ac/rtld: report better error messages for LDS overallocation

[mesa.git] / src / amd / common / ac_nir_to_llvm.c

diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
index dd673cf5ea0e43a59f0f561c1e52858e82901b9e..dffaeedfbb42b5c5e81d0ca4f0f65a647a956086 100644 (file)
--- a/src/amd/common/ac_nir_to_llvm.c
+++ b/src/amd/common/ac_nir_to_llvm.c
@@ -1622,13 +1622,11 @@ static void visit_store_ssbo(struct ac_nir_context *ctx,
                if (num_bytes == 1) {
                        ac_build_tbuffer_store_byte(&ctx->ac, rsrc, data,
                                                    offset, ctx->ac.i32_0,
-                                                   cache_policy & ac_glc,
-                                                   writeonly_memory);
+                                                   cache_policy & ac_glc);
                } else if (num_bytes == 2) {
                        ac_build_tbuffer_store_short(&ctx->ac, rsrc, data,
                                                     offset, ctx->ac.i32_0,
-                                                    cache_policy & ac_glc,
-                                                    writeonly_memory);
+                                                    cache_policy & ac_glc);
                } else {
                        int num_channels = num_bytes / 4;
 
@@ -1654,8 +1652,7 @@ static void visit_store_ssbo(struct ac_nir_context *ctx,
                                                    num_channels, offset,
                                                    ctx->ac.i32_0, 0,
                                                    cache_policy & ac_glc,
-                                                   false, writeonly_memory,
-                                                   false);
+                                                   false, false);
                }
        }
 }
@@ -1782,11 +1779,12 @@ static LLVMValueRef visit_load_buffer(struct ac_nir_context *ctx,
                                                         cache_policy & ac_glc);
                } else {
                        int num_channels = util_next_power_of_two(load_bytes) / 4;
+                       bool can_speculate = access & ACCESS_CAN_REORDER;
 
                        ret = ac_build_buffer_load(&ctx->ac, rsrc, num_channels,
                                                   vindex, offset, immoffset, 0,
                                                   cache_policy & ac_glc, 0,
-                                                  false, false);
+                                                  can_speculate, false);
                }
 
                LLVMTypeRef byte_vec = LLVMVectorType(ctx->ac.i8, ac_get_type_size(LLVMTypeOf(ret)));
@@ -2492,11 +2490,11 @@ static LLVMValueRef visit_image_load(struct ac_nir_context *ctx,
                vindex = LLVMBuildExtractElement(ctx->ac.builder, get_src(ctx, instr->src[1]),
                                                 ctx->ac.i32_0, "");
 
-               /* TODO: set "can_speculate" when OpenGL needs it. */
+               bool can_speculate = access & ACCESS_CAN_REORDER;
                res = ac_build_buffer_load_format(&ctx->ac, rsrc, vindex,
                                                  ctx->ac.i32_0, num_channels,
                                                  !!(args.cache_policy & ac_glc),
-                                                 false);
+                                                 can_speculate);
                res = ac_build_expand_to_vec4(&ctx->ac, res, num_channels);
 
                res = ac_trim_vector(&ctx->ac, res, instr->dest.ssa.num_components);
@@ -2556,8 +2554,7 @@ static void visit_image_store(struct ac_nir_context *ctx,
 
                ac_build_buffer_store_format(&ctx->ac, rsrc, src, vindex,
                                             ctx->ac.i32_0, src_channels,
-                                            args.cache_policy & ac_glc,
-                                            writeonly_memory);
+                                            args.cache_policy & ac_glc, false);
        } else {
                args.opcode = ac_image_store;
                args.data[0] = ac_to_float(&ctx->ac, get_src(ctx, instr->src[3]));
@@ -3458,6 +3455,26 @@ static void visit_intrinsic(struct ac_nir_context *ctx,
        case nir_intrinsic_quad_swap_diagonal:
                result = ac_build_quad_swizzle(&ctx->ac, get_src(ctx, instr->src[0]), 3, 2, 1 ,0);
                break;
+       case nir_intrinsic_quad_swizzle_amd: {
+               uint32_t mask = nir_intrinsic_swizzle_mask(instr);
+               result = ac_build_quad_swizzle(&ctx->ac, get_src(ctx, instr->src[0]),
+                                              mask & 0x3, (mask >> 2) & 0x3,
+                                              (mask >> 4) & 0x3, (mask >> 6) & 0x3);
+               break;
+       }
+       case nir_intrinsic_masked_swizzle_amd: {
+               uint32_t mask = nir_intrinsic_swizzle_mask(instr);
+               result = ac_build_ds_swizzle(&ctx->ac, get_src(ctx, instr->src[0]), mask);
+               break;
+       }
+       case nir_intrinsic_write_invocation_amd:
+               result = ac_build_writelane(&ctx->ac, get_src(ctx, instr->src[0]),
+                                           get_src(ctx, instr->src[1]),
+                                           get_src(ctx, instr->src[2]));
+               break;
+       case nir_intrinsic_mbcnt_amd:
+               result = ac_build_mbcnt(&ctx->ac, get_src(ctx, instr->src[0]));
+               break;
        default:
                fprintf(stderr, "Unknown intrinsic: ");
                nir_print_instr(&instr->instr, stderr);
@@ -3895,7 +3912,13 @@ static void visit_tex(struct ac_nir_context *ctx, nir_tex_instr *instr)
                args.offset = NULL;
        }
 
-       /* TODO TG4 support */
+       /* DMASK was repurposed for GATHER4. 4 components are always
+        * returned and DMASK works like a swizzle - it selects
+        * the component to fetch. The only valid DMASK values are
+        * 1=red, 2=green, 4=blue, 8=alpha. (e.g. 1 returns
+        * (red,red,red,red) etc.) The ISA document doesn't mention
+        * this.
+        */
        args.dmask = 0xf;
        if (instr->op == nir_texop_tg4) {
                if (instr->is_shadow)