[X86] Use X86ISD::CVTTP2SI and X86ISD::CVTTP2UI for lowering 128-bit cvttps2qq and...

Ideally ISD::FP_TO_SINT and ISD::FP_TO_UINT would only be used for cases with the same number of input and output elements.

Similar things have already been done for other convert intrinsics.

llvm-svn: 289316

diff --git a/llvm/lib/Target/X86/X86InstrAVX512.td b/llvm/lib/Target/X86/X86InstrAVX512.td
index fa6eba1..52a7cd9 100644 (file)
--- a/llvm/lib/Target/X86/X86InstrAVX512.td
+++ b/llvm/lib/Target/X86/X86InstrAVX512.td
@@ -6388,8 +6388,8 @@ multiclass avx512_cvtps2qq<bits<8> opc, string OpcodeStr,
 }
 
 // Convert Float to Signed/Unsigned Quardword with truncation
-multiclass avx512_cvttps2qq<bits<8> opc, string OpcodeStr,
-                                  SDNode OpNode, SDNode OpNodeRnd> {
+multiclass avx512_cvttps2qq<bits<8> opc, string OpcodeStr, SDNode OpNode,
+                            SDNode OpNode128, SDNode OpNodeRnd> {
   let Predicates = [HasDQI] in {
     defm Z : avx512_vcvt_fp<opc, OpcodeStr, v8i64_info, v8f32x_info, OpNode>,
              avx512_vcvt_fp_sae<opc, OpcodeStr, v8i64_info, v8f32x_info,
@@ -6398,7 +6398,7 @@ multiclass avx512_cvttps2qq<bits<8> opc, string OpcodeStr,
   let Predicates = [HasDQI, HasVLX] in {
     // Explicitly specified broadcast string, since we take only 2 elements
     // from v4f32x_info source
-    defm Z128 : avx512_vcvt_fp<opc, OpcodeStr, v2i64x_info, v4f32x_info, OpNode,
+    defm Z128 : avx512_vcvt_fp<opc, OpcodeStr, v2i64x_info, v4f32x_info, OpNode128,
                                "{1to2}", "", f64mem>, EVEX_V128;
     defm Z256 : avx512_vcvt_fp<opc, OpcodeStr, v4i64x_info, v4f32x_info, OpNode>,
                                EVEX_V256;
@@ -6496,14 +6496,14 @@ defm VCVTTPD2QQ : avx512_cvttpd2qq<0x7A, "vcvttpd2qq", fp_to_sint,
                                  X86cvttp2siRnd>, VEX_W,
                                  PD, EVEX_CD8<64, CD8VF>;
 
-defm VCVTTPS2QQ : avx512_cvttps2qq<0x7A, "vcvttps2qq", fp_to_sint,
+defm VCVTTPS2QQ : avx512_cvttps2qq<0x7A, "vcvttps2qq", fp_to_sint, X86cvttp2si,
                                  X86cvttp2siRnd>, PD, EVEX_CD8<32, CD8VH>;
 
 defm VCVTTPD2UQQ : avx512_cvttpd2qq<0x78, "vcvttpd2uqq", fp_to_uint,
                                  X86cvttp2uiRnd>, VEX_W,
                                  PD, EVEX_CD8<64, CD8VF>;
 
-defm VCVTTPS2UQQ : avx512_cvttps2qq<0x78, "vcvttps2uqq", fp_to_uint,
+defm VCVTTPS2UQQ : avx512_cvttps2qq<0x78, "vcvttps2uqq", fp_to_uint, X86cvttp2ui,
                                  X86cvttp2uiRnd>, PD, EVEX_CD8<32, CD8VH>;
 
 defm VCVTQQ2PD : avx512_cvtqq2pd<0xE6, "vcvtqq2pd", sint_to_fp,

diff --git a/llvm/lib/Target/X86/X86IntrinsicsInfo.h b/llvm/lib/Target/X86/X86IntrinsicsInfo.h
index 5e5369e..002e2e1 100644 (file)
--- a/llvm/lib/Target/X86/X86IntrinsicsInfo.h
+++ b/llvm/lib/Target/X86/X86IntrinsicsInfo.h
@@ -601,7 +601,7 @@ static const IntrinsicData  IntrinsicsWithoutChain[] = {
   X86_INTRINSIC_DATA(avx512_mask_cvttps2dq_512, INTR_TYPE_1OP_MASK,
                      ISD::FP_TO_SINT, X86ISD::CVTTP2SI_RND),
   X86_INTRINSIC_DATA(avx512_mask_cvttps2qq_128, INTR_TYPE_1OP_MASK,
-                     ISD::FP_TO_SINT, 0),
+                     X86ISD::CVTTP2SI, 0),
   X86_INTRINSIC_DATA(avx512_mask_cvttps2qq_256, INTR_TYPE_1OP_MASK,
                      ISD::FP_TO_SINT, 0),
   X86_INTRINSIC_DATA(avx512_mask_cvttps2qq_512, INTR_TYPE_1OP_MASK,
@@ -613,7 +613,7 @@ static const IntrinsicData  IntrinsicsWithoutChain[] = {
   X86_INTRINSIC_DATA(avx512_mask_cvttps2udq_512, INTR_TYPE_1OP_MASK,
                      ISD::FP_TO_UINT, X86ISD::CVTTP2UI_RND),
   X86_INTRINSIC_DATA(avx512_mask_cvttps2uqq_128, INTR_TYPE_1OP_MASK,
-                     ISD::FP_TO_UINT, 0),
+                     X86ISD::CVTTP2UI, 0),
   X86_INTRINSIC_DATA(avx512_mask_cvttps2uqq_256, INTR_TYPE_1OP_MASK,
                      ISD::FP_TO_UINT, 0),
   X86_INTRINSIC_DATA(avx512_mask_cvttps2uqq_512, INTR_TYPE_1OP_MASK,