amd: add ac_cu_info::has_mad32 flag and use in ACO

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38701>

src/amd/common/ac_gpu_info.c

@@ -310,6 +310,7 @@ ac_fill_cu_info(struct radeon_info *info, struct drm_amdgpu_info_device *device_
       info->family == CHIP_VEGA12 || info->family == CHIP_VEGA20 ||
        info->family == CHIP_MI100 || info->family == CHIP_MI200 ||
        info->family == CHIP_GFX940;
+   cu_info->has_mad32 = info->gfx_level == GFX9 ? info->family <= CHIP_MI200 : info->gfx_level < GFX10_3;
    cu_info->has_packed_math_16bit = info->gfx_level >= GFX9;
    cu_info->has_accelerated_dot_product =
       info->family == CHIP_VEGA20 ||

src/amd/common/ac_gpu_info.h

@@ -52,6 +52,8 @@ struct ac_cu_info {
     * Otherwise, unfused v_mad_mix* is available on GFX9.
     */
    bool has_fma_mix : 1;
+   /* Whether chips support unfused multiply-add instructions. */
+   bool has_mad32 : 1;
    /* Whether chips support double rate packed math instructions. */
    bool has_packed_math_16bit : 1;
    /* Whether chips support dot product instructions. A subset of these support a smaller

src/amd/compiler/aco_ir.cpp

@@ -112,6 +112,7 @@ init_program(Program* program, Stage stage, const struct aco_shader_info* info,
    program->dev.has_mac_legacy32 = program->gfx_level <= GFX7 || program->gfx_level == GFX10;
    program->dev.has_fmac_legacy32 = program->gfx_level >= GFX10_3 && program->gfx_level < GFX12;
    program->dev.fused_mad_mix = options->cu_info->has_fma_mix;
+   program->dev.has_mad32 = options->cu_info->has_mad32;
 
    if (program->gfx_level >= GFX12) {
       program->dev.scratch_global_offset_min = -8388608;

src/amd/compiler/aco_ir.h

@@ -2245,6 +2245,7 @@ struct DeviceInfo {
    bool has_fast_fma32 = false;
    bool has_mac_legacy32 = false;
    bool has_fmac_legacy32 = false;
+   bool has_mad32 = false;
    bool fused_mad_mix = false;
    bool xnack_enabled = false;
    bool sram_ecc_enabled = false;

src/amd/compiler/aco_optimizer.cpp

@@ -4202,8 +4202,7 @@ combine_instruction(opt_ctx& ctx, aco_ptr<Instruction>& instr)
       combine_instr_pattern{aco_opcode::src_op, aco_opcode::res_op, mask, swizzle, __VA_ARGS__})
 
    if (info.opcode == aco_opcode::v_add_f32) {
-      if (ctx.program->gfx_level < GFX10_3 && ctx.program->family != CHIP_GFX940 &&
-          ctx.fp_mode.denorm32 == 0) {
+      if (ctx.program->dev.has_mad32 && ctx.fp_mode.denorm32 == 0) {
          add_opt(v_mul_f32, v_mad_f32, 0x3, "120");
          add_opt(v_mul_legacy_f32, v_mad_legacy_f32, 0x3, "120");
       }