AMDGPU : Add intrinsic for instruction v_cvt_pk_u8_f32

[wdng]

Summary

AMDGPU : Add intrinsic for instruction v_cvt_pk_u8_f32

[arsenm]

There should probably be constant folding for this put into InstCombineCalls

include/llvm/IR/IntrinsicsAMDGPU.td

410

Remove the name comment

test/CodeGen/AMDGPU/v_cvt_pk_u8_f32.ll

8–12

You should have a test for 0..3

[wdng]

1. Remove comments
2. Add 0...3 LIT tests and one test for all.

[• tstellarAMD]

Just a few minor comments, with those changes LGTM.

include/llvm/IR/IntrinsicsAMDGPU.td

415–419

This seems like an unrelated whitespace change.

lib/Target/AMDGPU/SIInstructions.td

1568

Unnecessary whitespace change

test/CodeGen/AMDGPU/v_cvt_pk_u8_f32.ll

62

We don't need the #1 attribute on these test functions.

include/llvm/IR/IntrinsicsAMDGPU.td

   GCCBuiltin<"__builtin_amdgcn_lerp">,
   Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
 
+// llvm.amdgcn.cvt.pk.u8.f32
+def int_amdgcn_cvt_pk_u8_f32 :
+  GCCBuiltin<"__builtin_amdgcn_cvt_pk_u8_f32">,
+  Intrinsic<[llvm_i32_ty], [llvm_float_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+
 def int_amdgcn_icmp :
   Intrinsic<[llvm_i64_ty], [llvm_anyint_ty, LLVMMatchType<0>, llvm_i32_ty], 
             [IntrNoMem, IntrConvergent]>;

lib/Target/AMDGPU/SIInstrInfo.td

 def VOP_F64_F64_F64_F64 : VOPProfile <[f64, f64, f64, f64]>;
 def VOP_I32_I32_I32_I32 : VOPProfile <[i32, i32, i32, i32]>;
 def VOP_I64_I32_I32_I64 : VOPProfile <[i64, i32, i32, i64]>;
+def VOP_I32_F32_I32_I32 : VOPProfile <[i32, f32, i32, i32]>;
 
 // This class is used only with VOPC instructions. Use $sdst for out operand
 class SIInstAlias <string asm, Instruction inst, VOPProfile p> :

lib/Target/AMDGPU/SIInstructions.td

   VOP_I32_I32_I32_I32, int_amdgcn_lerp
 >;
 } // End isCommutable = 1
   
 //def V_LERP_U8 : VOP3_U8 <0x0000014d, "v_lerp_u8", []>;
 defm V_ALIGNBIT_B32 : VOP3Inst <vop3<0x14e, 0x1ce>, "v_alignbit_b32",
   VOP_I32_I32_I32_I32

 defm V_SAD_U32 : VOP3Inst <vop3<0x15d, 0x1dc>, "v_sad_u32",
   VOP_I32_I32_I32_I32
 >;
+
+defm V_CVT_PK_U8_F32 : VOP3Inst<vop3<0x15e, 0x1dd>, "v_cvt_pk_u8_f32",
+  VOP_I32_F32_I32_I32, int_amdgcn_cvt_pk_u8_f32
+>;
+
 //def V_CVT_PK_U8_F32 : VOP3_U8 <0x0000015e, "v_cvt_pk_u8_f32", []>;
 defm V_DIV_FIXUP_F32 : VOP3Inst <
   vop3<0x15f, 0x1de>, "v_div_fixup_f32", VOP_F32_F32_F32_F32, AMDGPUdiv_fixup

test/CodeGen/AMDGPU/v_cvt_pk_u8_f32.ll

+; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
+; RUN: llc -march=amdgcn -mcpu=fiji -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
+
+declare i32 @llvm.amdgcn.cvt.pk.u8.f32(float, i32, i32) #0
+
+; GCN-LABEL: {{^}}v_cvt_pk_u8_f32:
+; GCN: v_cvt_pk_u8_f32 v{{[0-9]+}}, s{{[0-9]+}}, 1, v{{[0-9]+}}
+define void @v_cvt_pk_u8_f32(i32 addrspace(1)* %out, float %src, i32 %reg) #1 {
+  %result = call i32 @llvm.amdgcn.cvt.pk.u8.f32(float %src, i32 1, i32 %reg) #0
+  store i32 %result, i32 addrspace(1)* %out, align 4
+  ret void
+}
+
+; GCN-LABEL: {{^}}v_cvt_pk_u8_f32_idx:
+; GCN: v_cvt_pk_u8_f32 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
+define void @v_cvt_pk_u8_f32_idx(i32 addrspace(1)* %out, float %src, i32 %idx, i32 %reg) #1 {
+  %result = call i32 @llvm.amdgcn.cvt.pk.u8.f32(float %src, i32 %idx, i32 %reg) #0
+  store i32 %result, i32 addrspace(1)* %out, align 4
+  ret void
+}
+
+attributes #0 = { nounwind readnone }
+attributes #1 = { nounwind }