[llvm] [TTI] Provide conservative costs for @llvm.speculative.load. (PR #180036)

[David Sherwood via llvm-commits]

================
@@ -55,16 +55,27 @@ define void @speculative_load_cost_fixed(ptr %p) {
 }
 
 define void @speculative_load_cost_scalable(ptr %p) {
-; COMMON-LABEL: 'speculative_load_cost_scalable'
-; COMMON-NEXT:  Cost Model: Invalid cost for instruction: %1 = call <vscale x 2 x i64> (ptr, i1, ...) @llvm.speculative.load.nxv2i64.p0(ptr %p, i1 false, i64 0)
-; COMMON-NEXT:  Cost Model: Invalid cost for instruction: %2 = call <vscale x 4 x i32> (ptr, i1, ...) @llvm.speculative.load.nxv4i32.p0(ptr %p, i1 false, i64 0)
-; COMMON-NEXT:  Cost Model: Invalid cost for instruction: %3 = call <vscale x 8 x i16> (ptr, i1, ...) @llvm.speculative.load.nxv8i16.p0(ptr %p, i1 false, i64 0)
-; COMMON-NEXT:  Cost Model: Invalid cost for instruction: %4 = call <vscale x 16 x i8> (ptr, i1, ...) @llvm.speculative.load.nxv16i8.p0(ptr %p, i1 false, i64 0)
-; COMMON-NEXT:  Cost Model: Invalid cost for instruction: %5 = call <vscale x 2 x double> (ptr, i1, ...) @llvm.speculative.load.nxv2f64.p0(ptr %p, i1 false, i64 0)
-; COMMON-NEXT:  Cost Model: Invalid cost for instruction: %6 = call <vscale x 4 x float> (ptr, i1, ...) @llvm.speculative.load.nxv4f32.p0(ptr %p, i1 false, i64 0)
-; COMMON-NEXT:  Cost Model: Invalid cost for instruction: %7 = call <vscale x 8 x float> (ptr, i1, ...) @llvm.speculative.load.nxv8f32.p0(ptr %p, i1 false, i64 0)
-; COMMON-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+; NEON-LABEL: 'speculative_load_cost_scalable'
+; NEON-NEXT:  Cost Model: Invalid cost for instruction: %1 = call <vscale x 2 x i64> (ptr, i1, ...) @llvm.speculative.load.nxv2i64.p0(ptr %p, i1 false, i64 0)
+; NEON-NEXT:  Cost Model: Invalid cost for instruction: %2 = call <vscale x 4 x i32> (ptr, i1, ...) @llvm.speculative.load.nxv4i32.p0(ptr %p, i1 false, i64 0)
+; NEON-NEXT:  Cost Model: Invalid cost for instruction: %3 = call <vscale x 8 x i16> (ptr, i1, ...) @llvm.speculative.load.nxv8i16.p0(ptr %p, i1 false, i64 0)
+; NEON-NEXT:  Cost Model: Invalid cost for instruction: %4 = call <vscale x 16 x i8> (ptr, i1, ...) @llvm.speculative.load.nxv16i8.p0(ptr %p, i1 false, i64 0)
+; NEON-NEXT:  Cost Model: Invalid cost for instruction: %5 = call <vscale x 2 x double> (ptr, i1, ...) @llvm.speculative.load.nxv2f64.p0(ptr %p, i1 false, i64 0)
+; NEON-NEXT:  Cost Model: Invalid cost for instruction: %6 = call <vscale x 4 x float> (ptr, i1, ...) @llvm.speculative.load.nxv4f32.p0(ptr %p, i1 false, i64 0)
+; NEON-NEXT:  Cost Model: Invalid cost for instruction: %7 = call <vscale x 8 x float> (ptr, i1, ...) @llvm.speculative.load.nxv8f32.p0(ptr %p, i1 false, i64 0)
+; NEON-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; SVE-LABEL: 'speculative_load_cost_scalable'
+; SVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %1 = call <vscale x 2 x i64> (ptr, i1, ...) @llvm.speculative.load.nxv2i64.p0(ptr %p, i1 false, i64 0)
+; SVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %2 = call <vscale x 4 x i32> (ptr, i1, ...) @llvm.speculative.load.nxv4i32.p0(ptr %p, i1 false, i64 0)
+; SVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %3 = call <vscale x 8 x i16> (ptr, i1, ...) @llvm.speculative.load.nxv8i16.p0(ptr %p, i1 false, i64 0)
+; SVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %4 = call <vscale x 16 x i8> (ptr, i1, ...) @llvm.speculative.load.nxv16i8.p0(ptr %p, i1 false, i64 0)
+; SVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %5 = call <vscale x 2 x double> (ptr, i1, ...) @llvm.speculative.load.nxv2f64.p0(ptr %p, i1 false, i64 0)
----------------
david-arm wrote:

This looks wrong because the type <vscale x 2 x double> could be larger than 16 bytes.

https://github.com/llvm/llvm-project/pull/180036