[AMDGPU] Add gfx1250 runlines to vop3 dpp tests. NFC. (#145089)

dpp8 disasm test does not work yet.

Author: rampitec

llvm/test/MC/AMDGPU/gfx12_asm_vop3_dpp16.s

@@ -1,8 +1,10 @@
 // NOTE: Assertions have been autogenerated by utils/update_mc_test_checks.py UTC_ARGS: --unique --version 5
-// RUN: not llvm-mc -triple=amdgcn -mcpu=gfx1200 -mattr=+wavefrontsize32,+real-true16 -show-encoding %s | FileCheck --check-prefixes=GFX12,W32 %s
-// RUN: not llvm-mc -triple=amdgcn -mcpu=gfx1200 -mattr=+wavefrontsize64,+real-true16 -show-encoding %s | FileCheck --check-prefixes=GFX12,W64 %s
+// RUN: not llvm-mc -triple=amdgcn -mcpu=gfx1200 -mattr=+wavefrontsize32,+real-true16 -show-encoding %s | FileCheck --check-prefixes=GFX12,GFX1200,W32 %s
+// RUN: not llvm-mc -triple=amdgcn -mcpu=gfx1200 -mattr=+wavefrontsize64,+real-true16 -show-encoding %s | FileCheck --check-prefixes=GFX12,GFX1200,W64 %s
 // RUN: not llvm-mc -triple=amdgcn -mcpu=gfx1200 -mattr=+wavefrontsize32,+real-true16 -filetype=null %s 2>&1 | FileCheck --check-prefixes=GFX12-ERR,W32-ERR --implicit-check-not=error: %s
 // RUN: not llvm-mc -triple=amdgcn -mcpu=gfx1200 -mattr=+wavefrontsize64,+real-true16 -filetype=null %s 2>&1 | FileCheck --check-prefixes=GFX12-ERR,W64-ERR --implicit-check-not=error: %s
+// RUN: not llvm-mc -triple=amdgcn -mcpu=gfx1250 -mattr=+wavefrontsize32,+real-true16 -show-encoding %s | FileCheck --check-prefixes=GFX12,W32 %s
+// RUN: not llvm-mc -triple=amdgcn -mcpu=gfx1250 -mattr=+wavefrontsize64,+real-true16 -show-encoding %s | FileCheck --check-prefixes=GFX12,W64 %s
 
 v_add3_u32_e64_dpp v5, v1, v2, v3 quad_perm:[3,2,1,0]
 // GFX12: v_add3_u32_e64_dpp v5, v1, v2, v3 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x55,0xd6,0xfa,0x04,0x0e,0x04,0x01,0x1b,0x00,0xff]
@@ -1237,37 +1239,37 @@ v_cvt_pk_bf8_f32_e64_dpp v1.l, -v2, |v3| quad_perm:[3,2,1,0] row_mask:0xe bank_m
 // GFX12: v_cvt_pk_bf8_f32_e64_dpp v1.l, -v2, |v3| quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0xd fi:1 ; encoding: [0x01,0x02,0x6a,0xd7,0xfa,0x06,0x02,0x20,0x02,0x1b,0x04,0xed]
 
 v_cvt_pk_fp8_f32_e64_dpp v1.l, -v2, |v3| quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0xd
-// GFX12: v_cvt_pk_fp8_f32_e64_dpp v1.l, -v2, |v3| quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0xd ; encoding: [0x01,0x02,0x69,0xd7,0xfa,0x06,0x02,0x20,0x02,0x1b,0x00,0xed]
+// GFX1200: v_cvt_pk_fp8_f32_e64_dpp v1.l, -v2, |v3| quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0xd ; encoding: [0x01,0x02,0x69,0xd7,0xfa,0x06,0x02,0x20,0x02,0x1b,0x00,0xed]
 
 v_cvt_pk_fp8_f32_e64_dpp v255.h, -v2, |v3| quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0xd
-// GFX12: v_cvt_pk_fp8_f32_e64_dpp v255.h, -v2, |v3| op_sel:[0,0,1] quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0xd ; encoding: [0xff,0x42,0x69,0xd7,0xfa,0x06,0x02,0x20,0x02,0x1b,0x00,0xed]
+// GFX1200: v_cvt_pk_fp8_f32_e64_dpp v255.h, -v2, |v3| op_sel:[0,0,1] quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0xd ; encoding: [0xff,0x42,0x69,0xd7,0xfa,0x06,0x02,0x20,0x02,0x1b,0x00,0xed]
 
 v_cvt_pk_fp8_f32_e64_dpp v1.l, -v2, |v3| quad_perm:[0,1,2,3]
-// GFX12: v_cvt_pk_fp8_f32_e64_dpp v1.l, -v2, |v3| quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x01,0x02,0x69,0xd7,0xfa,0x06,0x02,0x20,0x02,0xe4,0x00,0xff]
+// GFX1200: v_cvt_pk_fp8_f32_e64_dpp v1.l, -v2, |v3| quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x01,0x02,0x69,0xd7,0xfa,0x06,0x02,0x20,0x02,0xe4,0x00,0xff]
 
 v_cvt_pk_fp8_f32_e64_dpp v6.l, -v2, |v3| quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0xd
-// GFX12: v_cvt_pk_fp8_f32_e64_dpp v6.l, -v2, |v3| quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0xd ; encoding: [0x06,0x02,0x69,0xd7,0xfa,0x06,0x02,0x20,0x02,0x1b,0x00,0xed]
+// GFX1200: v_cvt_pk_fp8_f32_e64_dpp v6.l, -v2, |v3| quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0xd ; encoding: [0x06,0x02,0x69,0xd7,0xfa,0x06,0x02,0x20,0x02,0x1b,0x00,0xed]
 
 v_cvt_pk_fp8_f32_e64_dpp v1.l, -v6, |v3| quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0xd
-// GFX12: v_cvt_pk_fp8_f32_e64_dpp v1.l, -v6, |v3| quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0xd ; encoding: [0x01,0x02,0x69,0xd7,0xfa,0x06,0x02,0x20,0x06,0x1b,0x00,0xed]
+// GFX1200: v_cvt_pk_fp8_f32_e64_dpp v1.l, -v6, |v3| quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0xd ; encoding: [0x01,0x02,0x69,0xd7,0xfa,0x06,0x02,0x20,0x06,0x1b,0x00,0xed]
 
 v_cvt_pk_fp8_f32_e64_dpp v1.l, -v2, |v255| quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0xd
-// GFX12: v_cvt_pk_fp8_f32_e64_dpp v1.l, -v2, |v255| quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0xd ; encoding: [0x01,0x02,0x69,0xd7,0xfa,0xfe,0x03,0x20,0x02,0x1b,0x00,0xed]
+// GFX1200: v_cvt_pk_fp8_f32_e64_dpp v1.l, -v2, |v255| quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0xd ; encoding: [0x01,0x02,0x69,0xd7,0xfa,0xfe,0x03,0x20,0x02,0x1b,0x00,0xed]
 
 v_cvt_pk_fp8_f32_e64_dpp v1.l, -v2, |v3| quad_perm:[0,2,1,3] row_mask:0xe bank_mask:0xd
-// GFX12: v_cvt_pk_fp8_f32_e64_dpp v1.l, -v2, |v3| quad_perm:[0,2,1,3] row_mask:0xe bank_mask:0xd ; encoding: [0x01,0x02,0x69,0xd7,0xfa,0x06,0x02,0x20,0x02,0xd8,0x00,0xed]
+// GFX1200: v_cvt_pk_fp8_f32_e64_dpp v1.l, -v2, |v3| quad_perm:[0,2,1,3] row_mask:0xe bank_mask:0xd ; encoding: [0x01,0x02,0x69,0xd7,0xfa,0x06,0x02,0x20,0x02,0xd8,0x00,0xed]
 
 v_cvt_pk_fp8_f32_e64_dpp v1.l, -v2, |v3| quad_perm:[3,2,1,0] row_mask:0x2 bank_mask:0xd
-// GFX12: v_cvt_pk_fp8_f32_e64_dpp v1.l, -v2, |v3| quad_perm:[3,2,1,0] row_mask:0x2 bank_mask:0xd ; encoding: [0x01,0x02,0x69,0xd7,0xfa,0x06,0x02,0x20,0x02,0x1b,0x00,0x2d]
+// GFX1200: v_cvt_pk_fp8_f32_e64_dpp v1.l, -v2, |v3| quad_perm:[3,2,1,0] row_mask:0x2 bank_mask:0xd ; encoding: [0x01,0x02,0x69,0xd7,0xfa,0x06,0x02,0x20,0x02,0x1b,0x00,0x2d]
 
 v_cvt_pk_fp8_f32_e64_dpp v1.l, -v2, |v3| quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0x5
-// GFX12: v_cvt_pk_fp8_f32_e64_dpp v1.l, -v2, |v3| quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0x5 ; encoding: [0x01,0x02,0x69,0xd7,0xfa,0x06,0x02,0x20,0x02,0x1b,0x00,0xe5]
+// GFX1200: v_cvt_pk_fp8_f32_e64_dpp v1.l, -v2, |v3| quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0x5 ; encoding: [0x01,0x02,0x69,0xd7,0xfa,0x06,0x02,0x20,0x02,0x1b,0x00,0xe5]
 
 v_cvt_pk_fp8_f32_e64_dpp v1.l, -v2, |v3| quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0xd fi:1
-// GFX12: v_cvt_pk_fp8_f32_e64_dpp v1.l, -v2, |v3| quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0xd fi:1 ; encoding: [0x01,0x02,0x69,0xd7,0xfa,0x06,0x02,0x20,0x02,0x1b,0x04,0xed]
+// GFX1200: v_cvt_pk_fp8_f32_e64_dpp v1.l, -v2, |v3| quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0xd fi:1 ; encoding: [0x01,0x02,0x69,0xd7,0xfa,0x06,0x02,0x20,0x02,0x1b,0x04,0xed]
 
 v_cvt_pk_fp8_f32_e64_dpp v255.h, -v2, |v3| quad_perm:[0,1,2,3]
-// GFX12: v_cvt_pk_fp8_f32_e64_dpp v255.h, -v2, |v3| op_sel:[0,0,1] quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x42,0x69,0xd7,0xfa,0x06,0x02,0x20,0x02,0xe4,0x00,0xff]
+// GFX1200: v_cvt_pk_fp8_f32_e64_dpp v255.h, -v2, |v3| op_sel:[0,0,1] quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xff,0x42,0x69,0xd7,0xfa,0x06,0x02,0x20,0x02,0xe4,0x00,0xff]
 
 v_cvt_sr_bf8_f32_e64_dpp v1, -v2, v3 quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0xd
 // GFX12: v_cvt_sr_bf8_f32_e64_dpp v1, -v2, v3 quad_perm:[3,2,1,0] row_mask:0xe bank_mask:0xd ; encoding: [0x01,0x00,0x6c,0xd7,0xfa,0x06,0x02,0x20,0x02,0x1b,0x00,0xed]
@@ -5592,70 +5594,100 @@ v_pack_b32_f16_e64_dpp v255, -|v255.l|, -|v255.h| op_sel:[0,1,0] row_xmask:15 ro
 // GFX12: v_pack_b32_f16_e64_dpp v255, -|v255.l|, -|v255.h| op_sel:[0,1,0] row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:1 fi:1 ; encoding: [0xff,0x13,0x11,0xd7,0xfa,0xfe,0x03,0x60,0xff,0x6f,0x0d,0x30]
 
 v_dot2_f16_f16_e64_dpp v0.l, v1, v2, v3.l quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1
-// GFX12: v_dot2_f16_f16_e64_dpp v0.l, v1, v2, v3.l quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1 ; encoding: [0x00,0x00,0x66,0xd6,0xfa,0x04,0x0e,0x04,0x01,0xe4,0x04,0x00]
+// GFX1200: v_dot2_f16_f16_e64_dpp v0.l, v1, v2, v3.l quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1 ; encoding: [0x00,0x00,0x66,0xd6,0xfa,0x04,0x0e,0x04,0x01,0xe4,0x04,0x00]
+
+v_dot2_f16_f16_e64_dpp v0, v1, v2, v3 op_sel:[1,1,0,0] quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1
+// GFX12-ERR: :[[@LINE-1]]:39: error: invalid op_sel operand
+
+v_dot2_f16_f16_e64_dpp v0, s1, v2, v3 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1
+// GFX12-ERR: :[[@LINE-1]]:28: error: invalid operand for instruction
+
+v_dot2_f16_f16_e64_dpp v0, v1, s2, v3 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1
+// GFX1200: v_dot2_f16_f16_e64_dpp v0, v1, s2, v3 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1 ; encoding: [0x00,0x00,0x66,0xd6,0xfa,0x04,0x0c,0x04,0x01,0xe4,0x04,0x00]
+
+v_dot2_f16_f16_e64_dpp v0, v1, v2, v3 op_sel:[0,0,1,1] quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1
+// GFX1200: v_dot2_f16_f16_e64_dpp v0, v1, v2, v3 op_sel:[0,0,1,1] quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1 ; encoding: [0x00,0x60,0x66,0xd6,0xfa,0x04,0x0e,0x04,0x01,0xe4,0x04,0x00]
+
+v_dot2_f16_f16_e64_dpp v0, |v1|, -v2, -|s3| op_sel:[0,0,1,1] quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1
+// GFX1200: v_dot2_f16_f16_e64_dpp v0, |v1|, -v2, -|s3| op_sel:[0,0,1,1] quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1 ; encoding: [0x00,0x65,0x66,0xd6,0xfa,0x04,0x0e,0xc0,0x01,0xe4,0x04,0x00]
+
+v_dot2_f16_f16_e64_dpp v5, v1, v2, 0.5 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf
+// GFX1200: v_dot2_f16_f16_e64_dpp v5, v1, v2, 0.5 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x66,0xd6,0xfa,0x04,0xc2,0x03,0x01,0x1b,0x00,0xff]
+
+v_dot2_bf16_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1
+// GFX1200: v_dot2_bf16_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1 ; encoding: [0x00,0x00,0x67,0xd6,0xfa,0x04,0x0e,0x04,0x01,0xe4,0x04,0x00]
+
+v_dot2_bf16_bf16_e64_dpp v0, v1, v2, v3 op_sel:[1,1,0,0] quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1
+// GFX12-ERR: :[[@LINE-1]]:41: error: invalid op_sel operand
+
+v_dot2_bf16_bf16_e64_dpp v0, s1, v2, v3 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0
+// GFX12-ERR: :[[@LINE-1]]:30: error: invalid operand for instruction
 
 v_dot2_f16_f16_e64_dpp v0.l, s1, v2, v3.l quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1
 // GFX12-ERR: :[[@LINE-1]]:30: error: invalid operand for instruction
 
-v_dot2_f16_f16_e64_dpp v0.l, v1, v2, v3.l op_sel:[1,1,0,0] quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1
-// GFX12-ERR: :[[@LINE-1]]:43: error: invalid op_sel operand
+v_dot2_bf16_bf16_e64_dpp v0.l, v1, s2, v3.l quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0
+// GFX1200: v_dot2_bf16_bf16_e64_dpp v0.l, v1, s2, v3.l quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 ; encoding: [0x00,0x00,0x67,0xd6,0xfa,0x04,0x0c,0x04,0x01,0xe4,0x00,0x00]
+
+v_dot2_bf16_bf16_e64_dpp v0.l, v1, v2, v3.l op_sel:[0,0,1,1] quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1
+// GFX12-ERR: :[[@LINE-1]]:45: error: op_sel operand conflicts with 16-bit operand suffix
 
-v_dot2_f16_f16_e64_dpp v0.l, v1, s2, v3.l quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1
-// GFX12: v_dot2_f16_f16_e64_dpp v0.l, v1, s2, v3.l quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1 ; encoding: [0x00,0x00,0x66,0xd6,0xfa,0x04,0x0c,0x04,0x01,0xe4,0x04,0x00]
+v_dot2_bf16_bf16_e64_dpp v0, |v1|, -v2, -|s3| op_sel:[0,0,1,1] quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1
+// GFX1200: v_dot2_bf16_bf16_e64_dpp v0, |v1|, -v2, -|s3| op_sel:[0,0,1,1] quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1 ; encoding: [0x00,0x65,0x67,0xd6,0xfa,0x04,0x0e,0xc0,0x01,0xe4,0x04,0x00]
 
 v_dot2_f16_f16_e64_dpp v0.h, v1, v2, v3.h op_sel:[0,0,1,1] quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1
-// GFX12: v_dot2_f16_f16_e64_dpp v0.h, v1, v2, v3.h op_sel:[0,0,1,1] quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1 ; encoding: [0x00,0x60,0x66,0xd6,0xfa,0x04,0x0e,0x04,0x01,0xe4,0x04,0x00]
+// GFX1200: v_dot2_f16_f16_e64_dpp v0.h, v1, v2, v3.h op_sel:[0,0,1,1] quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1 ; encoding: [0x00,0x60,0x66,0xd6,0xfa,0x04,0x0e,0x04,0x01,0xe4,0x04,0x00]
+
+v_dot2_bf16_bf16_e64_dpp v5, v1, v2, 0 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf
+// GFX1200: v_dot2_bf16_bf16_e64_dpp v5, v1, v2, 0 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x67,0xd6,0xfa,0x04,0x02,0x02,0x01,0x1b,0x00,0xff]
 
 v_dot2_f16_f16_e64_dpp v0.h, |v1|, -v2, -|s3| op_sel:[0,0,1,1] quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1
-// GFX12: v_dot2_f16_f16_e64_dpp v0.h, |v1|, -v2, -|s3| op_sel:[0,0,1,1] quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1 ; encoding: [0x00,0x65,0x66,0xd6,0xfa,0x04,0x0e,0xc0,0x01,0xe4,0x04,0x00]
+// GFX1200: v_dot2_f16_f16_e64_dpp v0.h, |v1|, -v2, -|s3| op_sel:[0,0,1,1] quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1 ; encoding: [0x00,0x65,0x66,0xd6,0xfa,0x04,0x0e,0xc0,0x01,0xe4,0x04,0x00]
 
 v_dot2_f16_f16_e64_dpp v5.l, v1, v2, 0.5 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf
-// GFX12: v_dot2_f16_f16_e64_dpp v5.l, v1, v2, 0.5 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x66,0xd6,0xfa,0x04,0xc2,0x03,0x01,0x1b,0x00,0xff]
+// GFX1200: v_dot2_f16_f16_e64_dpp v5.l, v1, v2, 0.5 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x66,0xd6,0xfa,0x04,0xc2,0x03,0x01,0x1b,0x00,0xff]
 
 v_dot2_f16_f16_e64_dpp v5.l, v1, v2, v255.h quad_perm:[0,1,2,3]
-// GFX12: v_dot2_f16_f16_e64_dpp v5.l, v1, v2, v255.h op_sel:[0,0,1,0] quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x20,0x66,0xd6,0xfa,0x04,0xfe,0x07,0x01,0xe4,0x00,0xff]
+// GFX1200: v_dot2_f16_f16_e64_dpp v5.l, v1, v2, v255.h op_sel:[0,0,1,0] quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x20,0x66,0xd6,0xfa,0x04,0xfe,0x07,0x01,0xe4,0x00,0xff]
 
 v_dot2_f16_f16_e64_dpp v255.h, -|v255|, -|v255|, -|src_scc| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
-// GFX12: v_dot2_f16_f16_e64_dpp v255.h, -|v255|, -|v255|, -|src_scc| op_sel:[0,0,0,1] row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x47,0x66,0xd6,0xfa,0xfe,0xf7,0xe3,0xff,0x6f,0x05,0x30]
+// GFX1200: v_dot2_f16_f16_e64_dpp v255.h, -|v255|, -|v255|, -|src_scc| op_sel:[0,0,0,1] row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x47,0x66,0xd6,0xfa,0xfe,0xf7,0xe3,0xff,0x6f,0x05,0x30]
 
 v_dot2_bf16_bf16_e64_dpp v0.l, v1, v2, v3.l quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1
-// GFX12: v_dot2_bf16_bf16_e64_dpp v0.l, v1, v2, v3.l quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1 ; encoding: [0x00,0x00,0x67,0xd6,0xfa,0x04,0x0e,0x04,0x01,0xe4,0x04,0x00]
+// GFX1200: v_dot2_bf16_bf16_e64_dpp v0.l, v1, v2, v3.l quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1 ; encoding: [0x00,0x00,0x67,0xd6,0xfa,0x04,0x0e,0x04,0x01,0xe4,0x04,0x00]
 
 v_dot2_bf16_bf16_e64_dpp v0.l, s1, v2, v3.l quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0
 // GFX12-ERR: :[[@LINE-1]]:32: error: invalid operand for instruction
 
 v_dot2_bf16_bf16_e64_dpp v0.l, v1, v2, v3.l op_sel:[1,1,0,0] quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1
 // GFX12-ERR: :[[@LINE-1]]:45: error: invalid op_sel operand
 
-v_dot2_bf16_bf16_e64_dpp v0.l, v1, s2, v3.l quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0
-// GFX12: v_dot2_bf16_bf16_e64_dpp v0.l, v1, s2, v3.l quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 ; encoding: [0x00,0x00,0x67,0xd6,0xfa,0x04,0x0c,0x04,0x01,0xe4,0x00,0x00]
-
 v_dot2_bf16_bf16_e64_dpp v0.h, v1, v2, v3.h op_sel:[0,0,1,1] quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1
-// GFX12: v_dot2_bf16_bf16_e64_dpp v0.h, v1, v2, v3.h op_sel:[0,0,1,1] quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1 ; encoding: [0x00,0x60,0x67,0xd6,0xfa,0x04,0x0e,0x04,0x01,0xe4,0x04,0x00]
+// GFX1200: v_dot2_bf16_bf16_e64_dpp v0.h, v1, v2, v3.h op_sel:[0,0,1,1] quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1 ; encoding: [0x00,0x60,0x67,0xd6,0xfa,0x04,0x0e,0x04,0x01,0xe4,0x04,0x00]
 
 v_dot2_bf16_bf16_e64_dpp v0.h, |v1|, -v2, -|s3| op_sel:[0,0,1,1] quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1
-// GFX12: v_dot2_bf16_bf16_e64_dpp v0.h, |v1|, -v2, -|s3| op_sel:[0,0,1,1] quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1 ; encoding: [0x00,0x65,0x67,0xd6,0xfa,0x04,0x0e,0xc0,0x01,0xe4,0x04,0x00]
+// GFX1200: v_dot2_bf16_bf16_e64_dpp v0.h, |v1|, -v2, -|s3| op_sel:[0,0,1,1] quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 fi:1 ; encoding: [0x00,0x65,0x67,0xd6,0xfa,0x04,0x0e,0xc0,0x01,0xe4,0x04,0x00]
 
 v_dot2_bf16_bf16_e64_dpp v5.l, v1, v2, 0 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf
-// GFX12: v_dot2_bf16_bf16_e64_dpp v5.l, v1, v2, 0 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x67,0xd6,0xfa,0x04,0x02,0x02,0x01,0x1b,0x00,0xff]
+// GFX1200: v_dot2_bf16_bf16_e64_dpp v5.l, v1, v2, 0 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x67,0xd6,0xfa,0x04,0x02,0x02,0x01,0x1b,0x00,0xff]
 
 v_dot2_bf16_bf16_e64_dpp v5.l, v1, v2, v255.h row_mirror
-// GFX12: v_dot2_bf16_bf16_e64_dpp v5.l, v1, v2, v255.h op_sel:[0,0,1,0] row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x20,0x67,0xd6,0xfa,0x04,0xfe,0x07,0x01,0x40,0x01,0xff]
+// GFX1200: v_dot2_bf16_bf16_e64_dpp v5.l, v1, v2, v255.h op_sel:[0,0,1,0] row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x20,0x67,0xd6,0xfa,0x04,0xfe,0x07,0x01,0x40,0x01,0xff]
 
 v_dot2_bf16_bf16_e64_dpp v255.h, -|v255|, -|v255|, -|src_scc| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
-// GFX12: v_dot2_bf16_bf16_e64_dpp v255.h, -|v255|, -|v255|, -|src_scc| op_sel:[0,0,0,1] row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x47,0x67,0xd6,0xfa,0xfe,0xf7,0xe3,0xff,0x6f,0x05,0x30]
+// GFX1200: v_dot2_bf16_bf16_e64_dpp v255.h, -|v255|, -|v255|, -|src_scc| op_sel:[0,0,0,1] row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x47,0x67,0xd6,0xfa,0xfe,0xf7,0xe3,0xff,0x6f,0x05,0x30]
 
 v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf fi:1 ; encoding: [0x00,0x40,0x1a,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x04,0xff]
-// GFX12: v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf fi:1 ; encoding: [0x00,0x40,0x1a,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x04,0xff]
+// GFX1200: v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf fi:1 ; encoding: [0x00,0x40,0x1a,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x04,0xff]
 
 v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x45,0x1a,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe]
-// GFX12: v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x45,0x1a,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe]
+// GFX1200: v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x45,0x1a,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe]
 
 v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0xf bank_mask:0xf ; encoding: [0x00,0x40,0x1a,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x39,0x00,0xff]
-// GFX12: v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0xf bank_mask:0xf ; encoding: [0x00,0x40,0x1a,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x39,0x00,0xff]
+// GFX1200: v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0xf bank_mask:0xf ; encoding: [0x00,0x40,0x1a,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x39,0x00,0xff]
 
 v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0x0 bank_mask:0x0 ; encoding: [0x00,0x40,0x1a,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x39,0x00,0x00]
-// GFX12: v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0x0 bank_mask:0x0 ; encoding: [0x00,0x40,0x1a,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x39,0x00,0x00]
+// GFX1200: v_dot2_f32_bf16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0x0 bank_mask:0x0 ; encoding: [0x00,0x40,0x1a,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x39,0x00,0x00]
 
 v_minimum_f32 v5, v1, v2 quad_perm:[3,2,1,0]
 // GFX12: v_minimum_f32_e64_dpp v5, v1, v2 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x65,0xd7,0xfa,0x04,0x02,0x00,0x01,0x1b,0x00,0xff]