Changes in directory llvm/lib/Target/X86:
X86InstrSSE.td updated: 1.39 -> 1.40
---
Log message:
Added aliases to scalar SSE instructions, e.g. addss, to match x86 intrinsics.
The source operands type are v4sf with upper bits passes through.
Added matching code for these.
---
Diffs of the changes: (+201 -47)
X86InstrSSE.td | 248 ++++++++++++++++++++++++++++++++++++++++++++++-----------
1 files changed, 201 insertions(+), 47 deletions(-)
Index: llvm/lib/Target/X86/X86InstrSSE.td
diff -u llvm/lib/Target/X86/X86InstrSSE.td:1.39
llvm/lib/Target/X86/X86InstrSSE.td:1.40
--- llvm/lib/Target/X86/X86InstrSSE.td:1.39 Tue Mar 28 01:01:28 2006
+++ llvm/lib/Target/X86/X86InstrSSE.td Tue Mar 28 17:51:43 2006
@@ -174,53 +174,6 @@
[(set VR128:$dst,
(v2f64 (scalar_to_vector (loadf64 addr:$src))))]>;
-
-// Conversion instructions
-def CVTSS2SIrr: SSI<0x2D, MRMSrcReg, (ops R32:$dst, FR32:$src),
- "cvtss2si {$src, $dst|$dst, $src}", []>;
-def CVTSS2SIrm: SSI<0x2D, MRMSrcMem, (ops R32:$dst, f32mem:$src),
- "cvtss2si {$src, $dst|$dst, $src}", []>;
-
-def CVTTSS2SIrr: SSI<0x2C, MRMSrcReg, (ops R32:$dst, FR32:$src),
- "cvttss2si {$src, $dst|$dst, $src}",
- [(set R32:$dst, (fp_to_sint FR32:$src))]>;
-def CVTTSS2SIrm: SSI<0x2C, MRMSrcMem, (ops R32:$dst, f32mem:$src),
- "cvttss2si {$src, $dst|$dst, $src}",
- [(set R32:$dst, (fp_to_sint (loadf32 addr:$src)))]>;
-def CVTTSD2SIrr: SDI<0x2C, MRMSrcReg, (ops R32:$dst, FR64:$src),
- "cvttsd2si {$src, $dst|$dst, $src}",
- [(set R32:$dst, (fp_to_sint FR64:$src))]>;
-def CVTTSD2SIrm: SDI<0x2C, MRMSrcMem, (ops R32:$dst, f64mem:$src),
- "cvttsd2si {$src, $dst|$dst, $src}",
- [(set R32:$dst, (fp_to_sint (loadf64 addr:$src)))]>;
-def CVTSD2SSrr: SDI<0x5A, MRMSrcReg, (ops FR32:$dst, FR64:$src),
- "cvtsd2ss {$src, $dst|$dst, $src}",
- [(set FR32:$dst, (fround FR64:$src))]>;
-def CVTSD2SSrm: SDI<0x5A, MRMSrcMem, (ops FR32:$dst, f64mem:$src),
- "cvtsd2ss {$src, $dst|$dst, $src}",
- [(set FR32:$dst, (fround (loadf64 addr:$src)))]>;
-def CVTSI2SSrr: SSI<0x2A, MRMSrcReg, (ops FR32:$dst, R32:$src),
- "cvtsi2ss {$src, $dst|$dst, $src}",
- [(set FR32:$dst, (sint_to_fp R32:$src))]>;
-def CVTSI2SSrm: SSI<0x2A, MRMSrcMem, (ops FR32:$dst, i32mem:$src),
- "cvtsi2ss {$src, $dst|$dst, $src}",
- [(set FR32:$dst, (sint_to_fp (loadi32 addr:$src)))]>;
-def CVTSI2SDrr: SDI<0x2A, MRMSrcReg, (ops FR64:$dst, R32:$src),
- "cvtsi2sd {$src, $dst|$dst, $src}",
- [(set FR64:$dst, (sint_to_fp R32:$src))]>;
-def CVTSI2SDrm: SDI<0x2A, MRMSrcMem, (ops FR64:$dst, i32mem:$src),
- "cvtsi2sd {$src, $dst|$dst, $src}",
- [(set FR64:$dst, (sint_to_fp (loadi32 addr:$src)))]>;
-// SSE2 instructions with XS prefix
-def CVTSS2SDrr: I<0x5A, MRMSrcReg, (ops FR64:$dst, FR32:$src),
- "cvtss2sd {$src, $dst|$dst, $src}",
- [(set FR64:$dst, (fextend FR32:$src))]>, XS,
- Requires<[HasSSE2]>;
-def CVTSS2SDrm: I<0x5A, MRMSrcMem, (ops FR64:$dst, f32mem:$src),
- "cvtss2sd {$src, $dst|$dst, $src}",
- [(set FR64:$dst, (fextend (loadf32 addr:$src)))]>, XS,
- Requires<[HasSSE2]>;
-
// Arithmetic instructions
let isTwoAddress = 1 in {
let isCommutable = 1 in {
@@ -317,6 +270,207 @@
def MINSDrm : SDI<0x5D, MRMSrcMem, (ops FR64:$dst, f64mem:$src),
"minsd {$src, $dst|$dst, $src}", []>;
+
+// Aliases to match intrinsics which expect XMM operand(s).
+let isTwoAddress = 1 in {
+let isCommutable = 1 in {
+def Int_ADDSSrr : SSI<0x58, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
+ VR128:$src2),
+ "addss {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse_add_ss VR128:$src1,
+ VR128:$src2))]>;
+def Int_ADDSDrr : SDI<0x58, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
+ VR128:$src2),
+ "addsd {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse2_add_sd VR128:$src1,
+ VR128:$src2))]>;
+def Int_MULSSrr : SSI<0x59, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
+ VR128:$src2),
+ "mulss {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse_mul_ss VR128:$src1,
+ VR128:$src2))]>;
+def Int_MULSDrr : SDI<0x59, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
+ VR128:$src2),
+ "mulsd {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse2_mul_sd VR128:$src1,
+ VR128:$src2))]>;
+}
+
+def Int_ADDSSrm : SSI<0x58, MRMSrcMem, (ops VR128:$dst, VR128:$src1,
+ f32mem:$src2),
+ "addss {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse_add_ss VR128:$src1,
+ (load addr:$src2)))]>;
+def Int_ADDSDrm : SDI<0x58, MRMSrcMem, (ops VR128:$dst, VR128:$src1,
+ f64mem:$src2),
+ "addsd {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse2_add_sd VR128:$src1,
+ (load addr:$src2)))]>;
+def Int_MULSSrm : SSI<0x59, MRMSrcMem, (ops VR128:$dst, VR128:$src1,
+ f32mem:$src2),
+ "mulss {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse_mul_ss VR128:$src1,
+ (load addr:$src2)))]>;
+def Int_MULSDrm : SDI<0x59, MRMSrcMem, (ops VR128:$dst, VR128:$src1,
+ f64mem:$src2),
+ "mulsd {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse2_mul_sd VR128:$src1,
+ (load addr:$src2)))]>;
+
+def Int_DIVSSrr : SSI<0x5E, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
VR128:$src2),
+ "divss {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse_div_ss VR128:$src1,
+ VR128:$src2))]>;
+def Int_DIVSSrm : SSI<0x5E, MRMSrcMem, (ops VR128:$dst, VR128:$src1,
f32mem:$src2),
+ "divss {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse_div_ss VR128:$src1,
+ (load addr:$src2)))]>;
+def Int_DIVSDrr : SDI<0x5E, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
VR128:$src2),
+ "divsd {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse2_div_sd VR128:$src1,
+ VR128:$src2))]>;
+def Int_DIVSDrm : SDI<0x5E, MRMSrcMem, (ops VR128:$dst, VR128:$src1,
f64mem:$src2),
+ "divsd {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse2_div_sd VR128:$src1,
+ (load addr:$src2)))]>;
+
+def Int_SUBSSrr : SSI<0x5C, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
VR128:$src2),
+ "subss {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse_sub_ss VR128:$src1,
+ VR128:$src2))]>;
+def Int_SUBSSrm : SSI<0x5C, MRMSrcMem, (ops VR128:$dst, VR128:$src1,
f32mem:$src2),
+ "subss {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse_sub_ss VR128:$src1,
+ (load addr:$src2)))]>;
+def Int_SUBSDrr : SDI<0x5C, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
VR128:$src2),
+ "subsd {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse2_sub_sd VR128:$src1,
+ VR128:$src2))]>;
+def Int_SUBSDrm : SDI<0x5C, MRMSrcMem, (ops VR128:$dst, VR128:$src1,
f64mem:$src2),
+ "subsd {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse2_sub_sd VR128:$src1,
+ (load addr:$src2)))]>;
+}
+
+def Int_SQRTSSrr : SSI<0x51, MRMSrcReg, (ops VR128:$dst, VR128:$src),
+ "sqrtss {$src, $dst|$dst, $src}",
+ [(set VR128:$dst, (int_x86_sse_sqrt_ss VR128:$src))]>;
+def Int_SQRTSSrm : SSI<0x51, MRMSrcMem, (ops VR128:$dst, f32mem:$src),
+ "sqrtss {$src, $dst|$dst, $src}",
+ [(set VR128:$dst, (int_x86_sse_sqrt_ss
+ (load addr:$src)))]>;
+def Int_SQRTSDrr : SDI<0x51, MRMSrcReg, (ops VR128:$dst, VR128:$src),
+ "sqrtsd {$src, $dst|$dst, $src}",
+ [(set VR128:$dst, (int_x86_sse2_sqrt_sd VR128:$src))]>;
+def Int_SQRTSDrm : SDI<0x51, MRMSrcMem, (ops VR128:$dst, f64mem:$src),
+ "sqrtsd {$src, $dst|$dst, $src}",
+ [(set VR128:$dst, (int_x86_sse2_sqrt_sd
+ (load addr:$src)))]>;
+
+def Int_RSQRTSSrr : SSI<0x52, MRMSrcReg, (ops VR128:$dst, VR128:$src),
+ "rsqrtss {$src, $dst|$dst, $src}",
+ [(set VR128:$dst, (int_x86_sse_rsqrt_ss VR128:$src))]>;
+def Int_RSQRTSSrm : SSI<0x52, MRMSrcMem, (ops VR128:$dst, f32mem:$src),
+ "rsqrtss {$src, $dst|$dst, $src}",
+ [(set VR128:$dst, (int_x86_sse_rsqrt_ss
+ (load addr:$src)))]>;
+def Int_RCPSSrr : SSI<0x53, MRMSrcReg, (ops VR128:$dst, VR128:$src),
+ "rcpss {$src, $dst|$dst, $src}",
+ [(set VR128:$dst, (int_x86_sse_rcp_ss VR128:$src))]>;
+def Int_RCPSSrm : SSI<0x53, MRMSrcMem, (ops VR128:$dst, f32mem:$src),
+ "rcpss {$src, $dst|$dst, $src}",
+ [(set VR128:$dst, (int_x86_sse_rcp_ss
+ (load addr:$src)))]>;
+
+let isTwoAddress = 1 in {
+def Int_MAXSSrr : SSI<0x5F, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
+ VR128:$src2),
+ "maxss {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse_max_ss VR128:$src1,
+ VR128:$src2))]>;
+def Int_MAXSSrm : SSI<0x5F, MRMSrcMem, (ops VR128:$dst, VR128:$src1,
+ f32mem:$src2),
+ "maxss {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse_max_ss VR128:$src1,
+ (load addr:$src2)))]>;
+def Int_MAXSDrr : SDI<0x5F, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
+ VR128:$src2),
+ "maxsd {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse2_max_sd VR128:$src1,
+ VR128:$src2))]>;
+def Int_MAXSDrm : SDI<0x5F, MRMSrcMem, (ops VR128:$dst, VR128:$src1,
+ f64mem:$src2),
+ "maxsd {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse2_max_sd VR128:$src1,
+ (load addr:$src2)))]>;
+def Int_MINSSrr : SSI<0x5D, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
+ VR128:$src2),
+ "minss {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse_min_ss VR128:$src1,
+ VR128:$src2))]>;
+def Int_MINSSrm : SSI<0x5D, MRMSrcMem, (ops VR128:$dst, VR128:$src1,
+ f32mem:$src2),
+ "minss {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse_min_ss VR128:$src1,
+ (load addr:$src2)))]>;
+def Int_MINSDrr : SDI<0x5D, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
+ VR128:$src2),
+ "minsd {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse2_min_sd VR128:$src1,
+ VR128:$src2))]>;
+def Int_MINSDrm : SDI<0x5D, MRMSrcMem, (ops VR128:$dst, VR128:$src1,
+ f64mem:$src2),
+ "minsd {$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse2_min_sd VR128:$src1,
+ (load addr:$src2)))]>;
+}
+
+// Conversion instructions
+def CVTSS2SIrr: SSI<0x2D, MRMSrcReg, (ops R32:$dst, FR32:$src),
+ "cvtss2si {$src, $dst|$dst, $src}", []>;
+def CVTSS2SIrm: SSI<0x2D, MRMSrcMem, (ops R32:$dst, f32mem:$src),
+ "cvtss2si {$src, $dst|$dst, $src}", []>;
+
+def CVTTSS2SIrr: SSI<0x2C, MRMSrcReg, (ops R32:$dst, FR32:$src),
+ "cvttss2si {$src, $dst|$dst, $src}",
+ [(set R32:$dst, (fp_to_sint FR32:$src))]>;
+def CVTTSS2SIrm: SSI<0x2C, MRMSrcMem, (ops R32:$dst, f32mem:$src),
+ "cvttss2si {$src, $dst|$dst, $src}",
+ [(set R32:$dst, (fp_to_sint (loadf32 addr:$src)))]>;
+def CVTTSD2SIrr: SDI<0x2C, MRMSrcReg, (ops R32:$dst, FR64:$src),
+ "cvttsd2si {$src, $dst|$dst, $src}",
+ [(set R32:$dst, (fp_to_sint FR64:$src))]>;
+def CVTTSD2SIrm: SDI<0x2C, MRMSrcMem, (ops R32:$dst, f64mem:$src),
+ "cvttsd2si {$src, $dst|$dst, $src}",
+ [(set R32:$dst, (fp_to_sint (loadf64 addr:$src)))]>;
+def CVTSD2SSrr: SDI<0x5A, MRMSrcReg, (ops FR32:$dst, FR64:$src),
+ "cvtsd2ss {$src, $dst|$dst, $src}",
+ [(set FR32:$dst, (fround FR64:$src))]>;
+def CVTSD2SSrm: SDI<0x5A, MRMSrcMem, (ops FR32:$dst, f64mem:$src),
+ "cvtsd2ss {$src, $dst|$dst, $src}",
+ [(set FR32:$dst, (fround (loadf64 addr:$src)))]>;
+def CVTSI2SSrr: SSI<0x2A, MRMSrcReg, (ops FR32:$dst, R32:$src),
+ "cvtsi2ss {$src, $dst|$dst, $src}",
+ [(set FR32:$dst, (sint_to_fp R32:$src))]>;
+def CVTSI2SSrm: SSI<0x2A, MRMSrcMem, (ops FR32:$dst, i32mem:$src),
+ "cvtsi2ss {$src, $dst|$dst, $src}",
+ [(set FR32:$dst, (sint_to_fp (loadi32 addr:$src)))]>;
+def CVTSI2SDrr: SDI<0x2A, MRMSrcReg, (ops FR64:$dst, R32:$src),
+ "cvtsi2sd {$src, $dst|$dst, $src}",
+ [(set FR64:$dst, (sint_to_fp R32:$src))]>;
+def CVTSI2SDrm: SDI<0x2A, MRMSrcMem, (ops FR64:$dst, i32mem:$src),
+ "cvtsi2sd {$src, $dst|$dst, $src}",
+ [(set FR64:$dst, (sint_to_fp (loadi32 addr:$src)))]>;
+// SSE2 instructions with XS prefix
+def CVTSS2SDrr: I<0x5A, MRMSrcReg, (ops FR64:$dst, FR32:$src),
+ "cvtss2sd {$src, $dst|$dst, $src}",
+ [(set FR64:$dst, (fextend FR32:$src))]>, XS,
+ Requires<[HasSSE2]>;
+def CVTSS2SDrm: I<0x5A, MRMSrcMem, (ops FR64:$dst, f32mem:$src),
+ "cvtss2sd {$src, $dst|$dst, $src}",
+ [(set FR64:$dst, (fextend (loadf32 addr:$src)))]>, XS,
+ Requires<[HasSSE2]>;
+
// Comparison instructions
let isTwoAddress = 1 in {
def CMPSSrr : SSI<0xC2, MRMSrcReg,
_______________________________________________
llvm-commits mailing list
llvm-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
