Skip to content

[AMDGPU] Add support for v_cvt_pk_bf16_f32 on gfx1250 #150053

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.

Sign up for GitHub

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

Jump to bottom
Merged
merged 1 commit into from
Jul 23, 2025
Merged

[AMDGPU] Add support for v_cvt_pk_bf16_f32 on gfx1250 #150053

merged 1 commit into from
Jul 23, 2025

Conversation

shiltian
Copy link
Contributor

Co-authored-by: Mekhanoshin, Stanislav [email protected]

@shiltian shiltian requested review from changpeng and rampitec July 22, 2025 16:21
@llvmbot llvmbot added backend:AMDGPU llvm:mc Machine (object) code labels Jul 22, 2025
@shiltian Graphite App
Copy link
Contributor Author

shiltian commented Jul 22, 2025 •
edited
Loading

This stack of pull requests is managed by Graphite. Learn more about stacking.

@llvmbot
Copy link
Member

llvmbot commented Jul 22, 2025

@llvm/pr-subscribers-mc

Author: Shilei Tian (shiltian)

Changes

Co-authored-by: Mekhanoshin, Stanislav <[email protected]>

Patch is 39.48 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/150053.diff

11 Files Affected:

  • (modified) llvm/lib/Target/AMDGPU/VOP3Instructions.td (+1)
  • (modified) llvm/test/CodeGen/AMDGPU/bf16-conversions.ll (+149)
  • (modified) llvm/test/MC/AMDGPU/gfx1250_asm_vop3-fake16.s (+45)
  • (modified) llvm/test/MC/AMDGPU/gfx1250_asm_vop3.s (+45)
  • (added) llvm/test/MC/AMDGPU/gfx1250_asm_vop3_dpp16-fake16.s (+59)
  • (added) llvm/test/MC/AMDGPU/gfx1250_asm_vop3_dpp16.s (+59)
  • (added) llvm/test/MC/AMDGPU/gfx1250_asm_vop3_dpp8-fake16.s (+19)
  • (added) llvm/test/MC/AMDGPU/gfx1250_asm_vop3_dpp8.s (+19)
  • (modified) llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop3.txt (+46)
  • (added) llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop3_dpp16.txt (+45)
  • (added) llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop3_dpp8.txt (+15)
diff --git a/llvm/lib/Target/AMDGPU/VOP3Instructions.td b/llvm/lib/Target/AMDGPU/VOP3Instructions.td
index aee2f2cb3d771..b6f95686704be 100644
--- a/llvm/lib/Target/AMDGPU/VOP3Instructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP3Instructions.td
@@ -1918,6 +1918,7 @@ let AssemblerPredicate = isGFX11Plus in {
 
 // These instructions differ from GFX12 variant by supporting DPP:
 defm V_LSHL_ADD_U64                  : VOP3Only_Realtriple_gfx1250<0x252>;
+defm V_CVT_PK_BF16_F32               : VOP3Only_Realtriple_gfx1250<0x36d>;
 
 //===----------------------------------------------------------------------===//
 // GFX10.
diff --git a/llvm/test/CodeGen/AMDGPU/bf16-conversions.ll b/llvm/test/CodeGen/AMDGPU/bf16-conversions.ll
index 5b4866c386793..6823a472c3ac6 100644
--- a/llvm/test/CodeGen/AMDGPU/bf16-conversions.ll
+++ b/llvm/test/CodeGen/AMDGPU/bf16-conversions.ll
@@ -1,6 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
 ; RUN: llc -mtriple=amdgcn -mcpu=gfx942 < %s | FileCheck --check-prefixes=GCN,GFX-942 %s
 ; RUN: llc -mtriple=amdgcn -mcpu=gfx950 < %s | FileCheck --check-prefixes=GCN,GFX-950 %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1250 < %s | FileCheck --check-prefix=GFX1250 %s
 
 ; TODO: Add global-isel when it can support bf16
 
@@ -9,6 +10,11 @@ define amdgpu_ps float @v_test_cvt_bf16_f32_v(bfloat %v) {
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
 ; GCN-NEXT:    ; return to shader part epilog
+;
+; GFX1250-LABEL: v_test_cvt_bf16_f32_v:
+; GFX1250:       ; %bb.0:
+; GFX1250-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
+; GFX1250-NEXT:    ; return to shader part epilog
   %cvt = fpext bfloat %v to float
   ret float %cvt
 }
@@ -19,6 +25,13 @@ define amdgpu_ps float @v_test_cvt_bf16_f32_s(bfloat inreg %v) {
 ; GCN-NEXT:    s_lshl_b32 s0, s0, 16
 ; GCN-NEXT:    v_mov_b32_e32 v0, s0
 ; GCN-NEXT:    ; return to shader part epilog
+;
+; GFX1250-LABEL: v_test_cvt_bf16_f32_s:
+; GFX1250:       ; %bb.0:
+; GFX1250-NEXT:    s_lshl_b32 s0, s0, 16
+; GFX1250-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1250-NEXT:    v_mov_b32_e32 v0, s0
+; GFX1250-NEXT:    ; return to shader part epilog
   %cvt = fpext bfloat %v to float
   ret float %cvt
 }
@@ -47,6 +60,11 @@ define amdgpu_ps float @v_test_cvt_v2f32_v2bf16_v(<2 x float> %src) {
 ; GFX-950:       ; %bb.0:
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, v0, v1
 ; GFX-950-NEXT:    ; return to shader part epilog
+;
+; GFX1250-LABEL: v_test_cvt_v2f32_v2bf16_v:
+; GFX1250:       ; %bb.0:
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, v0, v1
+; GFX1250-NEXT:    ; return to shader part epilog
   %res = fptrunc <2 x float> %src to <2 x bfloat>
   %cast = bitcast <2 x bfloat> %res to float
   ret float %cast
@@ -80,6 +98,11 @@ define amdgpu_ps float @v_test_cvt_v2f32_v2bf16_s(<2 x float> inreg %src) {
 ; GFX-950-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, s0, v0
 ; GFX-950-NEXT:    ; return to shader part epilog
+;
+; GFX1250-LABEL: v_test_cvt_v2f32_v2bf16_s:
+; GFX1250:       ; %bb.0:
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, s0, s1
+; GFX1250-NEXT:    ; return to shader part epilog
   %res = fptrunc <2 x float> %src to <2 x bfloat>
   %cast = bitcast <2 x bfloat> %res to float
   ret float %cast
@@ -103,6 +126,13 @@ define amdgpu_ps float @v_test_cvt_f32_bf16_v(float %src) {
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
 ; GFX-950-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
 ; GFX-950-NEXT:    ; return to shader part epilog
+;
+; GFX1250-LABEL: v_test_cvt_f32_bf16_v:
+; GFX1250:       ; %bb.0:
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
+; GFX1250-NEXT:    ; return to shader part epilog
   %trunc = fptrunc float %src to bfloat
   %ext = fpext bfloat %trunc to float
   ret float %ext
@@ -172,6 +202,38 @@ define amdgpu_ps float @v_test_cvt_v2f64_v2bf16_v(<2 x double> %src) {
 ; GFX-950-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, v0, v4
 ; GFX-950-NEXT:    ; return to shader part epilog
+;
+; GFX1250-LABEL: v_test_cvt_v2f64_v2bf16_v:
+; GFX1250:       ; %bb.0:
+; GFX1250-NEXT:    v_cvt_f32_f64_e32 v8, v[2:3]
+; GFX1250-NEXT:    v_cvt_f32_f64_e32 v9, v[0:1]
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[4:5], v8
+; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[6:7], v9
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX1250-NEXT:    v_cmp_gt_f64_e64 s1, |v[2:3]|, |v[4:5]|
+; GFX1250-NEXT:    v_cmp_nlg_f64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1250-NEXT:    v_cmp_nlg_f64_e64 s0, v[0:1], v[6:7]
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1250-NEXT:    v_cndmask_b32_e64 v2, -1, 1, s1
+; GFX1250-NEXT:    v_cmp_gt_f64_e64 s1, |v[0:1]|, |v[6:7]|
+; GFX1250-NEXT:    v_dual_add_nc_u32 v1, v8, v2 :: v_dual_bitop2_b32 v10, 1, v8 bitop3:0x40
+; GFX1250-NEXT:    s_wait_alu 0xf1ff
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX1250-NEXT:    v_cndmask_b32_e64 v0, -1, 1, s1
+; GFX1250-NEXT:    v_and_b32_e32 v11, 1, v9
+; GFX1250-NEXT:    v_cmp_eq_u32_e64 s1, 1, v10
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX1250-NEXT:    v_add_nc_u32_e32 v0, v9, v0
+; GFX1250-NEXT:    v_cmp_eq_u32_e64 s2, 1, v11
+; GFX1250-NEXT:    s_or_b32 vcc_lo, s1, vcc_lo
+; GFX1250-NEXT:    v_cndmask_b32_e32 v1, v1, v8, vcc_lo
+; GFX1250-NEXT:    s_or_b32 vcc_lo, s2, s0
+; GFX1250-NEXT:    s_wait_alu 0xfffe
+; GFX1250-NEXT:    v_cndmask_b32_e32 v0, v0, v9, vcc_lo
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, v0, v1
+; GFX1250-NEXT:    ; return to shader part epilog
   %res = fptrunc <2 x double> %src to <2 x bfloat>
   %cast = bitcast <2 x bfloat> %res to float
   ret float %cast
@@ -201,6 +263,11 @@ define amdgpu_ps float @fptrunc_f32_f32_to_v2bf16(float %a, float %b) {
 ; GFX-950:       ; %bb.0: ; %entry
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, v0, v1
 ; GFX-950-NEXT:    ; return to shader part epilog
+;
+; GFX1250-LABEL: fptrunc_f32_f32_to_v2bf16:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, v0, v1
+; GFX1250-NEXT:    ; return to shader part epilog
 entry:
   %a.cvt = fptrunc float %a to bfloat
   %b.cvt = fptrunc float %b to bfloat
@@ -236,6 +303,11 @@ define amdgpu_ps float @fptrunc_f32_f32_to_v2bf16_mods(float %a, float %b) {
 ; GFX-950:       ; %bb.0: ; %entry
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, -v0, |v1|
 ; GFX-950-NEXT:    ; return to shader part epilog
+;
+; GFX1250-LABEL: fptrunc_f32_f32_to_v2bf16_mods:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, -v0, |v1|
+; GFX1250-NEXT:    ; return to shader part epilog
 entry:
   %a.neg = fneg float %a
   %a.cvt = fptrunc float %a.neg to bfloat
@@ -269,6 +341,13 @@ define amdgpu_ps void @fptrunc_f32_to_bf16(float %a, ptr %out) {
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
 ; GFX-950-NEXT:    flat_store_short v[2:3], v0
 ; GFX-950-NEXT:    s_endpgm
+;
+; GFX1250-LABEL: fptrunc_f32_to_bf16:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    v_dual_mov_b32 v3, v2 :: v_dual_mov_b32 v2, v1
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
+; GFX1250-NEXT:    flat_store_b16 v[2:3], v0
+; GFX1250-NEXT:    s_endpgm
 entry:
   %a.cvt = fptrunc float %a to bfloat
   store bfloat %a.cvt, ptr %out
@@ -298,6 +377,13 @@ define amdgpu_ps void @fptrunc_f32_to_bf16_abs(float %a, ptr %out) {
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, |v0|, s0
 ; GFX-950-NEXT:    flat_store_short v[2:3], v0
 ; GFX-950-NEXT:    s_endpgm
+;
+; GFX1250-LABEL: fptrunc_f32_to_bf16_abs:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    v_dual_mov_b32 v3, v2 :: v_dual_mov_b32 v2, v1
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, |v0|, s0
+; GFX1250-NEXT:    flat_store_b16 v[2:3], v0
+; GFX1250-NEXT:    s_endpgm
 entry:
   %a.abs = call float @llvm.fabs.f32(float %a)
   %a.cvt = fptrunc float %a.abs to bfloat
@@ -328,6 +414,13 @@ define amdgpu_ps void @fptrunc_f32_to_bf16_neg(float %a, ptr %out) {
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, -v0, s0
 ; GFX-950-NEXT:    flat_store_short v[2:3], v0
 ; GFX-950-NEXT:    s_endpgm
+;
+; GFX1250-LABEL: fptrunc_f32_to_bf16_neg:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    v_dual_mov_b32 v3, v2 :: v_dual_mov_b32 v2, v1
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, -v0, s0
+; GFX1250-NEXT:    flat_store_b16 v[2:3], v0
+; GFX1250-NEXT:    s_endpgm
 entry:
   %a.neg = fneg float %a
   %a.cvt = fptrunc float %a.neg to bfloat
@@ -373,6 +466,24 @@ define amdgpu_ps void @fptrunc_f64_to_bf16(double %a, ptr %out) {
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
 ; GFX-950-NEXT:    flat_store_short v[2:3], v0
 ; GFX-950-NEXT:    s_endpgm
+;
+; GFX1250-LABEL: fptrunc_f64_to_bf16:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    v_cvt_f32_f64_e32 v6, v[0:1]
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[4:5], v6
+; GFX1250-NEXT:    v_cmp_gt_f64_e64 s0, |v[0:1]|, |v[4:5]|
+; GFX1250-NEXT:    v_cmp_nlg_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT:    v_cndmask_b32_e64 v0, -1, 1, s0
+; GFX1250-NEXT:    v_dual_add_nc_u32 v0, v6, v0 :: v_dual_bitop2_b32 v7, 1, v6 bitop3:0x40
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1250-NEXT:    v_cmp_eq_u32_e64 s0, 1, v7
+; GFX1250-NEXT:    s_or_b32 vcc_lo, vcc_lo, s0
+; GFX1250-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc_lo
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
+; GFX1250-NEXT:    flat_store_b16 v[2:3], v0
+; GFX1250-NEXT:    s_endpgm
 entry:
   %a.cvt = fptrunc double %a to bfloat
   store bfloat %a.cvt, ptr %out
@@ -417,6 +528,25 @@ define amdgpu_ps void @fptrunc_f64_to_bf16_neg(double %a, ptr %out) {
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
 ; GFX-950-NEXT:    flat_store_short v[2:3], v0
 ; GFX-950-NEXT:    s_endpgm
+;
+; GFX1250-LABEL: fptrunc_f64_to_bf16_neg:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    v_cvt_f32_f64_e64 v6, -v[0:1]
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[4:5], v6
+; GFX1250-NEXT:    v_cmp_gt_f64_e64 s1, |v[0:1]|, |v[4:5]|
+; GFX1250-NEXT:    v_cmp_nlg_f64_e64 s0, -v[0:1], v[4:5]
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT:    v_cndmask_b32_e64 v0, -1, 1, s1
+; GFX1250-NEXT:    v_dual_add_nc_u32 v0, v6, v0 :: v_dual_bitop2_b32 v7, 1, v6 bitop3:0x40
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1250-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 1, v7
+; GFX1250-NEXT:    s_or_b32 vcc_lo, s0, vcc_lo
+; GFX1250-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc_lo
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
+; GFX1250-NEXT:    flat_store_b16 v[2:3], v0
+; GFX1250-NEXT:    s_endpgm
 entry:
   %a.neg = fneg double %a
   %a.cvt = fptrunc double %a.neg to bfloat
@@ -462,6 +592,25 @@ define amdgpu_ps void @fptrunc_f64_to_bf16_abs(double %a, ptr %out) {
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
 ; GFX-950-NEXT:    flat_store_short v[2:3], v0
 ; GFX-950-NEXT:    s_endpgm
+;
+; GFX1250-LABEL: fptrunc_f64_to_bf16_abs:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    v_cvt_f32_f64_e64 v6, |v[0:1]|
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[4:5], v6
+; GFX1250-NEXT:    v_cmp_gt_f64_e64 s1, |v[0:1]|, |v[4:5]|
+; GFX1250-NEXT:    v_cmp_nlg_f64_e64 s0, |v[0:1]|, v[4:5]
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT:    v_cndmask_b32_e64 v0, -1, 1, s1
+; GFX1250-NEXT:    v_dual_add_nc_u32 v0, v6, v0 :: v_dual_bitop2_b32 v7, 1, v6 bitop3:0x40
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1250-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 1, v7
+; GFX1250-NEXT:    s_or_b32 vcc_lo, s0, vcc_lo
+; GFX1250-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc_lo
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
+; GFX1250-NEXT:    flat_store_b16 v[2:3], v0
+; GFX1250-NEXT:    s_endpgm
 entry:
   %a.abs = call double @llvm.fabs.f64(double %a)
   %a.cvt = fptrunc double %a.abs to bfloat
diff --git a/llvm/test/MC/AMDGPU/gfx1250_asm_vop3-fake16.s b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3-fake16.s
index 0070c8ab9ee78..789d6f892762b 100644
--- a/llvm/test/MC/AMDGPU/gfx1250_asm_vop3-fake16.s
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3-fake16.s
@@ -15,3 +15,48 @@ v_lshl_add_u64 v[2:3], s[4:5], 4, v[2:3]
 
 v_lshl_add_u64 v[2:3], v[4:5], v7, 12345
 // GFX1250: v_lshl_add_u64 v[2:3], v[4:5], v7, 0x3039 ; encoding: [0x02,0x00,0x52,0xd6,0x04,0x0f,0xfe,0x03,0x39,0x30,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, v1, v2
+// GFX1250: v_cvt_pk_bf16_f32 v5, v1, v2            ; encoding: [0x05,0x00,0x6d,0xd7,0x01,0x05,0x02,0x00]
+
+v_cvt_pk_bf16_f32 v5, v255, v255
+// GFX1250: v_cvt_pk_bf16_f32 v5, v255, v255        ; encoding: [0x05,0x00,0x6d,0xd7,0xff,0xff,0x03,0x00]
+
+v_cvt_pk_bf16_f32 v5, s1, s2
+// GFX1250: v_cvt_pk_bf16_f32 v5, s1, s2            ; encoding: [0x05,0x00,0x6d,0xd7,0x01,0x04,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, s105, s105
+// GFX1250: v_cvt_pk_bf16_f32 v5, s105, s105        ; encoding: [0x05,0x00,0x6d,0xd7,0x69,0xd2,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, vcc_lo, ttmp15
+// GFX1250: v_cvt_pk_bf16_f32 v5, vcc_lo, ttmp15    ; encoding: [0x05,0x00,0x6d,0xd7,0x6a,0xf6,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, vcc_hi, 0xaf123456
+// GFX1250: v_cvt_pk_bf16_f32 v5, vcc_hi, 0xaf123456 ; encoding: [0x05,0x00,0x6d,0xd7,0x6b,0xfe,0x01,0x00,0x56,0x34,0x12,0xaf]
+
+v_cvt_pk_bf16_f32 v5, ttmp15, src_scc
+// GFX1250: v_cvt_pk_bf16_f32 v5, ttmp15, src_scc   ; encoding: [0x05,0x00,0x6d,0xd7,0x7b,0xfa,0x01,0x00]
+
+v_cvt_pk_bf16_f32 v5, m0, 0.5
+// GFX1250: v_cvt_pk_bf16_f32 v5, m0, 0.5           ; encoding: [0x05,0x00,0x6d,0xd7,0x7d,0xe0,0x01,0x00]
+
+v_cvt_pk_bf16_f32 v5, exec_lo, -1
+// GFX1250: v_cvt_pk_bf16_f32 v5, exec_lo, -1       ; encoding: [0x05,0x00,0x6d,0xd7,0x7e,0x82,0x01,0x00]
+
+v_cvt_pk_bf16_f32 v5, exec_hi, null
+// GFX1250: v_cvt_pk_bf16_f32 v5, exec_hi, null     ; encoding: [0x05,0x00,0x6d,0xd7,0x7f,0xf8,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, null, exec_lo
+// GFX1250: v_cvt_pk_bf16_f32 v5, null, exec_lo     ; encoding: [0x05,0x00,0x6d,0xd7,0x7c,0xfc,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, -1, exec_hi
+// GFX1250: v_cvt_pk_bf16_f32 v5, -1, exec_hi       ; encoding: [0x05,0x00,0x6d,0xd7,0xc1,0xfe,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, 0.5, m0 mul:2
+// GFX1250: v_cvt_pk_bf16_f32 v5, 0.5, m0 mul:2     ; encoding: [0x05,0x00,0x6d,0xd7,0xf0,0xfa,0x00,0x08]
+
+v_cvt_pk_bf16_f32 v5, src_scc, vcc_lo mul:4
+// GFX1250: v_cvt_pk_bf16_f32 v5, src_scc, vcc_lo mul:4 ; encoding: [0x05,0x00,0x6d,0xd7,0xfd,0xd4,0x00,0x10]
+
+v_cvt_pk_bf16_f32 v255, -|0xaf123456|, vcc_hi clamp div:2
+// GFX1250: v_cvt_pk_bf16_f32 v255, -|0xaf123456|, vcc_hi clamp div:2 ; encoding: [0xff,0x81,0x6d,0xd7,0xff,0xd6,0x00,0x38,0x56,0x34,0x12,0xaf]
diff --git a/llvm/test/MC/AMDGPU/gfx1250_asm_vop3.s b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3.s
index 553eacc8e7b61..e1165faf59d9c 100644
--- a/llvm/test/MC/AMDGPU/gfx1250_asm_vop3.s
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3.s
@@ -15,3 +15,48 @@ v_lshl_add_u64 v[2:3], s[4:5], 4, v[2:3]
 
 v_lshl_add_u64 v[2:3], v[4:5], v7, 12345
 // GFX1250: v_lshl_add_u64 v[2:3], v[4:5], v7, 0x3039 ; encoding: [0x02,0x00,0x52,0xd6,0x04,0x0f,0xfe,0x03,0x39,0x30,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, v1, v2
+// GFX1250: v_cvt_pk_bf16_f32 v5, v1, v2            ; encoding: [0x05,0x00,0x6d,0xd7,0x01,0x05,0x02,0x00]
+
+v_cvt_pk_bf16_f32 v5, v255, v255
+// GFX1250: v_cvt_pk_bf16_f32 v5, v255, v255        ; encoding: [0x05,0x00,0x6d,0xd7,0xff,0xff,0x03,0x00]
+
+v_cvt_pk_bf16_f32 v5, s1, s2
+// GFX1250: v_cvt_pk_bf16_f32 v5, s1, s2            ; encoding: [0x05,0x00,0x6d,0xd7,0x01,0x04,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, s105, s105
+// GFX1250: v_cvt_pk_bf16_f32 v5, s105, s105        ; encoding: [0x05,0x00,0x6d,0xd7,0x69,0xd2,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, vcc_lo, ttmp15
+// GFX1250: v_cvt_pk_bf16_f32 v5, vcc_lo, ttmp15    ; encoding: [0x05,0x00,0x6d,0xd7,0x6a,0xf6,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, vcc_hi, 0xaf123456
+// GFX1250: v_cvt_pk_bf16_f32 v5, vcc_hi, 0xaf123456 ; encoding: [0x05,0x00,0x6d,0xd7,0x6b,0xfe,0x01,0x00,0x56,0x34,0x12,0xaf]
+
+v_cvt_pk_bf16_f32 v5, ttmp15, src_scc
+// GFX1250: v_cvt_pk_bf16_f32 v5, ttmp15, src_scc   ; encoding: [0x05,0x00,0x6d,0xd7,0x7b,0xfa,0x01,0x00]
+
+v_cvt_pk_bf16_f32 v5, m0, 0.5
+// GFX1250: v_cvt_pk_bf16_f32 v5, m0, 0.5           ; encoding: [0x05,0x00,0x6d,0xd7,0x7d,0xe0,0x01,0x00]
+
+v_cvt_pk_bf16_f32 v5, exec_lo, -1
+// GFX1250: v_cvt_pk_bf16_f32 v5, exec_lo, -1       ; encoding: [0x05,0x00,0x6d,0xd7,0x7e,0x82,0x01,0x00]
+
+v_cvt_pk_bf16_f32 v5, exec_hi, null
+// GFX1250: v_cvt_pk_bf16_f32 v5, exec_hi, null     ; encoding: [0x05,0x00,0x6d,0xd7,0x7f,0xf8,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, null, exec_lo
+// GFX1250: v_cvt_pk_bf16_f32 v5, null, exec_lo     ; encoding: [0x05,0x00,0x6d,0xd7,0x7c,0xfc,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, -1, exec_hi
+// GFX1250: v_cvt_pk_bf16_f32 v5, -1, exec_hi       ; encoding: [0x05,0x00,0x6d,0xd7,0xc1,0xfe,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, 0.5, m0 mul:2
+// GFX1250: v_cvt_pk_bf16_f32 v5, 0.5, m0 mul:2     ; encoding: [0x05,0x00,0x6d,0xd7,0xf0,0xfa,0x00,0x08]
+
+v_cvt_pk_bf16_f32 v5, src_scc, vcc_lo mul:4
+// GFX1250: v_cvt_pk_bf16_f32 v5, src_scc, vcc_lo mul:4 ; encoding: [0x05,0x00,0x6d,0xd7,0xfd,0xd4,0x00,0x10]
+
+v_cvt_pk_bf16_f32 v255, -|0xaf123456|, vcc_hi clamp div:2
+// GFX1250: v_cvt_pk_bf16_f32 v255, -|0xaf123456|, vcc_hi clamp div:2 ; encoding: [0xff,0x81,0x6d,0xd7,0xff,0xd6,0x00,0x38,0x56,0x34,0x12,0xaf]
diff --git a/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_dpp16-fake16.s b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_dpp16-fake16.s
new file mode 100644
index 0000000000000..bc910b9dd18e9
--- /dev/null
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_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-prefix=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_pk_bf16_f32_e64_dpp v5, v1, v2 quad_perm:[3,2,1,0]
+// GFX1250: v_cvt_pk_bf16_f32_e64_dpp v5, v1, v2 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x6d,0xd7,0xfa,0x04,0x02,0x00,0x01,0x1b,0x00,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_pk_bf16_f32_e64_dpp v5, v1, v2 quad_perm:[0,1,2,3]
+// GFX1250: v_cvt_pk_bf16_f32_e64_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x6d,0xd7,0xfa,0x04,0x02,0x00,0x01,0xe4,0x00,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_pk_bf16_f32_e64_dpp v5, v1, v2 row_mirror
+// GFX1250: v_cvt_pk_bf16_f32_e64_dpp v5, v1, v2 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x6d,0xd7,0xfa,0x04,0x02,0x00,0x01,0x40,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_pk_bf16_f32_e64_dpp v5, v1, v2 row_half_mirror
+// GFX1250: v_cvt_pk_bf16_f32_e64_dpp v5, v1, v2 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x6d,0xd7,0xfa,0x04,0x02,0x00,0x01,0x41,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_pk_bf16_f32_e64_dpp v5, v1, v2 row_shl:1
+// GFX1250: v_cvt_pk_bf16_f32_e64_dpp v5, v1, v2 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x6d,0xd7,0xfa,0x04,0x02,0x00,0x01,0x01,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_pk_bf16_f32_e64_dpp v5, v1, v2 row_shl:15
+// GFX1250: v_cvt_pk_bf16_f32_e64_dpp v5...
[truncated]

@llvmbot
Copy link
Member

llvmbot commented Jul 22, 2025

@llvm/pr-subscribers-backend-amdgpu

Author: Shilei Tian (shiltian)

Changes

Co-authored-by: Mekhanoshin, Stanislav <[email protected]>

Patch is 39.48 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/150053.diff

11 Files Affected:

  • (modified) llvm/lib/Target/AMDGPU/VOP3Instructions.td (+1)
  • (modified) llvm/test/CodeGen/AMDGPU/bf16-conversions.ll (+149)
  • (modified) llvm/test/MC/AMDGPU/gfx1250_asm_vop3-fake16.s (+45)
  • (modified) llvm/test/MC/AMDGPU/gfx1250_asm_vop3.s (+45)
  • (added) llvm/test/MC/AMDGPU/gfx1250_asm_vop3_dpp16-fake16.s (+59)
  • (added) llvm/test/MC/AMDGPU/gfx1250_asm_vop3_dpp16.s (+59)
  • (added) llvm/test/MC/AMDGPU/gfx1250_asm_vop3_dpp8-fake16.s (+19)
  • (added) llvm/test/MC/AMDGPU/gfx1250_asm_vop3_dpp8.s (+19)
  • (modified) llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop3.txt (+46)
  • (added) llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop3_dpp16.txt (+45)
  • (added) llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vop3_dpp8.txt (+15)
diff --git a/llvm/lib/Target/AMDGPU/VOP3Instructions.td b/llvm/lib/Target/AMDGPU/VOP3Instructions.td
index aee2f2cb3d771..b6f95686704be 100644
--- a/llvm/lib/Target/AMDGPU/VOP3Instructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP3Instructions.td
@@ -1918,6 +1918,7 @@ let AssemblerPredicate = isGFX11Plus in {
 
 // These instructions differ from GFX12 variant by supporting DPP:
 defm V_LSHL_ADD_U64                  : VOP3Only_Realtriple_gfx1250<0x252>;
+defm V_CVT_PK_BF16_F32               : VOP3Only_Realtriple_gfx1250<0x36d>;
 
 //===----------------------------------------------------------------------===//
 // GFX10.
diff --git a/llvm/test/CodeGen/AMDGPU/bf16-conversions.ll b/llvm/test/CodeGen/AMDGPU/bf16-conversions.ll
index 5b4866c386793..6823a472c3ac6 100644
--- a/llvm/test/CodeGen/AMDGPU/bf16-conversions.ll
+++ b/llvm/test/CodeGen/AMDGPU/bf16-conversions.ll
@@ -1,6 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
 ; RUN: llc -mtriple=amdgcn -mcpu=gfx942 < %s | FileCheck --check-prefixes=GCN,GFX-942 %s
 ; RUN: llc -mtriple=amdgcn -mcpu=gfx950 < %s | FileCheck --check-prefixes=GCN,GFX-950 %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1250 < %s | FileCheck --check-prefix=GFX1250 %s
 
 ; TODO: Add global-isel when it can support bf16
 
@@ -9,6 +10,11 @@ define amdgpu_ps float @v_test_cvt_bf16_f32_v(bfloat %v) {
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
 ; GCN-NEXT:    ; return to shader part epilog
+;
+; GFX1250-LABEL: v_test_cvt_bf16_f32_v:
+; GFX1250:       ; %bb.0:
+; GFX1250-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
+; GFX1250-NEXT:    ; return to shader part epilog
   %cvt = fpext bfloat %v to float
   ret float %cvt
 }
@@ -19,6 +25,13 @@ define amdgpu_ps float @v_test_cvt_bf16_f32_s(bfloat inreg %v) {
 ; GCN-NEXT:    s_lshl_b32 s0, s0, 16
 ; GCN-NEXT:    v_mov_b32_e32 v0, s0
 ; GCN-NEXT:    ; return to shader part epilog
+;
+; GFX1250-LABEL: v_test_cvt_bf16_f32_s:
+; GFX1250:       ; %bb.0:
+; GFX1250-NEXT:    s_lshl_b32 s0, s0, 16
+; GFX1250-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1250-NEXT:    v_mov_b32_e32 v0, s0
+; GFX1250-NEXT:    ; return to shader part epilog
   %cvt = fpext bfloat %v to float
   ret float %cvt
 }
@@ -47,6 +60,11 @@ define amdgpu_ps float @v_test_cvt_v2f32_v2bf16_v(<2 x float> %src) {
 ; GFX-950:       ; %bb.0:
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, v0, v1
 ; GFX-950-NEXT:    ; return to shader part epilog
+;
+; GFX1250-LABEL: v_test_cvt_v2f32_v2bf16_v:
+; GFX1250:       ; %bb.0:
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, v0, v1
+; GFX1250-NEXT:    ; return to shader part epilog
   %res = fptrunc <2 x float> %src to <2 x bfloat>
   %cast = bitcast <2 x bfloat> %res to float
   ret float %cast
@@ -80,6 +98,11 @@ define amdgpu_ps float @v_test_cvt_v2f32_v2bf16_s(<2 x float> inreg %src) {
 ; GFX-950-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, s0, v0
 ; GFX-950-NEXT:    ; return to shader part epilog
+;
+; GFX1250-LABEL: v_test_cvt_v2f32_v2bf16_s:
+; GFX1250:       ; %bb.0:
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, s0, s1
+; GFX1250-NEXT:    ; return to shader part epilog
   %res = fptrunc <2 x float> %src to <2 x bfloat>
   %cast = bitcast <2 x bfloat> %res to float
   ret float %cast
@@ -103,6 +126,13 @@ define amdgpu_ps float @v_test_cvt_f32_bf16_v(float %src) {
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
 ; GFX-950-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
 ; GFX-950-NEXT:    ; return to shader part epilog
+;
+; GFX1250-LABEL: v_test_cvt_f32_bf16_v:
+; GFX1250:       ; %bb.0:
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
+; GFX1250-NEXT:    ; return to shader part epilog
   %trunc = fptrunc float %src to bfloat
   %ext = fpext bfloat %trunc to float
   ret float %ext
@@ -172,6 +202,38 @@ define amdgpu_ps float @v_test_cvt_v2f64_v2bf16_v(<2 x double> %src) {
 ; GFX-950-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, v0, v4
 ; GFX-950-NEXT:    ; return to shader part epilog
+;
+; GFX1250-LABEL: v_test_cvt_v2f64_v2bf16_v:
+; GFX1250:       ; %bb.0:
+; GFX1250-NEXT:    v_cvt_f32_f64_e32 v8, v[2:3]
+; GFX1250-NEXT:    v_cvt_f32_f64_e32 v9, v[0:1]
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[4:5], v8
+; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[6:7], v9
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX1250-NEXT:    v_cmp_gt_f64_e64 s1, |v[2:3]|, |v[4:5]|
+; GFX1250-NEXT:    v_cmp_nlg_f64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1250-NEXT:    v_cmp_nlg_f64_e64 s0, v[0:1], v[6:7]
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1250-NEXT:    v_cndmask_b32_e64 v2, -1, 1, s1
+; GFX1250-NEXT:    v_cmp_gt_f64_e64 s1, |v[0:1]|, |v[6:7]|
+; GFX1250-NEXT:    v_dual_add_nc_u32 v1, v8, v2 :: v_dual_bitop2_b32 v10, 1, v8 bitop3:0x40
+; GFX1250-NEXT:    s_wait_alu 0xf1ff
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX1250-NEXT:    v_cndmask_b32_e64 v0, -1, 1, s1
+; GFX1250-NEXT:    v_and_b32_e32 v11, 1, v9
+; GFX1250-NEXT:    v_cmp_eq_u32_e64 s1, 1, v10
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX1250-NEXT:    v_add_nc_u32_e32 v0, v9, v0
+; GFX1250-NEXT:    v_cmp_eq_u32_e64 s2, 1, v11
+; GFX1250-NEXT:    s_or_b32 vcc_lo, s1, vcc_lo
+; GFX1250-NEXT:    v_cndmask_b32_e32 v1, v1, v8, vcc_lo
+; GFX1250-NEXT:    s_or_b32 vcc_lo, s2, s0
+; GFX1250-NEXT:    s_wait_alu 0xfffe
+; GFX1250-NEXT:    v_cndmask_b32_e32 v0, v0, v9, vcc_lo
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, v0, v1
+; GFX1250-NEXT:    ; return to shader part epilog
   %res = fptrunc <2 x double> %src to <2 x bfloat>
   %cast = bitcast <2 x bfloat> %res to float
   ret float %cast
@@ -201,6 +263,11 @@ define amdgpu_ps float @fptrunc_f32_f32_to_v2bf16(float %a, float %b) {
 ; GFX-950:       ; %bb.0: ; %entry
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, v0, v1
 ; GFX-950-NEXT:    ; return to shader part epilog
+;
+; GFX1250-LABEL: fptrunc_f32_f32_to_v2bf16:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, v0, v1
+; GFX1250-NEXT:    ; return to shader part epilog
 entry:
   %a.cvt = fptrunc float %a to bfloat
   %b.cvt = fptrunc float %b to bfloat
@@ -236,6 +303,11 @@ define amdgpu_ps float @fptrunc_f32_f32_to_v2bf16_mods(float %a, float %b) {
 ; GFX-950:       ; %bb.0: ; %entry
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, -v0, |v1|
 ; GFX-950-NEXT:    ; return to shader part epilog
+;
+; GFX1250-LABEL: fptrunc_f32_f32_to_v2bf16_mods:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, -v0, |v1|
+; GFX1250-NEXT:    ; return to shader part epilog
 entry:
   %a.neg = fneg float %a
   %a.cvt = fptrunc float %a.neg to bfloat
@@ -269,6 +341,13 @@ define amdgpu_ps void @fptrunc_f32_to_bf16(float %a, ptr %out) {
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
 ; GFX-950-NEXT:    flat_store_short v[2:3], v0
 ; GFX-950-NEXT:    s_endpgm
+;
+; GFX1250-LABEL: fptrunc_f32_to_bf16:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    v_dual_mov_b32 v3, v2 :: v_dual_mov_b32 v2, v1
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
+; GFX1250-NEXT:    flat_store_b16 v[2:3], v0
+; GFX1250-NEXT:    s_endpgm
 entry:
   %a.cvt = fptrunc float %a to bfloat
   store bfloat %a.cvt, ptr %out
@@ -298,6 +377,13 @@ define amdgpu_ps void @fptrunc_f32_to_bf16_abs(float %a, ptr %out) {
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, |v0|, s0
 ; GFX-950-NEXT:    flat_store_short v[2:3], v0
 ; GFX-950-NEXT:    s_endpgm
+;
+; GFX1250-LABEL: fptrunc_f32_to_bf16_abs:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    v_dual_mov_b32 v3, v2 :: v_dual_mov_b32 v2, v1
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, |v0|, s0
+; GFX1250-NEXT:    flat_store_b16 v[2:3], v0
+; GFX1250-NEXT:    s_endpgm
 entry:
   %a.abs = call float @llvm.fabs.f32(float %a)
   %a.cvt = fptrunc float %a.abs to bfloat
@@ -328,6 +414,13 @@ define amdgpu_ps void @fptrunc_f32_to_bf16_neg(float %a, ptr %out) {
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, -v0, s0
 ; GFX-950-NEXT:    flat_store_short v[2:3], v0
 ; GFX-950-NEXT:    s_endpgm
+;
+; GFX1250-LABEL: fptrunc_f32_to_bf16_neg:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    v_dual_mov_b32 v3, v2 :: v_dual_mov_b32 v2, v1
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, -v0, s0
+; GFX1250-NEXT:    flat_store_b16 v[2:3], v0
+; GFX1250-NEXT:    s_endpgm
 entry:
   %a.neg = fneg float %a
   %a.cvt = fptrunc float %a.neg to bfloat
@@ -373,6 +466,24 @@ define amdgpu_ps void @fptrunc_f64_to_bf16(double %a, ptr %out) {
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
 ; GFX-950-NEXT:    flat_store_short v[2:3], v0
 ; GFX-950-NEXT:    s_endpgm
+;
+; GFX1250-LABEL: fptrunc_f64_to_bf16:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    v_cvt_f32_f64_e32 v6, v[0:1]
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[4:5], v6
+; GFX1250-NEXT:    v_cmp_gt_f64_e64 s0, |v[0:1]|, |v[4:5]|
+; GFX1250-NEXT:    v_cmp_nlg_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT:    v_cndmask_b32_e64 v0, -1, 1, s0
+; GFX1250-NEXT:    v_dual_add_nc_u32 v0, v6, v0 :: v_dual_bitop2_b32 v7, 1, v6 bitop3:0x40
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1250-NEXT:    v_cmp_eq_u32_e64 s0, 1, v7
+; GFX1250-NEXT:    s_or_b32 vcc_lo, vcc_lo, s0
+; GFX1250-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc_lo
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
+; GFX1250-NEXT:    flat_store_b16 v[2:3], v0
+; GFX1250-NEXT:    s_endpgm
 entry:
   %a.cvt = fptrunc double %a to bfloat
   store bfloat %a.cvt, ptr %out
@@ -417,6 +528,25 @@ define amdgpu_ps void @fptrunc_f64_to_bf16_neg(double %a, ptr %out) {
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
 ; GFX-950-NEXT:    flat_store_short v[2:3], v0
 ; GFX-950-NEXT:    s_endpgm
+;
+; GFX1250-LABEL: fptrunc_f64_to_bf16_neg:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    v_cvt_f32_f64_e64 v6, -v[0:1]
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[4:5], v6
+; GFX1250-NEXT:    v_cmp_gt_f64_e64 s1, |v[0:1]|, |v[4:5]|
+; GFX1250-NEXT:    v_cmp_nlg_f64_e64 s0, -v[0:1], v[4:5]
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT:    v_cndmask_b32_e64 v0, -1, 1, s1
+; GFX1250-NEXT:    v_dual_add_nc_u32 v0, v6, v0 :: v_dual_bitop2_b32 v7, 1, v6 bitop3:0x40
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1250-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 1, v7
+; GFX1250-NEXT:    s_or_b32 vcc_lo, s0, vcc_lo
+; GFX1250-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc_lo
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
+; GFX1250-NEXT:    flat_store_b16 v[2:3], v0
+; GFX1250-NEXT:    s_endpgm
 entry:
   %a.neg = fneg double %a
   %a.cvt = fptrunc double %a.neg to bfloat
@@ -462,6 +592,25 @@ define amdgpu_ps void @fptrunc_f64_to_bf16_abs(double %a, ptr %out) {
 ; GFX-950-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
 ; GFX-950-NEXT:    flat_store_short v[2:3], v0
 ; GFX-950-NEXT:    s_endpgm
+;
+; GFX1250-LABEL: fptrunc_f64_to_bf16_abs:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    v_cvt_f32_f64_e64 v6, |v[0:1]|
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT:    v_cvt_f64_f32_e32 v[4:5], v6
+; GFX1250-NEXT:    v_cmp_gt_f64_e64 s1, |v[0:1]|, |v[4:5]|
+; GFX1250-NEXT:    v_cmp_nlg_f64_e64 s0, |v[0:1]|, v[4:5]
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT:    v_cndmask_b32_e64 v0, -1, 1, s1
+; GFX1250-NEXT:    v_dual_add_nc_u32 v0, v6, v0 :: v_dual_bitop2_b32 v7, 1, v6 bitop3:0x40
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1250-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 1, v7
+; GFX1250-NEXT:    s_or_b32 vcc_lo, s0, vcc_lo
+; GFX1250-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc_lo
+; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
+; GFX1250-NEXT:    flat_store_b16 v[2:3], v0
+; GFX1250-NEXT:    s_endpgm
 entry:
   %a.abs = call double @llvm.fabs.f64(double %a)
   %a.cvt = fptrunc double %a.abs to bfloat
diff --git a/llvm/test/MC/AMDGPU/gfx1250_asm_vop3-fake16.s b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3-fake16.s
index 0070c8ab9ee78..789d6f892762b 100644
--- a/llvm/test/MC/AMDGPU/gfx1250_asm_vop3-fake16.s
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3-fake16.s
@@ -15,3 +15,48 @@ v_lshl_add_u64 v[2:3], s[4:5], 4, v[2:3]
 
 v_lshl_add_u64 v[2:3], v[4:5], v7, 12345
 // GFX1250: v_lshl_add_u64 v[2:3], v[4:5], v7, 0x3039 ; encoding: [0x02,0x00,0x52,0xd6,0x04,0x0f,0xfe,0x03,0x39,0x30,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, v1, v2
+// GFX1250: v_cvt_pk_bf16_f32 v5, v1, v2            ; encoding: [0x05,0x00,0x6d,0xd7,0x01,0x05,0x02,0x00]
+
+v_cvt_pk_bf16_f32 v5, v255, v255
+// GFX1250: v_cvt_pk_bf16_f32 v5, v255, v255        ; encoding: [0x05,0x00,0x6d,0xd7,0xff,0xff,0x03,0x00]
+
+v_cvt_pk_bf16_f32 v5, s1, s2
+// GFX1250: v_cvt_pk_bf16_f32 v5, s1, s2            ; encoding: [0x05,0x00,0x6d,0xd7,0x01,0x04,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, s105, s105
+// GFX1250: v_cvt_pk_bf16_f32 v5, s105, s105        ; encoding: [0x05,0x00,0x6d,0xd7,0x69,0xd2,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, vcc_lo, ttmp15
+// GFX1250: v_cvt_pk_bf16_f32 v5, vcc_lo, ttmp15    ; encoding: [0x05,0x00,0x6d,0xd7,0x6a,0xf6,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, vcc_hi, 0xaf123456
+// GFX1250: v_cvt_pk_bf16_f32 v5, vcc_hi, 0xaf123456 ; encoding: [0x05,0x00,0x6d,0xd7,0x6b,0xfe,0x01,0x00,0x56,0x34,0x12,0xaf]
+
+v_cvt_pk_bf16_f32 v5, ttmp15, src_scc
+// GFX1250: v_cvt_pk_bf16_f32 v5, ttmp15, src_scc   ; encoding: [0x05,0x00,0x6d,0xd7,0x7b,0xfa,0x01,0x00]
+
+v_cvt_pk_bf16_f32 v5, m0, 0.5
+// GFX1250: v_cvt_pk_bf16_f32 v5, m0, 0.5           ; encoding: [0x05,0x00,0x6d,0xd7,0x7d,0xe0,0x01,0x00]
+
+v_cvt_pk_bf16_f32 v5, exec_lo, -1
+// GFX1250: v_cvt_pk_bf16_f32 v5, exec_lo, -1       ; encoding: [0x05,0x00,0x6d,0xd7,0x7e,0x82,0x01,0x00]
+
+v_cvt_pk_bf16_f32 v5, exec_hi, null
+// GFX1250: v_cvt_pk_bf16_f32 v5, exec_hi, null     ; encoding: [0x05,0x00,0x6d,0xd7,0x7f,0xf8,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, null, exec_lo
+// GFX1250: v_cvt_pk_bf16_f32 v5, null, exec_lo     ; encoding: [0x05,0x00,0x6d,0xd7,0x7c,0xfc,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, -1, exec_hi
+// GFX1250: v_cvt_pk_bf16_f32 v5, -1, exec_hi       ; encoding: [0x05,0x00,0x6d,0xd7,0xc1,0xfe,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, 0.5, m0 mul:2
+// GFX1250: v_cvt_pk_bf16_f32 v5, 0.5, m0 mul:2     ; encoding: [0x05,0x00,0x6d,0xd7,0xf0,0xfa,0x00,0x08]
+
+v_cvt_pk_bf16_f32 v5, src_scc, vcc_lo mul:4
+// GFX1250: v_cvt_pk_bf16_f32 v5, src_scc, vcc_lo mul:4 ; encoding: [0x05,0x00,0x6d,0xd7,0xfd,0xd4,0x00,0x10]
+
+v_cvt_pk_bf16_f32 v255, -|0xaf123456|, vcc_hi clamp div:2
+// GFX1250: v_cvt_pk_bf16_f32 v255, -|0xaf123456|, vcc_hi clamp div:2 ; encoding: [0xff,0x81,0x6d,0xd7,0xff,0xd6,0x00,0x38,0x56,0x34,0x12,0xaf]
diff --git a/llvm/test/MC/AMDGPU/gfx1250_asm_vop3.s b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3.s
index 553eacc8e7b61..e1165faf59d9c 100644
--- a/llvm/test/MC/AMDGPU/gfx1250_asm_vop3.s
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3.s
@@ -15,3 +15,48 @@ v_lshl_add_u64 v[2:3], s[4:5], 4, v[2:3]
 
 v_lshl_add_u64 v[2:3], v[4:5], v7, 12345
 // GFX1250: v_lshl_add_u64 v[2:3], v[4:5], v7, 0x3039 ; encoding: [0x02,0x00,0x52,0xd6,0x04,0x0f,0xfe,0x03,0x39,0x30,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, v1, v2
+// GFX1250: v_cvt_pk_bf16_f32 v5, v1, v2            ; encoding: [0x05,0x00,0x6d,0xd7,0x01,0x05,0x02,0x00]
+
+v_cvt_pk_bf16_f32 v5, v255, v255
+// GFX1250: v_cvt_pk_bf16_f32 v5, v255, v255        ; encoding: [0x05,0x00,0x6d,0xd7,0xff,0xff,0x03,0x00]
+
+v_cvt_pk_bf16_f32 v5, s1, s2
+// GFX1250: v_cvt_pk_bf16_f32 v5, s1, s2            ; encoding: [0x05,0x00,0x6d,0xd7,0x01,0x04,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, s105, s105
+// GFX1250: v_cvt_pk_bf16_f32 v5, s105, s105        ; encoding: [0x05,0x00,0x6d,0xd7,0x69,0xd2,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, vcc_lo, ttmp15
+// GFX1250: v_cvt_pk_bf16_f32 v5, vcc_lo, ttmp15    ; encoding: [0x05,0x00,0x6d,0xd7,0x6a,0xf6,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, vcc_hi, 0xaf123456
+// GFX1250: v_cvt_pk_bf16_f32 v5, vcc_hi, 0xaf123456 ; encoding: [0x05,0x00,0x6d,0xd7,0x6b,0xfe,0x01,0x00,0x56,0x34,0x12,0xaf]
+
+v_cvt_pk_bf16_f32 v5, ttmp15, src_scc
+// GFX1250: v_cvt_pk_bf16_f32 v5, ttmp15, src_scc   ; encoding: [0x05,0x00,0x6d,0xd7,0x7b,0xfa,0x01,0x00]
+
+v_cvt_pk_bf16_f32 v5, m0, 0.5
+// GFX1250: v_cvt_pk_bf16_f32 v5, m0, 0.5           ; encoding: [0x05,0x00,0x6d,0xd7,0x7d,0xe0,0x01,0x00]
+
+v_cvt_pk_bf16_f32 v5, exec_lo, -1
+// GFX1250: v_cvt_pk_bf16_f32 v5, exec_lo, -1       ; encoding: [0x05,0x00,0x6d,0xd7,0x7e,0x82,0x01,0x00]
+
+v_cvt_pk_bf16_f32 v5, exec_hi, null
+// GFX1250: v_cvt_pk_bf16_f32 v5, exec_hi, null     ; encoding: [0x05,0x00,0x6d,0xd7,0x7f,0xf8,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, null, exec_lo
+// GFX1250: v_cvt_pk_bf16_f32 v5, null, exec_lo     ; encoding: [0x05,0x00,0x6d,0xd7,0x7c,0xfc,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, -1, exec_hi
+// GFX1250: v_cvt_pk_bf16_f32 v5, -1, exec_hi       ; encoding: [0x05,0x00,0x6d,0xd7,0xc1,0xfe,0x00,0x00]
+
+v_cvt_pk_bf16_f32 v5, 0.5, m0 mul:2
+// GFX1250: v_cvt_pk_bf16_f32 v5, 0.5, m0 mul:2     ; encoding: [0x05,0x00,0x6d,0xd7,0xf0,0xfa,0x00,0x08]
+
+v_cvt_pk_bf16_f32 v5, src_scc, vcc_lo mul:4
+// GFX1250: v_cvt_pk_bf16_f32 v5, src_scc, vcc_lo mul:4 ; encoding: [0x05,0x00,0x6d,0xd7,0xfd,0xd4,0x00,0x10]
+
+v_cvt_pk_bf16_f32 v255, -|0xaf123456|, vcc_hi clamp div:2
+// GFX1250: v_cvt_pk_bf16_f32 v255, -|0xaf123456|, vcc_hi clamp div:2 ; encoding: [0xff,0x81,0x6d,0xd7,0xff,0xd6,0x00,0x38,0x56,0x34,0x12,0xaf]
diff --git a/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_dpp16-fake16.s b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_dpp16-fake16.s
new file mode 100644
index 0000000000000..bc910b9dd18e9
--- /dev/null
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vop3_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-prefix=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_pk_bf16_f32_e64_dpp v5, v1, v2 quad_perm:[3,2,1,0]
+// GFX1250: v_cvt_pk_bf16_f32_e64_dpp v5, v1, v2 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x6d,0xd7,0xfa,0x04,0x02,0x00,0x01,0x1b,0x00,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_pk_bf16_f32_e64_dpp v5, v1, v2 quad_perm:[0,1,2,3]
+// GFX1250: v_cvt_pk_bf16_f32_e64_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x6d,0xd7,0xfa,0x04,0x02,0x00,0x01,0xe4,0x00,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_pk_bf16_f32_e64_dpp v5, v1, v2 row_mirror
+// GFX1250: v_cvt_pk_bf16_f32_e64_dpp v5, v1, v2 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x6d,0xd7,0xfa,0x04,0x02,0x00,0x01,0x40,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_pk_bf16_f32_e64_dpp v5, v1, v2 row_half_mirror
+// GFX1250: v_cvt_pk_bf16_f32_e64_dpp v5, v1, v2 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x6d,0xd7,0xfa,0x04,0x02,0x00,0x01,0x41,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_pk_bf16_f32_e64_dpp v5, v1, v2 row_shl:1
+// GFX1250: v_cvt_pk_bf16_f32_e64_dpp v5, v1, v2 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x6d,0xd7,0xfa,0x04,0x02,0x00,0x01,0x01,0x01,0xff]
+// GFX12-ERR: :[[@LINE-2]]:1: error: instruction not supported on this GPU
+
+v_cvt_pk_bf16_f32_e64_dpp v5, v1, v2 row_shl:15
+// GFX1250: v_cvt_pk_bf16_f32_e64_dpp v5...
[truncated]

@shiltian
Copy link
Contributor Author

@piotrAMD FYI, the check lines of llvm/test/CodeGen/AMDGPU/bf16-conversions.ll has been re-generated.

@shiltian shiltian mentioned this pull request Jul 22, 2025
Merged
@shiltian shiltian force-pushed the users/shiltian/v_cvt_pk_bf16_f32 branch from cd79e69 to 32c56c8 Compare July 22, 2025 17:31
@shiltian Graphite App
Copy link
Contributor Author

shiltian commented Jul 22, 2025 •
edited
Loading

Merge activity

  • Jul 22, 11:55 PM UTC: A user started a stack merge that includes this pull request via Graphite.
  • Jul 22, 11:58 PM UTC: Graphite rebased this pull request as part of a merge.
  • Jul 23, 12:00 AM UTC: @shiltian merged this pull request with Graphite.

@shiltian shiltian force-pushed the users/shiltian/v_cvt_pk_bf16_f32 branch from 32c56c8 to b37ed4d Compare July 22, 2025 23:57
@shiltian shiltian merged commit fbeb801 into main Jul 23, 2025
7 of 9 checks passed
@shiltian shiltian deleted the users/shiltian/v_cvt_pk_bf16_f32 branch July 23, 2025 00:00
mahesh-attarde pushed a commit to mahesh-attarde/llvm-project that referenced this pull request Jul 28, 2025
@shiltian @rampitec @mahesh-attarde
[AMDGPU] Add support for v_cvt_pk_bf16_f32 on gfx1250 (llvm#150053)
a24f81b 
Co-authored-by: Mekhanoshin, Stanislav <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@changpeng changpeng changpeng approved these changes

@rampitec rampitec Awaiting requested review from rampitec

Assignees
No one assigned
Labels
backend:AMDGPU llvm:mc Machine (object) code
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@shiltian @llvmbot @changpeng