From 6f729b28204c9f0252da4dbe57b992ce362b3fe2 Mon Sep 17 00:00:00 2001 From: Evan Cheng Date: Wed, 19 Mar 2008 23:38:52 +0000 Subject: [PATCH] Add intrinsics to match mmx shift builtin's with immediate operand. llvm-svn: 48569 --- include/llvm/IntrinsicsX86.td | 24 ++++++++++++++++++++++++ lib/Target/X86/X86InstrMMX.td | 22 +++++++++++----------- lib/VMCore/AutoUpgrade.cpp | 3 ++- test/CodeGen/X86/mmx-shift.ll | 20 ++++++++++++++++++++ 4 files changed, 57 insertions(+), 12 deletions(-) create mode 100644 test/CodeGen/X86/mmx-shift.ll diff --git a/include/llvm/IntrinsicsX86.td b/include/llvm/IntrinsicsX86.td index 5bcfa7922ba..e40097e897c 100644 --- a/include/llvm/IntrinsicsX86.td +++ b/include/llvm/IntrinsicsX86.td @@ -916,29 +916,53 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". def int_x86_mmx_psll_w : GCCBuiltin<"__builtin_ia32_psllw">, Intrinsic<[llvm_v4i16_ty, llvm_v4i16_ty, llvm_v1i64_ty], [IntrNoMem]>; + def int_x86_mmx_pslli_w : GCCBuiltin<"__builtin_ia32_psllwi">, + Intrinsic<[llvm_v4i16_ty, llvm_v4i16_ty, + llvm_i32_ty], [IntrNoMem]>; def int_x86_mmx_psll_d : GCCBuiltin<"__builtin_ia32_pslld">, Intrinsic<[llvm_v2i32_ty, llvm_v2i32_ty, llvm_v1i64_ty], [IntrNoMem]>; + def int_x86_mmx_pslli_d : GCCBuiltin<"__builtin_ia32_pslldi">, + Intrinsic<[llvm_v2i32_ty, llvm_v2i32_ty, + llvm_i32_ty], [IntrNoMem]>; def int_x86_mmx_psll_q : GCCBuiltin<"__builtin_ia32_psllq">, Intrinsic<[llvm_v1i64_ty, llvm_v1i64_ty, llvm_v1i64_ty], [IntrNoMem]>; + def int_x86_mmx_pslli_q : GCCBuiltin<"__builtin_ia32_psllqi">, + Intrinsic<[llvm_v1i64_ty, llvm_v1i64_ty, + llvm_i32_ty], [IntrNoMem]>; def int_x86_mmx_psrl_w : GCCBuiltin<"__builtin_ia32_psrlw">, Intrinsic<[llvm_v4i16_ty, llvm_v4i16_ty, llvm_v1i64_ty], [IntrNoMem]>; + def int_x86_mmx_psrli_w : GCCBuiltin<"__builtin_ia32_psrlwi">, + Intrinsic<[llvm_v4i16_ty, llvm_v4i16_ty, + llvm_i32_ty], [IntrNoMem]>; def int_x86_mmx_psrl_d : GCCBuiltin<"__builtin_ia32_psrld">, Intrinsic<[llvm_v2i32_ty, llvm_v2i32_ty, llvm_v1i64_ty], [IntrNoMem]>; + def int_x86_mmx_psrli_d : GCCBuiltin<"__builtin_ia32_psrldi">, + Intrinsic<[llvm_v2i32_ty, llvm_v2i32_ty, + llvm_i32_ty], [IntrNoMem]>; def int_x86_mmx_psrl_q : GCCBuiltin<"__builtin_ia32_psrlq">, Intrinsic<[llvm_v1i64_ty, llvm_v1i64_ty, llvm_v1i64_ty], [IntrNoMem]>; + def int_x86_mmx_psrli_q : GCCBuiltin<"__builtin_ia32_psrlqi">, + Intrinsic<[llvm_v1i64_ty, llvm_v1i64_ty, + llvm_i32_ty], [IntrNoMem]>; def int_x86_mmx_psra_w : GCCBuiltin<"__builtin_ia32_psraw">, Intrinsic<[llvm_v4i16_ty, llvm_v4i16_ty, llvm_v1i64_ty], [IntrNoMem]>; + def int_x86_mmx_psrai_w : GCCBuiltin<"__builtin_ia32_psrawi">, + Intrinsic<[llvm_v4i16_ty, llvm_v4i16_ty, + llvm_i32_ty], [IntrNoMem]>; def int_x86_mmx_psra_d : GCCBuiltin<"__builtin_ia32_psrad">, Intrinsic<[llvm_v2i32_ty, llvm_v2i32_ty, llvm_v1i64_ty], [IntrNoMem]>; + def int_x86_mmx_psrai_d : GCCBuiltin<"__builtin_ia32_psradi">, + Intrinsic<[llvm_v2i32_ty, llvm_v2i32_ty, + llvm_i32_ty], [IntrNoMem]>; } // Pack ops. diff --git a/lib/Target/X86/X86InstrMMX.td b/lib/Target/X86/X86InstrMMX.td index 3d8bd1f579c..b079d17fd6c 100644 --- a/lib/Target/X86/X86InstrMMX.td +++ b/lib/Target/X86/X86InstrMMX.td @@ -116,7 +116,8 @@ let isTwoAddress = 1 in { } multiclass MMXI_binop_rmi_int opc, bits<8> opc2, Format ImmForm, - string OpcodeStr, Intrinsic IntId> { + string OpcodeStr, Intrinsic IntId, + Intrinsic ImmIntId> { def rr : MMXI; @@ -126,8 +127,7 @@ let isTwoAddress = 1 in { (bitconvert (load_mmx addr:$src2))))]>; def ri : MMXIi8; + [(set VR64:$dst, (ImmIntId VR64:$src1, imm:$src2))]>; } } @@ -268,23 +268,23 @@ let isTwoAddress = 1 in { // Shift Instructions defm MMX_PSRLW : MMXI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw", - int_x86_mmx_psrl_w>; + int_x86_mmx_psrl_w, int_x86_mmx_psrli_w>; defm MMX_PSRLD : MMXI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld", - int_x86_mmx_psrl_d>; + int_x86_mmx_psrl_d, int_x86_mmx_psrli_d>; defm MMX_PSRLQ : MMXI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq", - int_x86_mmx_psrl_q>; + int_x86_mmx_psrl_q, int_x86_mmx_psrli_q>; defm MMX_PSLLW : MMXI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw", - int_x86_mmx_psll_w>; + int_x86_mmx_psll_w, int_x86_mmx_pslli_w>; defm MMX_PSLLD : MMXI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld", - int_x86_mmx_psll_d>; + int_x86_mmx_psll_d, int_x86_mmx_pslli_d>; defm MMX_PSLLQ : MMXI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq", - int_x86_mmx_psll_q>; + int_x86_mmx_psll_q, int_x86_mmx_pslli_q>; defm MMX_PSRAW : MMXI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw", - int_x86_mmx_psra_w>; + int_x86_mmx_psra_w, int_x86_mmx_psrai_w>; defm MMX_PSRAD : MMXI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad", - int_x86_mmx_psra_d>; + int_x86_mmx_psra_d, int_x86_mmx_psrai_d>; // Comparison Instructions defm MMX_PCMPEQB : MMXI_binop_rm_int<0x74, "pcmpeqb", int_x86_mmx_pcmpeq_b>; diff --git a/lib/VMCore/AutoUpgrade.cpp b/lib/VMCore/AutoUpgrade.cpp index 343a4b6431b..a46138a612b 100644 --- a/lib/VMCore/AutoUpgrade.cpp +++ b/lib/VMCore/AutoUpgrade.cpp @@ -122,7 +122,8 @@ static bool UpgradeIntrinsicFunction1(Function *F, Function *&NewFn) { if (Name.compare(5,10,"x86.mmx.ps",10) == 0 && (Name.compare(13,4,"psll", 4) == 0 || Name.compare(13,4,"psra", 4) == 0 || - Name.compare(13,4,"psrl", 4) == 0)) { + Name.compare(13,4,"psrl", 4) == 0) && + Name[17] != 'i') { const llvm::Type *VT = VectorType::get(IntegerType::get(64), 1); diff --git a/test/CodeGen/X86/mmx-shift.ll b/test/CodeGen/X86/mmx-shift.ll new file mode 100644 index 00000000000..1ebc0435c81 --- /dev/null +++ b/test/CodeGen/X86/mmx-shift.ll @@ -0,0 +1,20 @@ +; RUN: llvm-as < %s | llc -march=x86 -mattr=+mmx | grep psllq | grep 32 +; RUN: llvm-as < %s | llc -march=x86 -mattr=+mmx | grep psrad + +define i64 @t1(<1 x i64> %mm1) nounwind { +entry: + %tmp6 = tail call <1 x i64> @llvm.x86.mmx.pslli.q( <1 x i64> %mm1, i32 32 ) ; <<1 x i64>> [#uses=1] + %retval1112 = bitcast <1 x i64> %tmp6 to i64 ; [#uses=1] + ret i64 %retval1112 +} + +declare <1 x i64> @llvm.x86.mmx.pslli.q(<1 x i64>, i32) nounwind readnone + +define i64 @t2(<2 x i32> %mm1, <2 x i32> %mm2) nounwind { +entry: + %tmp7 = tail call <2 x i32> @llvm.x86.mmx.psra.d( <2 x i32> %mm1, <2 x i32> %mm2 ) nounwind readnone ; <<2 x i32>> [#uses=1] + %retval1112 = bitcast <2 x i32> %tmp7 to i64 ; [#uses=1] + ret i64 %retval1112 +} + +declare <2 x i32> @llvm.x86.mmx.psra.d(<2 x i32>, <2 x i32>) nounwind readnone