Adjust x86 strided vectorization tests for cost comparison.
Strided vectorization tests in GCC are updated with explicit cost comparison flags and adjusted code generation expectations.
The gcc.target/i386/vect-strided-?.c tests have been modified with explicit cost comparison settings. Variants with cost comparison disabled are retained, while new variants (-?b.c) enable cost comparison and adjust expected code generation accordingly. This allows for testing the vectorizer’s behavior when cost comparison influences the choice between V2SImode and V4SImode with shuffles.
In Details
This commit addresses the impact of cost comparison on strided vectorization in x86. When cost comparison is enabled and MMX-with-SSE vector width is available, the vectorizer now prefers V2SImode over V4SImode with shuffles. The original tests are modified to disable vector cost comparison, while new -?b.c variants are created to enable it. This change highlights the interaction between ix86_vector_costs, target-specific vector widths, and shuffle costs.
For Context
Strided vectorization optimizes memory accesses with non-unit strides. GCC's vectorizer employs cost models to evaluate different vectorization strategies. This commit updates tests to account for the impact of cost comparison on the choice of vector modes (V2SI vs V4SI) in strided vectorization scenarios on x86, using MMX and SSE instruction sets. The cost model helps the compiler decide which approach yields better performance.