Skip to content

AMDGPU: MC support for v_cvt_scalef32_pk_f32_[fp|bf]8 of gfx950. #117383

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Nov 25, 2024
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 17 additions & 0 deletions llvm/lib/Target/AMDGPU/VOP3Instructions.td
Original file line number Diff line number Diff line change
Expand Up @@ -899,16 +899,31 @@ def VOP3_CVT_SCALE_FP8BF8_F32_Profile : VOP3_Profile<VOPProfile<[i32, f32, f32,
let HasOMod = 0;
}

def VOP3_CVT_SCALE_PK_F32_FP8BF8_Profile : VOP3_Profile<VOPProfile<[v2f32, i32, f32, untyped]>,
VOP3_OPSEL> {
let InsVOP3OpSel = (ins FP32InputMods:$src0_modifiers, Src0RC64:$src0,
FP32InputMods:$src1_modifiers, Src1RC64:$src1,
op_sel0:$op_sel);
let HasClamp = 0;
let HasSrc2 = 0;
let HasSrc2Mods = 0;
let HasExtVOP3DPP = 0;
let HasOpSel = 1;
let HasOMod = 0;
}

let SubtargetPredicate = HasFP8ConversionScaleInsts, mayRaiseFPException = 0 in {
defm V_CVT_SCALEF32_F16_FP8 : VOP3Inst<"v_cvt_scalef32_f16_fp8", VOP3_CVT_SCALE_F1632_FP8BF8_Profile<f16>>;
defm V_CVT_SCALEF32_F32_FP8 : VOP3Inst<"v_cvt_scalef32_f32_fp8", VOP3_CVT_SCALE_F1632_FP8BF8_Profile<f32>>;
defm V_CVT_SCALEF32_PK_FP8_F32 : VOP3Inst<"v_cvt_scalef32_pk_fp8_f32", VOP3_CVT_SCALE_FP8BF8_F32_Profile>;
defm V_CVT_SCALEF32_PK_F32_FP8 : VOP3Inst<"v_cvt_scalef32_pk_f32_fp8", VOP3_CVT_SCALE_PK_F32_FP8BF8_Profile>;
}

let SubtargetPredicate = HasBF8ConversionScaleInsts, mayRaiseFPException = 0 in {
defm V_CVT_SCALEF32_F16_BF8 : VOP3Inst<"v_cvt_scalef32_f16_bf8", VOP3_CVT_SCALE_F1632_FP8BF8_Profile<f16>>;
defm V_CVT_SCALEF32_F32_BF8 : VOP3Inst<"v_cvt_scalef32_f32_bf8", VOP3_CVT_SCALE_F1632_FP8BF8_Profile<f32>>;
defm V_CVT_SCALEF32_PK_BF8_F32 : VOP3Inst<"v_cvt_scalef32_pk_bf8_f32", VOP3_CVT_SCALE_FP8BF8_F32_Profile>;
defm V_CVT_SCALEF32_PK_F32_BF8 : VOP3Inst<"v_cvt_scalef32_pk_f32_bf8", VOP3_CVT_SCALE_PK_F32_FP8BF8_Profile>;
}

let SubtargetPredicate = isGFX10Plus in {
Expand Down Expand Up @@ -1837,9 +1852,11 @@ let OtherPredicates = [HasFP8ConversionScaleInsts] in {
defm V_CVT_SCALEF32_F16_FP8 : VOP3OpSel_Real_gfx9 <0x24a>;
defm V_CVT_SCALEF32_F32_FP8 : VOP3OpSel_Real_gfx9 <0x23b>;
defm V_CVT_SCALEF32_PK_FP8_F32 : VOP3OpSel_Real_gfx9 <0x235>;
defm V_CVT_SCALEF32_PK_F32_FP8 : VOP3OpSel_Real_gfx9 <0x239>;
}
let OtherPredicates = [HasBF8ConversionScaleInsts] in {
defm V_CVT_SCALEF32_F16_BF8 : VOP3OpSel_Real_gfx9 <0x24b>;
defm V_CVT_SCALEF32_F32_BF8 : VOP3OpSel_Real_gfx9 <0x23c>;
defm V_CVT_SCALEF32_PK_BF8_F32 : VOP3OpSel_Real_gfx9 <0x236>;
defm V_CVT_SCALEF32_PK_F32_BF8 : VOP3OpSel_Real_gfx9 <0x23a>;
}
48 changes: 48 additions & 0 deletions llvm/test/MC/AMDGPU/gfx950_asm_features.s
Original file line number Diff line number Diff line change
Expand Up @@ -549,3 +549,51 @@ v_cvt_scalef32_pk_bf8_f32 v1, v1, -v2, |v3| op_sel:[0,0,0,1]
// NOT-GFX950: :[[@LINE+2]]:{{[0-9]+}}: error:
// GFX950: v_cvt_scalef32_pk_bf8_f32 v1, v1, s2, 3 op_sel:[0,0,0,1] ; encoding: [0x01,0x40,0x36,0xd2,0x01,0x05,0x0c,0x02]
v_cvt_scalef32_pk_bf8_f32 v1, v1, s2, 3 op_sel:[0,0,0,1]

// NOT-GFX950: :[[@LINE+2]]:{{[0-9]+}}: error:
// GFX950: v_cvt_scalef32_pk_f32_fp8 v[2:3], v2, v3 ; encoding: [0x02,0x00,0x39,0xd2,0x02,0x07,0x02,0x00]
v_cvt_scalef32_pk_f32_fp8 v[2:3], v2, v3

// NOT-GFX950: :[[@LINE+2]]:{{[0-9]+}}: error:
// GFX950: v_cvt_scalef32_pk_f32_fp8 v[2:3], v2, s3 ; encoding: [0x02,0x00,0x39,0xd2,0x02,0x07,0x00,0x00]
v_cvt_scalef32_pk_f32_fp8 v[2:3], v2, s3

// NOT-GFX950: :[[@LINE+2]]:{{[0-9]+}}: error:
// GFX950: v_cvt_scalef32_pk_f32_fp8 v[2:3], s2, 3 ; encoding: [0x02,0x00,0x39,0xd2,0x02,0x06,0x01,0x00]
v_cvt_scalef32_pk_f32_fp8 v[2:3], s2, 3

// NOT-GFX950: :[[@LINE+2]]:{{[0-9]+}}: error:
// GFX950: v_cvt_scalef32_pk_f32_fp8 v[2:3], v2, v3 op_sel:[1,0,0] ; encoding: [0x02,0x08,0x39,0xd2,0x02,0x07,0x02,0x00]
v_cvt_scalef32_pk_f32_fp8 v[2:3], v2, v3 op_sel:[1,0,0]

// NOT-GFX950: :[[@LINE+2]]:{{[0-9]+}}: error:
// GFX950: v_cvt_scalef32_pk_f32_fp8 v[2:3], v2, s3 op_sel:[1,0,0] ; encoding: [0x02,0x08,0x39,0xd2,0x02,0x07,0x00,0x00]
v_cvt_scalef32_pk_f32_fp8 v[2:3], v2, s3 op_sel:[1,0,0]

// NOT-GFX950: :[[@LINE+2]]:{{[0-9]+}}: error:
// GFX950: v_cvt_scalef32_pk_f32_fp8 v[2:3], s2, 3 op_sel:[1,0,0] ; encoding: [0x02,0x08,0x39,0xd2,0x02,0x06,0x01,0x00]
v_cvt_scalef32_pk_f32_fp8 v[2:3], s2, 3 op_sel:[1,0,0]

// NOT-GFX950: :[[@LINE+2]]:{{[0-9]+}}: error:
// GFX950: v_cvt_scalef32_pk_f32_bf8 v[2:3], v2, v3 ; encoding: [0x02,0x00,0x3a,0xd2,0x02,0x07,0x02,0x00]
v_cvt_scalef32_pk_f32_bf8 v[2:3], v2, v3

// NOT-GFX950: :[[@LINE+2]]:{{[0-9]+}}: error:
// GFX950: v_cvt_scalef32_pk_f32_bf8 v[2:3], v2, s3 ; encoding: [0x02,0x00,0x3a,0xd2,0x02,0x07,0x00,0x00]
v_cvt_scalef32_pk_f32_bf8 v[2:3], v2, s3

// NOT-GFX950: :[[@LINE+2]]:{{[0-9]+}}: error:
// GFX950: v_cvt_scalef32_pk_f32_bf8 v[2:3], s2, 3 ; encoding: [0x02,0x00,0x3a,0xd2,0x02,0x06,0x01,0x00]
v_cvt_scalef32_pk_f32_bf8 v[2:3], s2, 3

// NOT-GFX950: :[[@LINE+2]]:{{[0-9]+}}: error:
// GFX950: v_cvt_scalef32_pk_f32_bf8 v[2:3], v2, v3 op_sel:[1,0,0] ; encoding: [0x02,0x08,0x3a,0xd2,0x02,0x07,0x02,0x00]
v_cvt_scalef32_pk_f32_bf8 v[2:3], v2, v3 op_sel:[1,0,0]

// NOT-GFX950: :[[@LINE+2]]:{{[0-9]+}}: error:
// GFX950: v_cvt_scalef32_pk_f32_bf8 v[2:3], v2, s3 op_sel:[1,0,0] ; encoding: [0x02,0x08,0x3a,0xd2,0x02,0x07,0x00,0x00]
v_cvt_scalef32_pk_f32_bf8 v[2:3], v2, s3 op_sel:[1,0,0]

// NOT-GFX950: :[[@LINE+2]]:{{[0-9]+}}: error:
// GFX950: v_cvt_scalef32_pk_f32_bf8 v[2:3], s2, 3 op_sel:[1,0,0] ; encoding: [0x02,0x08,0x3a,0xd2,0x02,0x06,0x01,0x00]
v_cvt_scalef32_pk_f32_bf8 v[2:3], s2, 3 op_sel:[1,0,0]
36 changes: 36 additions & 0 deletions llvm/test/MC/Disassembler/AMDGPU/gfx950_dasm_vop3.txt
Original file line number Diff line number Diff line change
Expand Up @@ -359,3 +359,39 @@

# GFX950: v_cvt_scalef32_pk_bf8_f32 v1, v1, s2, 3 op_sel:[0,0,0,1] ; encoding: [0x01,0x40,0x36,0xd2,0x01,0x05,0x0c,0x02]
0x01,0x40,0x36,0xd2,0x01,0x05,0x0c,0x02

# GFX950: v_cvt_scalef32_pk_f32_fp8 v[2:3], v2, v3 ; encoding: [0x02,0x00,0x39,0xd2,0x02,0x07,0x02,0x00]
0x02,0x00,0x39,0xd2,0x02,0x07,0x02,0x00

# GFX950: v_cvt_scalef32_pk_f32_fp8 v[2:3], v2, s3 ; encoding: [0x02,0x00,0x39,0xd2,0x02,0x07,0x00,0x00]
0x02,0x00,0x39,0xd2,0x02,0x07,0x00,0x00

# GFX950: v_cvt_scalef32_pk_f32_fp8 v[2:3], s2, 3 ; encoding: [0x02,0x00,0x39,0xd2,0x02,0x06,0x01,0x00]
0x02,0x00,0x39,0xd2,0x02,0x06,0x01,0x00

# GFX950: v_cvt_scalef32_pk_f32_fp8 v[2:3], v2, v3 op_sel:[1,0,0] ; encoding: [0x02,0x08,0x39,0xd2,0x02,0x07,0x02,0x00]
0x02,0x08,0x39,0xd2,0x02,0x07,0x02,0x00

# GFX950: v_cvt_scalef32_pk_f32_fp8 v[2:3], v2, s3 op_sel:[1,0,0] ; encoding: [0x02,0x08,0x39,0xd2,0x02,0x07,0x00,0x00]
0x02,0x08,0x39,0xd2,0x02,0x07,0x00,0x00

# GFX950: v_cvt_scalef32_pk_f32_fp8 v[2:3], s2, 3 op_sel:[1,0,0] ; encoding: [0x02,0x08,0x39,0xd2,0x02,0x06,0x01,0x00]
0x02,0x08,0x39,0xd2,0x02,0x06,0x01,0x00

# GFX950: v_cvt_scalef32_pk_f32_bf8 v[2:3], v2, v3 ; encoding: [0x02,0x00,0x3a,0xd2,0x02,0x07,0x02,0x00]
0x02,0x00,0x3a,0xd2,0x02,0x07,0x02,0x00

# GFX950: v_cvt_scalef32_pk_f32_bf8 v[2:3], v2, s3 ; encoding: [0x02,0x00,0x3a,0xd2,0x02,0x07,0x00,0x00]
0x02,0x00,0x3a,0xd2,0x02,0x07,0x00,0x00

# GFX950: v_cvt_scalef32_pk_f32_bf8 v[2:3], s2, 3 ; encoding: [0x02,0x00,0x3a,0xd2,0x02,0x06,0x01,0x00]
0x02,0x00,0x3a,0xd2,0x02,0x06,0x01,0x00

# GFX950: v_cvt_scalef32_pk_f32_bf8 v[2:3], v2, v3 op_sel:[1,0,0] ; encoding: [0x02,0x08,0x3a,0xd2,0x02,0x07,0x02,0x00]
0x02,0x08,0x3a,0xd2,0x02,0x07,0x02,0x00

# GFX950: v_cvt_scalef32_pk_f32_bf8 v[2:3], v2, s3 op_sel:[1,0,0] ; encoding: [0x02,0x08,0x3a,0xd2,0x02,0x07,0x00,0x00]
0x02,0x08,0x3a,0xd2,0x02,0x07,0x00,0x00

# GFX950: v_cvt_scalef32_pk_f32_bf8 v[2:3], s2, 3 op_sel:[1,0,0] ; encoding: [0x02,0x08,0x3a,0xd2,0x02,0x06,0x01,0x00]
0x02,0x08,0x3a,0xd2,0x02,0x06,0x01,0x00
Loading