v5.l, v1.l row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]
-v_rsq_f16_e64_dpp v5, v1 row_shl:1
-// GFX12: v_rsq_f16_e64_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]
+v_rsq_f16_e64_dpp v5.l, v1.l row_shl:15
+// GFX12: v_rsq_f16_e64_dpp v5.l, v1.l row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]
-v_rsq_f16_e64_dpp v5, v1 row_shl:15
-// GFX12: v_rsq_f16_e64_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]
+v_rsq_f16_e64_dpp v5.l, v1.l row_shr:1
+// GFX12: v_rsq_f16_e64_dpp v5.l, v1.l row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]
-v_rsq_f16_e64_dpp v5, v1 row_shr:1
-// GFX12: v_rsq_f16_e64_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]
+v_rsq_f16_e64_dpp v5.l, v1.l row_shr:15
+// GFX12: v_rsq_f16_e64_dpp v5.l, v1.l row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]
-v_rsq_f16_e64_dpp v5, v1 row_shr:15
-// GFX12: v_rsq_f16_e64_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]
+v_rsq_f16_e64_dpp v5.l, v1.l row_ror:1
+// GFX12: v_rsq_f16_e64_dpp v5.l, v1.l row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]
-v_rsq_f16_e64_dpp v5, v1 row_ror:1
-// GFX12: v_rsq_f16_e64_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]
+v_rsq_f16_e64_dpp v5.l, v1.l row_ror:15
+// GFX12: v_rsq_f16_e64_dpp v5.l, v1.l row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]
-v_rsq_f16_e64_dpp v5, v1 row_ror:15
-// GFX12: v_rsq_f16_e64_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]
+v_rsq_f16_e64_dpp v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf
+// GFX12: v_rsq_f16_e64_dpp v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]
-v_rsq_f16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
-// GFX12: v_rsq_f16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]
+v_rsq_f16_e64_dpp v5.l, v1.l mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX12: v_rsq_f16_e64_dpp v5.l, v1.l mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
-v_rsq_f16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
-// GFX12: v_rsq_f16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
+v_rsq_f16_e64_dpp v5.l, v1.l mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_rsq_f16_e64_dpp v5.l, v1.l mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
-v_rsq_f16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
-// GFX12: v_rsq_f16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
+v_rsq_f16_e64_dpp v255.l, -|v255.l| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_rsq_f16_e64_dpp v255.l, -|v255.l| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xd6,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
-v_rsq_f16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
-// GFX12: v_rsq_f16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xd6,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
+v_rsq_f16_e64_dpp v5.h, v1.h mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX12: v_rsq_f16_e64_dpp v5.h, v1.h op_sel:[1,1] mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x48,0xd6,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
+
+v_rsq_f16_e64_dpp v5.l, v1.h mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_rsq_f16_e64_dpp v5.l, v1.h op_sel:[1,0] mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x08,0xd6,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
+
+v_rsq_f16_e64_dpp v255.h, -|v255.l| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_rsq_f16_e64_dpp v255.h, -|v255.l| op_sel:[0,1] clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0xc1,0xd6,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
v_rsq_f32_e64_dpp v5, v1 quad_perm:[3,2,1,0]
// GFX12: v_rsq_f32_e64_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xae,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
@@ -2695,47 +2749,56 @@ v_sin_f32_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl
v_sin_f32_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
// GFX12: v_sin_f32_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xb5,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
-v_sqrt_f16_e64_dpp v5, v1 quad_perm:[3,2,1,0]
-// GFX12: v_sqrt_f16_e64_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+v_sqrt_f16_e64_dpp v5.l, v1.l quad_perm:[3,2,1,0]
+// GFX12: v_sqrt_f16_e64_dpp v5.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+
+v_sqrt_f16_e64_dpp v5.l, v1.l quad_perm:[0,1,2,3]
+// GFX12: v_sqrt_f16_e64_dpp v5.l, v1.l quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]
+
+v_sqrt_f16_e64_dpp v5.l, v1.l row_mirror
+// GFX12: v_sqrt_f16_e64_dpp v5.l, v1.l row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]
+
+v_sqrt_f16_e64_dpp v5.l, v1.l row_half_mirror
+// GFX12: v_sqrt_f16_e64_dpp v5.l, v1.l row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]
-v_sqrt_f16_e64_dpp v5, v1 quad_perm:[0,1,2,3]
-// GFX12: v_sqrt_f16_e64_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]
+v_sqrt_f16_e64_dpp v5.l, v1.l row_shl:1
+// GFX12: v_sqrt_f16_e64_dpp v5.l, v1.l row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]
-v_sqrt_f16_e64_dpp v5, v1 row_mirror
-// GFX12: v_sqrt_f16_e64_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]
+v_sqrt_f16_e64_dpp v5.l, v1.l row_shl:15
+// GFX12: v_sqrt_f16_e64_dpp v5.l, v1.l row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]
-v_sqrt_f16_e64_dpp v5, v1 row_half_mirror
-// GFX12: v_sqrt_f16_e64_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]
+v_sqrt_f16_e64_dpp v5.l, v1.l row_shr:1
+// GFX12: v_sqrt_f16_e64_dpp v5.l, v1.l row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]
-v_sqrt_f16_e64_dpp v5, v1 row_shl:1
-// GFX12: v_sqrt_f16_e64_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]
+v_sqrt_f16_e64_dpp v5.l, v1.l row_shr:15
+// GFX12: v_sqrt_f16_e64_dpp v5.l, v1.l row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]
-v_sqrt_f16_e64_dpp v5, v1 row_shl:15
-// GFX12: v_sqrt_f16_e64_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]
+v_sqrt_f16_e64_dpp v5.l, v1.l row_ror:1
+// GFX12: v_sqrt_f16_e64_dpp v5.l, v1.l row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]
-v_sqrt_f16_e64_dpp v5, v1 row_shr:1
-// GFX12: v_sqrt_f16_e64_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]
+v_sqrt_f16_e64_dpp v5.l, v1.l row_ror:15
+// GFX12: v_sqrt_f16_e64_dpp v5.l, v1.l row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]
-v_sqrt_f16_e64_dpp v5, v1 row_shr:15
-// GFX12: v_sqrt_f16_e64_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]
+v_sqrt_f16_e64_dpp v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf
+// GFX12: v_sqrt_f16_e64_dpp v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]
-v_sqrt_f16_e64_dpp v5, v1 row_ror:1
-// GFX12: v_sqrt_f16_e64_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]
+v_sqrt_f16_e64_dpp v5.l, v1.l mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX12: v_sqrt_f16_e64_dpp v5.l, v1.l mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
-v_sqrt_f16_e64_dpp v5, v1 row_ror:15
-// GFX12: v_sqrt_f16_e64_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]
+v_sqrt_f16_e64_dpp v5.l, v1.l mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_sqrt_f16_e64_dpp v5.l, v1.l mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
-v_sqrt_f16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
-// GFX12: v_sqrt_f16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]
+v_sqrt_f16_e64_dpp v255.l, -|v255.l| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_sqrt_f16_e64_dpp v255.l, -|v255.l| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xd5,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
-v_sqrt_f16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
-// GFX12: v_sqrt_f16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
+v_sqrt_f16_e64_dpp v5.h, v1.h mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX12: v_sqrt_f16_e64_dpp v5.h, v1.h op_sel:[1,1] mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x48,0xd5,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
-v_sqrt_f16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
-// GFX12: v_sqrt_f16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
+v_sqrt_f16_e64_dpp v5.l, v1.h mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX12: v_sqrt_f16_e64_dpp v5.l, v1.h op_sel:[1,0] mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x08,0xd5,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
-v_sqrt_f16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
-// GFX12: v_sqrt_f16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xd5,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
+v_sqrt_f16_e64_dpp v255.h, -|v255.l| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX12: v_sqrt_f16_e64_dpp v255.h, -|v255.l| op_sel:[0,1] clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0xc1,0xd5,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
v_sqrt_f32_e64_dpp v5, v1 quad_perm:[3,2,1,0]
// GFX12: v_sqrt_f32_e64_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xb3,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
diff --git a/llvm/test/MC/AMDGPU/gfx12_asm_vop3_from_vop1_dpp8.s b/llvm/test/MC/AMDGPU/gfx12_asm_vop3_from_vop1_dpp8.s
index 1b1a91fbf1c8c..d2fc865cc8b4c 100644
--- a/llvm/test/MC/AMDGPU/gfx12_asm_vop3_from_vop1_dpp8.s
+++ b/llvm/test/MC/AMDGPU/gfx12_asm_vop3_from_vop1_dpp8.s
@@ -10,17 +10,26 @@ v_bfrev_b32_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1
v_bfrev_b32_e64_dpp v255, v255 dpp8:[0,0,0,0,0,0,0,0] fi:0
// GFX12: v_bfrev_b32_e64_dpp v255, v255 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x00,0xb8,0xd5,0xe9,0x00,0x00,0x00,0xff,0x00,0x00,0x00]
-v_ceil_f16_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_ceil_f16_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xdc,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+v_ceil_f16_e64_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_ceil_f16_e64_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xdc,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
-v_ceil_f16_e64_dpp v5, v1 mul:2 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_ceil_f16_e64_dpp v5, v1 mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xdc,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
+v_ceil_f16_e64_dpp v5.l, v1.l mul:2 dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_ceil_f16_e64_dpp v5.l, v1.l mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xdc,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
-v_ceil_f16_e64_dpp v5, v1 mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
-// GFX12: v_ceil_f16_e64_dpp v5, v1 mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x00,0xdc,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
+v_ceil_f16_e64_dpp v5.l, v1.l mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_ceil_f16_e64_dpp v5.l, v1.l mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x00,0xdc,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
-v_ceil_f16_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
-// GFX12: v_ceil_f16_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xdc,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
+v_ceil_f16_e64_dpp v255.l, -|v255.l| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_ceil_f16_e64_dpp v255.l, -|v255.l| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xdc,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
+
+v_ceil_f16_e64_dpp v5.h, v1.h mul:2 dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_ceil_f16_e64_dpp v5.h, v1.h op_sel:[1,1] mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xdc,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
+
+v_ceil_f16_e64_dpp v5.l, v1.h mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_ceil_f16_e64_dpp v5.l, v1.h op_sel:[1,0] mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x08,0xdc,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
+
+v_ceil_f16_e64_dpp v255.h, -|v255.l| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_ceil_f16_e64_dpp v255.h, -|v255.l| op_sel:[0,1] clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0xc1,0xdc,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
v_ceil_f32_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0]
// GFX12: v_ceil_f32_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xa2,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
@@ -442,17 +451,26 @@ v_cvt_u32_u16_e64_dpp v255, v255.l dpp8:[0,0,0,0,0,0,0,0] fi:0
v_cvt_u32_u16_e64_dpp v255, v255.h dpp8:[0,0,0,0,0,0,0,0] fi:0
// GFX12: v_cvt_u32_u16_e64_dpp v255, v255.h op_sel:[1,0] dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x08,0xeb,0xd5,0xe9,0x00,0x00,0x00,0xff,0x00,0x00,0x00]
-v_exp_f16_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_exp_f16_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd8,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+v_exp_f16_e64_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_exp_f16_e64_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd8,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+
+v_exp_f16_e64_dpp v5.l, v1.l mul:2 dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_exp_f16_e64_dpp v5.l, v1.l mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd8,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
+
+v_exp_f16_e64_dpp v5.l, v1.l mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_exp_f16_e64_dpp v5.l, v1.l mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x00,0xd8,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
+
+v_exp_f16_e64_dpp v255.l, -|v255.l| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_exp_f16_e64_dpp v255.l, -|v255.l| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xd8,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
-v_exp_f16_e64_dpp v5, v1 mul:2 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_exp_f16_e64_dpp v5, v1 mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd8,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
+v_exp_f16_e64_dpp v5.h, v1.h mul:2 dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_exp_f16_e64_dpp v5.h, v1.h op_sel:[1,1] mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xd8,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
-v_exp_f16_e64_dpp v5, v1 mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
-// GFX12: v_exp_f16_e64_dpp v5, v1 mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x00,0xd8,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
+v_exp_f16_e64_dpp v5.l, v1.h mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_exp_f16_e64_dpp v5.l, v1.h op_sel:[1,0] mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x08,0xd8,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
-v_exp_f16_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
-// GFX12: v_exp_f16_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xd8,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
+v_exp_f16_e64_dpp v255.h, -|v255.l| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_exp_f16_e64_dpp v255.h, -|v255.l| op_sel:[0,1] clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0xc1,0xd8,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
v_exp_f32_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0]
// GFX12: v_exp_f32_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xa5,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
@@ -493,17 +511,26 @@ v_ffbl_b32_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1
v_ffbl_b32_e64_dpp v255, v255 dpp8:[0,0,0,0,0,0,0,0] fi:0
// GFX12: v_ctz_i32_b32_e64_dpp v255, v255 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x00,0xba,0xd5,0xe9,0x00,0x00,0x00,0xff,0x00,0x00,0x00]
-v_floor_f16_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_floor_f16_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xdb,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+v_floor_f16_e64_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_floor_f16_e64_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xdb,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
-v_floor_f16_e64_dpp v5, v1 mul:2 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_floor_f16_e64_dpp v5, v1 mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xdb,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
+v_floor_f16_e64_dpp v5.l, v1.l mul:2 dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_floor_f16_e64_dpp v5.l, v1.l mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xdb,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
-v_floor_f16_e64_dpp v5, v1 mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
-// GFX12: v_floor_f16_e64_dpp v5, v1 mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x00,0xdb,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
+v_floor_f16_e64_dpp v5.l, v1.l mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_floor_f16_e64_dpp v5.l, v1.l mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x00,0xdb,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
-v_floor_f16_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
-// GFX12: v_floor_f16_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xdb,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
+v_floor_f16_e64_dpp v255.l, -|v255.l| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_floor_f16_e64_dpp v255.l, -|v255.l| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xdb,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
+
+v_floor_f16_e64_dpp v5.h, v1.h mul:2 dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_floor_f16_e64_dpp v5.h, v1.h op_sel:[1,1] mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xdb,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
+
+v_floor_f16_e64_dpp v5.l, v1.h mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_floor_f16_e64_dpp v5.l, v1.h op_sel:[1,0] mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x08,0xdb,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
+
+v_floor_f16_e64_dpp v255.h, -|v255.l| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_floor_f16_e64_dpp v255.h, -|v255.l| op_sel:[0,1] clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0xc1,0xdb,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
v_floor_f32_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0]
// GFX12: v_floor_f32_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xa4,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
@@ -610,17 +637,26 @@ v_frexp_mant_f32_e64_dpp v5, v1 mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
v_frexp_mant_f32_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
// GFX12: v_frexp_mant_f32_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xc0,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
-v_log_f16_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_log_f16_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd7,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+v_log_f16_e64_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_log_f16_e64_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd7,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+
+v_log_f16_e64_dpp v5.l, v1.l mul:2 dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_log_f16_e64_dpp v5.l, v1.l mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd7,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
+
+v_log_f16_e64_dpp v5.l, v1.l mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_log_f16_e64_dpp v5.l, v1.l mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x00,0xd7,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
-v_log_f16_e64_dpp v5, v1 mul:2 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_log_f16_e64_dpp v5, v1 mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd7,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
+v_log_f16_e64_dpp v255.l, -|v255.l| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_log_f16_e64_dpp v255.l, -|v255.l| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xd7,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
-v_log_f16_e64_dpp v5, v1 mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
-// GFX12: v_log_f16_e64_dpp v5, v1 mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x00,0xd7,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
+v_log_f16_e64_dpp v5.h, v1.h mul:2 dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_log_f16_e64_dpp v5.h, v1.h op_sel:[1,1] mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xd7,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
-v_log_f16_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
-// GFX12: v_log_f16_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xd7,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
+v_log_f16_e64_dpp v5.l, v1.h mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_log_f16_e64_dpp v5.l, v1.h op_sel:[1,0] mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x08,0xd7,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
+
+v_log_f16_e64_dpp v255.h, -|v255.l| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_log_f16_e64_dpp v255.h, -|v255.l| op_sel:[0,1] clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0xc1,0xd7,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
v_log_f32_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0]
// GFX12: v_log_f32_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xa7,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
@@ -679,17 +715,26 @@ v_not_b32_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1
v_not_b32_e64_dpp v255, v255 dpp8:[0,0,0,0,0,0,0,0] fi:0
// GFX12: v_not_b32_e64_dpp v255, v255 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x00,0xb7,0xd5,0xe9,0x00,0x00,0x00,0xff,0x00,0x00,0x00]
-v_rcp_f16_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_rcp_f16_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd4,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+v_rcp_f16_e64_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_rcp_f16_e64_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd4,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+
+v_rcp_f16_e64_dpp v5.l, v1.l mul:2 dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_rcp_f16_e64_dpp v5.l, v1.l mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd4,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
+
+v_rcp_f16_e64_dpp v5.l, v1.l mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_rcp_f16_e64_dpp v5.l, v1.l mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x00,0xd4,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
-v_rcp_f16_e64_dpp v5, v1 mul:2 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_rcp_f16_e64_dpp v5, v1 mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd4,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
+v_rcp_f16_e64_dpp v255.l, -|v255.l| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_rcp_f16_e64_dpp v255.l, -|v255.l| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xd4,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
-v_rcp_f16_e64_dpp v5, v1 mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
-// GFX12: v_rcp_f16_e64_dpp v5, v1 mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x00,0xd4,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
+v_rcp_f16_e64_dpp v5.h, v1.h mul:2 dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_rcp_f16_e64_dpp v5.h, v1.h op_sel:[1,1] mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xd4,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
-v_rcp_f16_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
-// GFX12: v_rcp_f16_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xd4,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
+v_rcp_f16_e64_dpp v5.l, v1.h mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_rcp_f16_e64_dpp v5.l, v1.h op_sel:[1,0] mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x08,0xd4,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
+
+v_rcp_f16_e64_dpp v255.h, -|v255.l| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_rcp_f16_e64_dpp v255.h, -|v255.l| op_sel:[0,1] clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0xc1,0xd4,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
v_rcp_f32_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0]
// GFX12: v_rcp_f32_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xaa,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
@@ -748,17 +793,26 @@ v_rndne_f32_e64_dpp v5, v1 mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
v_rndne_f32_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
// GFX12: v_rndne_f32_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xa3,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
-v_rsq_f16_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_rsq_f16_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd6,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+v_rsq_f16_e64_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_rsq_f16_e64_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd6,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+
+v_rsq_f16_e64_dpp v5.l, v1.l mul:2 dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_rsq_f16_e64_dpp v5.l, v1.l mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd6,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
-v_rsq_f16_e64_dpp v5, v1 mul:2 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_rsq_f16_e64_dpp v5, v1 mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd6,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
+v_rsq_f16_e64_dpp v5.l, v1.l mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_rsq_f16_e64_dpp v5.l, v1.l mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x00,0xd6,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
-v_rsq_f16_e64_dpp v5, v1 mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
-// GFX12: v_rsq_f16_e64_dpp v5, v1 mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x00,0xd6,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
+v_rsq_f16_e64_dpp v255.l, -|v255.l| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_rsq_f16_e64_dpp v255.l, -|v255.l| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xd6,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
-v_rsq_f16_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
-// GFX12: v_rsq_f16_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xd6,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
+v_rsq_f16_e64_dpp v5.h, v1.h mul:2 dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_rsq_f16_e64_dpp v5.h, v1.h op_sel:[1,1] mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xd6,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
+
+v_rsq_f16_e64_dpp v5.l, v1.h mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_rsq_f16_e64_dpp v5.l, v1.h op_sel:[1,0] mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x08,0xd6,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
+
+v_rsq_f16_e64_dpp v255.h, -|v255.l| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_rsq_f16_e64_dpp v255.h, -|v255.l| op_sel:[0,1] clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0xc1,0xd6,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
v_rsq_f32_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0]
// GFX12: v_rsq_f32_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xae,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
@@ -817,17 +871,26 @@ v_sin_f32_e64_dpp v5, v1 mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
v_sin_f32_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
// GFX12: v_sin_f32_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xb5,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
-v_sqrt_f16_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_sqrt_f16_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd5,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+v_sqrt_f16_e64_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_sqrt_f16_e64_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd5,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+
+v_sqrt_f16_e64_dpp v5.l, v1.l mul:2 dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_sqrt_f16_e64_dpp v5.l, v1.l mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd5,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
+
+v_sqrt_f16_e64_dpp v5.l, v1.l mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_sqrt_f16_e64_dpp v5.l, v1.l mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x00,0xd5,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
+
+v_sqrt_f16_e64_dpp v255.l, -|v255.l| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_sqrt_f16_e64_dpp v255.l, -|v255.l| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xd5,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
-v_sqrt_f16_e64_dpp v5, v1 mul:2 dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: v_sqrt_f16_e64_dpp v5, v1 mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xd5,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
+v_sqrt_f16_e64_dpp v5.h, v1.h mul:2 dpp8:[7,6,5,4,3,2,1,0]
+// GFX12: v_sqrt_f16_e64_dpp v5.h, v1.h op_sel:[1,1] mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xd5,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
-v_sqrt_f16_e64_dpp v5, v1 mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
-// GFX12: v_sqrt_f16_e64_dpp v5, v1 mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x00,0xd5,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
+v_sqrt_f16_e64_dpp v5.l, v1.h mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX12: v_sqrt_f16_e64_dpp v5.l, v1.h op_sel:[1,0] mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x08,0xd5,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
-v_sqrt_f16_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
-// GFX12: v_sqrt_f16_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xd5,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
+v_sqrt_f16_e64_dpp v255.h, -|v255.l| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX12: v_sqrt_f16_e64_dpp v255.h, -|v255.l| op_sel:[0,1] clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0xc1,0xd5,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
v_sqrt_f32_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0]
// GFX12: v_sqrt_f32_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xb3,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
More information about the llvm-commits
mailing list