Add x86 SIMD encoding for icmp sgt

cranelift/codegen/meta/src/isa/x86/encodings.rs

@@ -685,6 +685,7 @@ pub(crate) fn define(
     let use_sse41 = settings.predicate_by_name("use_sse41");
     let use_ssse3_simd = settings.predicate_by_name("use_ssse3_simd");
     let use_sse41_simd = settings.predicate_by_name("use_sse41_simd");
+    let use_sse42_simd = settings.predicate_by_name("use_sse42_simd");
 
     // Definitions.
     let mut e = PerCpuModeEncodings::new();
@@ -2025,21 +2026,25 @@ pub(crate) fn define(
         e.enc_32_64(x86_psra, rec_fa.opcodes(*opcodes));
     }
 
-    // SIMD icmp using PCMPEQ*
+    // SIMD integer comparisons
-    for ty in ValueType::all_lane_types().filter(|t| t.is_int() && allowed_simd_type(t)) {
+    {
-        let (opcodes, isa_predicate): (&[_], _) = match ty.lane_bits() {
+        use IntCC::*;
-            8 => (&PCMPEQB, None),
+        for (ty, cc, opcodes, isa_predicate) in &[
-            16 => (&PCMPEQW, None),
+            (I8, Equal, &PCMPEQB[..], None),
-            32 => (&PCMPEQD, None),
+            (I16, Equal, &PCMPEQW[..], None),
-            64 => (&PCMPEQQ, Some(use_sse41_simd)),
+            (I32, Equal, &PCMPEQD[..], None),
-            _ => panic!("invalid size for SIMD icmp"),
+            (I64, Equal, &PCMPEQQ[..], Some(use_sse41_simd)),
-        };
+            (I8, SignedGreaterThan, &PCMPGTB[..], None),
-
+            (I16, SignedGreaterThan, &PCMPGTW[..], None),
+            (I32, SignedGreaterThan, &PCMPGTD[..], None),
+            (I64, SignedGreaterThan, &PCMPGTQ, Some(use_sse42_simd)),
+        ] {
             let instruction = icmp
-            .bind(Immediate::IntCC(IntCC::Equal))
+                .bind(Immediate::IntCC(*cc))
-            .bind(vector(ty, sse_vector_size));
+                .bind(vector(*ty, sse_vector_size));
             let template = rec_icscc_fpr.nonrex().opcodes(opcodes);
-        e.enc_32_64_maybe_isap(instruction, template, isa_predicate);
+            e.enc_32_64_maybe_isap(instruction, template, *isa_predicate);
+        }
     }
 
     // Reference type instructions

cranelift/codegen/meta/src/isa/x86/opcodes.rs

@@ -281,6 +281,18 @@ pub static PCMPEQQ: [u8; 4] = [0x66, 0x0f, 0x38, 0x29];
 /// Compare packed data for equal (SSE2).
 pub static PCMPEQW: [u8; 3] = [0x66, 0x0f, 0x75];
 
+/// Compare packed signed byte integers for greater than (SSE2).
+pub static PCMPGTB: [u8; 3] = [0x66, 0x0f, 0x64];
+
+/// Compare packed signed doubleword integers for greater than (SSE2).
+pub static PCMPGTD: [u8; 3] = [0x66, 0x0f, 0x66];
+
+/// Compare packed signed quadword integers for greater than (SSE4.2).
+pub static PCMPGTQ: [u8; 4] = [0x66, 0x0f, 0x38, 0x37];
+
+/// Compare packed signed word integers for greater than (SSE2).
+pub static PCMPGTW: [u8; 3] = [0x66, 0x0f, 0x65];
+
 /// Extract doubleword or quadword, depending on REX.W (SSE4.1).
 pub static PEXTR: [u8; 4] = [0x66, 0x0f, 0x3a, 0x16];
 

cranelift/filetests/filetests/isa/x86/simd-comparison-binemit.clif

@@ -0,0 +1,27 @@
+test binemit
+set enable_simd
+target x86_64 skylake
+
+function %icmp_sgt_i8x16(i8x16, i8x16) -> b8x16 {
+ebb0(v0: i8x16 [%xmm2], v1: i8x16 [%xmm1]):
+[-, %xmm2]  v2 = icmp sgt v0, v1     ; bin: 66 0f 64 d1
+            return v2
+}
+
+function %icmp_sgt_i16x8(i16x8, i16x8) -> b16x8 {
+ebb0(v0: i16x8 [%xmm4], v1: i16x8 [%xmm3]):
+[-, %xmm4]  v2 = icmp sgt v0, v1     ; bin: 66 0f 65 e3
+            return v2
+}
+
+function %icmp_sgt_i32x4(i32x4, i32x4) -> b32x4 {
+ebb0(v0: i32x4 [%xmm6], v1: i32x4 [%xmm5]):
+[-, %xmm6]  v2 = icmp sgt v0, v1     ; bin: 66 0f 66 f5
+            return v2
+}
+
+function %icmp_sgt_i64x2(i64x2, i64x2) -> b64x2 {
+ebb0(v0: i64x2 [%xmm0], v1: i64x2 [%xmm7]):
+[-, %xmm0]  v2 = icmp sgt v0, v1     ; bin: 66 0f 38 37 c7
+            return v2
+}

cranelift/filetests/filetests/isa/x86/simd-comparison-run.clif

@@ -43,3 +43,26 @@ ebb0:
     return v5
 }
 ; run
+
+function %icmp_sgt_i8x16() -> b1 {
+ebb0:
+    v0 = vconst.i8x16 [0 1 2 0 0 0 0 0 0 0 0 0 0 0 0 0]
+    v1 = vconst.i8x16 [1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0xff]
+    v2 = icmp sgt v0, v1
+    v3 = raw_bitcast.i8x16 v2
+    v4 = vconst.i8x16 [0 0 0xff 0 0 0 0 0 0 0 0 0 0 0 0 0xff]
+    v7 = icmp eq v3, v4
+    v8 = vall_true v7
+    return v8
+}
+; run
+
+function %icmp_sgt_i64x2() -> b1 {
+ebb0:
+    v0 = vconst.i64x2 [0 -42]
+    v1 = vconst.i64x2 [-1 -43]
+    v2 = icmp sgt v0, v1
+    v8 = vall_true v2
+    return v8
+}
+; run