aboutsummaryrefslogtreecommitdiff
path: root/llvm/lib/Target/AMDGPU/VOP1Instructions.td
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/lib/Target/AMDGPU/VOP1Instructions.td')
-rw-r--r--llvm/lib/Target/AMDGPU/VOP1Instructions.td14
1 files changed, 14 insertions, 0 deletions
diff --git a/llvm/lib/Target/AMDGPU/VOP1Instructions.td b/llvm/lib/Target/AMDGPU/VOP1Instructions.td
index 1bbbb61..f621f85 100644
--- a/llvm/lib/Target/AMDGPU/VOP1Instructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP1Instructions.td
@@ -803,6 +803,9 @@ let SubtargetPredicate = isGFX1250Plus in {
def : Cvt_F_F8_Pat_ByteSel<int_amdgcn_cvt_f16_fp8, V_CVT_F16_FP8_fake16_e64, 1>;
def : Cvt_F_F8_Pat_ByteSel<int_amdgcn_cvt_f16_bf8, V_CVT_F16_BF8_fake16_e64, 1>;
}
+
+ defm V_SAT_PK4_I4_I8 : VOP1Inst_t16<"v_sat_pk4_i4_i8", VOP1_I16_I32, int_amdgcn_sat_pk4_i4_i8>;
+ defm V_SAT_PK4_U4_U8 : VOP1Inst_t16<"v_sat_pk4_u4_u8", VOP1_I16_I32, int_amdgcn_sat_pk4_u4_u8>;
} // End SubtargetPredicate = isGFX1250Plus
let SubtargetPredicate = isGFX10Plus in {
@@ -1080,6 +1083,13 @@ multiclass VOP1_Real_FULL_t16_and_fake16_gfx1250<
VOP1_Real_FULL_with_name<GFX1250Gen, op, opName#"_fake16", asmName>;
}
+multiclass VOP1_Real_OpSelIsDPP_gfx1250<bits<9> op> : VOP1_Real_e32<GFX1250Gen, op> {
+ defvar ps = !cast<VOP_Pseudo>(NAME#"_e64");
+ def _e64_gfx1250 :
+ VOP3_Real_Gen<ps, 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">;
@@ -1147,8 +1157,12 @@ defm V_MOV_B64 : VOP1_Real_FULL <GFX1250Gen, 0x1d>;
defm V_TANH_F32 : VOP1_Real_FULL<GFX1250Gen, 0x01e>;
defm V_TANH_F16 : VOP1_Real_FULL_t16_and_fake16_gfx1250<0x01f>;
+defm V_PERMLANE16_SWAP_B32 : VOP1_Real_OpSelIsDPP_gfx1250<0x049>;
defm V_TANH_BF16 : VOP1_Real_FULL_t16_and_fake16_gfx1250<0x04a>;
+defm V_PRNG_B32 : VOP1_Real_FULL<GFX1250Gen, 0x04b>;
defm V_CVT_F32_BF16 : VOP1_Real_FULL_t16_and_fake16_gfx1250<0x072, "v_cvt_f32_bf16", "V_CVT_F32_BF16_gfx1250">;
+defm V_SAT_PK4_I4_I8 : VOP1_Real_FULL_t16_and_fake16_gfx1250<0x073>;
+defm V_SAT_PK4_U4_U8 : VOP1_Real_FULL_t16_and_fake16_gfx1250<0x074>;
defm V_CVT_PK_F16_FP8 : VOP1_Real_FULL_t16_and_fake16_gfx1250<0x075>;
defm V_CVT_PK_F16_BF8 : VOP1_Real_FULL_t16_and_fake16_gfx1250<0x076>;
defm V_CVT_F16_FP8 : VOP1_Real_FULL_t16_and_fake16_gfx1250<0x077>;