[AMDGPU] gfx1251 VOP1 dpp support #159637
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This stack of pull requests is managed by Graphite. Learn more about stacking. |
|
@llvm/pr-subscribers-backend-amdgpu Author: Stanislav Mekhanoshin (rampitec) ChangesPatch is 30.39 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/159637.diff 8 Files Affected:
diff --git a/llvm/lib/Target/AMDGPU/VOP1Instructions.td b/llvm/lib/Target/AMDGPU/VOP1Instructions.td
index f816d7de27ee4..6230c17e20804 100644
--- a/llvm/lib/Target/AMDGPU/VOP1Instructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP1Instructions.td
@@ -212,6 +212,11 @@ def VOP_I16_F16_SPECIAL_OMOD_fake16 : VOPProfile_Fake16<VOP_I16_F16> {
}
+def VOP_F64_F64_NO_DPP : VOPProfile <[f64, f64, untyped, untyped]> {
+ let HasExtVOP3DPP = 0;
+ let HasExt64BitDPP = 0;
+}
+
//===----------------------------------------------------------------------===//
// VOP1 Instructions
//===----------------------------------------------------------------------===//
@@ -344,9 +349,9 @@ defm V_SQRT_F32 : VOP1Inst <"v_sqrt_f32", VOP_F32_F32, int_amdgcn_sqrt>;
} // End TRANS = 1, SchedRW = [WriteTrans32]
let TRANS = 1, SchedRW = [WriteTrans64] in {
-defm V_RCP_F64 : VOP1Inst <"v_rcp_f64", VOP_F64_F64, AMDGPUrcp>;
-defm V_RSQ_F64 : VOP1Inst <"v_rsq_f64", VOP_F64_F64, AMDGPUrsq>;
-defm V_SQRT_F64 : VOP1Inst <"v_sqrt_f64", VOP_F64_F64, int_amdgcn_sqrt>;
+defm V_RCP_F64 : VOP1Inst <"v_rcp_f64", VOP_F64_F64_NO_DPP, AMDGPUrcp>;
+defm V_RSQ_F64 : VOP1Inst <"v_rsq_f64", VOP_F64_F64_NO_DPP, AMDGPUrsq>;
+defm V_SQRT_F64 : VOP1Inst <"v_sqrt_f64", VOP_F64_F64_NO_DPP, int_amdgcn_sqrt>;
} // End TRANS = 1, SchedRW = [WriteTrans64]
let TRANS = 1, SchedRW = [WriteTrans32] in {
@@ -1025,6 +1030,11 @@ multiclass VOP1_Real_FULL_with_name<GFXGen Gen, bits<9> op, string opName,
multiclass VOP1_Real_NO_DPP<GFXGen Gen, bits<9> op> :
VOP1_Real_e32<Gen, op>, VOP1_Real_e64<Gen, op>;
+multiclass VOP1_Real_with_DPP16<GFXGen Gen, bits<9> op> :
+ VOP1_Real_NO_DPP<Gen, op>,
+ VOP1_Real_dpp<Gen, op>,
+ VOP3_Real_dpp_Base<Gen, {0, 1, 1, op{6-0}}>;
+
multiclass VOP1_Real_FULL_t16_gfx11_gfx12<bits<9> op, string asmName,
string opName = NAME> :
VOP1_Real_FULL_with_name<GFX11Gen, op, opName, asmName>,
@@ -1057,6 +1067,11 @@ multiclass VOP1_Real_FULL_t16_and_fake16_gfx1250<
VOP1_Real_FULL_with_name<GFX1250Gen, op, opName#"_fake16", asmName>;
}
+multiclass VOP1_Real_FULL_with_name_gfx11_gfx12_not_gfx1250<bits<9> op, string opName,
+ string asmName> :
+ VOP1_Real_FULL_with_name<GFX11Gen, op, opName, asmName>,
+ VOP1_Real_FULL_with_name<GFX12Not12_50Gen, op, opName, asmName>;
+
multiclass VOP1_Real_OpSelIsDPP_gfx1250<bits<9> op> : VOP1_Real_e32<GFX1250Gen, op> {
defvar ps = !cast<VOP_Pseudo>(NAME#"_e64");
def _e64_gfx1250 :
@@ -1064,10 +1079,10 @@ multiclass VOP1_Real_OpSelIsDPP_gfx1250<bits<9> op> : VOP1_Real_e32<GFX1250Gen,
VOP3OpSelIsDPP_gfx12<{0, 1, 1, op{6-0}}, ps.Pfl>;
}
-defm V_CVT_F32_FP8 : VOP1_Real_FULL_with_name<GFX12Not12_50Gen, 0x06c, "V_CVT_F32_FP8_OP_SEL", "v_cvt_f32_fp8">;
-defm V_CVT_F32_FP8 : VOP1_Real_FULL_with_name<GFX1250Gen, 0x06c, "V_CVT_F32_FP8_gfx1250", "v_cvt_f32_fp8">;
+defm V_CVT_F32_FP8 : VOP1_Real_FULL_with_name_gfx11_gfx12_not_gfx1250<0x06c, "V_CVT_F32_FP8_OP_SEL", "v_cvt_f32_fp8">;
+defm V_CVT_F32_FP8 : VOP1_Real_FULL_with_name<GFX1250Gen, 0x06c, "V_CVT_F32_FP8_gfx1250", "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">;
+defm V_CVT_F32_BF8 : VOP1_Real_FULL_with_name<GFX12Gen, 0x06d, "V_CVT_F32_BF8_OP_SEL", "v_cvt_f32_bf8">;
defm V_CVT_PK_F32_FP8_fake16 : VOP1_Real_e32_with_name<GFX12Gen, 0x06e, "V_CVT_PK_F32_FP8_fake16", "v_cvt_pk_f32_fp8">;
defm V_CVT_PK_F32_FP8_t16 : VOP1_Real_e32_with_name<GFX12Gen, 0x06e, "V_CVT_PK_F32_FP8_t16", "v_cvt_pk_f32_fp8">;
@@ -1252,17 +1267,17 @@ let AssemblerPredicate = isGFX7Only, DecoderNamespace = "GFX7" in {
multiclass VOP1_Real_gfx7<bits<9> op> :
VOP1_Real_e32_gfx7<op>, VOP1_Real_e64_gfx7<op>;
-multiclass VOP1_Real_gfx7_gfx10_NO_DPP_gfx11_gfx12<bits<9> op> :
+multiclass VOP1_Real_gfx7_gfx10_NO_DPP_gfx11_with_DPP16_gfx12<bits<9> op> :
VOP1_Real_gfx7<op>, VOP1_Real_gfx10<op>, VOP1_Real_NO_DPP<GFX11Gen, op>,
- VOP1_Real_NO_DPP<GFX12Gen, op>;
+ VOP1_Real_with_DPP16<GFX12Gen, op>;
defm V_LOG_LEGACY_F32 : VOP1_Real_gfx7<0x045>;
defm V_EXP_LEGACY_F32 : VOP1_Real_gfx7<0x046>;
-defm V_TRUNC_F64 : VOP1_Real_gfx7_gfx10_NO_DPP_gfx11_gfx12<0x017>;
-defm V_CEIL_F64 : VOP1_Real_gfx7_gfx10_NO_DPP_gfx11_gfx12<0x018>;
-defm V_RNDNE_F64 : VOP1_Real_gfx7_gfx10_NO_DPP_gfx11_gfx12<0x019>;
-defm V_FLOOR_F64 : VOP1_Real_gfx7_gfx10_NO_DPP_gfx11_gfx12<0x01a>;
+defm V_TRUNC_F64 : VOP1_Real_gfx7_gfx10_NO_DPP_gfx11_with_DPP16_gfx12<0x017>;
+defm V_CEIL_F64 : VOP1_Real_gfx7_gfx10_NO_DPP_gfx11_with_DPP16_gfx12<0x018>;
+defm V_RNDNE_F64 : VOP1_Real_gfx7_gfx10_NO_DPP_gfx11_with_DPP16_gfx12<0x019>;
+defm V_FLOOR_F64 : VOP1_Real_gfx7_gfx10_NO_DPP_gfx11_with_DPP16_gfx12<0x01a>;
//===----------------------------------------------------------------------===//
// GFX6, GFX7, GFX10, GFX11, GFX12
@@ -1300,6 +1315,10 @@ multiclass VOP1_Real_gfx6_gfx7_gfx10_NO_DPP_gfx11_gfx12<bits<9> op> :
VOP1_Real_gfx6_gfx7_gfx10<op>, VOP1_Real_NO_DPP<GFX11Gen, op>,
VOP1_Real_NO_DPP<GFX12Gen, op>;
+multiclass VOP1_Real_gfx6_gfx7_gfx10_NO_DPP_gfx11_with_DPP16_gfx12<bits<9> op> :
+ VOP1_Real_gfx6_gfx7_gfx10<op>, VOP1_Real_NO_DPP<GFX11Gen, op>,
+ VOP1_Real_with_DPP16<GFX12Gen, op>;
+
multiclass VOP1Only_Real_gfx6_gfx7_gfx10_gfx11_gfx12<bits<9> op> :
VOP1Only_Real_gfx6_gfx7<op>, VOP1Only_Real_gfx10_gfx11_gfx12<op>;
@@ -1314,8 +1333,8 @@ defm V_RSQ_CLAMP_F64 : VOP1_Real_gfx6_gfx7<0x032>;
defm V_NOP : VOP1_Real_gfx6_gfx7_gfx10_NO_DPP_gfx11_gfx12<0x000>;
defm V_MOV_B32 : VOP1_Real_gfx6_gfx7_gfx10_FULL_gfx11_gfx12<0x001>;
defm V_READFIRSTLANE_B32 : VOP1Only_Real_gfx6_gfx7_gfx10_gfx11_gfx12<0x002>;
-defm V_CVT_I32_F64 : VOP1_Real_gfx6_gfx7_gfx10_NO_DPP_gfx11_gfx12<0x003>;
-defm V_CVT_F64_I32 : VOP1_Real_gfx6_gfx7_gfx10_NO_DPP_gfx11_gfx12<0x004>;
+defm V_CVT_I32_F64 : VOP1_Real_gfx6_gfx7_gfx10_NO_DPP_gfx11_with_DPP16_gfx12<0x003>;
+defm V_CVT_F64_I32 : VOP1_Real_gfx6_gfx7_gfx10_NO_DPP_gfx11_with_DPP16_gfx12<0x004>;
defm V_CVT_F32_I32 : VOP1_Real_gfx6_gfx7_gfx10_FULL_gfx11_gfx12<0x005>;
defm V_CVT_F32_U32 : VOP1_Real_gfx6_gfx7_gfx10_FULL_gfx11_gfx12<0x006>;
defm V_CVT_U32_F32 : VOP1_Real_gfx6_gfx7_gfx10_FULL_gfx11_gfx12<0x007>;
@@ -1325,14 +1344,14 @@ defm V_CVT_F32_F16 : VOP1_Real_gfx6_gfx7_gfx10<0x00b>;
defm V_CVT_RPI_I32_F32 : VOP1_Real_gfx6_gfx7_gfx10<0x00c>;
defm V_CVT_FLR_I32_F32 : VOP1_Real_gfx6_gfx7_gfx10<0x00d>;
defm V_CVT_OFF_F32_I4 : VOP1_Real_gfx6_gfx7_gfx10_FULL_gfx11_gfx12<0x00e>;
-defm V_CVT_F32_F64 : VOP1_Real_gfx6_gfx7_gfx10_NO_DPP_gfx11_gfx12<0x00f>;
-defm V_CVT_F64_F32 : VOP1_Real_gfx6_gfx7_gfx10_NO_DPP_gfx11_gfx12<0x010>;
+defm V_CVT_F32_F64 : VOP1_Real_gfx6_gfx7_gfx10_NO_DPP_gfx11_with_DPP16_gfx12<0x00f>;
+defm V_CVT_F64_F32 : VOP1_Real_gfx6_gfx7_gfx10_NO_DPP_gfx11_with_DPP16_gfx12<0x010>;
defm V_CVT_F32_UBYTE0 : VOP1_Real_gfx6_gfx7_gfx10_FULL_gfx11_gfx12<0x011>;
defm V_CVT_F32_UBYTE1 : VOP1_Real_gfx6_gfx7_gfx10_FULL_gfx11_gfx12<0x012>;
defm V_CVT_F32_UBYTE2 : VOP1_Real_gfx6_gfx7_gfx10_FULL_gfx11_gfx12<0x013>;
defm V_CVT_F32_UBYTE3 : VOP1_Real_gfx6_gfx7_gfx10_FULL_gfx11_gfx12<0x014>;
-defm V_CVT_U32_F64 : VOP1_Real_gfx6_gfx7_gfx10_NO_DPP_gfx11_gfx12<0x015>;
-defm V_CVT_F64_U32 : VOP1_Real_gfx6_gfx7_gfx10_NO_DPP_gfx11_gfx12<0x016>;
+defm V_CVT_U32_F64 : VOP1_Real_gfx6_gfx7_gfx10_NO_DPP_gfx11_with_DPP16_gfx12<0x015>;
+defm V_CVT_F64_U32 : VOP1_Real_gfx6_gfx7_gfx10_NO_DPP_gfx11_with_DPP16_gfx12<0x016>;
defm V_FRACT_F32 : VOP1_Real_gfx6_gfx7_gfx10_FULL_gfx11_gfx12<0x020>;
defm V_TRUNC_F32 : VOP1_Real_gfx6_gfx7_gfx10_FULL_gfx11_gfx12<0x021>;
defm V_CEIL_F32 : VOP1_Real_gfx6_gfx7_gfx10_FULL_gfx11_gfx12<0x022>;
@@ -1354,9 +1373,9 @@ defm V_BFREV_B32 : VOP1_Real_gfx6_gfx7_gfx10_FULL_gfx11_gfx12<0x038>;
defm V_FFBH_U32 : VOP1_Real_gfx6_gfx7_gfx10<0x039>;
defm V_FFBL_B32 : VOP1_Real_gfx6_gfx7_gfx10<0x03a>;
defm V_FFBH_I32 : VOP1_Real_gfx6_gfx7_gfx10<0x03b>;
-defm V_FREXP_EXP_I32_F64 : VOP1_Real_gfx6_gfx7_gfx10_NO_DPP_gfx11_gfx12<0x03c>;
-defm V_FREXP_MANT_F64 : VOP1_Real_gfx6_gfx7_gfx10_NO_DPP_gfx11_gfx12<0x03d>;
-defm V_FRACT_F64 : VOP1_Real_gfx6_gfx7_gfx10_NO_DPP_gfx11_gfx12<0x03e>;
+defm V_FREXP_EXP_I32_F64 : VOP1_Real_gfx6_gfx7_gfx10_NO_DPP_gfx11_with_DPP16_gfx12<0x03c>;
+defm V_FREXP_MANT_F64 : VOP1_Real_gfx6_gfx7_gfx10_NO_DPP_gfx11_with_DPP16_gfx12<0x03d>;
+defm V_FRACT_F64 : VOP1_Real_gfx6_gfx7_gfx10_NO_DPP_gfx11_with_DPP16_gfx12<0x03e>;
defm V_FREXP_EXP_I32_F32 : VOP1_Real_gfx6_gfx7_gfx10_FULL_gfx11_gfx12<0x03f>;
defm V_FREXP_MANT_F32 : VOP1_Real_gfx6_gfx7_gfx10_FULL_gfx11_gfx12<0x040>;
defm V_CLREXCP : VOP1_Real_gfx6_gfx7_gfx10<0x041>;
@@ -1410,7 +1429,9 @@ multiclass VOP1_Real_vi <bits<10> op> {
if !cast<VOP1_Pseudo>(NAME#"_e32").Pfl.HasExtDPP then
def _dpp_vi :
VOP_DPP_Real<!cast<VOP1_DPP_Pseudo>(NAME#"_dpp"), SIEncodingFamily.VI>,
- VOP1_DPPe<op{7-0}, !cast<VOP1_DPP_Pseudo>(NAME#"_dpp")>;
+ VOP1_DPPe<op{7-0}, !cast<VOP1_DPP_Pseudo>(NAME#"_dpp")> {
+ let AssemblerPredicate = isGFX8GFX9;
+ }
}
defm V_NOP : VOP1_Real_vi <0x0>;
diff --git a/llvm/test/CodeGen/AMDGPU/dpp64_combine.mir b/llvm/test/CodeGen/AMDGPU/dpp64_combine.mir
index 84da231c95a62..8094dbaf418b8 100644
--- a/llvm/test/CodeGen/AMDGPU/dpp64_combine.mir
+++ b/llvm/test/CodeGen/AMDGPU/dpp64_combine.mir
@@ -1,5 +1,6 @@
# RUN: llc -mtriple=amdgcn -mcpu=gfx90a -run-pass=gcn-dpp-combine -verify-machineinstrs -o - %s | FileCheck %s --check-prefix=GCN
# RUN: llc -mtriple=amdgcn -mcpu=gfx942 -run-pass=gcn-dpp-combine -verify-machineinstrs -o - %s | FileCheck %s --check-prefix=GCN
+# RUN: llc -mtriple=amdgcn -mcpu=gfx1251 -run-pass=gcn-dpp-combine -o - %s | FileCheck %s --check-prefix=GCN
---
# GCN-LABEL: name: dpp64_old_impdef
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mov.dpp.gfx1251.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mov.dpp.gfx1251.ll
new file mode 100644
index 0000000000000..7a2f8faae9e89
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mov.dpp.gfx1251.ll
@@ -0,0 +1,25 @@
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GCN,GFX12 %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1250 < %s | FileCheck -check-prefixes=GCN,GFX12 %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1251 < %s | FileCheck -check-prefixes=GCN,GFX1251 %s
+
+; GCN-LABEL: {{^}}mov_dpp64_test:
+; GCN: v_mov_b32_dpp v{{[0-9]+}}, v{{[0-9]+}} quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
+; GCN: v_mov_b32_dpp v{{[0-9]+}}, v{{[0-9]+}} quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
+define amdgpu_kernel void @mov_dpp64_test(ptr addrspace(1) %out, i64 %in1) {
+ %tmp0 = call i64 @llvm.amdgcn.mov.dpp.i64(i64 %in1, i32 1, i32 1, i32 1, i1 0) #0
+ store i64 %tmp0, ptr addrspace(1) %out
+ ret void
+}
+
+; GCN-LABEL: {{^}}mov_dpp64_row_share_test:
+; GFX12-COUNT-2: v_mov_b32_dpp v{{[0-9]+}}, v{{[0-9]+}} row_share:1 row_mask:0x1 bank_mask:0x1
+; GFX1251: v_mov_b64_dpp v[{{[0-9:]+}}], v[{{[0-9:]+}}] row_share:1 row_mask:0x1 bank_mask:0x1
+define amdgpu_kernel void @mov_dpp64_row_share_test(ptr addrspace(1) %out, i64 %in1) {
+ %tmp0 = call i64 @llvm.amdgcn.mov.dpp.i64(i64 %in1, i32 337, i32 1, i32 1, i1 0) #0
+ store i64 %tmp0, ptr addrspace(1) %out
+ ret void
+}
+
+declare i64 @llvm.amdgcn.mov.dpp.i64(i64, i32, i32, i32, i1) #0
+
+attributes #0 = { nounwind readnone convergent }
diff --git a/llvm/test/MC/AMDGPU/gfx1251_asm_vop1_dpp16.s b/llvm/test/MC/AMDGPU/gfx1251_asm_vop1_dpp16.s
new file mode 100644
index 0000000000000..bb1ccaf53ce32
--- /dev/null
+++ b/llvm/test/MC/AMDGPU/gfx1251_asm_vop1_dpp16.s
@@ -0,0 +1,98 @@
+// RUN: llvm-mc -triple=amdgcn -mcpu=gfx1251 -show-encoding %s | FileCheck --check-prefixes=GFX1251 %s
+// RUN: not llvm-mc -triple=amdgcn -mcpu=gfx1250 -show-encoding %s 2>&1 | FileCheck --check-prefix=GFX1250-ERR --implicit-check-not=error: --strict-whitespace %s
+
+v_mov_b64 v[4:5], v[2:3] row_share:0 row_mask:0xf bank_mask:0xf
+// GFX1251: v_mov_b64_dpp v[4:5], v[2:3] row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x3a,0x08,0x7e,0x02,0x50,0x01,0xff]
+// GFX1250-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: not a valid operand.
+// GFX1250-ERR-NEXT:{{^}}v_mov_b64 v[4:5], v[2:3] row_share:0 row_mask:0xf bank_mask:0xf
+// GFX1250-ERR-NEXT:{{^}} ^
+
+v_mov_b64 v[4:5], v[2:3] row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX1251: v_mov_b64_dpp v[4:5], v[2:3] row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0x3a,0x08,0x7e,0x02,0x5f,0x01,0x01]
+// GFX1250-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: not a valid operand.
+// GFX1250-ERR-NEXT:{{^}}v_mov_b64 v[4:5], v[2:3] row_share:15 row_mask:0x0 bank_mask:0x1
+// GFX1250-ERR-NEXT:{{^}} ^
+
+v_mov_b64 v[254:255], v[254:255] row_share:3 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX1251: v_mov_b64_dpp v[254:255], v[254:255] row_share:3 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xfa,0x3a,0xfc,0x7f,0xfe,0x53,0x05,0x30]
+// GFX1250-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: not a valid operand.
+// GFX1250-ERR-NEXT:{{^}}v_mov_b64 v[254:255], v[254:255] row_share:3 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
+// GFX1250-ERR-NEXT:{{^}} ^
+
+v_cvt_i32_f64 v2, v[4:5] row_share:1
+// GFX1251: v_cvt_i32_f64_dpp v2, v[4:5] row_share:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x06,0x04,0x7e,0x04,0x51,0x01,0xff]
+// GFX1250-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: not a valid operand.
+// GFX1250-ERR-NEXT:{{^}}v_cvt_i32_f64 v2, v[4:5] row_share:1
+// GFX1250-ERR-NEXT:{{^}} ^
+
+v_cvt_f64_i32 v[4:5], v2 row_share:1
+// GFX1251: v_cvt_f64_i32_dpp v[4:5], v2 row_share:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x08,0x08,0x7e,0x02,0x51,0x01,0xff]
+// GFX1250-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: not a valid operand.
+// GFX1250-ERR-NEXT:{{^}}v_cvt_f64_i32 v[4:5], v2 row_share:1
+// GFX1250-ERR-NEXT:{{^}} ^
+
+v_cvt_f32_f64 v2, v[4:5] row_share:1
+// GFX1251: v_cvt_f32_f64_dpp v2, v[4:5] row_share:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x1e,0x04,0x7e,0x04,0x51,0x01,0xff]
+// GFX1250-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: not a valid operand.
+// GFX1250-ERR-NEXT:{{^}}v_cvt_f32_f64 v2, v[4:5] row_share:1
+// GFX1250-ERR-NEXT:{{^}} ^
+
+v_cvt_f64_f32 v[4:5], v2 row_share:1
+// GFX1251: v_cvt_f64_f32_dpp v[4:5], v2 row_share:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x20,0x08,0x7e,0x02,0x51,0x01,0xff]
+// GFX1250-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: not a valid operand.
+// GFX1250-ERR-NEXT:{{^}}v_cvt_f64_f32 v[4:5], v2 row_share:1
+// GFX1250-ERR-NEXT:{{^}} ^
+
+v_cvt_u32_f64 v2, v[4:5] row_share:1
+// GFX1251: v_cvt_u32_f64_dpp v2, v[4:5] row_share:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x2a,0x04,0x7e,0x04,0x51,0x01,0xff]
+// GFX1250-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: not a valid operand.
+// GFX1250-ERR-NEXT:{{^}}v_cvt_u32_f64 v2, v[4:5] row_share:1
+// GFX1250-ERR-NEXT:{{^}} ^
+
+v_cvt_f64_u32 v[4:5], v2 row_share:1
+// GFX1251: v_cvt_f64_u32_dpp v[4:5], v2 row_share:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x2c,0x08,0x7e,0x02,0x51,0x01,0xff]
+// GFX1250-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: not a valid operand.
+// GFX1250-ERR-NEXT:{{^}}v_cvt_f64_u32 v[4:5], v2 row_share:1
+// GFX1250-ERR-NEXT:{{^}} ^
+
+v_trunc_f64 v[2:3], v[4:5] row_share:1
+// GFX1251: v_trunc_f64_dpp v[2:3], v[4:5] row_share:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x2e,0x04,0x7e,0x04,0x51,0x01,0xff]
+// GFX1250-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: not a valid operand.
+// GFX1250-ERR-NEXT:{{^}}v_trunc_f64 v[2:3], v[4:5] row_share:1
+// GFX1250-ERR-NEXT:{{^}} ^
+
+v_ceil_f64 v[2:3], v[4:5] row_share:1
+// GFX1251: v_ceil_f64_dpp v[2:3], v[4:5] row_share:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x30,0x04,0x7e,0x04,0x51,0x01,0xff]
+// GFX1250-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: not a valid operand.
+// GFX1250-ERR-NEXT:{{^}}v_ceil_f64 v[2:3], v[4:5] row_share:1
+// GFX1250-ERR-NEXT:{{^}} ^
+
+v_rndne_f64 v[2:3], v[4:5] row_share:1
+// GFX1251: v_rndne_f64_dpp v[2:3], v[4:5] row_share:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x32,0x04,0x7e,0x04,0x51,0x01,0xff]
+// GFX1250-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: not a valid operand.
+// GFX1250-ERR-NEXT:{{^}}v_rndne_f64 v[2:3], v[4:5] row_share:1
+// GFX1250-ERR-NEXT:{{^}} ^
+
+v_floor_f64 v[2:3], v[4:5] row_share:1
+// GFX1251: v_floor_f64_dpp v[2:3], v[4:5] row_share:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x34,0x04,0x7e,0x04,0x51,0x01,0xff]
+// GFX1250-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: not a valid operand.
+// GFX1250-ERR-NEXT:{{^}}v_floor_f64 v[2:3], v[4:5] row_share:1
+// GFX1250-ERR-NEXT:{{^}} ^
+
+v_frexp_exp_i32_f64 v2, v[4:5] row_share:1
+// GFX1251: v_frexp_exp_i32_f64_dpp v2, v[4:5] row_share:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x78,0x04,0x7e,0x04,0x51,0x01,0xff]
+// GFX1250-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: not a valid operand.
+// GFX1250-ERR-NEXT:{{^}}v_frexp_exp_i32_f64 v2, v[4:5] row_share:1
+// GFX1250-ERR-NEXT:{{^}} ^
+
+v_frexp_mant_f64 v[2:3], v[4:5] row_share:1
+// GFX1251: v_frexp_mant_f64_dpp v[2:3], v[4:5] row_share:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x7a,0x04,0x7e,0x04,0x51,0x01,0xff]
+// GFX1250-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: not a valid operand.
+// GFX1250-ERR-NEXT:{{^}}v_frexp_mant_f64 v[2:3], v[4:5] row_share:1
+// GFX1250-ERR-NEXT:{{^}} ^
+
+v_fract_f64 v[2:3], v[4:5] row_share:1
+// GFX1251: v_fract_f64_dpp v[2:3], v[4:5] row_share:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x7c,0x04,0x7e,0x04,0x51,0x01,0xff]
+// GFX1250-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: not a valid operand.
+// GFX1250-ERR-NEXT:{{^}}v_fract_f64 v[2:3], v[4:5] row_share:1
+// GFX1250-ERR-NEXT:{{^}} ^
diff --git a/llvm/test/MC/AMDGPU/gfx1251_asm_vop1_err.s b/llvm/test/MC/AMDGPU/gfx1251_asm_vop1_err.s
new file mode 100644
index 0000000000000..1d88e9cb59c8e
--- /dev/null
+++ b/llvm/test/MC/AMDGPU/gfx1251_asm_vop1_err.s
@@ -0,0 +1,156 @@
+// RUN: not llvm-mc -triple=amdgcn -mcpu=gfx1251 -show-encoding %s 2>&1 | FileCheck --check-prefix=GFX1251-ERR --implicit-check-not=error: --strict-whitespace %s
+
+v_mov_b64 v[4:5], v[2:3] dpp8:[7,6,5,4,3,2,1,0]
+// GFX1251-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: not a valid operand.
+// GFX1251-ERR-NEXT:{{^}}v_mov_b64 v[4:5], v[2:3] dpp8:[7,6,5,4,3,2,1,0]
+// GFX1251-ERR-NEXT:{{^}} ^
+
+v_cvt_i32_f64 v2, v[4:5] dpp8:[7,6,5,4,3,2,1,0]
+// GFX1251-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: not a valid operand.
+// GFX1251-ERR-NEXT:{{^}}v_cvt_i32_f64 v2, v[4:5] dpp8:[7,6,5,4,3,2,1,0]
+// GFX1251-ERR-NEXT:{{^}} ^
+
+v_cvt_f64_i32 v[4:5], v2 dpp8:[7,6,5,4,3,2,1,0]
+// GFX1251-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: not a valid operand.
+// GFX1251-ERR-NEXT:{{^}}v_cvt_f64_i32 v[4:5], v2 dpp8:[7,6,5,4,3,2,1,0]
+// GFX1251-ERR-NEXT:{{^}} ^
+
+v_cvt_f32_f64 v2, v[4:5] dpp8:[7,6,5,4,3,2,1,0]
+// GFX1251-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: not a valid operand.
+// GFX1251-ERR-NEXT:{{^}}v_cvt_f32_f64 v2, v[4:5] dpp8:[7,6,5,4,3,2,1,0]
+// GFX1251-ERR-NEXT:{{^}} ^
+
+v_cvt_f64_f32 v[4:5], v2 dpp8:[7,6,5,4,3,2,1,0]
+// GFX1251-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: not a valid operand.
+// GFX1251-ERR-NEXT:{{^}}v_cvt_f64_f32 v[4:5], v2 dpp8:[7,6,5,4,3,2,1,0]
+// GFX1251-ERR-NEXT:{{^}} ^
+
+v_cvt_u32_f64 v2, v[4:5] dpp8:[7,6,5,4,3,2,1,0]
+// GFX1251-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: not a valid operand.
+// GFX1251-ERR-NEXT:{{^}}v_cvt_u32_f64 v2, v[4:5] dpp8:[7,6,5,4,3,2,1,0]
+// GFX1251-ERR-NEXT:{{^}} ^
+
+v_cvt_f64_u32 v[4:5], v2 dpp8:[7,6,5,4,3,2,1,0]
+// GFX1251-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: not a valid operand.
+// GFX1251-ERR-NEXT:{{^}}v_cvt_f64_u32 v[4:5], v2 dpp8:[7,6,5,4,3,2,1,0]
+// GFX1251-ERR-NEXT:{{^}} ^
+
+v_trunc_f64 v[2:3], v[4:5] dpp8:[7,6,5,4,3,2,1,0]...
[truncated]
|
|
VOP3 forms are pending corresponding VOP3 changes. |
shiltian
approved these changes
Sep 18, 2025
rampitec
deleted the
users/rampitec/09-18-_amdgpu_gfx1251_vop1_dpp_support
branch
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.