diff options
Diffstat (limited to 'src/gallium/drivers/radeonsi/si_shader.c')
| -rw-r--r-- | src/gallium/drivers/radeonsi/si_shader.c | 72 |
1 files changed, 63 insertions, 9 deletions
diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c index f72e5af31fd..c3fe13deeaa 100644 --- a/src/gallium/drivers/radeonsi/si_shader.c +++ b/src/gallium/drivers/radeonsi/si_shader.c @@ -1973,6 +1973,7 @@ static LLVMValueRef fetch_constant( unsigned swizzle) { struct si_shader_context *ctx = si_shader_context(bld_base); + struct si_shader_selector *sel = ctx->shader->selector; const struct tgsi_ind_register *ireg = ®->Indirect; unsigned buf, idx; @@ -1996,9 +1997,60 @@ static LLVMValueRef fetch_constant( return si_llvm_emit_fetch_64bit(bld_base, type, lo, hi); } + idx = reg->Register.Index * 4 + swizzle; + if (reg->Register.Indirect) { + addr = si_get_indirect_index(ctx, ireg, 16, idx * 4); + } else { + addr = LLVMConstInt(ctx->i32, idx * 4, 0); + } + + /* Fast path when user data SGPRs point to constant buffer 0 directly. */ + if (sel->info.const_buffers_declared == 1 && + sel->info.shader_buffers_declared == 0) { + LLVMValueRef ptr = + LLVMGetParam(ctx->main_fn, ctx->param_const_and_shader_buffers); + + /* This enables use of s_load_dword and flat_load_dword for const buffer 0 + * loads, and up to x4 load opcode merging. However, it leads to horrible + * code reducing SIMD wave occupancy from 8 to 2 in many cases. + * + * Using s_buffer_load_dword (x1) seems to be the best option right now. + */ +#if 0 /* keep this codepath disabled */ + if (!reg->Register.Indirect) { + addr = LLVMBuildLShr(ctx->ac.builder, addr, LLVMConstInt(ctx->i32, 2, 0), ""); + LLVMValueRef result = ac_build_load_invariant(&ctx->ac, ptr, addr); + return bitcast(bld_base, type, result); + } +#endif + + /* Do the bounds checking with a descriptor, because + * doing computation and manual bounds checking of 64-bit + * addresses generates horrible VALU code with very high + * VGPR usage and very low SIMD occupancy. + */ + ptr = LLVMBuildPtrToInt(ctx->ac.builder, ptr, ctx->i64, ""); + ptr = LLVMBuildBitCast(ctx->ac.builder, ptr, ctx->v2i32, ""); + + LLVMValueRef desc_elems[] = { + LLVMBuildExtractElement(ctx->ac.builder, ptr, ctx->i32_0, ""), + LLVMBuildExtractElement(ctx->ac.builder, ptr, ctx->i32_1, ""), + LLVMConstInt(ctx->i32, (sel->info.const_file_max[0] + 1) * 16, 0), + LLVMConstInt(ctx->i32, + S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X) | + S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y) | + S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z) | + S_008F0C_DST_SEL_W(V_008F0C_SQ_SEL_W) | + S_008F0C_NUM_FORMAT(V_008F0C_BUF_NUM_FORMAT_FLOAT) | + S_008F0C_DATA_FORMAT(V_008F0C_BUF_DATA_FORMAT_32), 0) + }; + LLVMValueRef desc = ac_build_gather_values(&ctx->ac, desc_elems, 4); + LLVMValueRef result = buffer_load_const(ctx, desc, addr); + return bitcast(bld_base, type, result); + } + assert(reg->Register.Dimension); buf = reg->Dimension.Index; - idx = reg->Register.Index * 4 + swizzle; if (reg->Dimension.Indirect) { LLVMValueRef ptr = LLVMGetParam(ctx->main_fn, ctx->param_const_and_shader_buffers); @@ -2012,12 +2064,6 @@ static LLVMValueRef fetch_constant( } else bufp = load_const_buffer_desc(ctx, buf); - if (reg->Register.Indirect) { - addr = si_get_indirect_index(ctx, ireg, 16, idx * 4); - } else { - addr = LLVMConstInt(ctx->i32, idx * 4, 0); - } - return bitcast(bld_base, type, buffer_load_const(ctx, bufp, addr)); } @@ -4255,10 +4301,18 @@ static void declare_per_stage_desc_pointers(struct si_shader_context *ctx, struct si_function_info *fninfo, bool assign_params) { + LLVMTypeRef const_shader_buf_type; + + if (ctx->shader->selector->info.const_buffers_declared == 1 && + ctx->shader->selector->info.shader_buffers_declared == 0) + const_shader_buf_type = ctx->f32; + else + const_shader_buf_type = ctx->v4i32; + unsigned const_and_shader_buffers = add_arg(fninfo, ARG_SGPR, - si_const_array(ctx->v4i32, - SI_NUM_SHADER_BUFFERS + SI_NUM_CONST_BUFFERS)); + si_const_array(const_shader_buf_type, 0)); + unsigned samplers_and_images = add_arg(fninfo, ARG_SGPR, si_const_array(ctx->v8i32, |