AMDGPU/EG,CM: Add fp16 conversion instructions

[jvesely]

Summary

Passes vstore_half(float) piglit on Turks.

[jvesely]

ping. this one is rather straightforward.

[arsenm]

LGTM

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

 def MUL_LIT_eg : MUL_LIT_Common<0x1F>;
 def LOG_CLAMPED_eg : LOG_CLAMPED_Common<0x82>;
 def MUL_UINT24_eg : R600_2OP <0xB5, "MUL_UINT24",
   [(set i32:$dst, (AMDGPUmul_u24 i32:$src0, i32:$src1))], VecALU
 >;
 def DOT4_eg : DOT4_Common<0xBE>;
 defm CUBE_eg : CUBE_Common<0xC0>;
 
-def BCNT_INT : R600_1OP_Helper <0xAA, "BCNT_INT", ctpop, VecALU>;
 
 def ADDC_UINT : R600_2OP_Helper <0x52, "ADDC_UINT", AMDGPUcarry>;
 def SUBB_UINT : R600_2OP_Helper <0x53, "SUBB_UINT", AMDGPUborrow>;
 
+def FLT32_TO_FLT16 : R600_1OP_Helper <0xA2, "FLT32_TO_FLT16", fp_to_f16, VecALU>;
+def FLT16_TO_FLT32 : R600_1OP_Helper <0xA3, "FLT16_TO_FLT32", f16_to_fp, VecALU>;
+def BCNT_INT : R600_1OP_Helper <0xAA, "BCNT_INT", ctpop, VecALU>;
 def FFBH_UINT : R600_1OP_Helper <0xAB, "FFBH_UINT", AMDGPUffbh_u32, VecALU>;
 def FFBL_INT : R600_1OP_Helper <0xAC, "FFBL_INT", cttz_zero_undef, VecALU>;
 
 let hasSideEffects = 1 in {
   def MOVA_INT_eg : R600_1OP <0xCC, "MOVA_INT", [], VecALU>;
 }
 
 def FLT_TO_INT_eg : FLT_TO_INT_Common<0x50> {

llvm/trunk/test/CodeGen/AMDGPU/fp16_to_fp.ll

-; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s
-; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s
-
-declare float @llvm.convert.from.fp16.f32(i16) nounwind readnone
-declare double @llvm.convert.from.fp16.f64(i16) nounwind readnone
-
-; SI-LABEL: {{^}}test_convert_fp16_to_fp32:
-; SI: buffer_load_ushort [[VAL:v[0-9]+]]
-; SI: v_cvt_f32_f16_e32 [[RESULT:v[0-9]+]], [[VAL]]
-; SI: buffer_store_dword [[RESULT]]
-define void @test_convert_fp16_to_fp32(float addrspace(1)* noalias %out, i16 addrspace(1)* noalias %in) nounwind {
-  %val = load i16, i16 addrspace(1)* %in, align 2
-  %cvt = call float @llvm.convert.from.fp16.f32(i16 %val) nounwind readnone
-  store float %cvt, float addrspace(1)* %out, align 4
-  ret void
-}
-
-
-; SI-LABEL: {{^}}test_convert_fp16_to_fp64:
-; SI: buffer_load_ushort [[VAL:v[0-9]+]]
-; SI: v_cvt_f32_f16_e32 [[RESULT32:v[0-9]+]], [[VAL]]
-; SI: v_cvt_f64_f32_e32 [[RESULT:v\[[0-9]+:[0-9]+\]]], [[RESULT32]]
-; SI: buffer_store_dwordx2 [[RESULT]]
-define void @test_convert_fp16_to_fp64(double addrspace(1)* noalias %out, i16 addrspace(1)* noalias %in) nounwind {
-  %val = load i16, i16 addrspace(1)* %in, align 2
-  %cvt = call double @llvm.convert.from.fp16.f64(i16 %val) nounwind readnone
-  store double %cvt, double addrspace(1)* %out, align 4
-  ret void
-}

llvm/trunk/test/CodeGen/AMDGPU/fp16_to_fp32.ll

+; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=FUNC %s
+; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=FUNC %s
+; RUN: llc -march=r600 -mcpu=cypress -verify-machineinstrs < %s | FileCheck -check-prefix=EG -check-prefix=EGCM -check-prefix=FUNC %s
+; RUN: llc -march=r600 -mcpu=cayman -verify-machineinstrs < %s | FileCheck -check-prefix=CM -check-prefix=EGCM -check-prefix=FUNC %s
+
+declare float @llvm.convert.from.fp16.f32(i16) nounwind readnone
+
+; FUNC-LABEL: {{^}}test_convert_fp16_to_fp32:
+; GCN: buffer_load_ushort [[VAL:v[0-9]+]]
+; GCN: v_cvt_f32_f16_e32 [[RESULT:v[0-9]+]], [[VAL]]
+; GCN: buffer_store_dword [[RESULT]]
+
+; EG: MEM_RAT_CACHELESS STORE_RAW [[RES:T[0-9]+\.[XYZW]]]
+; CM: MEM_RAT_CACHELESS STORE_DWORD [[RES:T[0-9]+\.[XYZW]]]
+; EGCM: VTX_READ_16 [[VAL:T[0-9]+\.[XYZW]]]
+; EGCM: FLT16_TO_FLT32{{[ *]*}}[[RES]], [[VAL]]
+define void @test_convert_fp16_to_fp32(float addrspace(1)* noalias %out, i16 addrspace(1)* noalias %in) nounwind {
+  %val = load i16, i16 addrspace(1)* %in, align 2
+  %cvt = call float @llvm.convert.from.fp16.f32(i16 %val) nounwind readnone
+  store float %cvt, float addrspace(1)* %out, align 4
+  ret void
+}

llvm/trunk/test/CodeGen/AMDGPU/fp16_to_fp64.ll

+; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=FUNC %s
+; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=FUNC %s
+
+declare double @llvm.convert.from.fp16.f64(i16) nounwind readnone
+
+; FUNC-LABEL: {{^}}test_convert_fp16_to_fp64:
+; GCN: buffer_load_ushort [[VAL:v[0-9]+]]
+; GCN: v_cvt_f32_f16_e32 [[RESULT32:v[0-9]+]], [[VAL]]
+; GCN: v_cvt_f64_f32_e32 [[RESULT:v\[[0-9]+:[0-9]+\]]], [[RESULT32]]
+; GCN: buffer_store_dwordx2 [[RESULT]]
+define void @test_convert_fp16_to_fp64(double addrspace(1)* noalias %out, i16 addrspace(1)* noalias %in) nounwind {
+  %val = load i16, i16 addrspace(1)* %in, align 2
+  %cvt = call double @llvm.convert.from.fp16.f64(i16 %val) nounwind readnone
+  store double %cvt, double addrspace(1)* %out, align 4
+  ret void
+}

llvm/trunk/test/CodeGen/AMDGPU/fp32_to_fp16.ll

-; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s
-; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s
+; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=FUNC %s
+; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=FUNC %s
+; RUN: llc -march=r600 -mcpu=cypress -verify-machineinstrs < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
 
 declare i16 @llvm.convert.to.fp16.f32(float) nounwind readnone
 
-; SI-LABEL: {{^}}test_convert_fp32_to_fp16:
-; SI: buffer_load_dword [[VAL:v[0-9]+]]
-; SI: v_cvt_f16_f32_e32 [[RESULT:v[0-9]+]], [[VAL]]
-; SI: buffer_store_short [[RESULT]]
+; FUNC-LABEL: {{^}}test_convert_fp32_to_fp16:
+; GCN: buffer_load_dword [[VAL:v[0-9]+]]
+; GCN: v_cvt_f16_f32_e32 [[RESULT:v[0-9]+]], [[VAL]]
+; GCN: buffer_store_short [[RESULT]]
+
+; EG: MEM_RAT MSKOR
+; EG: VTX_READ_32
+; EG: FLT32_TO_FLT16
 define void @test_convert_fp32_to_fp16(i16 addrspace(1)* noalias %out, float addrspace(1)* noalias %in) nounwind {
   %val = load float, float addrspace(1)* %in, align 4
   %cvt = call i16 @llvm.convert.to.fp16.f32(float %val) nounwind readnone
   store i16 %cvt, i16 addrspace(1)* %out, align 2
   ret void
 }