This is an archive of the discontinued LLVM Phabricator instance.

Differential D45251

[AMDGPU][MC][GFX9] Added instructions s_mul_hi_*32, s_lshl*_add_u32
ClosedPublic

Authored by dp on Apr 4 2018, 4:39 AM.

Details

Reviewers
arsenm
artem.tamazov
timcorringham
Commits
rG2f8e146ad3d4: [AMDGPU][MC][GFX9] Added instructions s_mul_hi_*32, s_lshl*_add_u32
rL329562: [AMDGPU][MC][GFX9] Added instructions s_mul_hi_*32, s_lshl*_add_u32
Summary

See the following bugs:

https://bugs.llvm.org/show_bug.cgi?id=36841
https://bugs.llvm.org/show_bug.cgi?id=36842

A list of added instructions:

s_mul_hi_u32
s_mul_hi_i32

s_lshl1_add_u32
s_lshl2_add_u32
s_lshl3_add_u32
s_lshl4_add_u32

Diff Detail

Repository
rL LLVM

Event Timeline

dp created this revision.Apr 4 2018, 4:39 AM
Herald added subscribers: t-tye, tpr, dstuttard and 4 others. · View Herald TranscriptApr 4 2018, 4:39 AM
artem.tamazov accepted this revision.Apr 4 2018, 7:02 AM

Looks good.

This revision is now accepted and ready to land.Apr 4 2018, 7:02 AM
Closed by commit rL329562: [AMDGPU][MC][GFX9] Added instructions s_mul_hi_*32, s_lshl*_add_u32 (authored by dpreobra). · Explain WhyApr 9 2018, 6:13 AM
This revision was automatically updated to reflect the committed changes.
Herald added a subscriber: llvm-commits. · View Herald TranscriptApr 9 2018, 6:13 AM

Revision Contents

PathSize
llvm/
trunk/
lib/
Target/
AMDGPU/
21 lines
test/
MC/
AMDGPU/
156 lines
Disassembler/
AMDGPU/
28 lines

Diff 141631

llvm/trunk/lib/Target/AMDGPU/SOPInstructions.td

Show First 20 Lines • Show All 489 Lines • ▼ Show 20 Lines > {
let has_sdst = 0; let has_sdst = 0;
} }
} }
let SubtargetPredicate = isGFX9 in { let SubtargetPredicate = isGFX9 in {
def S_PACK_LL_B32_B16 : SOP2_32<"s_pack_ll_b32_b16">; def S_PACK_LL_B32_B16 : SOP2_32<"s_pack_ll_b32_b16">;
def S_PACK_LH_B32_B16 : SOP2_32<"s_pack_lh_b32_b16">; def S_PACK_LH_B32_B16 : SOP2_32<"s_pack_lh_b32_b16">;
def S_PACK_HH_B32_B16 : SOP2_32<"s_pack_hh_b32_b16">; def S_PACK_HH_B32_B16 : SOP2_32<"s_pack_hh_b32_b16">;
let Defs = [SCC] in {
def S_LSHL1_ADD_U32 : SOP2_32<"s_lshl1_add_u32">;
def S_LSHL2_ADD_U32 : SOP2_32<"s_lshl2_add_u32">;
def S_LSHL3_ADD_U32 : SOP2_32<"s_lshl3_add_u32">;
def S_LSHL4_ADD_U32 : SOP2_32<"s_lshl4_add_u32">;
} // End Defs = [SCC]
def S_MUL_HI_U32 : SOP2_32<"s_mul_hi_u32">;
def S_MUL_HI_I32 : SOP2_32<"s_mul_hi_i32">;
} }
//===----------------------------------------------------------------------===// //===----------------------------------------------------------------------===//
// SOPK Instructions // SOPK Instructions
//===----------------------------------------------------------------------===// //===----------------------------------------------------------------------===//
class SOPK_Pseudo <string opName, dag outs, dag ins, class SOPK_Pseudo <string opName, dag outs, dag ins,
string asmOps, list<dag> pattern=[]> : string asmOps, list<dag> pattern=[]> :
▲ Show 20 Lines • Show All 843 Lines • ▼ Show 20 Lines
// SOP1 - GFX9. // SOP1 - GFX9.
//===----------------------------------------------------------------------===// //===----------------------------------------------------------------------===//
def S_ANDN1_SAVEEXEC_B64_vi : SOP1_Real_vi<0x33, S_ANDN1_SAVEEXEC_B64>; def S_ANDN1_SAVEEXEC_B64_vi : SOP1_Real_vi<0x33, S_ANDN1_SAVEEXEC_B64>;
def S_ORN1_SAVEEXEC_B64_vi : SOP1_Real_vi<0x34, S_ORN1_SAVEEXEC_B64>; def S_ORN1_SAVEEXEC_B64_vi : SOP1_Real_vi<0x34, S_ORN1_SAVEEXEC_B64>;
def S_ANDN1_WREXEC_B64_vi : SOP1_Real_vi<0x35, S_ANDN1_WREXEC_B64>; def S_ANDN1_WREXEC_B64_vi : SOP1_Real_vi<0x35, S_ANDN1_WREXEC_B64>;
def S_ANDN2_WREXEC_B64_vi : SOP1_Real_vi<0x36, S_ANDN2_WREXEC_B64>; def S_ANDN2_WREXEC_B64_vi : SOP1_Real_vi<0x36, S_ANDN2_WREXEC_B64>;
def S_BITREPLICATE_B64_B32_vi : SOP1_Real_vi<0x37, S_BITREPLICATE_B64_B32>; def S_BITREPLICATE_B64_B32_vi : SOP1_Real_vi<0x37, S_BITREPLICATE_B64_B32>;
//===----------------------------------------------------------------------===//
// SOP2 - GFX9.
//===----------------------------------------------------------------------===//
def S_LSHL1_ADD_U32_vi : SOP2_Real_vi<0x2e, S_LSHL1_ADD_U32>;
def S_LSHL2_ADD_U32_vi : SOP2_Real_vi<0x2f, S_LSHL2_ADD_U32>;
def S_LSHL3_ADD_U32_vi : SOP2_Real_vi<0x30, S_LSHL3_ADD_U32>;
def S_LSHL4_ADD_U32_vi : SOP2_Real_vi<0x31, S_LSHL4_ADD_U32>;
def S_MUL_HI_U32_vi : SOP2_Real_vi<0x2c, S_MUL_HI_U32>;
def S_MUL_HI_I32_vi : SOP2_Real_vi<0x2d, S_MUL_HI_I32>;

llvm/trunk/test/MC/AMDGPU/sop2.s

// RUN: llvm-mc -arch=amdgcn -show-encoding %s | FileCheck --check-prefix=GCN --check-prefix=SICI %s // RUN: not llvm-mc -arch=amdgcn -show-encoding %s | FileCheck --check-prefix=GCN --check-prefix=SICI %s
// RUN: llvm-mc -arch=amdgcn -mcpu=tahiti -show-encoding %s | FileCheck --check-prefix=GCN --check-prefix=SICI %s // RUN: not llvm-mc -arch=amdgcn -mcpu=tahiti -show-encoding %s | FileCheck --check-prefix=GCN --check-prefix=SICI %s
// RUN: llvm-mc -arch=amdgcn -mcpu=bonaire -show-encoding %s | FileCheck --check-prefix=GCN --check-prefix=SICI %s // RUN: not llvm-mc -arch=amdgcn -mcpu=bonaire -show-encoding %s | FileCheck --check-prefix=GCN --check-prefix=SICI %s
// RUN: not llvm-mc -arch=amdgcn -mcpu=fiji -show-encoding %s | FileCheck --check-prefix=GCN --check-prefix=VI %s // RUN: not llvm-mc -arch=amdgcn -mcpu=fiji -show-encoding %s | FileCheck --check-prefix=GCN --check-prefix=GFX89 %s
// RUN: not llvm-mc -arch=amdgcn -mcpu=fiji -show-encoding %s 2>&1 | FileCheck --check-prefix=NOVI %s // RUN: not llvm-mc -arch=amdgcn -mcpu=gfx900 -show-encoding %s | FileCheck --check-prefix=GCN --check-prefix=GFX89 --check-prefix=GFX9 %s
// RUN: not llvm-mc -arch=amdgcn -show-encoding %s 2>&1 | FileCheck --check-prefix=NOSICIVI %s
// RUN: not llvm-mc -arch=amdgcn -mcpu=tahiti -show-encoding %s 2>&1 | FileCheck --check-prefix=NOSICIVI %s
// RUN: not llvm-mc -arch=amdgcn -mcpu=bonaire -show-encoding %s 2>&1 | FileCheck --check-prefix=NOSICIVI %s
// RUN: not llvm-mc -arch=amdgcn -mcpu=fiji -show-encoding %s 2>&1 | FileCheck --check-prefix=NOSICIVI --check-prefix=NOVI --check-prefix=NOGFX89 %s
// RUN: not llvm-mc -arch=amdgcn -mcpu=gfx900 -show-encoding %s 2>&1 | FileCheck --check-prefix=NOGFX89 %s
s_add_u32 s1, s2, s3 s_add_u32 s1, s2, s3
// GCN: s_add_u32 s1, s2, s3 ; encoding: [0x02,0x03,0x01,0x80] // GCN: s_add_u32 s1, s2, s3 ; encoding: [0x02,0x03,0x01,0x80]
s_sub_u32 s1, s2, s3 s_sub_u32 s1, s2, s3
// GCN: s_sub_u32 s1, s2, s3 ; encoding: [0x02,0x03,0x81,0x80] // GCN: s_sub_u32 s1, s2, s3 ; encoding: [0x02,0x03,0x81,0x80]
s_add_i32 s1, s2, s3 s_add_i32 s1, s2, s3
Show All 23 Lines
s_cselect_b32 s1, s2, s3 s_cselect_b32 s1, s2, s3
// GCN: s_cselect_b32 s1, s2, s3 ; encoding: [0x02,0x03,0x01,0x85] // GCN: s_cselect_b32 s1, s2, s3 ; encoding: [0x02,0x03,0x01,0x85]
s_cselect_b64 s[2:3], s[4:5], s[6:7] s_cselect_b64 s[2:3], s[4:5], s[6:7]
// GCN: s_cselect_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x85] // GCN: s_cselect_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x85]
s_and_b32 s2, s4, s6 s_and_b32 s2, s4, s6
// SICI: s_and_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x87] // SICI: s_and_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x87]
// VI: s_and_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x86] // GFX89: s_and_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x86]
s_and_b64 s[2:3], s[4:5], s[6:7] s_and_b64 s[2:3], s[4:5], s[6:7]
// SICI: s_and_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x87] // SICI: s_and_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x87]
// VI: s_and_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x86] // GFX89: s_and_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x86]
s_or_b32 s2, s4, s6 s_or_b32 s2, s4, s6
// SICI: s_or_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x88] // SICI: s_or_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x88]
// VI: s_or_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x87] // GFX89: s_or_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x87]
s_or_b64 s[2:3], s[4:5], s[6:7] s_or_b64 s[2:3], s[4:5], s[6:7]
// SICI: s_or_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x88] // SICI: s_or_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x88]
// VI: s_or_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x87] // GFX89: s_or_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x87]
s_xor_b32 s2, s4, s6 s_xor_b32 s2, s4, s6
// SICI: s_xor_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x89] // SICI: s_xor_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x89]
// VI: s_xor_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x88] // GFX89: s_xor_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x88]
s_xor_b64 s[2:3], s[4:5], s[6:7] s_xor_b64 s[2:3], s[4:5], s[6:7]
// SICI: s_xor_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x89] // SICI: s_xor_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x89]
// VI: s_xor_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x88] // GFX89: s_xor_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x88]
s_andn2_b32 s2, s4, s6 s_andn2_b32 s2, s4, s6
// SICI: s_andn2_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8a] // SICI: s_andn2_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8a]
// VI: s_andn2_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x89] // GFX89: s_andn2_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x89]
s_andn2_b64 s[2:3], s[4:5], s[6:7] s_andn2_b64 s[2:3], s[4:5], s[6:7]
// SICI: s_andn2_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x8a] // SICI: s_andn2_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x8a]
// VI: s_andn2_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x89] // GFX89: s_andn2_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x89]
s_orn2_b32 s2, s4, s6 s_orn2_b32 s2, s4, s6
// SICI: s_orn2_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8b] // SICI: s_orn2_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8b]
// VI: s_orn2_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8a] // GFX89: s_orn2_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8a]
s_orn2_b64 s[2:3], s[4:5], s[6:7] s_orn2_b64 s[2:3], s[4:5], s[6:7]
// SICI: s_orn2_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x8b] // SICI: s_orn2_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x8b]
// VI: s_orn2_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x8a] // GFX89: s_orn2_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x8a]
s_nand_b32 s2, s4, s6 s_nand_b32 s2, s4, s6
// SICI: s_nand_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8c] // SICI: s_nand_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8c]
// VI: s_nand_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8b] // GFX89: s_nand_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8b]
s_nand_b64 s[2:3], s[4:5], s[6:7] s_nand_b64 s[2:3], s[4:5], s[6:7]
// SICI: s_nand_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x8c] // SICI: s_nand_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x8c]
// VI: s_nand_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x8b] // GFX89: s_nand_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x8b]
s_nor_b32 s2, s4, s6 s_nor_b32 s2, s4, s6
// SICI: s_nor_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8d] // SICI: s_nor_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8d]
// VI: s_nor_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8c] // GFX89: s_nor_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8c]
s_nor_b64 s[2:3], s[4:5], s[6:7] s_nor_b64 s[2:3], s[4:5], s[6:7]
// SICI: s_nor_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x8d] // SICI: s_nor_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x8d]
// VI: s_nor_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x8c] // GFX89: s_nor_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x8c]
s_xnor_b32 s2, s4, s6 s_xnor_b32 s2, s4, s6
// SICI: s_xnor_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8e] // SICI: s_xnor_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8e]
// VI: s_xnor_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8d] // GFX89: s_xnor_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8d]
s_xnor_b64 s[2:3], s[4:5], s[6:7] s_xnor_b64 s[2:3], s[4:5], s[6:7]
// SICI: s_xnor_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x8e] // SICI: s_xnor_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x8e]
// VI: s_xnor_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x8d] // GFX89: s_xnor_b64 s[2:3], s[4:5], s[6:7] ; encoding: [0x04,0x06,0x82,0x8d]
s_lshl_b32 s2, s4, s6 s_lshl_b32 s2, s4, s6
// SICI: s_lshl_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8f] // SICI: s_lshl_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8f]
// VI: s_lshl_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8e] // GFX89: s_lshl_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8e]
s_lshl_b64 s[2:3], s[4:5], s6 s_lshl_b64 s[2:3], s[4:5], s6
// SICI: s_lshl_b64 s[2:3], s[4:5], s6 ; encoding: [0x04,0x06,0x82,0x8f] // SICI: s_lshl_b64 s[2:3], s[4:5], s6 ; encoding: [0x04,0x06,0x82,0x8f]
// VI: s_lshl_b64 s[2:3], s[4:5], s6 ; encoding: [0x04,0x06,0x82,0x8e] // GFX89: s_lshl_b64 s[2:3], s[4:5], s6 ; encoding: [0x04,0x06,0x82,0x8e]
s_lshr_b32 s2, s4, s6 s_lshr_b32 s2, s4, s6
// SICI: s_lshr_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x90] // SICI: s_lshr_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x90]
// VI: s_lshr_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8f] // GFX89: s_lshr_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x8f]
s_lshr_b64 s[2:3], s[4:5], s6 s_lshr_b64 s[2:3], s[4:5], s6
// SICI: s_lshr_b64 s[2:3], s[4:5], s6 ; encoding: [0x04,0x06,0x82,0x90] // SICI: s_lshr_b64 s[2:3], s[4:5], s6 ; encoding: [0x04,0x06,0x82,0x90]
// VI: s_lshr_b64 s[2:3], s[4:5], s6 ; encoding: [0x04,0x06,0x82,0x8f] // GFX89: s_lshr_b64 s[2:3], s[4:5], s6 ; encoding: [0x04,0x06,0x82,0x8f]
s_ashr_i32 s2, s4, s6 s_ashr_i32 s2, s4, s6
// SICI: s_ashr_i32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x91] // SICI: s_ashr_i32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x91]
// VI: s_ashr_i32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x90] // GFX89: s_ashr_i32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x90]
s_ashr_i64 s[2:3], s[4:5], s6 s_ashr_i64 s[2:3], s[4:5], s6
// SICI: s_ashr_i64 s[2:3], s[4:5], s6 ; encoding: [0x04,0x06,0x82,0x91] // SICI: s_ashr_i64 s[2:3], s[4:5], s6 ; encoding: [0x04,0x06,0x82,0x91]
// VI: s_ashr_i64 s[2:3], s[4:5], s6 ; encoding: [0x04,0x06,0x82,0x90] // GFX89: s_ashr_i64 s[2:3], s[4:5], s6 ; encoding: [0x04,0x06,0x82,0x90]
s_bfm_b32 s2, s4, s6 s_bfm_b32 s2, s4, s6
// SICI: s_bfm_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x92] // SICI: s_bfm_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x92]
// VI: s_bfm_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x91] // GFX89: s_bfm_b32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x91]
s_bfm_b64 s[2:3], s4, s6 s_bfm_b64 s[2:3], s4, s6
// SICI: s_bfm_b64 s[2:3], s4, s6 ; encoding: [0x04,0x06,0x82,0x92] // SICI: s_bfm_b64 s[2:3], s4, s6 ; encoding: [0x04,0x06,0x82,0x92]
// VI: s_bfm_b64 s[2:3], s4, s6 ; encoding: [0x04,0x06,0x82,0x91] // GFX89: s_bfm_b64 s[2:3], s4, s6 ; encoding: [0x04,0x06,0x82,0x91]
s_mul_i32 s2, s4, s6 s_mul_i32 s2, s4, s6
// SICI: s_mul_i32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x93] // SICI: s_mul_i32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x93]
// VI: s_mul_i32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x92] // GFX89: s_mul_i32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x92]
s_bfe_u32 s2, s4, s6 s_bfe_u32 s2, s4, s6
// SICI: s_bfe_u32 s2, s4, s6 ; encoding: [0x04,0x06,0x82,0x93] // SICI: s_bfe_u32 s2, s4, s6 ; encoding: [0x04,0x06,0x82,0x93]
// VI: s_bfe_u32 s2, s4, s6 ; encoding: [0x04,0x06,0x82,0x92] // GFX89: s_bfe_u32 s2, s4, s6 ; encoding: [0x04,0x06,0x82,0x92]
s_bfe_i32 s2, s4, s6 s_bfe_i32 s2, s4, s6
// SICI: s_bfe_i32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x94] // SICI: s_bfe_i32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x94]
// VI: s_bfe_i32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x93] // GFX89: s_bfe_i32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x93]
s_bfe_u64 s[2:3], s[4:5], s6 s_bfe_u64 s[2:3], s[4:5], s6
// SICI: s_bfe_u64 s[2:3], s[4:5], s6 ; encoding: [0x04,0x06,0x82,0x94] // SICI: s_bfe_u64 s[2:3], s[4:5], s6 ; encoding: [0x04,0x06,0x82,0x94]
// VI: s_bfe_u64 s[2:3], s[4:5], s6 ; encoding: [0x04,0x06,0x82,0x93] // GFX89: s_bfe_u64 s[2:3], s[4:5], s6 ; encoding: [0x04,0x06,0x82,0x93]
s_bfe_i64 s[2:3], s[4:5], s6 s_bfe_i64 s[2:3], s[4:5], s6
// SICI: s_bfe_i64 s[2:3], s[4:5], s6 ; encoding: [0x04,0x06,0x02,0x95] // SICI: s_bfe_i64 s[2:3], s[4:5], s6 ; encoding: [0x04,0x06,0x02,0x95]
// VI: s_bfe_i64 s[2:3], s[4:5], s6 ; encoding: [0x04,0x06,0x02,0x94] // GFX89: s_bfe_i64 s[2:3], s[4:5], s6 ; encoding: [0x04,0x06,0x02,0x94]
s_cbranch_g_fork s[4:5], s[6:7] s_cbranch_g_fork s[4:5], s[6:7]
// SICI: s_cbranch_g_fork s[4:5], s[6:7] ; encoding: [0x04,0x06,0x80,0x95] // SICI: s_cbranch_g_fork s[4:5], s[6:7] ; encoding: [0x04,0x06,0x80,0x95]
// VI: s_cbranch_g_fork s[4:5], s[6:7] ; encoding: [0x04,0x06,0x80,0x94] // GFX89: s_cbranch_g_fork s[4:5], s[6:7] ; encoding: [0x04,0x06,0x80,0x94]
s_cbranch_g_fork 1, s[6:7] s_cbranch_g_fork 1, s[6:7]
// SICI: s_cbranch_g_fork 1, s[6:7] ; encoding: [0x81,0x06,0x80,0x95] // SICI: s_cbranch_g_fork 1, s[6:7] ; encoding: [0x81,0x06,0x80,0x95]
// VI: s_cbranch_g_fork 1, s[6:7] ; encoding: [0x81,0x06,0x80,0x94] // GFX89: s_cbranch_g_fork 1, s[6:7] ; encoding: [0x81,0x06,0x80,0x94]
s_cbranch_g_fork s[6:7], 2 s_cbranch_g_fork s[6:7], 2
// SICI: s_cbranch_g_fork s[6:7], 2 ; encoding: [0x06,0x82,0x80,0x95] // SICI: s_cbranch_g_fork s[6:7], 2 ; encoding: [0x06,0x82,0x80,0x95]
// VI: s_cbranch_g_fork s[6:7], 2 ; encoding: [0x06,0x82,0x80,0x94] // GFX89: s_cbranch_g_fork s[6:7], 2 ; encoding: [0x06,0x82,0x80,0x94]
s_absdiff_i32 s2, s4, s6 s_absdiff_i32 s2, s4, s6
// SICI: s_absdiff_i32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x96] // SICI: s_absdiff_i32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x96]
// VI: s_absdiff_i32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x95] // GFX89: s_absdiff_i32 s2, s4, s6 ; encoding: [0x04,0x06,0x02,0x95]
s_add_u32 s101, s102, s103 s_add_u32 s101, s102, s103
// SICI: s_add_u32 s101, s102, s103 ; encoding: [0x66,0x67,0x65,0x80] // SICI: s_add_u32 s101, s102, s103 ; encoding: [0x66,0x67,0x65,0x80]
// NOVI: error: not a valid operand // NOGFX89: error: not a valid operand
s_lshl1_add_u32 s5, s1, s2
// GFX9: s_lshl1_add_u32 s5, s1, s2 ; encoding: [0x01,0x02,0x05,0x97]
// NOSICIVI: error
s_lshl1_add_u32 s5, -1, s2
// GFX9: s_lshl1_add_u32 s5, -1, s2 ; encoding: [0xc1,0x02,0x05,0x97]
// NOSICIVI: error
s_lshl1_add_u32 s5, s1, 0
// GFX9: s_lshl1_add_u32 s5, s1, 0 ; encoding: [0x01,0x80,0x05,0x97]
// NOSICIVI: error
s_lshl1_add_u32 s5, s1, 0x3f717273
// GFX9: s_lshl1_add_u32 s5, s1, 0x3f717273 ; encoding: [0x01,0xff,0x05,0x97,0x73,0x72,0x71,0x3f]
// NOSICIVI: error
s_lshl2_add_u32 s101, s1, s2
// GFX9: s_lshl2_add_u32 s101, s1, s2 ; encoding: [0x01,0x02,0xe5,0x97]
// NOSICIVI: error
s_lshl2_add_u32 s5, 0xaf123456, s2
// GFX9: s_lshl2_add_u32 s5, 0xaf123456, s2 ; encoding: [0xff,0x02,0x85,0x97,0x56,0x34,0x12,0xaf]
// NOSICIVI: error
s_lshl3_add_u32 s5, 0x3f717273, s2
// GFX9: s_lshl3_add_u32 s5, 0x3f717273, s2 ; encoding: [0xff,0x02,0x05,0x98,0x73,0x72,0x71,0x3f]
// NOSICIVI: error
s_lshl3_add_u32 s5, s1, s101
// GFX9: s_lshl3_add_u32 s5, s1, s101 ; encoding: [0x01,0x65,0x05,0x98]
// NOSICIVI: error
s_lshl4_add_u32 s5, s1, 0xaf123456
// GFX9: s_lshl4_add_u32 s5, s1, 0xaf123456 ; encoding: [0x01,0xff,0x85,0x98,0x56,0x34,0x12,0xaf]
// NOSICIVI: error
s_lshl4_add_u32 s5, -1, s2
// GFX9: s_lshl4_add_u32 s5, -1, s2 ; encoding: [0xc1,0x02,0x85,0x98]
// NOSICIVI: error
s_mul_hi_i32 s5, s101, s2
// GFX9: s_mul_hi_i32 s5, s101, s2 ; encoding: [0x65,0x02,0x85,0x96]
// NOSICIVI: error
s_mul_hi_i32 s5, 0, s2
// GFX9: s_mul_hi_i32 s5, 0, s2 ; encoding: [0x80,0x02,0x85,0x96]
// NOSICIVI: error
s_mul_hi_i32 s5, 0x3f717273, s2
// GFX9: s_mul_hi_i32 s5, 0x3f717273, s2 ; encoding: [0xff,0x02,0x85,0x96,0x73,0x72,0x71,0x3f]
// NOSICIVI: error
s_mul_hi_i32 s5, s1, s101
// GFX9: s_mul_hi_i32 s5, s1, s101 ; encoding: [0x01,0x65,0x85,0x96]
// NOSICIVI: error
s_mul_hi_i32 s5, s1, 0
// GFX9: s_mul_hi_i32 s5, s1, 0 ; encoding: [0x01,0x80,0x85,0x96]
// NOSICIVI: error
s_mul_hi_u32 s5, s1, 0x3f717273
// GFX9: s_mul_hi_u32 s5, s1, 0x3f717273 ; encoding: [0x01,0xff,0x05,0x96,0x73,0x72,0x71,0x3f]
// NOSICIVI: error
s_mul_hi_u32 s5, s1, s101
// GFX9: s_mul_hi_u32 s5, s1, s101 ; encoding: [0x01,0x65,0x05,0x96]
// NOSICIVI: error
s_mul_hi_u32 s5, s1, 0
// GFX9: s_mul_hi_u32 s5, s1, 0 ; encoding: [0x01,0x80,0x05,0x96]
// NOSICIVI: error

llvm/trunk/test/MC/Disassembler/AMDGPU/sop2_gfx9.txt

# RUN: llvm-mc -arch=amdgcn -mcpu=gfx900 -disassemble -show-encoding < %s | FileCheck %s -check-prefix=GFX9
# GFX9: s_lshl1_add_u32 s5, 0, s2 ; encoding: [0x80,0x02,0x05,0x97]
0x80,0x02,0x05,0x97
# GFX9: s_lshl2_add_u32 s5, 0xaf123456, s2 ; encoding: [0xff,0x02,0x85,0x97,0x56,0x34,0x12,0xaf]
0xff,0x02,0x85,0x97,0x56,0x34,0x12,0xaf
# GFX9: s_lshl3_add_u32 s5, s1, -1 ; encoding: [0x01,0xc1,0x05,0x98]
0x01,0xc1,0x05,0x98
# GFX9: s_lshl4_add_u32 s5, s1, 0xaf123456 ; encoding: [0x01,0xff,0x85,0x98,0x56,0x34,0x12,0xaf]
0x01,0xff,0x85,0x98,0x56,0x34,0x12,0xaf
# GFX9: s_mul_hi_i32 s5, s101, s2 ; encoding: [0x65,0x02,0x85,0x96]
0x65,0x02,0x85,0x96
# GFX9: s_mul_hi_i32 s5, 0, s2 ; encoding: [0x80,0x02,0x85,0x96]
0x80,0x02,0x85,0x96
# GFX9: s_mul_hi_i32 s5, 0xaf123456, s2 ; encoding: [0xff,0x02,0x85,0x96,0x56,0x34,0x12,0xaf]
0xff,0x02,0x85,0x96,0x56,0x34,0x12,0xaf
# GFX9: s_mul_hi_u32 s5, s1, -1 ; encoding: [0x01,0xc1,0x05,0x96]
0x01,0xc1,0x05,0x96
# GFX9: s_mul_hi_u32 s5, s1, 0xaf123456 ; encoding: [0x01,0xff,0x05,0x96,0x56,0x34,0x12,0xaf]
0x01,0xff,0x05,0x96,0x56,0x34,0x12,0xaf