[llvm] [AMDGPU] Add the support for `v_cvt_f32_bf16 ` on gfx1250 (PR #145632)
via llvm-commits
llvm-commits at lists.llvm.org
Wed Jun 25 08:59:49 PDT 2025
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-amdgpu
@llvm/pr-subscribers-mc
Author: Shilei Tian (shiltian)
<details>
<summary>Changes</summary>
Co-authored-by: Shilei Tian <i@<!-- -->tianshilei.me>
---
Patch is 56.70 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/145632.diff
21 Files Affected:
- (modified) llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp (+12-3)
- (modified) llvm/lib/Target/AMDGPU/VOP1Instructions.td (+23-2)
- (added) llvm/test/MC/AMDGPU/gfx1250_asm_vop1-fake16.s (+47)
- (added) llvm/test/MC/AMDGPU/gfx1250_asm_vop1.s (+50)
- (added) llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp16-fake16.s (+59)
- (added) llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp16.s (+63)
- (added) llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp8-fake16.s (+15)
- (added) llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp8.s (+19)
- (added) llvm/test/MC/AMDGPU/gfx1250_asm_vop1_err.s (+16)
- (added) llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1-fake16.s (+77)
- (added) llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1.s (+80)
- (added) llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1_dpp16-fake16.s (+47)
- (added) llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1_dpp16.s (+51)
- (added) llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1_dpp8-fake16.s (+7)
- (added) llvm/test/MC/AMDGPU/gfx1250_asm_vop3_from_vop1_dpp8.s (+11)
- (added) llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop1.txt (+51)
- (added) llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop1_dpp16.txt (+48)
- (added) llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop1_dpp8.txt (+15)
- (added) llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop3_from_vop1.txt (+49)
- (added) llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop3_from_vop1_dpp16.txt (+51)
- (added) llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop3_from_vop1_dpp8.txt (+11)
``````````diff
diff --git a/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp b/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp
index 07a4292ef28bc..59c72fcbff18a 100644
--- a/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp
+++ b/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp
@@ -599,6 +599,11 @@ DecodeStatus AMDGPUDisassembler::getInstruction(MCInst &MI, uint64_t &Size,
DecW, Address, CS))
break;
+ if (isGFX1250() &&
+ tryDecodeInst(DecoderTableGFX125096, DecoderTableGFX1250_FAKE1696, MI,
+ DecW, Address, CS))
+ break;
+
if (isGFX12() &&
tryDecodeInst(DecoderTableGFX1296, DecoderTableGFX12_FAKE1696, MI,
DecW, Address, CS))
@@ -661,6 +666,11 @@ DecodeStatus AMDGPUDisassembler::getInstruction(MCInst &MI, uint64_t &Size,
if (isGFX10() && tryDecodeInst(DecoderTableGFX1064, MI, QW, Address, CS))
break;
+ if (isGFX1250() &&
+ tryDecodeInst(DecoderTableGFX125064, DecoderTableGFX1250_FAKE1664, MI,
+ QW, Address, CS))
+ break;
+
if (isGFX12() &&
tryDecodeInst(DecoderTableGFX1264, DecoderTableGFX12_FAKE1664, MI, QW,
Address, CS))
@@ -717,10 +727,9 @@ DecodeStatus AMDGPUDisassembler::getInstruction(MCInst &MI, uint64_t &Size,
Address, CS))
break;
- // FIXME: Should use DecoderTableGFX1250_FAKE1632, but it is not generated
- // yet.
if (isGFX1250() &&
- tryDecodeInst(DecoderTableGFX125032, MI, DW, Address, CS))
+ tryDecodeInst(DecoderTableGFX125032, DecoderTableGFX1250_FAKE1632, MI,
+ DW, Address, CS))
break;
if (isGFX12() &&
diff --git a/llvm/lib/Target/AMDGPU/VOP1Instructions.td b/llvm/lib/Target/AMDGPU/VOP1Instructions.td
index 02b912bcfb9e0..4ef7a34261b6f 100644
--- a/llvm/lib/Target/AMDGPU/VOP1Instructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP1Instructions.td
@@ -259,6 +259,12 @@ foreach vt = Reg32Types.types in {
>;
}
+let HasOMod = 0, HasClamp = 0 in {
+ def VOPProfile_CVT_F32_BF16_gfx1250_t16 : VOPProfile_True16 <VOP_F32_BF16>;
+ let HasOpSel = 1, EmitDstSel = 0 in
+ def VOPProfile_CVT_F32_BF16_gfx1250_fake16 : VOPProfile_Fake16 <VOP_F32_BF16>;
+} // End HasOMod = 0, HasClamp = 0
+
let isReMaterializable = 1 in {
let SchedRW = [WriteDoubleCvt] in {
// OMod clears exceptions when set in this instruction
@@ -309,8 +315,14 @@ let OtherPredicates = [UseRealTrue16Insts] in
let OtherPredicates = [UseFakeTrue16Insts] in
defm V_CVT_F32_F16_fake16 : VOP1Inst <"v_cvt_f32_f16_fake16", VOPProfile_Fake16<VOP_F32_F16>, any_fpextend>;
-let SubtargetPredicate = HasBF16ConversionInsts in
-defm V_CVT_F32_BF16 : VOP1Inst_t16 <"v_cvt_f32_bf16", VOP_F32_BF16>;
+let SubtargetPredicate = HasGFX950Insts, OtherPredicates = [HasBF16ConversionInsts] in {
+ defm V_CVT_F32_BF16 : VOP1Inst_t16 <"v_cvt_f32_bf16", VOP_F32_BF16>;
+}
+let SubtargetPredicate = isGFX1250Plus, OtherPredicates = [HasBF16ConversionInsts] in {
+ defm V_CVT_F32_BF16_gfx1250 : VOP1Inst_t16_with_profiles <"v_cvt_f32_bf16_gfx1250", VOP_F32_BF16,
+ VOPProfile_CVT_F32_BF16_gfx1250_t16,
+ VOPProfile_CVT_F32_BF16_gfx1250_fake16>;
+}
let ReadsModeReg = 0, mayRaiseFPException = 0 in {
defm V_CVT_RPI_I32_F32 : VOP1Inst <"v_cvt_rpi_i32_f32", VOP_I32_F32, cvt_rpi_i32_f32>;
@@ -980,6 +992,13 @@ multiclass VOP1_Real_NO_DPP_OP_SEL_with_name<GFXGen Gen, bits<9> op,
VOP1_Real_e32_with_name<Gen, op, opName, asmName>,
VOP3_Real_with_name<Gen, {0, 1, 1, op{6-0}}, opName, asmName>;
+multiclass VOP1_Real_FULL_t16_and_fake16_gfx1250<
+ bits<9> op, string asmName = !tolower(NAME), string opName = NAME> {
+ defm opName#"_t16" :
+ VOP1_Real_FULL_with_name<GFX1250Gen, op, opName#"_t16", asmName>;
+ defm opName#"_fake16":
+ VOP1_Real_FULL_with_name<GFX1250Gen, op, opName#"_fake16", asmName>;
+}
defm V_CVT_F32_FP8 : VOP1_Real_FULL_with_name<GFX12Gen, 0x06c, "V_CVT_F32_FP8_OP_SEL", "v_cvt_f32_fp8">;
defm V_CVT_F32_BF8 : VOP1_Real_FULL_with_name<GFX12Gen, 0x06d, "V_CVT_F32_BF8_OP_SEL", "v_cvt_f32_bf8">;
@@ -1042,6 +1061,8 @@ defm V_CVT_NORM_U16_F16 : VOP1_Real_FULL_t16_and_fake16_gfx11_gfx12<0x064>;
defm V_CVT_F16_F32 : VOP1_Real_FULL_t16_and_fake16_gfx11_gfx12<0x00a>;
defm V_CVT_F32_F16 : VOP1_Real_FULL_t16_and_fake16_gfx11_gfx12<0x00b>;
+defm V_CVT_F32_BF16 : VOP1_Real_FULL_t16_and_fake16_gfx1250<0x072, "v_cvt_f32_bf16", "V_CVT_F32_BF16_gfx1250">;
+
//===----------------------------------------------------------------------===//
// GFX10.
//===----------------------------------------------------------------------===//
diff --git a/llvm/test/MC/AMDGPU/gfx1250_asm_vop1-fake16.s b/llvm/test/MC/AMDGPU/gfx1250_asm_vop1-fake16.s
new file mode 100644
index 0000000000000..9b2e506d4e043
--- /dev/null
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop1-fake16.s
@@ -0,0 +1,47 @@
+// NOTE: Assertions have been autogenerated by utils/update_mc_test_checks.py UTC_ARGS: --version 5
+// RUN: llvm-mc -triple=amdgcn -mcpu=gfx1250 -mattr=-real-true16 -show-encoding %s | FileCheck --check-prefix=GFX1250 %s
+
+v_cvt_f32_bf16 v5, v1
+// GFX1250: v_cvt_f32_bf16_e32 v5, v1 ; encoding: [0x01,0xe5,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, v127
+// GFX1250: v_cvt_f32_bf16_e32 v5, v127 ; encoding: [0x7f,0xe5,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, s1
+// GFX1250: v_cvt_f32_bf16_e32 v5, s1 ; encoding: [0x01,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, s105
+// GFX1250: v_cvt_f32_bf16_e32 v5, s105 ; encoding: [0x69,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, vcc_lo
+// GFX1250: v_cvt_f32_bf16_e32 v5, vcc_lo ; encoding: [0x6a,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, vcc_hi
+// GFX1250: v_cvt_f32_bf16_e32 v5, vcc_hi ; encoding: [0x6b,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, ttmp15
+// GFX1250: v_cvt_f32_bf16_e32 v5, ttmp15 ; encoding: [0x7b,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, m0
+// GFX1250: v_cvt_f32_bf16_e32 v5, m0 ; encoding: [0x7d,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, exec_lo
+// GFX1250: v_cvt_f32_bf16_e32 v5, exec_lo ; encoding: [0x7e,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, exec_hi
+// GFX1250: v_cvt_f32_bf16_e32 v5, exec_hi ; encoding: [0x7f,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, null
+// GFX1250: v_cvt_f32_bf16_e32 v5, null ; encoding: [0x7c,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, -1
+// GFX1250: v_cvt_f32_bf16_e32 v5, -1 ; encoding: [0xc1,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, 0.5
+// GFX1250: v_cvt_f32_bf16_e32 v5, 0.5 ; encoding: [0xf0,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, src_scc
+// GFX1250: v_cvt_f32_bf16_e32 v5, src_scc ; encoding: [0xfd,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v127, 0x8000
+// GFX1250: v_cvt_f32_bf16_e32 v127, 0x8000 ; encoding: [0xff,0xe4,0xfe,0x7e,0x00,0x80,0x00,0x00]
diff --git a/llvm/test/MC/AMDGPU/gfx1250_asm_vop1.s b/llvm/test/MC/AMDGPU/gfx1250_asm_vop1.s
new file mode 100644
index 0000000000000..1d1badc4f009b
--- /dev/null
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop1.s
@@ -0,0 +1,50 @@
+// NOTE: Assertions have been autogenerated by utils/update_mc_test_checks.py UTC_ARGS: --version 5
+// RUN: llvm-mc -triple=amdgcn -mcpu=gfx1250 -mattr=+real-true16 -show-encoding %s | FileCheck --check-prefix=GFX1250 %s
+
+v_cvt_f32_bf16 v5, v1
+// GFX1250: v_cvt_f32_bf16_e32 v5, v1 ; encoding: [0x01,0xe5,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, v127
+// GFX1250: v_cvt_f32_bf16_e32 v5, v127 ; encoding: [0x7f,0xe5,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, s1
+// GFX1250: v_cvt_f32_bf16_e32 v5, s1 ; encoding: [0x01,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, s105
+// GFX1250: v_cvt_f32_bf16_e32 v5, s105 ; encoding: [0x69,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, vcc_lo
+// GFX1250: v_cvt_f32_bf16_e32 v5, vcc_lo ; encoding: [0x6a,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, vcc_hi
+// GFX1250: v_cvt_f32_bf16_e32 v5, vcc_hi ; encoding: [0x6b,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, ttmp15
+// GFX1250: v_cvt_f32_bf16_e32 v5, ttmp15 ; encoding: [0x7b,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, m0
+// GFX1250: v_cvt_f32_bf16_e32 v5, m0 ; encoding: [0x7d,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, exec_lo
+// GFX1250: v_cvt_f32_bf16_e32 v5, exec_lo ; encoding: [0x7e,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, exec_hi
+// GFX1250: v_cvt_f32_bf16_e32 v5, exec_hi ; encoding: [0x7f,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, null
+// GFX1250: v_cvt_f32_bf16_e32 v5, null ; encoding: [0x7c,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, -1
+// GFX1250: v_cvt_f32_bf16_e32 v5, -1 ; encoding: [0xc1,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, 0.5
+// GFX1250: v_cvt_f32_bf16_e32 v5, 0.5 ; encoding: [0xf0,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v5, src_scc
+// GFX1250: v_cvt_f32_bf16_e32 v5, src_scc ; encoding: [0xfd,0xe4,0x0a,0x7e]
+
+v_cvt_f32_bf16 v127, 0x8000
+// GFX1250: v_cvt_f32_bf16_e32 v127, 0x8000 ; encoding: [0xff,0xe4,0xfe,0x7e,0x00,0x80,0x00,0x00]
+
+v_cvt_f32_bf16 v5, v1.h
+// GFX1250: v_cvt_f32_bf16_e32 v5, v1.h ; encoding: [0x81,0xe5,0x0a,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
new file mode 100644
index 0000000000000..18b6c91e2fb8c
--- /dev/null
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp16-fake16.s
@@ -0,0 +1,59 @@
+// NOTE: Assertions have been autogenerated by utils/update_mc_test_checks.py UTC_ARGS: --version 5
+// RUN: llvm-mc -triple=amdgcn -mcpu=gfx1250 -mattr=-real-true16 -show-encoding %s | FileCheck --check-prefixes=GFX1250 %s
+// RUN: not llvm-mc -triple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 -show-encoding %s 2>&1 | FileCheck --check-prefix=GFX12-ERR --implicit-check-not=error: --strict-whitespace %s
+
+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
+
+v_cvt_f32_bf16 v5, v1 quad_perm:[0,1,2,3]
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0xe4,0x00,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v5, v1 row_mirror
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x40,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v5, v1 row_half_mirror
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x41,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v5, v1 row_shl:1
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x01,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v5, v1 row_shl:15
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x0f,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v5, v1 row_shr:1
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x11,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v5, v1 row_shr:15
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x1f,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v5, v1 row_ror:1
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x21,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v5, v1 row_ror:15
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x2f,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x50,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x5f,0x01,0x01]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x60,0x09,0x13]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX1250: v_cvt_f32_bf16_dpp v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xe4,0xfe,0x7e,0x7f,0x6f,0x35,0x30]
+// 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
new file mode 100644
index 0000000000000..0f71c46eb4725
--- /dev/null
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp16.s
@@ -0,0 +1,63 @@
+// NOTE: Assertions have been autogenerated by utils/update_mc_test_checks.py UTC_ARGS: --version 5
+// RUN: llvm-mc -triple=amdgcn -mcpu=gfx1250 -mattr=+real-true16 -show-encoding %s | FileCheck --check-prefixes=GFX1250 %s
+// RUN: not llvm-mc -triple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 -show-encoding %s 2>&1 | FileCheck --check-prefix=GFX12-ERR --implicit-check-not=error: --strict-whitespace %s
+
+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
+
+v_cvt_f32_bf16 v5, v1 quad_perm:[0,1,2,3]
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0xe4,0x00,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v5, v1 row_mirror
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x40,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v5, v1 row_half_mirror
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x41,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v5, v1 row_shl:1
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x01,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v5, v1 row_shl:15
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x0f,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v5, v1 row_shr:1
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x11,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v5, v1 row_shr:15
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x1f,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v5, v1 row_ror:1
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x21,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v5, v1 row_ror:15
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x2f,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x50,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x5f,0x01,0x01]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 ; encoding: [0xfa,0xe4,0x0a,0x7e,0x01,0x60,0x09,0x13]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX1250: v_cvt_f32_bf16_dpp v127, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0xe4,0xfe,0x7e,0x7f,0x6f,0x35,0x30]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v5, v1.h quad_perm:[3,2,1,0]
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1.h quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0xe4,0x0a,0x7e,0x81,0x1b,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
new file mode 100644
index 0000000000000..c973022dbeca6
--- /dev/null
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp8-fake16.s
@@ -0,0 +1,15 @@
+// NOTE: Assertions have been autogenerated by utils/update_mc_test_checks.py UTC_ARGS: --version 5
+// RUN: llvm-mc -triple=amdgcn -mcpu=gfx1250 -mattr=-real-true16 -show-encoding %s | FileCheck --check-prefixes=GFX1250 %s
+// RUN: not llvm-mc -triple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 -show-encoding %s 2>&1 | FileCheck --check-prefix=GFX12-ERR --implicit-check-not=error: --strict-whitespace %s
+
+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
+
+v_cvt_f32_bf16 v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1
+// GFX1250: v_cvt_f32_bf16_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0] fi:1 ; encoding: [0xea,0xe4,0x0a,0x7e,0x01,0x77,0x39,0x05]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_f32_bf16 v127, v127 dpp8:[0,0,0,0,0,0,0,0] fi:0
+// GFX1250: v_cvt_f32_bf16_dpp v127, v127 dpp8:[0,0,0,0,0,0,0,0] ; encoding: [0xe9,0xe4,0xfe,0x7e,0x7f,0x00,0x00,0x00]
+// 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
new file mode 100644
index 0000000000000..8a5f3cba2fbc0
--- /dev/null
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop1_dpp8.s
@@ -0,0 +1,19 @@
+// NOTE: Assertions have been autogenerated by utils/update_mc_test_checks.py UTC_ARGS: --version 5
+// RUN: llvm-mc -triple=amdgcn -mcpu=gfx1250 -mattr=+real-true16 -show-encoding %s | FileCheck --check-prefixes=GFX1250 %s
+// RUN: not llvm-mc -triple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 -show-encoding %s 2>&1 | FileCheck --check-prefix=GFX12-ERR --implicit-check-not=error: --strict-whitespace %s
+
+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]
...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/145632
More information about the llvm-commits
mailing list