[AMDGPU] Add support for `v_rcp_bf16` on gfx1250 (#148916)
Co-authored-by: Mekhanoshin, Stanislav <Stanislav.Mekhanoshin@amd.com>
diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index ab432b7..71e4b34 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -669,6 +669,7 @@
TARGET_BUILTIN(__builtin_amdgcn_s_wait_tensorcnt, "vIUs", "n", "gfx1250-insts")
TARGET_BUILTIN(__builtin_amdgcn_tanh_bf16, "yy", "nc", "bf16-trans-insts")
+TARGET_BUILTIN(__builtin_amdgcn_rcp_bf16, "yy", "nc", "bf16-trans-insts")
TARGET_BUILTIN(__builtin_amdgcn_cvt_f16_fp8, "hiIi", "nc", "gfx1250-insts")
TARGET_BUILTIN(__builtin_amdgcn_cvt_f16_bf8, "hiIi", "nc", "gfx1250-insts")
diff --git a/clang/lib/CodeGen/TargetBuiltins/AMDGPU.cpp b/clang/lib/CodeGen/TargetBuiltins/AMDGPU.cpp
index bd44874..0d8c2ed 100644
--- a/clang/lib/CodeGen/TargetBuiltins/AMDGPU.cpp
+++ b/clang/lib/CodeGen/TargetBuiltins/AMDGPU.cpp
@@ -411,6 +411,7 @@
case AMDGPU::BI__builtin_amdgcn_rcp:
case AMDGPU::BI__builtin_amdgcn_rcpf:
case AMDGPU::BI__builtin_amdgcn_rcph:
+ case AMDGPU::BI__builtin_amdgcn_rcp_bf16:
return emitBuiltinWithOneOverloadedType<1>(*this, E, Intrinsic::amdgcn_rcp);
case AMDGPU::BI__builtin_amdgcn_sqrt:
case AMDGPU::BI__builtin_amdgcn_sqrtf:
diff --git a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250.cl b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250.cl
index 830ceab..e50f02a 100644
--- a/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250.cl
+++ b/clang/test/CodeGenOpenCL/builtins-amdgcn-gfx1250.cl
@@ -61,6 +61,25 @@
*out = __builtin_amdgcn_tanh_bf16(a);
}
+// CHECK-LABEL: @test_rcp_bf16(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[OUT_ADDR:%.*]] = alloca ptr addrspace(1), align 8, addrspace(5)
+// CHECK-NEXT: [[A_ADDR:%.*]] = alloca bfloat, align 2, addrspace(5)
+// CHECK-NEXT: [[OUT_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[OUT_ADDR]] to ptr
+// CHECK-NEXT: [[A_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[A_ADDR]] to ptr
+// CHECK-NEXT: store ptr addrspace(1) [[OUT:%.*]], ptr [[OUT_ADDR_ASCAST]], align 8
+// CHECK-NEXT: store bfloat [[A:%.*]], ptr [[A_ADDR_ASCAST]], align 2
+// CHECK-NEXT: [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR_ASCAST]], align 2
+// CHECK-NEXT: [[TMP1:%.*]] = call bfloat @llvm.amdgcn.rcp.bf16(bfloat [[TMP0]])
+// CHECK-NEXT: [[TMP2:%.*]] = load ptr addrspace(1), ptr [[OUT_ADDR_ASCAST]], align 8
+// CHECK-NEXT: store bfloat [[TMP1]], ptr addrspace(1) [[TMP2]], align 2
+// CHECK-NEXT: ret void
+//
+void test_rcp_bf16(global __bf16* out, __bf16 a)
+{
+ *out = __builtin_amdgcn_rcp_bf16(a);
+}
+
// CHECK-LABEL: @test_cvt_f16_fp8(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[OUT_ADDR:%.*]] = alloca ptr addrspace(1), align 8, addrspace(5)
diff --git a/llvm/lib/Target/AMDGPU/VOP1Instructions.td b/llvm/lib/Target/AMDGPU/VOP1Instructions.td
index 57857c6..28f239b 100644
--- a/llvm/lib/Target/AMDGPU/VOP1Instructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP1Instructions.td
@@ -529,6 +529,7 @@
let SubtargetPredicate = HasBF16TransInsts in {
defm V_TANH_BF16 : VOP1Inst_t16 <"v_tanh_bf16", VOP_BF16_BF16, int_amdgcn_tanh>;
+defm V_RCP_BF16 : VOP1Inst_t16 <"v_rcp_bf16", VOP_BF16_BF16, AMDGPUrcp>;
}
} // End TRANS = 1, SchedRW = [WriteTrans32]
defm V_FREXP_MANT_F16 : VOP1Inst_t16 <"v_frexp_mant_f16", VOP_F16_F16, int_amdgcn_frexp_mant>;
@@ -1137,6 +1138,7 @@
defm V_CVT_PK_F16_BF8 : VOP1_Real_FULL_t16_and_fake16_gfx1250<0x076>;
defm V_CVT_F16_FP8 : VOP1_Real_FULL_t16_and_fake16_gfx1250<0x077>;
defm V_CVT_F16_BF8 : VOP1_Real_FULL_t16_and_fake16_gfx1250<0x078>;
+defm V_RCP_BF16 : VOP1_Real_FULL_t16_and_fake16_gfx1250<0x079>;
//===----------------------------------------------------------------------===//
// GFX10.
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.rcp.bf16.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.rcp.bf16.ll
new file mode 100644
index 0000000..3c49d0b
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.rcp.bf16.ll
@@ -0,0 +1,36 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; xUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=+real-true16 < %s | FileCheck -check-prefix=SDAG-TRUE16 %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=-real-true16 < %s | FileCheck -check-prefix=SDAG-FAKE16 %s
+; xUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=+real-true16 < %s | FileCheck -check-prefix=GI-TRUE16 %s
+; xUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=-real-true16 < %s | FileCheck -check-prefix=GI-FAKE16 %s
+
+; FIXME: t16 doesn't work at the moment because the store of s16 under t16 mode fails to select.
+; FIXME: GlobalISel does not work with bf16
+
+declare bfloat @llvm.amdgcn.rcp.bf16(bfloat) #0
+
+define amdgpu_kernel void @rcp_bf16(ptr addrspace(1) %out, bfloat %src) #1 {
+; SDAG-TRUE16-LABEL: rcp_bf16:
+; SDAG-TRUE16: ; %bb.0:
+; SDAG-TRUE16-NEXT: s_load_b96 s[0:2], s[4:5], 0x0
+; SDAG-TRUE16-NEXT: v_mov_b32_e32 v1, 0
+; SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0
+; SDAG-TRUE16-NEXT: v_rcp_bf16_e32 v0.l, s2
+; SDAG-TRUE16-NEXT: flat_store_b16 v1, v0, s[0:1]
+; SDAG-TRUE16-NEXT: s_endpgm
+;
+; SDAG-FAKE16-LABEL: rcp_bf16:
+; SDAG-FAKE16: ; %bb.0:
+; SDAG-FAKE16-NEXT: s_load_b96 s[0:2], s[4:5], 0x0
+; SDAG-FAKE16-NEXT: v_mov_b32_e32 v1, 0
+; SDAG-FAKE16-NEXT: s_wait_kmcnt 0x0
+; SDAG-FAKE16-NEXT: v_rcp_bf16_e32 v0, s2
+; SDAG-FAKE16-NEXT: global_store_b16 v1, v0, s[0:1]
+; SDAG-FAKE16-NEXT: s_endpgm
+ %rcp = call bfloat @llvm.amdgcn.rcp.bf16(bfloat %src) #0
+ store bfloat %rcp, ptr addrspace(1) %out, align 2
+ ret void
+}
+
+attributes #0 = { nounwind readnone }
+attributes #1 = { nounwind }
diff --git a/llvm/test/MC/AMDGPU/gfx1250_asm_vop1-fake16.s b/llvm/test/MC/AMDGPU/gfx1250_asm_vop1-fake16.s
index cf8ad0e..ce8f54a 100644
--- a/llvm/test/MC/AMDGPU/gfx1250_asm_vop1-fake16.s
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop1-fake16.s
@@ -73,6 +73,51 @@
v_tanh_bf16 v127, 0x8000
// GFX1250: v_tanh_bf16_e32 v127, 0x8000 ; encoding: [0xff,0x94,0xfe,0x7e,0x00,0x80,0x00,0x00]
+v_rcp_bf16 v5, v1
+// GFX1250: v_rcp_bf16_e32 v5, v1 ; encoding: [0x01,0xf3,0x0a,0x7e]
+
+v_rcp_bf16 v5, v127
+// GFX1250: v_rcp_bf16_e32 v5, v127 ; encoding: [0x7f,0xf3,0x0a,0x7e]
+
+v_rcp_bf16 v5, s1
+// GFX1250: v_rcp_bf16_e32 v5, s1 ; encoding: [0x01,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v5, s105
+// GFX1250: v_rcp_bf16_e32 v5, s105 ; encoding: [0x69,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v5, vcc_lo
+// GFX1250: v_rcp_bf16_e32 v5, vcc_lo ; encoding: [0x6a,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v5, vcc_hi
+// GFX1250: v_rcp_bf16_e32 v5, vcc_hi ; encoding: [0x6b,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v5, ttmp15
+// GFX1250: v_rcp_bf16_e32 v5, ttmp15 ; encoding: [0x7b,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v5, m0
+// GFX1250: v_rcp_bf16_e32 v5, m0 ; encoding: [0x7d,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v5, exec_lo
+// GFX1250: v_rcp_bf16_e32 v5, exec_lo ; encoding: [0x7e,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v5, exec_hi
+// GFX1250: v_rcp_bf16_e32 v5, exec_hi ; encoding: [0x7f,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v5, null
+// GFX1250: v_rcp_bf16_e32 v5, null ; encoding: [0x7c,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v5, -1
+// GFX1250: v_rcp_bf16_e32 v5, -1 ; encoding: [0xc1,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v5, 0.5
+// GFX1250: v_rcp_bf16_e32 v5, 0.5 ; encoding: [0xf0,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v5, src_scc
+// GFX1250: v_rcp_bf16_e32 v5, src_scc ; encoding: [0xfd,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v127, 0x8000
+// GFX1250: v_rcp_bf16_e32 v127, 0x8000 ; encoding: [0xff,0xf2,0xfe,0x7e,0x00,0x80,0x00,0x00]
+
v_cvt_f32_bf16 v5, v1
// GFX1250: v_cvt_f32_bf16_e32 v5, v1 ; encoding: [0x01,0xe5,0x0a,0x7e]
diff --git a/llvm/test/MC/AMDGPU/gfx1250_asm_vop1.s b/llvm/test/MC/AMDGPU/gfx1250_asm_vop1.s
index 055f540..7001a1f 100644
--- a/llvm/test/MC/AMDGPU/gfx1250_asm_vop1.s
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop1.s
@@ -76,6 +76,54 @@
v_tanh_bf16 v5.h, v1.h
// GFX1250: v_tanh_bf16_e32 v5.h, v1.h ; encoding: [0x81,0x95,0x0a,0x7f]
+v_rcp_bf16 v5, v1
+// GFX1250: v_rcp_bf16_e32 v5, v1 ; encoding: [0x01,0xf3,0x0a,0x7e]
+
+v_rcp_bf16 v5, v127
+// GFX1250: v_rcp_bf16_e32 v5, v127 ; encoding: [0x7f,0xf3,0x0a,0x7e]
+
+v_rcp_bf16 v5, s1
+// GFX1250: v_rcp_bf16_e32 v5, s1 ; encoding: [0x01,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v5, s105
+// GFX1250: v_rcp_bf16_e32 v5, s105 ; encoding: [0x69,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v5, vcc_lo
+// GFX1250: v_rcp_bf16_e32 v5, vcc_lo ; encoding: [0x6a,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v5, vcc_hi
+// GFX1250: v_rcp_bf16_e32 v5, vcc_hi ; encoding: [0x6b,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v5, ttmp15
+// GFX1250: v_rcp_bf16_e32 v5, ttmp15 ; encoding: [0x7b,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v5, m0
+// GFX1250: v_rcp_bf16_e32 v5, m0 ; encoding: [0x7d,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v5, exec_lo
+// GFX1250: v_rcp_bf16_e32 v5, exec_lo ; encoding: [0x7e,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v5, exec_hi
+// GFX1250: v_rcp_bf16_e32 v5, exec_hi ; encoding: [0x7f,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v5, null
+// GFX1250: v_rcp_bf16_e32 v5, null ; encoding: [0x7c,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v5, -1
+// GFX1250: v_rcp_bf16_e32 v5, -1 ; encoding: [0xc1,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v5, 0.5
+// GFX1250: v_rcp_bf16_e32 v5, 0.5 ; encoding: [0xf0,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v5, src_scc
+// GFX1250: v_rcp_bf16_e32 v5, src_scc ; encoding: [0xfd,0xf2,0x0a,0x7e]
+
+v_rcp_bf16 v127, 0x8000
+// GFX1250: v_rcp_bf16_e32 v127, 0x8000 ; encoding: [0xff,0xf2,0xfe,0x7e,0x00,0x80,0x00,0x00]
+
+v_rcp_bf16 v5.h, v1.h
+// GFX1250: v_rcp_bf16_e32 v5.h, v1.h ; encoding: [0x81,0xf3,0x0a,0x7f]
+
v_cvt_f32_bf16 v5, v1
// GFX1250: v_cvt_f32_bf16_e32 v5, v1 ; encoding: [0x01,0xe5,0x0a,0x7e]
diff --git a/llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp16-fake16.s b/llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp16-fake16.s
index 4e5754f..3de8fc2 100644
--- a/llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp16-fake16.s
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp16-fake16.s
@@ -58,6 +58,62 @@
// GFX1250: v_tanh_bf16_dpp v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0x94,0xfe,0x7e,0x7f,0x6f,0x35,0x30]
// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+v_rcp_bf16 v5, v1 quad_perm:[3,2,1,0]
+// GFX1250: v_rcp_bf16_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x1b,0x00,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 quad_perm:[0,1,2,3]
+// GFX1250: v_rcp_bf16_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0xe4,0x00,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 row_mirror
+// GFX1250: v_rcp_bf16_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x40,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 row_half_mirror
+// GFX1250: v_rcp_bf16_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x41,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 row_shl:1
+// GFX1250: v_rcp_bf16_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x01,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 row_shl:15
+// GFX1250: v_rcp_bf16_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x0f,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 row_shr:1
+// GFX1250: v_rcp_bf16_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x11,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 row_shr:15
+// GFX1250: v_rcp_bf16_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x1f,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 row_ror:1
+// GFX1250: v_rcp_bf16_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x21,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 row_ror:15
+// GFX1250: v_rcp_bf16_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x2f,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
+// GFX1250: v_rcp_bf16_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x50,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX1250: v_rcp_bf16_dpp v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x5f,0x01,0x01]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX1250: v_rcp_bf16_dpp v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x60,0x09,0x13]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX1250: v_rcp_bf16_dpp v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xf2,0xfe,0x7e,0x7f,0x6f,0x35,0x30]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
v_cvt_f32_bf16 v5, v1 quad_perm:[3,2,1,0]
// GFX1250: v_cvt_f32_bf16_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x1b,0x00,0xff]
// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
diff --git a/llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp16.s b/llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp16.s
index a678725..4632b15 100644
--- a/llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp16.s
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp16.s
@@ -62,6 +62,66 @@
// GFX1250: v_tanh_bf16_dpp v5.h, v1.h quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x94,0x0a,0x7f,0x81,0x1b,0x00,0xff]
// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+v_rcp_bf16 v5, v1 quad_perm:[3,2,1,0]
+// GFX1250: v_rcp_bf16_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x1b,0x00,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 quad_perm:[0,1,2,3]
+// GFX1250: v_rcp_bf16_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0xe4,0x00,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 row_mirror
+// GFX1250: v_rcp_bf16_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x40,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 row_half_mirror
+// GFX1250: v_rcp_bf16_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x41,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 row_shl:1
+// GFX1250: v_rcp_bf16_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x01,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 row_shl:15
+// GFX1250: v_rcp_bf16_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x0f,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 row_shr:1
+// GFX1250: v_rcp_bf16_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x11,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 row_shr:15
+// GFX1250: v_rcp_bf16_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x1f,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 row_ror:1
+// GFX1250: v_rcp_bf16_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x21,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 row_ror:15
+// GFX1250: v_rcp_bf16_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x2f,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
+// GFX1250: v_rcp_bf16_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x50,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX1250: v_rcp_bf16_dpp v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x5f,0x01,0x01]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX1250: v_rcp_bf16_dpp v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x60,0x09,0x13]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX1250: v_rcp_bf16_dpp v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xf2,0xfe,0x7e,0x7f,0x6f,0x35,0x30]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5.h, v1.h quad_perm:[3,2,1,0]
+// GFX1250: v_rcp_bf16_dpp v5.h, v1.h quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7f,0x81,0x1b,0x00,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
v_cvt_f32_bf16 v5, v1 quad_perm:[3,2,1,0]
// GFX1250: v_cvt_f32_bf16_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x1b,0x00,0xff]
// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
diff --git a/llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp8-fake16.s b/llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp8-fake16.s
index e6c35d5..1a6028a 100644
--- a/llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp8-fake16.s
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp8-fake16.s
@@ -14,6 +14,18 @@
// GFX1250: v_tanh_bf16_dpp v127, v127 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0x94,0xfe,0x7e,0x7f,0x00,0x00,0x00]
// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+v_rcp_bf16 v5, v1 dpp8:[7,6,5,4,3,2,1,0]
+// GFX1250: v_rcp_bf16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xf2,0x0a,0x7e,0x01,0x77,0x39,0x05]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX1250: v_rcp_bf16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xf2,0x0a,0x7e,0x01,0x77,0x39,0x05]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v127, v127 dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX1250: v_rcp_bf16_dpp v127, v127 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xf2,0xfe,0x7e,0x7f,0x00,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
v_cvt_f32_bf16 v5, v1 dpp8:[7,6,5,4,3,2,1,0]
// GFX1250: v_cvt_f32_bf16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xe4,0x0a,0x7e,0x01,0x77,0x39,0x05]
// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
diff --git a/llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp8.s b/llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp8.s
index 98bd1fd..13085da 100644
--- a/llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp8.s
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp8.s
@@ -18,6 +18,22 @@
// GFX1250: v_tanh_bf16_dpp v5.h, v1.h dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0x94,0x0a,0x7f,0x81,0x77,0x39,0x05]
// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+v_rcp_bf16 v5, v1 dpp8:[7,6,5,4,3,2,1,0]
+// GFX1250: v_rcp_bf16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xf2,0x0a,0x7e,0x01,0x77,0x39,0x05]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX1250: v_rcp_bf16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xf2,0x0a,0x7e,0x01,0x77,0x39,0x05]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v127, v127 dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX1250: v_rcp_bf16_dpp v127, v127 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xf2,0xfe,0x7e,0x7f,0x00,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16 v5.h, v1.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX1250: v_rcp_bf16_dpp v5.h, v1.h dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xf2,0x0a,0x7f,0x81,0x77,0x39,0x05]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
v_cvt_f32_bf16 v5, v1 dpp8:[7,6,5,4,3,2,1,0]
// GFX1250: v_cvt_f32_bf16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xe4,0x0a,0x7e,0x01,0x77,0x39,0x05]
// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
diff --git a/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1-fake16.s b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1-fake16.s
index 7196f40..a8bf9cb 100644
--- a/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1-fake16.s
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1-fake16.s
@@ -127,6 +127,51 @@
v_cvt_f32_fp8 v1, v3 byte_sel:2 clamp
// GFX1250: v_cvt_f32_fp8_e64 v1, v3 byte_sel:2 clamp ; encoding: [0x01,0x88,0xec,0xd5,0x03,0x01,0x00,0x00]
+v_rcp_bf16_e64 v5, v1
+// GFX1250: v_rcp_bf16_e64 v5, v1 ; encoding: [0x05,0x00,0xf9,0xd5,0x01,0x01,0x00,0x00]
+
+v_rcp_bf16_e64 v5, v255
+// GFX1250: v_rcp_bf16_e64 v5, v255 ; encoding: [0x05,0x00,0xf9,0xd5,0xff,0x01,0x00,0x00]
+
+v_rcp_bf16_e64 v5, s1
+// GFX1250: v_rcp_bf16_e64 v5, s1 ; encoding: [0x05,0x00,0xf9,0xd5,0x01,0x00,0x00,0x00]
+
+v_rcp_bf16_e64 v5, s105
+// GFX1250: v_rcp_bf16_e64 v5, s105 ; encoding: [0x05,0x00,0xf9,0xd5,0x69,0x00,0x00,0x00]
+
+v_rcp_bf16_e64 v5, vcc_lo
+// GFX1250: v_rcp_bf16_e64 v5, vcc_lo ; encoding: [0x05,0x00,0xf9,0xd5,0x6a,0x00,0x00,0x00]
+
+v_rcp_bf16_e64 v5, vcc_hi
+// GFX1250: v_rcp_bf16_e64 v5, vcc_hi ; encoding: [0x05,0x00,0xf9,0xd5,0x6b,0x00,0x00,0x00]
+
+v_rcp_bf16_e64 v5, ttmp15
+// GFX1250: v_rcp_bf16_e64 v5, ttmp15 ; encoding: [0x05,0x00,0xf9,0xd5,0x7b,0x00,0x00,0x00]
+
+v_rcp_bf16_e64 v5, m0
+// GFX1250: v_rcp_bf16_e64 v5, m0 ; encoding: [0x05,0x00,0xf9,0xd5,0x7d,0x00,0x00,0x00]
+
+v_rcp_bf16_e64 v5, exec_lo
+// GFX1250: v_rcp_bf16_e64 v5, exec_lo ; encoding: [0x05,0x00,0xf9,0xd5,0x7e,0x00,0x00,0x00]
+
+v_rcp_bf16_e64 v5, exec_hi
+// GFX1250: v_rcp_bf16_e64 v5, exec_hi ; encoding: [0x05,0x00,0xf9,0xd5,0x7f,0x00,0x00,0x00]
+
+v_rcp_bf16_e64 v5, null
+// GFX1250: v_rcp_bf16_e64 v5, null ; encoding: [0x05,0x00,0xf9,0xd5,0x7c,0x00,0x00,0x00]
+
+v_rcp_bf16_e64 v5, -1
+// GFX1250: v_rcp_bf16_e64 v5, -1 ; encoding: [0x05,0x00,0xf9,0xd5,0xc1,0x00,0x00,0x00]
+
+v_rcp_bf16_e64 v5, 0.5 mul:2
+// GFX1250: v_rcp_bf16_e64 v5, 0.5 mul:2 ; encoding: [0x05,0x00,0xf9,0xd5,0xf0,0x00,0x00,0x08]
+
+v_rcp_bf16_e64 v5, src_scc mul:4
+// GFX1250: v_rcp_bf16_e64 v5, src_scc mul:4 ; encoding: [0x05,0x00,0xf9,0xd5,0xfd,0x00,0x00,0x10]
+
+v_rcp_bf16_e64 v255, -|0x8000| clamp div:2
+// GFX1250: v_rcp_bf16_e64 v255, -|0x8000| clamp div:2 ; encoding: [0xff,0x81,0xf9,0xd5,0xff,0x00,0x00,0x38,0x00,0x80,0x00,0x00]
+
v_cvt_f32_bf16_e64 v5, v1
// GFX1250: v_cvt_f32_bf16_e64 v5, v1 ; encoding: [0x05,0x00,0xf2,0xd5,0x01,0x01,0x00,0x00]
diff --git a/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1.s b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1.s
index 3ea695f..09b7764 100644
--- a/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1.s
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1.s
@@ -130,6 +130,54 @@
v_cvt_f32_fp8 v1, v3 byte_sel:2 clamp
// GFX1250: v_cvt_f32_fp8_e64 v1, v3 byte_sel:2 clamp ; encoding: [0x01,0x88,0xec,0xd5,0x03,0x01,0x00,0x00]
+v_rcp_bf16_e64 v5, v1
+// GFX1250: v_rcp_bf16_e64 v5, v1 ; encoding: [0x05,0x00,0xf9,0xd5,0x01,0x01,0x00,0x00]
+
+v_rcp_bf16_e64 v5, v255
+// GFX1250: v_rcp_bf16_e64 v5, v255 ; encoding: [0x05,0x00,0xf9,0xd5,0xff,0x01,0x00,0x00]
+
+v_rcp_bf16_e64 v5, s1
+// GFX1250: v_rcp_bf16_e64 v5, s1 ; encoding: [0x05,0x00,0xf9,0xd5,0x01,0x00,0x00,0x00]
+
+v_rcp_bf16_e64 v5, s105
+// GFX1250: v_rcp_bf16_e64 v5, s105 ; encoding: [0x05,0x00,0xf9,0xd5,0x69,0x00,0x00,0x00]
+
+v_rcp_bf16_e64 v5, vcc_lo
+// GFX1250: v_rcp_bf16_e64 v5, vcc_lo ; encoding: [0x05,0x00,0xf9,0xd5,0x6a,0x00,0x00,0x00]
+
+v_rcp_bf16_e64 v5, vcc_hi
+// GFX1250: v_rcp_bf16_e64 v5, vcc_hi ; encoding: [0x05,0x00,0xf9,0xd5,0x6b,0x00,0x00,0x00]
+
+v_rcp_bf16_e64 v5, ttmp15
+// GFX1250: v_rcp_bf16_e64 v5, ttmp15 ; encoding: [0x05,0x00,0xf9,0xd5,0x7b,0x00,0x00,0x00]
+
+v_rcp_bf16_e64 v5, m0
+// GFX1250: v_rcp_bf16_e64 v5, m0 ; encoding: [0x05,0x00,0xf9,0xd5,0x7d,0x00,0x00,0x00]
+
+v_rcp_bf16_e64 v5, exec_lo
+// GFX1250: v_rcp_bf16_e64 v5, exec_lo ; encoding: [0x05,0x00,0xf9,0xd5,0x7e,0x00,0x00,0x00]
+
+v_rcp_bf16_e64 v5, exec_hi
+// GFX1250: v_rcp_bf16_e64 v5, exec_hi ; encoding: [0x05,0x00,0xf9,0xd5,0x7f,0x00,0x00,0x00]
+
+v_rcp_bf16_e64 v5, null
+// GFX1250: v_rcp_bf16_e64 v5, null ; encoding: [0x05,0x00,0xf9,0xd5,0x7c,0x00,0x00,0x00]
+
+v_rcp_bf16_e64 v5, -1
+// GFX1250: v_rcp_bf16_e64 v5, -1 ; encoding: [0x05,0x00,0xf9,0xd5,0xc1,0x00,0x00,0x00]
+
+v_rcp_bf16_e64 v5, 0.5 mul:2
+// GFX1250: v_rcp_bf16_e64 v5, 0.5 mul:2 ; encoding: [0x05,0x00,0xf9,0xd5,0xf0,0x00,0x00,0x08]
+
+v_rcp_bf16_e64 v5, src_scc mul:4
+// GFX1250: v_rcp_bf16_e64 v5, src_scc mul:4 ; encoding: [0x05,0x00,0xf9,0xd5,0xfd,0x00,0x00,0x10]
+
+v_rcp_bf16_e64 v255, -|0x8000| clamp div:2
+// GFX1250: v_rcp_bf16_e64 v255, -|0x8000| clamp div:2 ; encoding: [0xff,0x81,0xf9,0xd5,0xff,0x00,0x00,0x38,0x00,0x80,0x00,0x00]
+
+v_rcp_bf16 v5.h, v128.h
+// GFX1250: v_rcp_bf16_e64 v5.h, v128.h op_sel:[1,1] ; encoding: [0x05,0x48,0xf9,0xd5,0x80,0x01,0x00,0x00]
+
v_cvt_f32_bf16_e64 v5, v1
// GFX1250: v_cvt_f32_bf16_e64 v5, v1 ; encoding: [0x05,0x00,0xf2,0xd5,0x01,0x01,0x00,0x00]
diff --git a/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1_dpp16-fake16.s b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1_dpp16-fake16.s
index f7cb223..f112761 100644
--- a/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1_dpp16-fake16.s
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1_dpp16-fake16.s
@@ -58,6 +58,62 @@
// GFX1250: v_tanh_bf16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xca,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+v_rcp_bf16_e64_dpp v5, v1 quad_perm:[3,2,1,0]
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 quad_perm:[0,1,2,3]
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 row_mirror
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 row_half_mirror
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 row_shl:1
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 row_shl:15
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 row_shr:1
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 row_shr:15
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 row_ror:1
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 row_ror:15
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX1250: v_rcp_bf16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xf9,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
v_cvt_f32_bf16_e64_dpp v5, v1 quad_perm:[3,2,1,0]
// GFX1250: v_cvt_f32_bf16_e64_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf2,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
diff --git a/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1_dpp16.s b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1_dpp16.s
index dfa9741..b591fa1 100644
--- a/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1_dpp16.s
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1_dpp16.s
@@ -62,6 +62,66 @@
// GFX1250: v_tanh_bf16_e64_dpp v5.h, v128.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x48,0xca,0xd5,0xfa,0x00,0x00,0x00,0x80,0x1b,0x00,0xff]
// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+v_rcp_bf16_e64_dpp v5, v1 quad_perm:[3,2,1,0]
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 quad_perm:[0,1,2,3]
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 row_mirror
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 row_half_mirror
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 row_shl:1
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 row_shl:15
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 row_shr:1
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 row_shr:15
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 row_ror:1
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 row_ror:15
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX1250: v_rcp_bf16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xf9,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5.h, v128.h quad_perm:[3,2,1,0]
+// GFX1250: v_rcp_bf16_e64_dpp v5.h, v128.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x48,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x80,0x1b,0x00,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
v_cvt_f32_bf16_e64_dpp v5, v1 quad_perm:[3,2,1,0]
// GFX1250: v_cvt_f32_bf16_e64_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf2,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
diff --git a/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1_dpp8-fake16.s b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1_dpp8-fake16.s
index 686209d..1cbd20b 100644
--- a/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1_dpp8-fake16.s
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1_dpp8-fake16.s
@@ -18,6 +18,22 @@
// GFX1250: v_tanh_bf16_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xca,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+v_rcp_bf16_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0]
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xf9,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 mul:2 dpp8:[7,6,5,4,3,2,1,0]
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xf9,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x00,0xf9,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX1250: v_rcp_bf16_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xf9,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
v_cvt_f32_bf16_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0]
// GFX1250: v_cvt_f32_bf16_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xf2,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
diff --git a/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1_dpp8.s b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1_dpp8.s
index 52cbe92..4553476 100644
--- a/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1_dpp8.s
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1_dpp8.s
@@ -22,6 +22,26 @@
// GFX1250: v_tanh_bf16_e64_dpp v5.h, v128.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xca,0xd5,0xe9,0x00,0x00,0x00,0x80,0x77,0x39,0x05]
// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+v_rcp_bf16_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0]
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xf9,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 mul:2 dpp8:[7,6,5,4,3,2,1,0]
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xf9,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5, v1 mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX1250: v_rcp_bf16_e64_dpp v5, v1 mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x00,0xf9,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX1250: v_rcp_bf16_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xf9,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_rcp_bf16_e64_dpp v5.h, v128.h dpp8:[7,6,5,4,3,2,1,0]
+// GFX1250: v_rcp_bf16_e64_dpp v5.h, v128.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xf9,0xd5,0xe9,0x00,0x00,0x00,0x80,0x77,0x39,0x05]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
v_cvt_f32_bf16_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0]
// GFX1250: v_cvt_f32_bf16_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xf2,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop1.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop1.txt
index 1b1f65b..7debbd4 100644
--- a/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop1.txt
+++ b/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop1.txt
@@ -89,6 +89,69 @@
0x81,0x95,0x0a,0x7f
# GFX1250-REAL16: v_tanh_bf16_e32 v5.h, v1.h ; encoding: [0x81,0x95,0x0a,0x7f]
+0xff,0xf2,0xfe,0x7e,0x00,0x80,0x00,0x00
+# GFX1250-REAL16: v_rcp_bf16_e32 v127.l, 0x8000 ; encoding: [0xff,0xf2,0xfe,0x7e,0x00,0x80,0x00,0x00]
+# GFX1250-FAKE16: v_rcp_bf16_e32 v127, 0x8000 ; encoding: [0xff,0xf2,0xfe,0x7e,0x00,0x80,0x00,0x00]
+
+0xc1,0xf2,0x0a,0x7e
+# GFX1250-REAL16: v_rcp_bf16_e32 v5.l, -1 ; encoding: [0xc1,0xf2,0x0a,0x7e]
+# GFX1250-FAKE16: v_rcp_bf16_e32 v5, -1 ; encoding: [0xc1,0xf2,0x0a,0x7e]
+
+0xf0,0xf2,0x0a,0x7e
+# GFX1250-REAL16: v_rcp_bf16_e32 v5.l, 0.5 ; encoding: [0xf0,0xf2,0x0a,0x7e]
+# GFX1250-FAKE16: v_rcp_bf16_e32 v5, 0.5 ; encoding: [0xf0,0xf2,0x0a,0x7e]
+
+0x7f,0xf2,0x0a,0x7e
+# GFX1250-REAL16: v_rcp_bf16_e32 v5.l, exec_hi ; encoding: [0x7f,0xf2,0x0a,0x7e]
+# GFX1250-FAKE16: v_rcp_bf16_e32 v5, exec_hi ; encoding: [0x7f,0xf2,0x0a,0x7e]
+
+0x7e,0xf2,0x0a,0x7e
+# GFX1250-REAL16: v_rcp_bf16_e32 v5.l, exec_lo ; encoding: [0x7e,0xf2,0x0a,0x7e]
+# GFX1250-FAKE16: v_rcp_bf16_e32 v5, exec_lo ; encoding: [0x7e,0xf2,0x0a,0x7e]
+
+0x7d,0xf2,0x0a,0x7e
+# GFX1250-REAL16: v_rcp_bf16_e32 v5.l, m0 ; encoding: [0x7d,0xf2,0x0a,0x7e]
+# GFX1250-FAKE16: v_rcp_bf16_e32 v5, m0 ; encoding: [0x7d,0xf2,0x0a,0x7e]
+
+0x7c,0xf2,0x0a,0x7e
+# GFX1250-REAL16: v_rcp_bf16_e32 v5.l, null ; encoding: [0x7c,0xf2,0x0a,0x7e]
+# GFX1250-FAKE16: v_rcp_bf16_e32 v5, null ; encoding: [0x7c,0xf2,0x0a,0x7e]
+
+0x01,0xf2,0x0a,0x7e
+# GFX1250-REAL16: v_rcp_bf16_e32 v5.l, s1 ; encoding: [0x01,0xf2,0x0a,0x7e]
+# GFX1250-FAKE16: v_rcp_bf16_e32 v5, s1 ; encoding: [0x01,0xf2,0x0a,0x7e]
+
+0x69,0xf2,0x0a,0x7e
+# GFX1250-REAL16: v_rcp_bf16_e32 v5.l, s105 ; encoding: [0x69,0xf2,0x0a,0x7e]
+# GFX1250-FAKE16: v_rcp_bf16_e32 v5, s105 ; encoding: [0x69,0xf2,0x0a,0x7e]
+
+0xfd,0xf2,0x0a,0x7e
+# GFX1250-REAL16: v_rcp_bf16_e32 v5.l, src_scc ; encoding: [0xfd,0xf2,0x0a,0x7e]
+# GFX1250-FAKE16: v_rcp_bf16_e32 v5, src_scc ; encoding: [0xfd,0xf2,0x0a,0x7e]
+
+0x7b,0xf2,0x0a,0x7e
+# GFX1250-REAL16: v_rcp_bf16_e32 v5.l, ttmp15 ; encoding: [0x7b,0xf2,0x0a,0x7e]
+# GFX1250-FAKE16: v_rcp_bf16_e32 v5, ttmp15 ; encoding: [0x7b,0xf2,0x0a,0x7e]
+
+0x01,0xf3,0x0a,0x7e
+# GFX1250-REAL16: v_rcp_bf16_e32 v5.l, v1.l ; encoding: [0x01,0xf3,0x0a,0x7e]
+# GFX1250-FAKE16: v_rcp_bf16_e32 v5, v1 ; encoding: [0x01,0xf3,0x0a,0x7e]
+
+0x7f,0xf3,0x0a,0x7e
+# GFX1250-REAL16: v_rcp_bf16_e32 v5.l, v127.l ; encoding: [0x7f,0xf3,0x0a,0x7e]
+# GFX1250-FAKE16: v_rcp_bf16_e32 v5, v127 ; encoding: [0x7f,0xf3,0x0a,0x7e]
+
+0x6b,0xf2,0x0a,0x7e
+# GFX1250-REAL16: v_rcp_bf16_e32 v5.l, vcc_hi ; encoding: [0x6b,0xf2,0x0a,0x7e]
+# GFX1250-FAKE16: v_rcp_bf16_e32 v5, vcc_hi ; encoding: [0x6b,0xf2,0x0a,0x7e]
+
+0x6a,0xf2,0x0a,0x7e
+# GFX1250-REAL16: v_rcp_bf16_e32 v5.l, vcc_lo ; encoding: [0x6a,0xf2,0x0a,0x7e]
+# GFX1250-FAKE16: v_rcp_bf16_e32 v5, vcc_lo ; encoding: [0x6a,0xf2,0x0a,0x7e]
+
+0x81,0xf3,0x0a,0x7f
+# GFX1250-REAL16: v_rcp_bf16_e32 v5.h, v1.h ; encoding: [0x81,0xf3,0x0a,0x7f]
+
0xff,0xe4,0xfe,0x7e,0x00,0x80,0x00,0x00
# GFX1250: v_cvt_f32_bf16_e32 v127, 0x8000 ; encoding: [0xff,0xe4,0xfe,0x7e,0x00,0x80,0x00,0x00]
diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop1_dpp16.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop1_dpp16.txt
index db36451..098a0ba 100644
--- a/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop1_dpp16.txt
+++ b/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop1_dpp16.txt
@@ -61,6 +61,65 @@
0xfa,0x94,0x0a,0x7f,0x81,0x1b,0x00,0xff
# GFX1250-REAL16: v_tanh_bf16_dpp v5.h, v1.h quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x94,0x0a,0x7f,0x81,0x1b,0x00,0xff]
+0xfa,0xf2,0xfe,0x7e,0x7f,0x6f,0x35,0x30
+# GFX1250-REAL16: v_rcp_bf16_dpp v127.l, -|v127.l| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xf2,0xfe,0x7e,0x7f,0x6f,0x35,0x30]
+# GFX1250-FAKE16: v_rcp_bf16_dpp v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xf2,0xfe,0x7e,0x7f,0x6f,0x35,0x30]
+
+0xfa,0xf2,0x0a,0x7e,0x01,0xe4,0x00,0xff
+# GFX1250-REAL16: v_rcp_bf16_dpp v5.l, v1.l quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0xe4,0x00,0xff]
+# GFX1250-FAKE16: v_rcp_bf16_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0xe4,0x00,0xff]
+
+0xfa,0xf2,0x0a,0x7e,0x01,0x1b,0x00,0xff
+# GFX1250-REAL16: v_rcp_bf16_dpp v5.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x1b,0x00,0xff]
+# GFX1250-FAKE16: v_rcp_bf16_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x1b,0x00,0xff]
+
+0xfa,0xf2,0x0a,0x7e,0x01,0x41,0x01,0xff
+# GFX1250-REAL16: v_rcp_bf16_dpp v5.l, v1.l row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x41,0x01,0xff]
+# GFX1250-FAKE16: v_rcp_bf16_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x41,0x01,0xff]
+
+0xfa,0xf2,0x0a,0x7e,0x01,0x40,0x01,0xff
+# GFX1250-REAL16: v_rcp_bf16_dpp v5.l, v1.l row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x40,0x01,0xff]
+# GFX1250-FAKE16: v_rcp_bf16_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x40,0x01,0xff]
+
+0xfa,0xf2,0x0a,0x7e,0x01,0x21,0x01,0xff
+# GFX1250-REAL16: v_rcp_bf16_dpp v5.l, v1.l row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x21,0x01,0xff]
+# GFX1250-FAKE16: v_rcp_bf16_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x21,0x01,0xff]
+
+0xfa,0xf2,0x0a,0x7e,0x01,0x2f,0x01,0xff
+# GFX1250-REAL16: v_rcp_bf16_dpp v5.l, v1.l row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x2f,0x01,0xff]
+# GFX1250-FAKE16: v_rcp_bf16_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x2f,0x01,0xff]
+
+0xfa,0xf2,0x0a,0x7e,0x01,0x50,0x01,0xff
+# GFX1250-REAL16: v_rcp_bf16_dpp v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x50,0x01,0xff]
+# GFX1250-FAKE16: v_rcp_bf16_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x50,0x01,0xff]
+
+0xfa,0xf2,0x0a,0x7e,0x01,0x5f,0x01,0x01
+# GFX1250-REAL16: v_rcp_bf16_dpp v5.l, v1.l row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x5f,0x01,0x01]
+# GFX1250-FAKE16: v_rcp_bf16_dpp v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x5f,0x01,0x01]
+
+0xfa,0xf2,0x0a,0x7e,0x01,0x01,0x01,0xff
+# GFX1250-REAL16: v_rcp_bf16_dpp v5.l, v1.l row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x01,0x01,0xff]
+# GFX1250-FAKE16: v_rcp_bf16_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x01,0x01,0xff]
+
+0xfa,0xf2,0x0a,0x7e,0x01,0x0f,0x01,0xff
+# GFX1250-REAL16: v_rcp_bf16_dpp v5.l, v1.l row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x0f,0x01,0xff]
+# GFX1250-FAKE16: v_rcp_bf16_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x0f,0x01,0xff]
+
+0xfa,0xf2,0x0a,0x7e,0x01,0x11,0x01,0xff
+# GFX1250-REAL16: v_rcp_bf16_dpp v5.l, v1.l row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x11,0x01,0xff]
+# GFX1250-FAKE16: v_rcp_bf16_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x11,0x01,0xff]
+
+0xfa,0xf2,0x0a,0x7e,0x01,0x1f,0x01,0xff
+# GFX1250-REAL16: v_rcp_bf16_dpp v5.l, v1.l row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x1f,0x01,0xff]
+# GFX1250-FAKE16: v_rcp_bf16_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x1f,0x01,0xff]
+
+0xfa,0xf2,0x0a,0x7e,0x01,0x60,0x09,0x13
+# GFX1250-REAL16: v_rcp_bf16_dpp v5.l, v1.l row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x60,0x09,0x13]
+# GFX1250-FAKE16: v_rcp_bf16_dpp v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xf2,0x0a,0x7e,0x01,0x60,0x09,0x13]
+
+0xfa,0xf2,0x0a,0x7f,0x81,0x1b,0x00,0xff
+# GFX1250-REAL16: v_rcp_bf16_dpp v5.h, v1.h quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xf2,0x0a,0x7f,0x81,0x1b,0x00,0xff]
+
0xfa,0xe4,0xfe,0x7e,0x7f,0x6f,0x35,0x30
# GFX1250: v_cvt_f32_bf16_dpp v127, -|v127.l| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xe4,0xfe,0x7e,0x7f,0x6f,0x35,0x30]
diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop1_dpp8.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop1_dpp8.txt
index 4598efc..b097a4e 100644
--- a/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop1_dpp8.txt
+++ b/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop1_dpp8.txt
@@ -18,6 +18,22 @@
# GFX1250-REAL16: v_tanh_bf16_dpp v5.h, v1.h dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0x94,0x0a,0x7f,0x81,0x77,0x39,0x05]
# GFX1250-FAKE16: v_add_f64_e32 v[156:157], v[129:130], v[187:188] ; encoding: [0x81,0x77,0x39,0x05]
+0xe9,0xf2,0xfe,0x7e,0x7f,0x00,0x00,0x00
+# GFX1250-REAL16: v_rcp_bf16_dpp v127.l, v127.l dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xf2,0xfe,0x7e,0x7f,0x00,0x00,0x00]
+# GFX1250-FAKE16: v_rcp_bf16_dpp v127, v127 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xf2,0xfe,0x7e,0x7f,0x00,0x00,0x00]
+
+0xe9,0xf2,0x0a,0x7e,0x01,0x77,0x39,0x05
+# GFX1250-REAL16: v_rcp_bf16_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xf2,0x0a,0x7e,0x01,0x77,0x39,0x05]
+# GFX1250-FAKE16: v_rcp_bf16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xf2,0x0a,0x7e,0x01,0x77,0x39,0x05]
+
+0xea,0xf2,0x0a,0x7e,0x01,0x77,0x39,0x05
+# GFX1250-REAL16: v_rcp_bf16_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xf2,0x0a,0x7e,0x01,0x77,0x39,0x05]
+# GFX1250-FAKE16: v_rcp_bf16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xf2,0x0a,0x7e,0x01,0x77,0x39,0x05]
+
+0xe9,0xf2,0x0a,0x7f,0x81,0x77,0x39,0x05
+# GFX1250-REAL16: v_rcp_bf16_dpp v5.h, v1.h dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xf2,0x0a,0x7f,0x81,0x77,0x39,0x05]
+# GFX1250-FAKE16: v_add_f64_e32 v[156:157], v[129:130], v[187:188] ; encoding: [0x81,0x77,0x39,0x05]
+
0xe9,0xe4,0xfe,0x7e,0x7f,0x00,0x00,0x00
# GFX1250: v_cvt_f32_bf16_dpp v127, v127.l dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xe4,0xfe,0x7e,0x7f,0x00,0x00,0x00]
diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop3_from_vop1.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop3_from_vop1.txt
index 3e345d7..3231b2f 100644
--- a/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop3_from_vop1.txt
+++ b/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop3_from_vop1.txt
@@ -66,6 +66,70 @@
# GFX1250-REAL16: v_tanh_bf16_e64 v5.l, v128.h op_sel:[1,0] ; encoding: [0x05,0x08,0xca,0xd5,0x80,0x01,0x00,0x00]
# GFX1250-FAKE16: v_tanh_bf16_e64 v5, v128 ; encoding: [0x05,0x00,0xca,0xd5,0x80,0x01,0x00,0x00]
+0xff,0x81,0xf9,0xd5,0xff,0x00,0x00,0x38,0x00,0x80,0x00,0x00
+# GFX1250-REAL16: v_rcp_bf16_e64 v255.l, -|0x8000| clamp div:2 ; encoding: [0xff,0x81,0xf9,0xd5,0xff,0x00,0x00,0x38,0x00,0x80,0x00,0x00]
+# GFX1250-FAKE16: v_rcp_bf16_e64 v255, -|0x8000| clamp div:2 ; encoding: [0xff,0x81,0xf9,0xd5,0xff,0x00,0x00,0x38,0x00,0x80,0x00,0x00]
+
+0x05,0x00,0xf9,0xd5,0xc1,0x00,0x00,0x00
+# GFX1250-REAL16: v_rcp_bf16_e64 v5.l, -1 ; encoding: [0x05,0x00,0xf9,0xd5,0xc1,0x00,0x00,0x00]
+# GFX1250-FAKE16: v_rcp_bf16_e64 v5, -1 ; encoding: [0x05,0x00,0xf9,0xd5,0xc1,0x00,0x00,0x00]
+
+0x05,0x00,0xf9,0xd5,0xf0,0x00,0x00,0x08
+# GFX1250-REAL16: v_rcp_bf16_e64 v5.l, 0.5 mul:2 ; encoding: [0x05,0x00,0xf9,0xd5,0xf0,0x00,0x00,0x08]
+# GFX1250-FAKE16: v_rcp_bf16_e64 v5, 0.5 mul:2 ; encoding: [0x05,0x00,0xf9,0xd5,0xf0,0x00,0x00,0x08]
+
+0x05,0x00,0xf9,0xd5,0x7f,0x00,0x00,0x00
+# GFX1250-REAL16: v_rcp_bf16_e64 v5.l, exec_hi ; encoding: [0x05,0x00,0xf9,0xd5,0x7f,0x00,0x00,0x00]
+# GFX1250-FAKE16: v_rcp_bf16_e64 v5, exec_hi ; encoding: [0x05,0x00,0xf9,0xd5,0x7f,0x00,0x00,0x00]
+
+0x05,0x00,0xf9,0xd5,0x7e,0x00,0x00,0x00
+# GFX1250-REAL16: v_rcp_bf16_e64 v5.l, exec_lo ; encoding: [0x05,0x00,0xf9,0xd5,0x7e,0x00,0x00,0x00]
+# GFX1250-FAKE16: v_rcp_bf16_e64 v5, exec_lo ; encoding: [0x05,0x00,0xf9,0xd5,0x7e,0x00,0x00,0x00]
+
+0x05,0x00,0xf9,0xd5,0x7d,0x00,0x00,0x00
+# GFX1250-REAL16: v_rcp_bf16_e64 v5.l, m0 ; encoding: [0x05,0x00,0xf9,0xd5,0x7d,0x00,0x00,0x00]
+# GFX1250-FAKE16: v_rcp_bf16_e64 v5, m0 ; encoding: [0x05,0x00,0xf9,0xd5,0x7d,0x00,0x00,0x00]
+
+0x05,0x00,0xf9,0xd5,0x7c,0x00,0x00,0x00
+# GFX1250-REAL16: v_rcp_bf16_e64 v5.l, null ; encoding: [0x05,0x00,0xf9,0xd5,0x7c,0x00,0x00,0x00]
+# GFX1250-FAKE16: v_rcp_bf16_e64 v5, null ; encoding: [0x05,0x00,0xf9,0xd5,0x7c,0x00,0x00,0x00]
+
+0x05,0x00,0xf9,0xd5,0x01,0x00,0x00,0x00
+# GFX1250-REAL16: v_rcp_bf16_e64 v5.l, s1 ; encoding: [0x05,0x00,0xf9,0xd5,0x01,0x00,0x00,0x00]
+# GFX1250-FAKE16: v_rcp_bf16_e64 v5, s1 ; encoding: [0x05,0x00,0xf9,0xd5,0x01,0x00,0x00,0x00]
+
+0x05,0x00,0xf9,0xd5,0x69,0x00,0x00,0x00
+# GFX1250-REAL16: v_rcp_bf16_e64 v5.l, s105 ; encoding: [0x05,0x00,0xf9,0xd5,0x69,0x00,0x00,0x00]
+# GFX1250-FAKE16: v_rcp_bf16_e64 v5, s105 ; encoding: [0x05,0x00,0xf9,0xd5,0x69,0x00,0x00,0x00]
+
+0x05,0x00,0xf9,0xd5,0xfd,0x00,0x00,0x10
+# GFX1250-REAL16: v_rcp_bf16_e64 v5.l, src_scc mul:4 ; encoding: [0x05,0x00,0xf9,0xd5,0xfd,0x00,0x00,0x10]
+# GFX1250-FAKE16: v_rcp_bf16_e64 v5, src_scc mul:4 ; encoding: [0x05,0x00,0xf9,0xd5,0xfd,0x00,0x00,0x10]
+
+0x05,0x00,0xf9,0xd5,0x7b,0x00,0x00,0x00
+# GFX1250-REAL16: v_rcp_bf16_e64 v5.l, ttmp15 ; encoding: [0x05,0x00,0xf9,0xd5,0x7b,0x00,0x00,0x00]
+# GFX1250-FAKE16: v_rcp_bf16_e64 v5, ttmp15 ; encoding: [0x05,0x00,0xf9,0xd5,0x7b,0x00,0x00,0x00]
+
+0x05,0x00,0xf9,0xd5,0x01,0x01,0x00,0x00
+# GFX1250-REAL16: v_rcp_bf16_e64 v5.l, v1.l ; encoding: [0x05,0x00,0xf9,0xd5,0x01,0x01,0x00,0x00]
+# GFX1250-FAKE16: v_rcp_bf16_e64 v5, v1 ; encoding: [0x05,0x00,0xf9,0xd5,0x01,0x01,0x00,0x00]
+
+0x05,0x00,0xf9,0xd5,0xff,0x01,0x00,0x00
+# GFX1250-REAL16: v_rcp_bf16_e64 v5.l, v255.l ; encoding: [0x05,0x00,0xf9,0xd5,0xff,0x01,0x00,0x00]
+# GFX1250-FAKE16: v_rcp_bf16_e64 v5, v255 ; encoding: [0x05,0x00,0xf9,0xd5,0xff,0x01,0x00,0x00]
+
+0x05,0x00,0xf9,0xd5,0x6b,0x00,0x00,0x00
+# GFX1250-REAL16: v_rcp_bf16_e64 v5.l, vcc_hi ; encoding: [0x05,0x00,0xf9,0xd5,0x6b,0x00,0x00,0x00]
+# GFX1250-FAKE16: v_rcp_bf16_e64 v5, vcc_hi ; encoding: [0x05,0x00,0xf9,0xd5,0x6b,0x00,0x00,0x00]
+
+0x05,0x00,0xf9,0xd5,0x6a,0x00,0x00,0x00
+# GFX1250-REAL16: v_rcp_bf16_e64 v5.l, vcc_lo ; encoding: [0x05,0x00,0xf9,0xd5,0x6a,0x00,0x00,0x00]
+# GFX1250-FAKE16: v_rcp_bf16_e64 v5, vcc_lo ; encoding: [0x05,0x00,0xf9,0xd5,0x6a,0x00,0x00,0x00]
+
+0x05,0x48,0xf9,0xd5,0x80,0x01,0x00,0x00
+# GFX1250-REAL16: v_rcp_bf16_e64 v5.h, v128.h op_sel:[1,1] ; encoding: [0x05,0x48,0xf9,0xd5,0x80,0x01,0x00,0x00]
+# GFX1250-FAKE16: v_rcp_bf16_e64 v5, v128 ; encoding: [0x05,0x00,0xf9,0xd5,0x80,0x01,0x00,0x00]
+
0x01,0x00,0xed,0xd5,0x83,0x00,0x00,0x00
# GFX1250: v_cvt_f32_bf8_e64 v1, 3 ; encoding: [0x01,0x00,0xed,0xd5,0x83,0x00,0x00,0x00]
diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop3_from_vop1_dpp16.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop3_from_vop1_dpp16.txt
index bae8909..ae51033 100644
--- a/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop3_from_vop1_dpp16.txt
+++ b/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop3_from_vop1_dpp16.txt
@@ -2,6 +2,66 @@
# RUN: llvm-mc -triple=amdgcn -mcpu=gfx1250 -mattr=+real-true16 -disassemble -show-encoding < %s | FileCheck -check-prefixes=GFX1250-REAL16 %s
# RUN: llvm-mc -triple=amdgcn -mcpu=gfx1250 -mattr=-real-true16 -disassemble -show-encoding < %s | FileCheck -check-prefixes=GFX1250-FAKE16 %s
+0xff,0x81,0xf9,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30
+# GFX1250-REAL16: v_rcp_bf16_e64_dpp v255.l, -|v255.l| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xf9,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
+# GFX1250-FAKE16: v_rcp_bf16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xf9,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
+
+0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01
+# GFX1250-REAL16: v_rcp_bf16_e64_dpp v5.l, v1.l mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
+# GFX1250-FAKE16: v_rcp_bf16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]
+
+0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13
+# GFX1250-REAL16: v_rcp_bf16_e64_dpp v5.l, v1.l mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
+# GFX1250-FAKE16: v_rcp_bf16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]
+
+0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff
+# GFX1250-REAL16: v_rcp_bf16_e64_dpp v5.l, v1.l quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]
+# GFX1250-FAKE16: v_rcp_bf16_e64_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]
+
+0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff
+# GFX1250-REAL16: v_rcp_bf16_e64_dpp v5.l, v1.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+# GFX1250-FAKE16: v_rcp_bf16_e64_dpp v5, v1 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
+
+0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff
+# GFX1250-REAL16: v_rcp_bf16_e64_dpp v5.l, v1.l row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]
+# GFX1250-FAKE16: v_rcp_bf16_e64_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]
+
+0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff
+# GFX1250-REAL16: v_rcp_bf16_e64_dpp v5.l, v1.l row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]
+# GFX1250-FAKE16: v_rcp_bf16_e64_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]
+
+0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff
+# GFX1250-REAL16: v_rcp_bf16_e64_dpp v5.l, v1.l row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]
+# GFX1250-FAKE16: v_rcp_bf16_e64_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]
+
+0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff
+# GFX1250-REAL16: v_rcp_bf16_e64_dpp v5.l, v1.l row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]
+# GFX1250-FAKE16: v_rcp_bf16_e64_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]
+
+0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff
+# GFX1250-REAL16: v_rcp_bf16_e64_dpp v5.l, v1.l row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]
+# GFX1250-FAKE16: v_rcp_bf16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]
+
+0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff
+# GFX1250-REAL16: v_rcp_bf16_e64_dpp v5.l, v1.l row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]
+# GFX1250-FAKE16: v_rcp_bf16_e64_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]
+
+0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff
+# GFX1250-REAL16: v_rcp_bf16_e64_dpp v5.l, v1.l row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]
+# GFX1250-FAKE16: v_rcp_bf16_e64_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]
+
+0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff
+# GFX1250-REAL16: v_rcp_bf16_e64_dpp v5.l, v1.l row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]
+# GFX1250-FAKE16: v_rcp_bf16_e64_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]
+
+0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff
+# GFX1250-REAL16: v_rcp_bf16_e64_dpp v5.l, v1.l row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]
+# GFX1250-FAKE16: v_rcp_bf16_e64_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]
+
+0x05,0x48,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x80,0x1b,0x00,0xff
+# GFX1250-REAL16: v_rcp_bf16_e64_dpp v5.h, v128.h op_sel:[1,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x48,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x80,0x1b,0x00,0xff]
+# GFX1250-FAKE16: v_rcp_bf16_e64_dpp v5, v128 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0xf9,0xd5,0xfa,0x00,0x00,0x00,0x80,0x1b,0x00,0xff]
+
0xff,0x81,0xca,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30
# GFX1250-REAL16: v_tanh_bf16_e64_dpp v255.l, -|v255.l| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xca,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
# GFX1250-FAKE16: v_tanh_bf16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x81,0xca,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]
diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop3_from_vop1_dpp8.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop3_from_vop1_dpp8.txt
index 52f7932..fe62688 100644
--- a/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop3_from_vop1_dpp8.txt
+++ b/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop3_from_vop1_dpp8.txt
@@ -2,6 +2,26 @@
# RUN: llvm-mc -triple=amdgcn -mcpu=gfx1250 -mattr=+real-true16 -disassemble -show-encoding < %s | FileCheck -check-prefixes=GFX1250-REAL16 %s
# RUN: llvm-mc -triple=amdgcn -mcpu=gfx1250 -mattr=-real-true16 -disassemble -show-encoding < %s | FileCheck -check-prefixes=GFX1250-FAKE16 %s
+0xff,0x81,0xf9,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00
+# GFX1250-REAL16: v_rcp_bf16_e64_dpp v255.l, -|v255.l| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xf9,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
+# GFX1250-FAKE16: v_rcp_bf16_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xf9,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
+
+0x05,0x00,0xf9,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05
+# GFX1250-REAL16: v_rcp_bf16_e64_dpp v5.l, v1.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xf9,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+# GFX1250-FAKE16: v_rcp_bf16_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xf9,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
+
+0x05,0x00,0xf9,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05
+# GFX1250-REAL16: v_rcp_bf16_e64_dpp v5.l, v1.l mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xf9,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
+# GFX1250-FAKE16: v_rcp_bf16_e64_dpp v5, v1 mul:2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xf9,0xd5,0xe9,0x00,0x00,0x08,0x01,0x77,0x39,0x05]
+
+0x05,0x00,0xf9,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05
+# GFX1250-REAL16: v_rcp_bf16_e64_dpp v5.l, v1.l mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x00,0xf9,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
+# GFX1250-FAKE16: v_rcp_bf16_e64_dpp v5, v1 mul:4 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0x05,0x00,0xf9,0xd5,0xea,0x00,0x00,0x10,0x01,0x77,0x39,0x05]
+
+0x05,0x48,0xf9,0xd5,0xe9,0x00,0x00,0x00,0x80,0x77,0x39,0x05
+# GFX1250-REAL16: v_rcp_bf16_e64_dpp v5.h, v128.h op_sel:[1,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x48,0xf9,0xd5,0xe9,0x00,0x00,0x00,0x80,0x77,0x39,0x05]
+# GFX1250-FAKE16: v_rcp_bf16_e64_dpp v5, v128 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x05,0x00,0xf9,0xd5,0xe9,0x00,0x00,0x00,0x80,0x77,0x39,0x05]
+
0xff,0x81,0xca,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00
# GFX1250-REAL16: v_tanh_bf16_e64_dpp v255.l, -|v255.l| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xca,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
# GFX1250-FAKE16: v_tanh_bf16_e64_dpp v255, -|v255| clamp div:2 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xff,0x81,0xca,0xd5,0xe9,0x00,0x00,0x38,0xff,0x00,0x00,0x00]
